Adding upstream version 115.7.0esr.upstream/115.7.0esrupstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

37 files changed, 9449 insertions, 0 deletions

diff --git a/third_party/rust/sync15/.cargo-checksum.json b/third_party/rust/sync15/.cargo-checksum.json
new file mode 100644
index 0000000000..1d4ab35f5f
--- /dev/null
+++ b/third_party/rust/sync15/.cargo-checksum.json
@@ -0,0 +1 @@
+{"files":{"Cargo.toml":"0102c944a52a98d1b7c6ba7a5f3ec69e28e5a05f0c58a743f0c5ba020988122a","README.md":"6d4ff5b079ac5340d18fa127f583e7ad793c5a2328b8ecd12c3fc723939804f2","src/bso/content.rs":"d2d650f4932e9a7068a25dd7df0085b92cd8976a0635320e6ae306d5a425075c","src/bso/crypto.rs":"27602dcccb37d3a55620ee4e16b705da455d49af575de115c7c79c0178eb1d6d","src/bso/mod.rs":"09e723dc7e99295ecafdcadffaf604d66ea27cf2b7f1fd9ab3cac4f4698ff6a7","src/bso/test_utils.rs":"4ec5a2df5e1c0ec14dc770681e959bdcef6ef04f6fde435999197f46a8ae4831","src/client/coll_state.rs":"4301526b987532cd7ebd5d089980c9524aa29360e08fa0c1f5437696ed354822","src/client/coll_update.rs":"627e2266c5c8f1c5e0bc83061fa14c1287c7d17c78b47882543521f75543b499","src/client/collection_keys.rs":"c27b2277a3a52033b58ab01490fc2ea7007494195dd5e6dc2c6931a4ca96795a","src/client/mod.rs":"8f588d4a035cf79d96f2500f06d5651c1a7c566127c456ffa5429811ddce3fd6","src/client/request.rs":"8841524e37d8195867bdf6ba98c75f610cf47a4644adeebd6372cc6713f2260a","src/client/state.rs":"4e31193ef2471c1dfabf1c6a391bcb95e14ddb45855786a4194ff187d5c9347c","src/client/status.rs":"f445a8765dac9789444e23b5145148413407bb1d18a15ef56682243997f591bf","src/client/storage_client.rs":"8de72d4ba3ca4f68c8e1898466de83a2b543545a18679800cb4f7fbda2dc3183","src/client/sync.rs":"a04478c4e31bbf39f593decfc4844bfb7e678907031f0a93d1e005cf0047ebb4","src/client/sync_multiple.rs":"3729d4afd90ab1bd9982a3506252c99d8f37619cc1792ef4feba352ad01a7192","src/client/token.rs":"b268759d31e0fe17e0e2a428694cd9a317fcfbdd52f023d5d8c7cc6f00f1a102","src/client/util.rs":"71cc70ee41f821f53078675e636e9fad9c6046fa1a989e37f5487e340a2277d6","src/client_types.rs":"3c3cac1540b92482f43660d9e43bdde8481c4cc1a98253a68c80e791231f5976","src/clients_engine/engine.rs":"0eaa078978c95fc96284b48a7458ebff49f0aa70b1148ef019b9350b5ba4d0d4","src/clients_engine/mod.rs":"461729e6f89b66b2cbd89b041a03d4d6a8ba582284ed4f3015cb13e1a0c6da97","src/clients_engine/record.rs":"69357413571d688eea3a5207f9b88088cde285b9373c7bd4ea1e018dbc823dd2","src/clients_engine/ser.rs":"9796e44ed7daf04f22afbb51238ac25fd0de1438b72181351b4ca29fd70fd429","src/device_type.rs":"fe217453f19b374abcc10e9f503b25f4f712b555497bebe5aefcf2e9b258d28e","src/enc_payload.rs":"aa3eea7df49b24cd59831680a47c417b73a3e36e6b0f3f4baf14ca66bd68be6b","src/engine/bridged_engine.rs":"5c0748358a7b3040442bc81122400ab7b26acf728b194dbeee8a2166dc653e11","src/engine/changeset.rs":"949c520e508da0158b2df9fa20235f942ad8096cebbfc942b0fae41c35830c6b","src/engine/mod.rs":"cb638c2170c3785785877dfd059f26fbbfeae5edc599745dd861e4c299d310ad","src/engine/request.rs":"5923025fb9550178339f880a1bf8526d8e853e7a0b2bce6d9d687cc808ac0085","src/engine/sync_engine.rs":"9a5c6993dcceec8f82ead97a6d1840c7ed7dc6b326f7234c77f18848b6baf836","src/error.rs":"a45cfe02e6301f473c34678b694943c1a04308b8c292c6e0448bf495194c3b5e","src/key_bundle.rs":"ff8b10b95add934ecbc434b37ed089805886828ed159fd38bd692d1f01d06f7f","src/lib.rs":"eca1fa801820238141c4badeeec45d430aaec8b2ce088446ef94456d51014889","src/record_types.rs":"02bb3d352fb808131d298f9b90d9c95b7e9e0138b97c5401f3b9fdacc5562f44","src/server_timestamp.rs":"0020f31971ccbfc485894cabc3087459d42252b86d7de07f2136997864b0373b","src/telemetry.rs":"35e0313a052f16326e451e3d6e371337c1d71a471f32234ad9649fc1fa9f2237"},"package":null}
\ No newline at end of file
diff --git a/third_party/rust/sync15/Cargo.toml b/third_party/rust/sync15/Cargo.toml
new file mode 100644
index 0000000000..895a5c3816
--- /dev/null
+++ b/third_party/rust/sync15/Cargo.toml
@@ -0,0 +1,87 @@
+# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO
+#
+# When uploading crates to the registry Cargo will automatically
+# "normalize" Cargo.toml files for maximal compatibility
+# with all versions of Cargo and also rewrite `path` dependencies
+# to registry (e.g., crates.io) dependencies.
+#
+# If you are reading this file be aware that the original Cargo.toml
+# will likely look very different (and much more reasonable).
+# See Cargo.toml.orig for the original contents.
+
+[package]
+edition = "2021"
+name = "sync15"
+version = "0.1.0"
+authors = ["application-services@mozilla.com"]
+exclude = [
+    "/android",
+    "/ios",
+]
+readme = "README.md"
+license = "MPL-2.0"
+
+[dependencies]
+anyhow = "1.0"
+ffi-support = "0.4"
+lazy_static = "1.4"
+log = "0.4"
+serde_derive = "1"
+serde_json = "1"
+thiserror = "1.0"
+
+[dependencies.base16]
+version = "0.2"
+optional = true
+
+[dependencies.base64]
+version = "0.13"
+optional = true
+
+[dependencies.error-support]
+path = "../support/error"
+
+[dependencies.interrupt-support]
+path = "../support/interrupt"
+
+[dependencies.rc_crypto]
+path = "../support/rc_crypto"
+features = ["hawk"]
+optional = true
+
+[dependencies.serde]
+version = "1"
+features = ["derive"]
+
+[dependencies.sync-guid]
+path = "../support/guid"
+features = ["random"]
+
+[dependencies.url]
+version = "2.1"
+optional = true
+
+[dependencies.viaduct]
+path = "../viaduct"
+optional = true
+
+[dev-dependencies.env_logger]
+version = "0.7"
+default-features = false
+
+[features]
+crypto = [
+    "rc_crypto",
+    "base16",
+    "base64",
+]
+default = ["sync-engine"]
+random-guid = ["sync-guid/random"]
+standalone-sync = ["sync-client"]
+sync-client = [
+    "sync-engine",
+    "crypto",
+    "viaduct",
+    "url",
+]
+sync-engine = ["random-guid"]
diff --git a/third_party/rust/sync15/README.md b/third_party/rust/sync15/README.md
new file mode 100644
index 0000000000..a638435908
--- /dev/null
+++ b/third_party/rust/sync15/README.md
@@ -0,0 +1,128 @@
+# Low-level sync-1.5 helper component
+
+This component contains utility code to be shared between different
+data stores that want to sync against a Firefox Sync v1.5 sync server.
+It handles things like encrypting/decrypting records, obtaining and
+using storage node auth tokens, and so-on.
+
+There are 2 key concepts to understand here - the implementation itself, and
+a rust trait for a "syncable store" where component-specific logic lives - but
+before we dive into them, some preamble might help put things into context.
+
+## Nomenclature
+
+* The term "store" is generally used as the interface to the database - ie, the
+  thing that gets and saves items. It can also be seen as supplying the API
+  used by most consumers of the component. Note that the "places" component
+  is alone in using the term "api" for this object.
+
+* The term "engine" (or ideally, "sync engine") is used for the thing that
+  actually does the syncing for a store. Sync engines implement the SyncEngine
+  trait - the trait is either implemented directly by a store, or a new object
+  that has a reference to a store.
+
+## Introduction and History
+
+For many years Sync has worked exclusively against a "sync v1.5 server". This
+[is a REST API described here](https://mozilla-services.readthedocs.io/en/latest/storage/apis-1.5.html).
+The important part is that the API is conceptually quite simple - there are
+arbitrary "collections" containing "records" indexed by a GUID, and lacking
+traditonal database concepts like joins. Because the record is encrypted,
+there's very little scope for the server to be much smarter. Thus it's
+reasonably easy to create a fairly generic abstraction over the API that can be
+easily reused.
+
+Back in the deep past, we found ourselves with 2 different components that
+needed to sync against a sync v1.5 server. The apps using these components
+didn't have schedulers or any UI for choosing what to sync - so these
+components just looked at the existing state of the engines on the server and
+synced if they were enabled.
+
+This was also pre-megazord - the idea was that apps could choose from a "menu"
+of components to include - so we didn't really want to bind these components
+together. Therefore, there was no concept of "sync all" - instead, each of the
+components had to be synced individually. So this component started out as more
+of a "library" than a "component" which individual components could reuse - and
+each of these components was a "syncable store" (ie, a store which could supply
+ a "sync engine").
+
+Fast forward to Fenix and we needed a UI for managing all the engines supported
+there, and a single "sync now" experience etc - so we also have a sync_manager
+component - [see its README for more](../components/sync_manager/README.md).
+But even though it exists, there are still some parts of this component that
+reflect these early days - for example, it's still possible to sync just a
+single component using sync15 (ie, without going via the "sync manager"),
+although this isn't used and should be removed - the "sync manager" allows you
+to choose which engines to sync, so that should be used exclusively.
+
+## Metadata
+
+There's some metadata associated with a sync. Some of the metadata is "global"
+to the app (eg, the enabled state of engines, information about what servers to
+use, etc) and some is specific to an engine (eg, timestamp of the
+server's collection for this engine, guids for the collections, etc).
+
+We made the decision early on that no storage should be done by this
+component:
+
+* The "global" metadata should be stored by the application - but because it
+  doesn't need to interpret the data, we do this with an opaque string (that
+  is JSON, but the app should never assume or introspect that)
+
+* Each engine should store its own metadata, so we don't end up in the
+  situation where, say, a database is moved between profiles causing the
+  metadata to refer to a completely different data set. So each engine
+  stores its metadata in the same database as the data itself, so if the
+  database is moved or copied, the metadata comes with it)
+
+## Sync Implementation
+
+The core implementation does all of the interaction with things like the
+tokenserver, the `meta/global` and `info/collections` collections, etc. It
+does all network interaction (ie, individual engines don't need to interact with
+the network at all), tracks things like whether the server is asking us to
+"backoff" due to operational concerns, manages encryption keys and the
+encryption itself, etc. The general flow of a sync - which interacts with the
+`SyncEngine` trait - is:
+
+* Does all pre-sync setup, such as checking `meta/global`, and whether the
+  sync IDs on the server match the sync IDs we last saw (ie, to check whether
+  something drastic has happened since we last synced)
+* Asks the engine about how to formulate the URL query params to obtain the
+  records the engine cares about. In most cases, this will simply be "records
+  since the last modified timestamp of the last sync".
+* Downloads and decrypts these records.
+* Passes these records to the engine for processing, and obtains records that
+  should be uploaded to the server.
+* Encrypts these outgoing records and uploads them.
+* Tells the engine about the result of the upload (ie, the last-modified
+  timestamp of the POST so it can be saved as engine metadata)
+
+As above, the sync15 component really only deals with a single engine at a time.
+See the "sync manager" for how multiple engine are managed (but the tl;dr is
+that the "sync manager" leans on this very heavily, but knows about multiple
+engine and manages shared state)
+
+## The `SyncEngine` trait
+
+The SyncEngine trait is where all logic specific to a collection lives. A "sync
+engine" implements (or provides) this trait to implement actual syncing.
+
+For <handwave> reasons, it actually lives in the
+[sync-traits helper](https://github.com/mozilla/application-services/blob/main/components/support/sync15-traits/src/engine.rs)
+but for the purposes of this document, you should consider it as owned by sync15.
+
+This is actually quite a simple trait - at a high level, it's really just
+concerned with:
+
+* Get or set some metadata the sync15 component has decided should be saved or
+  fetched.
+
+* In a normal sync, take some "incoming" records, process them, and return
+  the "outgoing" records we should send to the server.
+
+* In some edge-cases, either "wipe" (ie, actually delete everything, which
+  almost never happens) or "reset" (ie, pretend this engine has never before
+  been synced)
+
+And that's it!
diff --git a/third_party/rust/sync15/src/bso/content.rs b/third_party/rust/sync15/src/bso/content.rs
new file mode 100644
index 0000000000..f7aa6f608b
--- /dev/null
+++ b/third_party/rust/sync15/src/bso/content.rs
@@ -0,0 +1,388 @@
+/* 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/.
+ */
+
+//! This module enhances the IncomingBso and OutgoingBso records to deal with
+//! arbitrary <T> types, which we call "content"
+//! It can:
+//! * Parse JSON into some <T> while handling tombstones and invalid json.
+//! * Turn arbitrary <T> objects with an `id` field into an OutgoingBso.
+
+use super::{IncomingBso, IncomingContent, IncomingKind, OutgoingBso, OutgoingEnvelope};
+use crate::Guid;
+use error_support::report_error;
+use serde::Serialize;
+
+// The only errors we return here are serde errors.
+type Result<T> = std::result::Result<T, serde_json::Error>;
+
+impl<T> IncomingContent<T> {
+    /// Returns Some(content) if [self.kind] is [IncomingKind::Content], None otherwise.
+    pub fn content(self) -> Option<T> {
+        match self.kind {
+            IncomingKind::Content(t) => Some(t),
+            _ => None,
+        }
+    }
+}
+
+// We don't want to force our T to be Debug, but we can be Debug if T is.
+impl<T: std::fmt::Debug> std::fmt::Debug for IncomingKind<T> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            IncomingKind::Content(r) => {
+                write!(f, "IncomingKind::Content<{:?}>", r)
+            }
+            IncomingKind::Tombstone => write!(f, "IncomingKind::Tombstone"),
+            IncomingKind::Malformed => write!(f, "IncomingKind::Malformed"),
+        }
+    }
+}
+
+impl IncomingBso {
+    /// Convert an [IncomingBso] to an [IncomingContent] possibly holding a T.
+    pub fn into_content<T: for<'de> serde::Deserialize<'de>>(self) -> IncomingContent<T> {
+        match serde_json::from_str(&self.payload) {
+            Ok(json) => {
+                // We got a good serde_json::Value, see if it's a <T>.
+                let kind = json_to_kind(json, &self.envelope.id);
+                IncomingContent {
+                    envelope: self.envelope,
+                    kind,
+                }
+            }
+            Err(e) => {
+                // payload isn't valid json.
+                log::warn!("Invalid incoming cleartext {}: {}", self.envelope.id, e);
+                IncomingContent {
+                    envelope: self.envelope,
+                    kind: IncomingKind::Malformed,
+                }
+            }
+        }
+    }
+}
+
+impl OutgoingBso {
+    /// Creates a new tombstone record.
+    /// Not all collections expect tombstones.
+    pub fn new_tombstone(envelope: OutgoingEnvelope) -> Self {
+        Self {
+            envelope,
+            payload: serde_json::json!({"deleted": true}).to_string(),
+        }
+    }
+
+    /// Creates a outgoing record from some <T>, which can be made into a JSON object
+    /// with a valid `id`. This is the most convenient way to create an outgoing
+    /// item from a <T> when the default envelope is suitable.
+    /// Will panic if there's no good `id` in the json.
+    pub fn from_content_with_id<T>(record: T) -> Result<Self>
+    where
+        T: Serialize,
+    {
+        let (json, id) = content_with_id_to_json(record)?;
+        Ok(Self {
+            envelope: id.into(),
+            payload: serde_json::to_string(&json)?,
+        })
+    }
+
+    /// Create an Outgoing record with an explicit envelope. Will panic if the
+    /// payload has an ID but it doesn't match the envelope.
+    pub fn from_content<T>(envelope: OutgoingEnvelope, record: T) -> Result<Self>
+    where
+        T: Serialize,
+    {
+        let json = content_to_json(record, &envelope.id)?;
+        Ok(Self {
+            envelope,
+            payload: serde_json::to_string(&json)?,
+        })
+    }
+}
+
+// Helpers for packing and unpacking serde objects to and from a <T>. In particular:
+// * Helping deal complications around raw json payload not having 'id' (the envelope is
+//   canonical) but needing it to exist when dealing with serde locally.
+//   For example, a record on the server after being decrypted looks like:
+//   `{"id": "a-guid", payload: {"field": "value"}}`
+//   But the `T` for this typically looks like `struct T { id: Guid, field: String}`
+//   So before we try and deserialize this record into a T, we copy the `id` field
+//   from the envelope into the payload, and when serializing from a T we do the
+//   reverse (ie, ensure the `id` in the payload is removed and placed in the envelope)
+// * Tombstones.
+
+// Deserializing json into a T
+fn json_to_kind<T>(mut json: serde_json::Value, id: &Guid) -> IncomingKind<T>
+where
+    T: for<'de> serde::Deserialize<'de>,
+{
+    // It's possible that the payload does not carry 'id', but <T> always does - so grab it from the
+    // envelope and put it into the json before deserializing the record.
+    if let serde_json::Value::Object(ref mut map) = json {
+        if map.contains_key("deleted") {
+            return IncomingKind::Tombstone;
+        }
+        match map.get("id") {
+            Some(serde_json::Value::String(content_id)) => {
+                // It exists in the payload! We treat a mismatch as malformed.
+                if content_id != id {
+                    log::trace!(
+                        "malformed incoming record: envelope id: {} payload id: {}",
+                        content_id,
+                        id
+                    );
+                    report_error!(
+                        "incoming-invalid-mismatched-ids",
+                        "Envelope and payload don't agree on the ID"
+                    );
+                    return IncomingKind::Malformed;
+                }
+                if !id.is_valid_for_sync_server() {
+                    log::trace!("malformed incoming record: id is not valid: {}", id);
+                    report_error!(
+                        "incoming-invalid-bad-payload-id",
+                        "ID in the payload is invalid"
+                    );
+                    return IncomingKind::Malformed;
+                }
+            }
+            Some(v) => {
+                // It exists in the payload but is not a string - they can't possibly be
+                // the same as the envelope uses a String, so must be malformed.
+                log::trace!("malformed incoming record: id is not a string: {}", v);
+                report_error!("incoming-invalid-wrong_type", "ID is not a string");
+                return IncomingKind::Malformed;
+            }
+            None => {
+                // Doesn't exist in the payload - add it before trying to deser a T.
+                if !id.is_valid_for_sync_server() {
+                    log::trace!("malformed incoming record: id is not valid: {}", id);
+                    report_error!(
+                        "incoming-invalid-bad-envelope-id",
+                        "ID in envelope is not valid"
+                    );
+                    return IncomingKind::Malformed;
+                }
+                map.insert("id".to_string(), id.to_string().into());
+            }
+        }
+    };
+    match serde_json::from_value(json) {
+        Ok(v) => IncomingKind::Content(v),
+        Err(e) => {
+            report_error!("invalid-incoming-content", "Invalid incoming T: {}", e);
+            IncomingKind::Malformed
+        }
+    }
+}
+
+// Serializing <T> into json with special handling of `id` (the `id` from the payload
+// is used as the envelope ID)
+fn content_with_id_to_json<T>(record: T) -> Result<(serde_json::Value, Guid)>
+where
+    T: Serialize,
+{
+    let mut json = serde_json::to_value(record)?;
+    let id = match json.as_object_mut() {
+        Some(ref mut map) => {
+            match map.get("id").as_ref().and_then(|v| v.as_str()) {
+                Some(id) => {
+                    let id: Guid = id.into();
+                    assert!(id.is_valid_for_sync_server(), "record's ID is invalid");
+                    id
+                }
+                // In practice, this is a "static" error and not influenced by runtime behavior
+                None => panic!("record does not have an ID in the payload"),
+            }
+        }
+        None => panic!("record is not a json object"),
+    };
+    Ok((json, id))
+}
+
+// Serializing <T> into json with special handling of `id` (if `id` in serialized
+// JSON already exists, we panic if it doesn't match the envelope. If the serialized
+// content does not have an `id`, it is added from the envelope)
+// is used as the envelope ID)
+fn content_to_json<T>(record: T, id: &Guid) -> Result<serde_json::Value>
+where
+    T: Serialize,
+{
+    let mut payload = serde_json::to_value(record)?;
+    if let Some(ref mut map) = payload.as_object_mut() {
+        if let Some(content_id) = map.get("id").as_ref().and_then(|v| v.as_str()) {
+            assert_eq!(content_id, id);
+            assert!(id.is_valid_for_sync_server(), "record's ID is invalid");
+        } else {
+            map.insert("id".to_string(), serde_json::Value::String(id.to_string()));
+        }
+    };
+    Ok(payload)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::bso::IncomingBso;
+    use serde::{Deserialize, Serialize};
+    use serde_json::json;
+
+    #[derive(Default, Debug, PartialEq, Serialize, Deserialize)]
+    struct TestStruct {
+        id: Guid,
+        data: u32,
+    }
+    #[test]
+    fn test_content_deser() {
+        env_logger::try_init().ok();
+        let json = json!({
+            "id": "test",
+            "payload": json!({"data": 1}).to_string(),
+        });
+        let incoming: IncomingBso = serde_json::from_value(json).unwrap();
+        assert_eq!(incoming.envelope.id, "test");
+        let record = incoming.into_content::<TestStruct>().content().unwrap();
+        let expected = TestStruct {
+            id: Guid::new("test"),
+            data: 1,
+        };
+        assert_eq!(record, expected);
+    }
+
+    #[test]
+    fn test_content_deser_empty_id() {
+        env_logger::try_init().ok();
+        let json = json!({
+            "id": "",
+            "payload": json!({"data": 1}).to_string(),
+        });
+        let incoming: IncomingBso = serde_json::from_value(json).unwrap();
+        // The envelope has an invalid ID, but it's not handled until we try and deserialize
+        // it into a T
+        assert_eq!(incoming.envelope.id, "");
+        let content = incoming.into_content::<TestStruct>();
+        assert!(matches!(content.kind, IncomingKind::Malformed));
+    }
+
+    #[test]
+    fn test_content_deser_invalid() {
+        env_logger::try_init().ok();
+        // And a non-empty but still invalid guid.
+        let json = json!({
+            "id": "X".repeat(65),
+            "payload": json!({"data": 1}).to_string(),
+        });
+        let incoming: IncomingBso = serde_json::from_value(json).unwrap();
+        let content = incoming.into_content::<TestStruct>();
+        assert!(matches!(content.kind, IncomingKind::Malformed));
+    }
+
+    #[test]
+    fn test_content_deser_not_string() {
+        env_logger::try_init().ok();
+        // A non-string id.
+        let json = json!({
+            "id": "0",
+            "payload": json!({"id": 0, "data": 1}).to_string(),
+        });
+        let incoming: IncomingBso = serde_json::from_value(json).unwrap();
+        let content = incoming.into_content::<serde_json::Value>();
+        assert!(matches!(content.kind, IncomingKind::Malformed));
+    }
+
+    #[test]
+    fn test_content_ser_with_id() {
+        env_logger::try_init().ok();
+        // When serializing, expect the ID to be in the top-level payload (ie,
+        // in the envelope) but should not appear in the cleartext `payload` part of
+        // the payload.
+        let val = TestStruct {
+            id: Guid::new("test"),
+            data: 1,
+        };
+        let outgoing = OutgoingBso::from_content_with_id(val).unwrap();
+
+        // The envelope should have our ID.
+        assert_eq!(outgoing.envelope.id, Guid::new("test"));
+
+        // and make sure `cleartext` part of the payload the data and the id.
+        let ct_value = serde_json::from_str::<serde_json::Value>(&outgoing.payload).unwrap();
+        assert_eq!(ct_value, json!({"data": 1, "id": "test"}));
+    }
+
+    #[test]
+    fn test_content_ser_with_envelope() {
+        env_logger::try_init().ok();
+        // When serializing, expect the ID to be in the top-level payload (ie,
+        // in the envelope) but should not appear in the cleartext `payload`
+        let val = TestStruct {
+            id: Guid::new("test"),
+            data: 1,
+        };
+        let envelope: OutgoingEnvelope = Guid::new("test").into();
+        let outgoing = OutgoingBso::from_content(envelope, val).unwrap();
+
+        // The envelope should have our ID.
+        assert_eq!(outgoing.envelope.id, Guid::new("test"));
+
+        // and make sure `cleartext` part of the payload has data and the id.
+        let ct_value = serde_json::from_str::<serde_json::Value>(&outgoing.payload).unwrap();
+        assert_eq!(ct_value, json!({"data": 1, "id": "test"}));
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_content_ser_no_ids() {
+        env_logger::try_init().ok();
+        #[derive(Serialize)]
+        struct StructWithNoId {
+            data: u32,
+        }
+        let val = StructWithNoId { data: 1 };
+        let _ = OutgoingBso::from_content_with_id(val);
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_content_ser_not_object() {
+        env_logger::try_init().ok();
+        let _ = OutgoingBso::from_content_with_id(json!("string"));
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_content_ser_mismatched_ids() {
+        env_logger::try_init().ok();
+        let val = TestStruct {
+            id: Guid::new("test"),
+            data: 1,
+        };
+        let envelope: OutgoingEnvelope = Guid::new("different").into();
+        let _ = OutgoingBso::from_content(envelope, val);
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_content_empty_id() {
+        env_logger::try_init().ok();
+        let val = TestStruct {
+            id: Guid::new(""),
+            data: 1,
+        };
+        let _ = OutgoingBso::from_content_with_id(val);
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_content_invalid_id() {
+        env_logger::try_init().ok();
+        let val = TestStruct {
+            id: Guid::new(&"X".repeat(65)),
+            data: 1,
+        };
+        let _ = OutgoingBso::from_content_with_id(val);
+    }
+}
diff --git a/third_party/rust/sync15/src/bso/crypto.rs b/third_party/rust/sync15/src/bso/crypto.rs
new file mode 100644
index 0000000000..d572c4692b
--- /dev/null
+++ b/third_party/rust/sync15/src/bso/crypto.rs
@@ -0,0 +1,197 @@
+/* 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/. */
+
+//! Support for "encrypted bso"s, as received by the storage servers.
+//! This module decrypts them into IncomingBso's suitable for use by the
+//! engines.
+use super::{IncomingBso, IncomingEnvelope, OutgoingBso, OutgoingEnvelope};
+use crate::error;
+use crate::key_bundle::KeyBundle;
+use crate::EncryptedPayload;
+use serde::{de::DeserializeOwned, Deserialize, Serialize};
+
+// The BSO implementation we use for encrypted payloads.
+// Note that this is almost identical to the IncomingBso implementations, except
+// instead of a String payload we use an EncryptedPayload. Obviously we *could*
+// just use a String payload and transform it into an EncryptedPayload - any maybe we
+// should - but this is marginally optimal in terms of deserialization.
+#[derive(Deserialize, Debug)]
+pub struct IncomingEncryptedBso {
+    #[serde(flatten)]
+    pub envelope: IncomingEnvelope,
+    #[serde(
+        with = "as_json",
+        bound(deserialize = "EncryptedPayload: DeserializeOwned")
+    )]
+    pub(crate) payload: EncryptedPayload,
+}
+
+impl IncomingEncryptedBso {
+    pub fn new(envelope: IncomingEnvelope, payload: EncryptedPayload) -> Self {
+        Self { envelope, payload }
+    }
+    /// Decrypt a BSO, consuming it into a clear-text version.
+    pub fn into_decrypted(self, key: &KeyBundle) -> error::Result<IncomingBso> {
+        Ok(IncomingBso::new(self.envelope, self.payload.decrypt(key)?))
+    }
+}
+
+#[derive(Serialize, Debug)]
+pub struct OutgoingEncryptedBso {
+    #[serde(flatten)]
+    pub envelope: OutgoingEnvelope,
+    #[serde(with = "as_json", bound(serialize = "EncryptedPayload: Serialize"))]
+    payload: EncryptedPayload,
+}
+
+impl OutgoingEncryptedBso {
+    pub fn new(envelope: OutgoingEnvelope, payload: EncryptedPayload) -> Self {
+        Self { envelope, payload }
+    }
+
+    #[inline]
+    pub fn serialized_payload_len(&self) -> usize {
+        self.payload.serialized_len()
+    }
+}
+
+impl OutgoingBso {
+    pub fn into_encrypted(self, key: &KeyBundle) -> error::Result<OutgoingEncryptedBso> {
+        Ok(OutgoingEncryptedBso {
+            envelope: self.envelope,
+            payload: EncryptedPayload::from_cleartext(key, self.payload)?,
+        })
+    }
+}
+
+// The BSOs we write to the servers expect a "payload" attribute which is a JSON serialized
+// string, rather than the JSON representation of the object.
+// ie, the serialized object is expected to look like:
+// `{"id": "some-guid", "payload": "{\"IV\": ... }"}` <-- payload is a string.
+// However, if we just serialize it directly, we end up with:
+// `{"id": "some-guid", "payload": {"IV":  ... }}` <-- payload is an object.
+// The magic here means we can serialize and deserialize directly into/from the object, correctly
+// working with the payload as a string, instead of needing to explicitly stringify/parse the
+// payload as an extra step.
+//
+// This would work for any <T>, but we only use it for EncryptedPayload - the way our cleartext
+// BSOs work mean it's not necessary there as they define the payload as a String - ie, they do
+// explicitly end up doing 2 JSON operations as an ergonomic design choice.
+mod as_json {
+    use serde::de::{self, Deserialize, DeserializeOwned, Deserializer};
+    use serde::ser::{self, Serialize, Serializer};
+
+    pub fn serialize<T, S>(t: &T, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        T: Serialize,
+        S: Serializer,
+    {
+        let j = serde_json::to_string(t).map_err(ser::Error::custom)?;
+        serializer.serialize_str(&j)
+    }
+
+    pub fn deserialize<'de, T, D>(deserializer: D) -> Result<T, D::Error>
+    where
+        T: DeserializeOwned,
+        D: Deserializer<'de>,
+    {
+        let j = String::deserialize(deserializer)?;
+        serde_json::from_str(&j).map_err(de::Error::custom)
+    }
+}
+
+// Lots of stuff for testing the sizes of encrypted records, because the servers have
+// certain limits in terms of max-POST sizes, forcing us to chunk uploads, but
+// we need to calculate based on encrypted record size rather than the raw <T> size.
+//
+// This is a little cludgey but I couldn't think of another way to have easy deserialization
+// without a bunch of wrapper types, while still only serializing a single time in the
+// postqueue.
+#[cfg(test)]
+impl OutgoingEncryptedBso {
+    /// Return the length of the serialized payload.
+    pub fn payload_serialized_len(&self) -> usize {
+        self.payload.serialized_len()
+    }
+
+    // self.payload is private, but tests want to create funky things.
+    // XXX - test only, but test in another crate :(
+    //#[cfg(test)]
+    pub fn make_test_bso(ciphertext: String) -> Self {
+        Self {
+            envelope: OutgoingEnvelope {
+                id: "".into(),
+                sortindex: None,
+                ttl: None,
+            },
+            payload: EncryptedPayload {
+                iv: "".into(),
+                hmac: "".into(),
+                ciphertext,
+            },
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::bso::OutgoingEnvelope;
+
+    #[test]
+    fn test_deserialize_enc() {
+        let serialized = r#"{
+            "id": "1234",
+            "collection": "passwords",
+            "modified": 12344321.0,
+            "payload": "{\"IV\": \"aaaaa\", \"hmac\": \"bbbbb\", \"ciphertext\": \"ccccc\"}"
+        }"#;
+        let record: IncomingEncryptedBso = serde_json::from_str(serialized).unwrap();
+        assert_eq!(&record.envelope.id, "1234");
+        assert_eq!((record.envelope.modified.0 - 12_344_321_000).abs(), 0);
+        assert_eq!(record.envelope.sortindex, None);
+        assert_eq!(&record.payload.iv, "aaaaa");
+        assert_eq!(&record.payload.hmac, "bbbbb");
+        assert_eq!(&record.payload.ciphertext, "ccccc");
+    }
+
+    #[test]
+    fn test_deserialize_autofields() {
+        let serialized = r#"{
+            "id": "1234",
+            "collection": "passwords",
+            "modified": 12344321.0,
+            "sortindex": 100,
+            "ttl": 99,
+            "payload": "{\"IV\": \"aaaaa\", \"hmac\": \"bbbbb\", \"ciphertext\": \"ccccc\"}"
+        }"#;
+        let record: IncomingEncryptedBso = serde_json::from_str(serialized).unwrap();
+        assert_eq!(record.envelope.sortindex, Some(100));
+        assert_eq!(record.envelope.ttl, Some(99));
+    }
+
+    #[test]
+    fn test_serialize_enc() {
+        let goal = r#"{"id":"1234","payload":"{\"IV\":\"aaaaa\",\"hmac\":\"bbbbb\",\"ciphertext\":\"ccccc\"}"}"#;
+        let record = OutgoingEncryptedBso {
+            envelope: OutgoingEnvelope {
+                id: "1234".into(),
+                ..Default::default()
+            },
+            payload: EncryptedPayload {
+                iv: "aaaaa".into(),
+                hmac: "bbbbb".into(),
+                ciphertext: "ccccc".into(),
+            },
+        };
+        let actual = serde_json::to_string(&record).unwrap();
+        assert_eq!(actual, goal);
+
+        let val_str_payload: serde_json::Value = serde_json::from_str(goal).unwrap();
+        assert_eq!(
+            val_str_payload["payload"].as_str().unwrap().len(),
+            record.payload.serialized_len()
+        )
+    }
+}
diff --git a/third_party/rust/sync15/src/bso/mod.rs b/third_party/rust/sync15/src/bso/mod.rs
new file mode 100644
index 0000000000..251c11fb3b
--- /dev/null
+++ b/third_party/rust/sync15/src/bso/mod.rs
@@ -0,0 +1,204 @@
+/* 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/. */
+
+/// This module defines our core "bso" abstractions.
+/// In the terminology of this crate:
+/// * "bso" is an acronym for "basic storage object" and used extensively in the sync server docs.
+///    the record always has a well-defined "envelope" with metadata (eg, the ID of the record,
+///    the server timestamp of the resource,  etc) and a field called `payload`.
+///    A bso is serialized to and from JSON.
+/// * There's a "cleartext" bso:
+/// ** The payload is a String, which itself is JSON encoded (ie, this string `payload` is
+///    always double JSON encoded in a server record)
+/// ** This supplies helper methods for working with the "content" (some arbitrary <T>) in the
+///    payload.
+/// * There's an "encrypted" bso
+/// ** The payload is an [crate::enc_payload::EncryptedPayload]
+/// ** Only clients use this; as soon as practical we decrypt and as late as practical we encrypt
+///    to and from encrypted bsos.
+/// ** The encrypted bsos etc are all in the [crypto] module and require the `crypto` feature.
+///
+/// Let's look at some real-world examples:
+/// # meta/global
+/// A "bso" (ie, record with an "envelope" and a "payload" with a JSON string) - but the payload
+/// is cleartext.
+/// ```json
+/// {
+///   "id":"global",
+///   "modified":1661564513.50,
+///   "payload": "{\"syncID\":\"p1z5_oDdOfLF\",\"storageVersion\":5,\"engines\":{\"passwords\":{\"version\":1,\"syncID\":\"6Y6JJkB074cF\"} /* snip */},\"declined\":[]}"
+/// }```
+///
+/// # encrypted bsos:
+/// Encrypted BSOs are still a "bso" (ie, a record with a field names `payload` which is a string)
+/// but the payload is in the form of an EncryptedPayload.
+/// For example, crypto/keys:
+/// ```json
+/// {
+///   "id":"keys",
+///   "modified":1661564513.74,
+///   "payload":"{\"IV\":\"snip-base-64==\",\"hmac\":\"snip-hex\",\"ciphertext\":\"snip-base64==\"}"
+/// }```
+/// (Note that as described above, most code working with bsos *do not* use that `payload`
+/// directly, but instead a decrypted cleartext bso.
+///
+/// Note all collection responses are the same shape as `crypto/keys` - a `payload` field with a
+/// JSON serialized EncryptedPayload, it's just that the final <T> content differs for each
+/// collection (eg, tabs and bookmarks have quite different <T>s JSON-encoded in the
+/// String payload.)
+///
+/// For completeness, some other "non-BSO" records - no "id", "modified" or "payload" fields in
+/// the response, just plain-old clear-text JSON.
+/// # Example
+/// ## `info/collections`
+/// ```json
+/// {
+///   "bookmarks":1661564648.65,
+///   "meta":1661564513.50,
+///   "addons":1661564649.09,
+///   "clients":1661564643.57,
+///   ...
+/// }```
+/// ## `info/configuration`
+/// ```json
+/// {
+///   "max_post_bytes":2097152,
+///   "max_post_records":100,
+///   "max_record_payload_bytes":2097152,
+///   ...
+/// }```
+///
+/// Given our definitions above, these are not any kind of "bso", so are
+/// not relevant to this module
+use crate::{Guid, ServerTimestamp};
+use serde::{Deserialize, Serialize};
+
+#[cfg(feature = "crypto")]
+mod crypto;
+#[cfg(feature = "crypto")]
+pub use crypto::{IncomingEncryptedBso, OutgoingEncryptedBso};
+
+mod content;
+
+// A feature for this would be ideal, but (a) the module is small and (b) it
+// doesn't really fit the "features" model for sync15 to have a dev-dependency
+// against itself but with a different feature set.
+pub mod test_utils;
+
+/// An envelope for an incoming item. Envelopes carry all the metadata for
+/// a Sync BSO record (`id`, `modified`, `sortindex`), *but not* the payload
+/// itself.
+#[derive(Debug, Clone, Deserialize)]
+pub struct IncomingEnvelope {
+    /// The ID of the record.
+    pub id: Guid,
+    // If we don't give it a default, a small handful of tests fail.
+    // XXX - we should probably fix the tests and kill this?
+    #[serde(default = "ServerTimestamp::default")]
+    pub modified: ServerTimestamp,
+    pub sortindex: Option<i32>,
+    pub ttl: Option<u32>,
+}
+
+/// An envelope for an outgoing item. This is conceptually identical to
+/// [IncomingEnvelope], but omits fields that are only set by the server,
+/// like `modified`.
+#[derive(Debug, Default, Clone, Serialize)]
+pub struct OutgoingEnvelope {
+    /// The ID of the record.
+    pub id: Guid,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub sortindex: Option<i32>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub ttl: Option<u32>,
+}
+
+/// Allow an outgoing envelope to be constructed with just a guid when default
+/// values for the other fields are OK.
+impl From<Guid> for OutgoingEnvelope {
+    fn from(id: Guid) -> Self {
+        OutgoingEnvelope {
+            id,
+            ..Default::default()
+        }
+    }
+}
+
+/// IncomingBso's can come from:
+/// * Directly from the server (ie, some records aren't encrypted, such as meta/global)
+/// * From environments where the encryption is done externally (eg, Rust syncing in Desktop
+///   Firefox has the encryption/decryption done by Firefox and the cleartext BSOs are passed in.
+/// * Read from the server as an EncryptedBso; see EncryptedBso description above.
+#[derive(Deserialize, Debug)]
+pub struct IncomingBso {
+    #[serde(flatten)]
+    pub envelope: IncomingEnvelope,
+    // payload is public for some edge-cases in some components, but in general,
+    // you should use into_content<> to get a record out of it.
+    pub payload: String,
+}
+
+impl IncomingBso {
+    pub fn new(envelope: IncomingEnvelope, payload: String) -> Self {
+        Self { envelope, payload }
+    }
+}
+
+#[derive(Serialize, Debug)]
+pub struct OutgoingBso {
+    #[serde(flatten)]
+    pub envelope: OutgoingEnvelope,
+    // payload is public for some edge-cases in some components, but in general,
+    // you should use into_content<> to get a record out of it.
+    pub payload: String,
+}
+
+impl OutgoingBso {
+    /// Most consumers will use `self.from_content` and `self.from_content_with_id`
+    /// but this exists for the few consumers for whom that doesn't make sense.
+    pub fn new<T: Serialize>(
+        envelope: OutgoingEnvelope,
+        val: &T,
+    ) -> Result<Self, serde_json::Error> {
+        Ok(Self {
+            envelope,
+            payload: serde_json::to_string(&val)?,
+        })
+    }
+}
+
+/// We also have the concept of "content", which helps work with a `T` which
+/// is represented inside the payload. Real-world examples of a `T` include
+/// Bookmarks or Tabs.
+/// See the content module for the implementations.
+///
+/// So this all flows together in the following way:
+/// * Incoming encrypted data:
+///   EncryptedIncomingBso -> IncomingBso -> [specific engine] -> IncomingContent<T>
+/// * Incoming cleartext data:
+///   IncomingBso -> IncomingContent<T>
+///   (Note that incoming cleartext only happens for a few collections managed by
+///   the sync client and never by specific engines - engine BSOs are always encryted)
+/// * Outgoing encrypted data:
+///   OutgoingBso (created in the engine) -> [this crate] -> EncryptedOutgoingBso
+///  * Outgoing cleartext data: just an OutgoingBso with no conversions needed.
+
+/// [IncomingContent] is the result of converting an [IncomingBso] into
+/// some <T> - it consumes the Bso, so you get the envelope, and the [IncomingKind]
+/// which reflects the state of parsing the json.
+#[derive(Debug)]
+pub struct IncomingContent<T> {
+    pub envelope: IncomingEnvelope,
+    pub kind: IncomingKind<T>,
+}
+
+/// The "kind" of incoming content after deserializing it.
+pub enum IncomingKind<T> {
+    /// A good, live T.
+    Content(T),
+    /// A record that used to be a T but has been replaced with a tombstone.
+    Tombstone,
+    /// Either not JSON, or can't be made into a T.
+    Malformed,
+}
diff --git a/third_party/rust/sync15/src/bso/test_utils.rs b/third_party/rust/sync15/src/bso/test_utils.rs
new file mode 100644
index 0000000000..55735afda2
--- /dev/null
+++ b/third_party/rust/sync15/src/bso/test_utils.rs
@@ -0,0 +1,61 @@
+/* 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/. */
+
+//! Utilities for tests to make IncomingBsos and Content from test data.
+use super::{IncomingBso, IncomingEnvelope, OutgoingBso};
+use crate::{Guid, ServerTimestamp};
+
+/// Tests often want an IncomingBso to test, and the easiest way is often to
+/// create an OutgoingBso convert it back to an incoming.
+impl OutgoingBso {
+    // These functions would ideally consume `self` and avoid the clones, but
+    // this is more convenient for some tests and the extra overhead doesn't
+    // really matter for tests.
+    /// When a test has an [OutgoingBso] and wants it as an [IncomingBso]
+    pub fn to_test_incoming(&self) -> IncomingBso {
+        self.to_test_incoming_ts(ServerTimestamp::default())
+    }
+
+    /// When a test has an [OutgoingBso] and wants it as an [IncomingBso] with a specific timestamp.
+    pub fn to_test_incoming_ts(&self, ts: ServerTimestamp) -> IncomingBso {
+        IncomingBso {
+            envelope: IncomingEnvelope {
+                id: self.envelope.id.clone(),
+                modified: ts,
+                sortindex: self.envelope.sortindex,
+                ttl: self.envelope.ttl,
+            },
+            payload: self.payload.clone(),
+        }
+    }
+
+    /// When a test has an [OutgoingBso] and wants it as an [IncomingBso] with a specific T.
+    pub fn to_test_incoming_t<T: for<'de> serde::Deserialize<'de>>(&self) -> T {
+        self.to_test_incoming().into_content().content().unwrap()
+    }
+}
+
+/// Helpers to create an IncomingBso from some T
+impl IncomingBso {
+    /// When a test has an T and wants it as an [IncomingBso]
+    pub fn from_test_content<T: serde::Serialize>(json: T) -> Self {
+        // Go via an OutgoingBso
+        OutgoingBso::from_content_with_id(json)
+            .unwrap()
+            .to_test_incoming()
+    }
+
+    /// When a test has an T and wants it as an [IncomingBso] with a specific timestamp.
+    pub fn from_test_content_ts<T: serde::Serialize>(json: T, ts: ServerTimestamp) -> Self {
+        // Go via an OutgoingBso
+        OutgoingBso::from_content_with_id(json)
+            .unwrap()
+            .to_test_incoming_ts(ts)
+    }
+
+    /// When a test wants a new incoming tombstone.
+    pub fn new_test_tombstone(guid: Guid) -> Self {
+        OutgoingBso::new_tombstone(guid.into()).to_test_incoming()
+    }
+}
diff --git a/third_party/rust/sync15/src/client/coll_state.rs b/third_party/rust/sync15/src/client/coll_state.rs
new file mode 100644
index 0000000000..53e3071f92
--- /dev/null
+++ b/third_party/rust/sync15/src/client/coll_state.rs
@@ -0,0 +1,353 @@
+/* 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/. */
+
+use super::request::InfoConfiguration;
+use super::{CollectionKeys, GlobalState};
+use crate::engine::{CollSyncIds, EngineSyncAssociation, SyncEngine};
+use crate::error;
+use crate::KeyBundle;
+use crate::ServerTimestamp;
+
+/// Holds state for a collection necessary to perform a sync of it. Lives for the lifetime
+/// of a single sync.
+#[derive(Debug, Clone)]
+pub struct CollState {
+    // Info about the server configuration/capabilities
+    pub config: InfoConfiguration,
+    // from meta/global, used for XIUS when we POST outgoing record based on this state.
+    pub last_modified: ServerTimestamp,
+    pub key: KeyBundle,
+}
+
+/// This mini state-machine helps build a CollState
+#[derive(Debug)]
+pub enum LocalCollState {
+    /// The state is unknown, with the EngineSyncAssociation the collection
+    /// reports.
+    Unknown { assoc: EngineSyncAssociation },
+
+    /// The engine has been declined. This is a "terminal" state.
+    Declined,
+
+    /// There's no such collection in meta/global. We could possibly update
+    /// meta/global, but currently all known collections are there by default,
+    /// so this is, basically, an error condition.
+    NoSuchCollection,
+
+    /// Either the global or collection sync ID has changed - we will reset the engine.
+    SyncIdChanged { ids: CollSyncIds },
+
+    /// The collection is ready to sync.
+    Ready { coll_state: CollState },
+}
+
+pub struct LocalCollStateMachine<'state> {
+    global_state: &'state GlobalState,
+    root_key: &'state KeyBundle,
+}
+
+impl<'state> LocalCollStateMachine<'state> {
+    fn advance(
+        &self,
+        from: LocalCollState,
+        engine: &dyn SyncEngine,
+    ) -> error::Result<LocalCollState> {
+        let name = &engine.collection_name().to_string();
+        let meta_global = &self.global_state.global;
+        match from {
+            LocalCollState::Unknown { assoc } => {
+                if meta_global.declined.contains(name) {
+                    return Ok(LocalCollState::Declined);
+                }
+                match meta_global.engines.get(name) {
+                    Some(engine_meta) => match assoc {
+                        EngineSyncAssociation::Disconnected => Ok(LocalCollState::SyncIdChanged {
+                            ids: CollSyncIds {
+                                global: meta_global.sync_id.clone(),
+                                coll: engine_meta.sync_id.clone(),
+                            },
+                        }),
+                        EngineSyncAssociation::Connected(ref ids)
+                            if ids.global == meta_global.sync_id
+                                && ids.coll == engine_meta.sync_id =>
+                        {
+                            // We are done - build the CollState
+                            let coll_keys = CollectionKeys::from_encrypted_payload(
+                                self.global_state.keys.clone(),
+                                self.global_state.keys_timestamp,
+                                self.root_key,
+                            )?;
+                            let key = coll_keys.key_for_collection(name).clone();
+                            let name = engine.collection_name();
+                            let config = self.global_state.config.clone();
+                            let last_modified = self
+                                .global_state
+                                .collections
+                                .get(name.as_ref())
+                                .cloned()
+                                .unwrap_or_default();
+                            let coll_state = CollState {
+                                config,
+                                last_modified,
+                                key,
+                            };
+                            Ok(LocalCollState::Ready { coll_state })
+                        }
+                        _ => Ok(LocalCollState::SyncIdChanged {
+                            ids: CollSyncIds {
+                                global: meta_global.sync_id.clone(),
+                                coll: engine_meta.sync_id.clone(),
+                            },
+                        }),
+                    },
+                    None => Ok(LocalCollState::NoSuchCollection),
+                }
+            }
+
+            LocalCollState::Declined => unreachable!("can't advance from declined"),
+
+            LocalCollState::NoSuchCollection => unreachable!("the collection is unknown"),
+
+            LocalCollState::SyncIdChanged { ids } => {
+                let assoc = EngineSyncAssociation::Connected(ids);
+                log::info!("Resetting {} engine", engine.collection_name());
+                engine.reset(&assoc)?;
+                Ok(LocalCollState::Unknown { assoc })
+            }
+
+            LocalCollState::Ready { .. } => unreachable!("can't advance from ready"),
+        }
+    }
+
+    // A little whimsy - a portmanteau of far and fast
+    fn run_and_run_as_farst_as_you_can(
+        &mut self,
+        engine: &dyn SyncEngine,
+    ) -> error::Result<Option<CollState>> {
+        let mut s = LocalCollState::Unknown {
+            assoc: engine.get_sync_assoc()?,
+        };
+        // This is a simple state machine and should never take more than
+        // 10 goes around.
+        let mut count = 0;
+        loop {
+            log::trace!("LocalCollState in {:?}", s);
+            match s {
+                LocalCollState::Ready { coll_state } => return Ok(Some(coll_state)),
+                LocalCollState::Declined | LocalCollState::NoSuchCollection => return Ok(None),
+                _ => {
+                    count += 1;
+                    if count > 10 {
+                        log::warn!("LocalCollStateMachine appears to be looping");
+                        return Ok(None);
+                    }
+                    // should we have better loop detection? Our limit of 10
+                    // goes is probably OK for now, but not really ideal.
+                    s = self.advance(s, engine)?;
+                }
+            };
+        }
+    }
+
+    pub fn get_state(
+        engine: &dyn SyncEngine,
+        global_state: &'state GlobalState,
+        root_key: &'state KeyBundle,
+    ) -> error::Result<Option<CollState>> {
+        let mut gingerbread_man = Self {
+            global_state,
+            root_key,
+        };
+        gingerbread_man.run_and_run_as_farst_as_you_can(engine)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::super::request::{InfoCollections, InfoConfiguration};
+    use super::super::CollectionKeys;
+    use super::*;
+    use crate::engine::CollectionRequest;
+    use crate::engine::{IncomingChangeset, OutgoingChangeset};
+    use crate::record_types::{MetaGlobalEngine, MetaGlobalRecord};
+    use crate::{telemetry, CollectionName};
+    use anyhow::Result;
+    use std::cell::{Cell, RefCell};
+    use std::collections::HashMap;
+    use sync_guid::Guid;
+
+    fn get_global_state(root_key: &KeyBundle) -> GlobalState {
+        let keys = CollectionKeys::new_random()
+            .unwrap()
+            .to_encrypted_payload(root_key)
+            .unwrap();
+        GlobalState {
+            config: InfoConfiguration::default(),
+            collections: InfoCollections::new(HashMap::new()),
+            global: MetaGlobalRecord {
+                sync_id: "syncIDAAAAAA".into(),
+                storage_version: 5usize,
+                engines: vec![(
+                    "bookmarks",
+                    MetaGlobalEngine {
+                        version: 1usize,
+                        sync_id: "syncIDBBBBBB".into(),
+                    },
+                )]
+                .into_iter()
+                .map(|(key, value)| (key.to_owned(), value))
+                .collect(),
+                declined: vec![],
+            },
+            global_timestamp: ServerTimestamp::default(),
+            keys,
+            keys_timestamp: ServerTimestamp::default(),
+        }
+    }
+
+    struct TestSyncEngine {
+        collection_name: &'static str,
+        assoc: Cell<EngineSyncAssociation>,
+        num_resets: RefCell<usize>,
+    }
+
+    impl TestSyncEngine {
+        fn new(collection_name: &'static str, assoc: EngineSyncAssociation) -> Self {
+            Self {
+                collection_name,
+                assoc: Cell::new(assoc),
+                num_resets: RefCell::new(0),
+            }
+        }
+        fn get_num_resets(&self) -> usize {
+            *self.num_resets.borrow()
+        }
+    }
+
+    impl SyncEngine for TestSyncEngine {
+        fn collection_name(&self) -> CollectionName {
+            self.collection_name.into()
+        }
+
+        fn apply_incoming(
+            &self,
+            _inbound: Vec<IncomingChangeset>,
+            _telem: &mut telemetry::Engine,
+        ) -> Result<OutgoingChangeset> {
+            unreachable!("these tests shouldn't call these");
+        }
+
+        fn sync_finished(
+            &self,
+            _new_timestamp: ServerTimestamp,
+            _records_synced: Vec<Guid>,
+        ) -> Result<()> {
+            unreachable!("these tests shouldn't call these");
+        }
+
+        fn get_collection_requests(
+            &self,
+            _server_timestamp: ServerTimestamp,
+        ) -> Result<Vec<CollectionRequest>> {
+            unreachable!("these tests shouldn't call these");
+        }
+
+        fn get_sync_assoc(&self) -> Result<EngineSyncAssociation> {
+            Ok(self.assoc.replace(EngineSyncAssociation::Disconnected))
+        }
+
+        fn reset(&self, new_assoc: &EngineSyncAssociation) -> Result<()> {
+            self.assoc.replace(new_assoc.clone());
+            *self.num_resets.borrow_mut() += 1;
+            Ok(())
+        }
+
+        fn wipe(&self) -> Result<()> {
+            unreachable!("these tests shouldn't call these");
+        }
+    }
+
+    #[test]
+    fn test_unknown() {
+        let root_key = KeyBundle::new_random().expect("should work");
+        let gs = get_global_state(&root_key);
+        let engine = TestSyncEngine::new("unknown", EngineSyncAssociation::Disconnected);
+        let cs = LocalCollStateMachine::get_state(&engine, &gs, &root_key).expect("should work");
+        assert!(cs.is_none(), "unknown collection name can't sync");
+        assert_eq!(engine.get_num_resets(), 0);
+    }
+
+    #[test]
+    fn test_known_no_state() {
+        let root_key = KeyBundle::new_random().expect("should work");
+        let gs = get_global_state(&root_key);
+        let engine = TestSyncEngine::new("bookmarks", EngineSyncAssociation::Disconnected);
+        let cs = LocalCollStateMachine::get_state(&engine, &gs, &root_key).expect("should work");
+        assert!(cs.is_some(), "collection can sync");
+        assert_eq!(
+            engine.assoc.replace(EngineSyncAssociation::Disconnected),
+            EngineSyncAssociation::Connected(CollSyncIds {
+                global: "syncIDAAAAAA".into(),
+                coll: "syncIDBBBBBB".into(),
+            })
+        );
+        assert_eq!(engine.get_num_resets(), 1);
+    }
+
+    #[test]
+    fn test_known_wrong_state() {
+        let root_key = KeyBundle::new_random().expect("should work");
+        let gs = get_global_state(&root_key);
+        let engine = TestSyncEngine::new(
+            "bookmarks",
+            EngineSyncAssociation::Connected(CollSyncIds {
+                global: "syncIDXXXXXX".into(),
+                coll: "syncIDYYYYYY".into(),
+            }),
+        );
+        let cs = LocalCollStateMachine::get_state(&engine, &gs, &root_key).expect("should work");
+        assert!(cs.is_some(), "collection can sync");
+        assert_eq!(
+            engine.assoc.replace(EngineSyncAssociation::Disconnected),
+            EngineSyncAssociation::Connected(CollSyncIds {
+                global: "syncIDAAAAAA".into(),
+                coll: "syncIDBBBBBB".into(),
+            })
+        );
+        assert_eq!(engine.get_num_resets(), 1);
+    }
+
+    #[test]
+    fn test_known_good_state() {
+        let root_key = KeyBundle::new_random().expect("should work");
+        let gs = get_global_state(&root_key);
+        let engine = TestSyncEngine::new(
+            "bookmarks",
+            EngineSyncAssociation::Connected(CollSyncIds {
+                global: "syncIDAAAAAA".into(),
+                coll: "syncIDBBBBBB".into(),
+            }),
+        );
+        let cs = LocalCollStateMachine::get_state(&engine, &gs, &root_key).expect("should work");
+        assert!(cs.is_some(), "collection can sync");
+        assert_eq!(engine.get_num_resets(), 0);
+    }
+
+    #[test]
+    fn test_declined() {
+        let root_key = KeyBundle::new_random().expect("should work");
+        let mut gs = get_global_state(&root_key);
+        gs.global.declined.push("bookmarks".to_string());
+        let engine = TestSyncEngine::new(
+            "bookmarks",
+            EngineSyncAssociation::Connected(CollSyncIds {
+                global: "syncIDAAAAAA".into(),
+                coll: "syncIDBBBBBB".into(),
+            }),
+        );
+        let cs = LocalCollStateMachine::get_state(&engine, &gs, &root_key).expect("should work");
+        assert!(cs.is_none(), "declined collection can sync");
+        assert_eq!(engine.get_num_resets(), 0);
+    }
+}
diff --git a/third_party/rust/sync15/src/client/coll_update.rs b/third_party/rust/sync15/src/client/coll_update.rs
new file mode 100644
index 0000000000..d09b1c2eb4
--- /dev/null
+++ b/third_party/rust/sync15/src/client/coll_update.rs
@@ -0,0 +1,127 @@
+/* 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/. */
+
+use super::{
+    request::{NormalResponseHandler, UploadInfo},
+    CollState, Sync15ClientResponse, Sync15StorageClient,
+};
+use crate::bso::OutgoingEncryptedBso;
+use crate::engine::{CollectionRequest, IncomingChangeset, OutgoingChangeset};
+use crate::error::{self, Error, Result};
+use crate::{CollectionName, KeyBundle, ServerTimestamp};
+
+pub fn encrypt_outgoing(
+    o: OutgoingChangeset,
+    key: &KeyBundle,
+) -> Result<Vec<OutgoingEncryptedBso>> {
+    o.changes
+        .into_iter()
+        .map(|change| change.into_encrypted(key))
+        .collect()
+}
+
+pub fn fetch_incoming(
+    client: &Sync15StorageClient,
+    state: &CollState,
+    collection_request: CollectionRequest,
+) -> Result<IncomingChangeset> {
+    let collection = collection_request.collection.clone();
+    let (records, timestamp) = match client.get_encrypted_records(collection_request)? {
+        Sync15ClientResponse::Success {
+            record,
+            last_modified,
+            ..
+        } => (record, last_modified),
+        other => return Err(other.create_storage_error()),
+    };
+    let mut result = IncomingChangeset::new(collection, timestamp);
+    result.changes.reserve(records.len());
+    for record in records {
+        // if we see a HMAC error, we've made an explicit decision to
+        // NOT handle it here, but restart the global state machine.
+        // That should cause us to re-read crypto/keys and things should
+        // work (although if for some reason crypto/keys was updated but
+        // not all storage was wiped we are probably screwed.)
+        result.changes.push(record.into_decrypted(&state.key)?);
+    }
+    Ok(result)
+}
+
+pub struct CollectionUpdate<'a> {
+    client: &'a Sync15StorageClient,
+    state: &'a CollState,
+    collection: CollectionName,
+    xius: ServerTimestamp,
+    to_update: Vec<OutgoingEncryptedBso>,
+    fully_atomic: bool,
+}
+
+impl<'a> CollectionUpdate<'a> {
+    pub fn new(
+        client: &'a Sync15StorageClient,
+        state: &'a CollState,
+        collection: CollectionName,
+        xius: ServerTimestamp,
+        records: Vec<OutgoingEncryptedBso>,
+        fully_atomic: bool,
+    ) -> CollectionUpdate<'a> {
+        CollectionUpdate {
+            client,
+            state,
+            collection,
+            xius,
+            to_update: records,
+            fully_atomic,
+        }
+    }
+
+    pub fn new_from_changeset(
+        client: &'a Sync15StorageClient,
+        state: &'a CollState,
+        changeset: OutgoingChangeset,
+        fully_atomic: bool,
+    ) -> Result<CollectionUpdate<'a>> {
+        let collection = changeset.collection.clone();
+        let to_update = encrypt_outgoing(changeset, &state.key)?;
+        Ok(CollectionUpdate::new(
+            client,
+            state,
+            collection,
+            state.last_modified,
+            to_update,
+            fully_atomic,
+        ))
+    }
+
+    /// Returns a list of the IDs that failed if allowed_dropped_records is true, otherwise
+    /// returns an empty vec.
+    pub fn upload(self) -> error::Result<UploadInfo> {
+        let mut failed = vec![];
+        let mut q = self.client.new_post_queue(
+            &self.collection,
+            &self.state.config,
+            self.xius,
+            NormalResponseHandler::new(!self.fully_atomic),
+        )?;
+
+        for record in self.to_update.into_iter() {
+            let enqueued = q.enqueue(&record)?;
+            if !enqueued && self.fully_atomic {
+                return Err(Error::RecordTooLargeError);
+            }
+        }
+
+        q.flush(true)?;
+        let mut info = q.completed_upload_info();
+        info.failed_ids.append(&mut failed);
+        if self.fully_atomic {
+            assert_eq!(
+                info.failed_ids.len(),
+                0,
+                "Bug: Should have failed by now if we aren't allowing dropped records"
+            );
+        }
+        Ok(info)
+    }
+}
diff --git a/third_party/rust/sync15/src/client/collection_keys.rs b/third_party/rust/sync15/src/client/collection_keys.rs
new file mode 100644
index 0000000000..f51894f756
--- /dev/null
+++ b/third_party/rust/sync15/src/client/collection_keys.rs
@@ -0,0 +1,61 @@
+/* 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/. */
+
+use crate::error::Result;
+use crate::record_types::CryptoKeysRecord;
+use crate::{EncryptedPayload, KeyBundle, ServerTimestamp};
+use std::collections::HashMap;
+
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub struct CollectionKeys {
+    pub timestamp: ServerTimestamp,
+    pub default: KeyBundle,
+    pub collections: HashMap<String, KeyBundle>,
+}
+
+impl CollectionKeys {
+    pub fn new_random() -> Result<CollectionKeys> {
+        let default = KeyBundle::new_random()?;
+        Ok(CollectionKeys {
+            timestamp: ServerTimestamp(0),
+            default,
+            collections: HashMap::new(),
+        })
+    }
+
+    pub fn from_encrypted_payload(
+        record: EncryptedPayload,
+        timestamp: ServerTimestamp,
+        root_key: &KeyBundle,
+    ) -> Result<CollectionKeys> {
+        let keys: CryptoKeysRecord = record.decrypt_into(root_key)?;
+        Ok(CollectionKeys {
+            timestamp,
+            default: KeyBundle::from_base64(&keys.default[0], &keys.default[1])?,
+            collections: keys
+                .collections
+                .into_iter()
+                .map(|kv| Ok((kv.0, KeyBundle::from_base64(&kv.1[0], &kv.1[1])?)))
+                .collect::<Result<HashMap<String, KeyBundle>>>()?,
+        })
+    }
+
+    pub fn to_encrypted_payload(&self, root_key: &KeyBundle) -> Result<EncryptedPayload> {
+        let record = CryptoKeysRecord {
+            id: "keys".into(),
+            collection: "crypto".into(),
+            default: self.default.to_b64_array(),
+            collections: self
+                .collections
+                .iter()
+                .map(|kv| (kv.0.clone(), kv.1.to_b64_array()))
+                .collect(),
+        };
+        EncryptedPayload::from_cleartext_payload(root_key, &record)
+    }
+
+    pub fn key_for_collection<'a>(&'a self, collection: &str) -> &'a KeyBundle {
+        self.collections.get(collection).unwrap_or(&self.default)
+    }
+}
diff --git a/third_party/rust/sync15/src/client/mod.rs b/third_party/rust/sync15/src/client/mod.rs
new file mode 100644
index 0000000000..c49486bfe6
--- /dev/null
+++ b/third_party/rust/sync15/src/client/mod.rs
@@ -0,0 +1,39 @@
+/* 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/. */
+
+//! A module for everything needed to be a "sync client" - ie, a device which
+//! can perform a full sync of any number of collections, including managing
+//! the server state.
+//!
+//! In general, the client is responsible for all communication with the sync server,
+//! including ensuring the state is correct, and encrypting/decrypting all records
+//! to and from the server. However, the actual syncing of the collections is
+//! delegated to an external [crate::engine](Sync Engine).
+//!
+//! One exception is that the "sync client" owns one sync engine - the
+//! [crate::clients_engine], which is managed internally.
+mod coll_state;
+mod coll_update;
+mod collection_keys;
+mod request;
+mod state;
+mod status;
+mod storage_client;
+mod sync;
+mod sync_multiple;
+mod token;
+mod util;
+
+pub(crate) use coll_state::{CollState, LocalCollStateMachine};
+pub(crate) use coll_update::{fetch_incoming, CollectionUpdate};
+pub(crate) use collection_keys::CollectionKeys;
+pub(crate) use request::InfoConfiguration;
+pub(crate) use state::GlobalState;
+pub use status::{ServiceStatus, SyncResult};
+pub use storage_client::{
+    SetupStorageClient, Sync15ClientResponse, Sync15StorageClient, Sync15StorageClientInit,
+};
+pub use sync_multiple::{
+    sync_multiple, sync_multiple_with_command_processor, MemoryCachedState, SyncRequestInfo,
+};
diff --git a/third_party/rust/sync15/src/client/request.rs b/third_party/rust/sync15/src/client/request.rs
new file mode 100644
index 0000000000..c69b630c8d
--- /dev/null
+++ b/third_party/rust/sync15/src/client/request.rs
@@ -0,0 +1,1199 @@
+/* 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/. */
+
+use super::storage_client::Sync15ClientResponse;
+use crate::bso::OutgoingEncryptedBso;
+use crate::error::{self, Error as ErrorKind, Result};
+use crate::ServerTimestamp;
+use serde_derive::*;
+use std::collections::HashMap;
+use std::default::Default;
+use std::ops::Deref;
+use sync_guid::Guid;
+use viaduct::status_codes;
+
+/// Manages a pair of (byte, count) limits for a PostQueue, such as
+/// (max_post_bytes, max_post_records) or (max_total_bytes, max_total_records).
+#[derive(Debug, Clone)]
+struct LimitTracker {
+    max_bytes: usize,
+    max_records: usize,
+    cur_bytes: usize,
+    cur_records: usize,
+}
+
+impl LimitTracker {
+    pub fn new(max_bytes: usize, max_records: usize) -> LimitTracker {
+        LimitTracker {
+            max_bytes,
+            max_records,
+            cur_bytes: 0,
+            cur_records: 0,
+        }
+    }
+
+    pub fn clear(&mut self) {
+        self.cur_records = 0;
+        self.cur_bytes = 0;
+    }
+
+    pub fn can_add_record(&self, payload_size: usize) -> bool {
+        // Desktop does the cur_bytes check as exclusive, but we shouldn't see any servers that
+        // don't have https://github.com/mozilla-services/server-syncstorage/issues/73
+        self.cur_records < self.max_records && self.cur_bytes + payload_size <= self.max_bytes
+    }
+
+    pub fn can_never_add(&self, record_size: usize) -> bool {
+        record_size >= self.max_bytes
+    }
+
+    pub fn record_added(&mut self, record_size: usize) {
+        assert!(
+            self.can_add_record(record_size),
+            "LimitTracker::record_added caller must check can_add_record"
+        );
+        self.cur_records += 1;
+        self.cur_bytes += record_size;
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone)]
+pub struct InfoConfiguration {
+    /// The maximum size in bytes of the overall HTTP request body that will be accepted by the
+    /// server.
+    #[serde(default = "default_max_request_bytes")]
+    pub max_request_bytes: usize,
+
+    /// The maximum number of records that can be uploaded to a collection in a single POST request.
+    #[serde(default = "usize::max_value")]
+    pub max_post_records: usize,
+
+    /// The maximum combined size in bytes of the record payloads that can be uploaded to a
+    /// collection in a single POST request.
+    #[serde(default = "usize::max_value")]
+    pub max_post_bytes: usize,
+
+    /// The maximum total number of records that can be uploaded to a collection as part of a
+    /// batched upload.
+    #[serde(default = "usize::max_value")]
+    pub max_total_records: usize,
+
+    /// The maximum total combined size in bytes of the record payloads that can be uploaded to a
+    /// collection as part of a batched upload.
+    #[serde(default = "usize::max_value")]
+    pub max_total_bytes: usize,
+
+    /// The maximum size of an individual BSO payload, in bytes.
+    #[serde(default = "default_max_record_payload_bytes")]
+    pub max_record_payload_bytes: usize,
+}
+
+// This is annoying but seems to be the only way to do it...
+fn default_max_request_bytes() -> usize {
+    260 * 1024
+}
+fn default_max_record_payload_bytes() -> usize {
+    256 * 1024
+}
+
+impl Default for InfoConfiguration {
+    #[inline]
+    fn default() -> InfoConfiguration {
+        InfoConfiguration {
+            max_request_bytes: default_max_request_bytes(),
+            max_record_payload_bytes: default_max_record_payload_bytes(),
+            max_post_records: usize::max_value(),
+            max_post_bytes: usize::max_value(),
+            max_total_records: usize::max_value(),
+            max_total_bytes: usize::max_value(),
+        }
+    }
+}
+
+#[derive(Clone, Debug, Default, Deserialize, Serialize)]
+pub struct InfoCollections(pub(crate) HashMap<String, ServerTimestamp>);
+
+impl InfoCollections {
+    pub fn new(collections: HashMap<String, ServerTimestamp>) -> InfoCollections {
+        InfoCollections(collections)
+    }
+}
+
+impl Deref for InfoCollections {
+    type Target = HashMap<String, ServerTimestamp>;
+
+    fn deref(&self) -> &HashMap<String, ServerTimestamp> {
+        &self.0
+    }
+}
+
+#[derive(Debug, Clone, Deserialize)]
+pub struct UploadResult {
+    batch: Option<String>,
+    /// Maps record id => why failed
+    #[serde(default = "HashMap::new")]
+    pub failed: HashMap<Guid, String>,
+    /// Vec of ids
+    #[serde(default = "Vec::new")]
+    pub success: Vec<Guid>,
+}
+
+pub type PostResponse = Sync15ClientResponse<UploadResult>;
+
+#[derive(Debug, Clone, Eq, PartialEq, Ord, PartialOrd, Hash)]
+pub enum BatchState {
+    Unsupported,
+    NoBatch,
+    InBatch(String),
+}
+
+#[derive(Debug)]
+pub struct PostQueue<Post, OnResponse> {
+    poster: Post,
+    on_response: OnResponse,
+    post_limits: LimitTracker,
+    batch_limits: LimitTracker,
+    max_payload_bytes: usize,
+    max_request_bytes: usize,
+    queued: Vec<u8>,
+    batch: BatchState,
+    last_modified: ServerTimestamp,
+}
+
+pub trait BatchPoster {
+    /// Note: Last argument (reference to the batch poster) is provided for the purposes of testing
+    /// Important: Poster should not report non-success HTTP statuses as errors!!
+    fn post<P, O>(
+        &self,
+        body: Vec<u8>,
+        xius: ServerTimestamp,
+        batch: Option<String>,
+        commit: bool,
+        queue: &PostQueue<P, O>,
+    ) -> Result<PostResponse>;
+}
+
+// We don't just use a FnMut here since we want to override it in mocking for RefCell<TestType>,
+// which we can't do for FnMut since neither FnMut nor RefCell are defined here. Also, this
+// is somewhat better for documentation.
+pub trait PostResponseHandler {
+    fn handle_response(&mut self, r: PostResponse, mid_batch: bool) -> Result<()>;
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct NormalResponseHandler {
+    pub failed_ids: Vec<Guid>,
+    pub successful_ids: Vec<Guid>,
+    pub allow_failed: bool,
+    pub pending_failed: Vec<Guid>,
+    pub pending_success: Vec<Guid>,
+}
+
+impl NormalResponseHandler {
+    pub fn new(allow_failed: bool) -> NormalResponseHandler {
+        NormalResponseHandler {
+            failed_ids: vec![],
+            successful_ids: vec![],
+            pending_failed: vec![],
+            pending_success: vec![],
+            allow_failed,
+        }
+    }
+}
+
+impl PostResponseHandler for NormalResponseHandler {
+    fn handle_response(&mut self, r: PostResponse, mid_batch: bool) -> error::Result<()> {
+        match r {
+            Sync15ClientResponse::Success { record, .. } => {
+                if !record.failed.is_empty() && !self.allow_failed {
+                    return Err(ErrorKind::RecordUploadFailed);
+                }
+                for id in record.success.iter() {
+                    self.pending_success.push(id.clone());
+                }
+                for kv in record.failed.iter() {
+                    self.pending_failed.push(kv.0.clone());
+                }
+                if !mid_batch {
+                    self.successful_ids.append(&mut self.pending_success);
+                    self.failed_ids.append(&mut self.pending_failed);
+                }
+                Ok(())
+            }
+            _ => Err(r.create_storage_error()),
+        }
+    }
+}
+
+impl<Poster, OnResponse> PostQueue<Poster, OnResponse>
+where
+    Poster: BatchPoster,
+    OnResponse: PostResponseHandler,
+{
+    pub fn new(
+        config: &InfoConfiguration,
+        ts: ServerTimestamp,
+        poster: Poster,
+        on_response: OnResponse,
+    ) -> PostQueue<Poster, OnResponse> {
+        PostQueue {
+            poster,
+            on_response,
+            last_modified: ts,
+            post_limits: LimitTracker::new(config.max_post_bytes, config.max_post_records),
+            batch_limits: LimitTracker::new(config.max_total_bytes, config.max_total_records),
+            batch: BatchState::NoBatch,
+            max_payload_bytes: config.max_record_payload_bytes,
+            max_request_bytes: config.max_request_bytes,
+            queued: Vec::new(),
+        }
+    }
+
+    #[inline]
+    fn in_batch(&self) -> bool {
+        !matches!(&self.batch, BatchState::Unsupported | BatchState::NoBatch)
+    }
+
+    pub fn enqueue(&mut self, record: &OutgoingEncryptedBso) -> Result<bool> {
+        let payload_length = record.serialized_payload_len();
+
+        if self.post_limits.can_never_add(payload_length)
+            || self.batch_limits.can_never_add(payload_length)
+            || payload_length >= self.max_payload_bytes
+        {
+            log::warn!(
+                "Single record too large to submit to server ({} b)",
+                payload_length
+            );
+            return Ok(false);
+        }
+
+        // Write directly into `queued` but undo if necessary (the vast majority of the time
+        // it won't be necessary). If we hit a problem we need to undo that, but the only error
+        // case we have to worry about right now is in flush()
+        let item_start = self.queued.len();
+
+        // This is conservative but can't hurt.
+        self.queued.reserve(payload_length + 2);
+
+        // Either the first character in an array, or a comma separating
+        // it from the previous item.
+        let c = if self.queued.is_empty() { b'[' } else { b',' };
+        self.queued.push(c);
+
+        // This unwrap is fine, since serde_json's failure case is HashMaps that have non-object
+        // keys, which is impossible. If you decide to change this part, you *need* to call
+        // `self.queued.truncate(item_start)` here in the failure case!
+        serde_json::to_writer(&mut self.queued, &record).unwrap();
+
+        let item_end = self.queued.len();
+
+        debug_assert!(
+            item_end >= payload_length,
+            "EncryptedPayload::serialized_len is bugged"
+        );
+
+        // The + 1 is only relevant for the final record, which will have a trailing ']'.
+        let item_len = item_end - item_start + 1;
+
+        if item_len >= self.max_request_bytes {
+            self.queued.truncate(item_start);
+            log::warn!(
+                "Single record too large to submit to server ({} b)",
+                item_len
+            );
+            return Ok(false);
+        }
+
+        let can_post_record = self.post_limits.can_add_record(payload_length);
+        let can_batch_record = self.batch_limits.can_add_record(payload_length);
+        let can_send_record = self.queued.len() < self.max_request_bytes;
+
+        if !can_post_record || !can_send_record || !can_batch_record {
+            log::debug!(
+                "PostQueue flushing! (can_post = {}, can_send = {}, can_batch = {})",
+                can_post_record,
+                can_send_record,
+                can_batch_record
+            );
+            // "unwrite" the record.
+            self.queued.truncate(item_start);
+            // Flush whatever we have queued.
+            self.flush(!can_batch_record)?;
+            // And write it again.
+            let c = if self.queued.is_empty() { b'[' } else { b',' };
+            self.queued.push(c);
+            serde_json::to_writer(&mut self.queued, &record).unwrap();
+        }
+
+        self.post_limits.record_added(payload_length);
+        self.batch_limits.record_added(payload_length);
+
+        Ok(true)
+    }
+
+    pub fn flush(&mut self, want_commit: bool) -> Result<()> {
+        if self.queued.is_empty() {
+            assert!(
+                !self.in_batch(),
+                "Bug: Somehow we're in a batch but have no queued records"
+            );
+            // Nothing to do!
+            return Ok(());
+        }
+
+        self.queued.push(b']');
+        let batch_id = match &self.batch {
+            // Not the first post and we know we have no batch semantics.
+            BatchState::Unsupported => None,
+            // First commit in possible batch
+            BatchState::NoBatch => Some("true".into()),
+            // In a batch and we have a batch id.
+            BatchState::InBatch(ref s) => Some(s.clone()),
+        };
+
+        log::info!(
+            "Posting {} records of {} bytes",
+            self.post_limits.cur_records,
+            self.queued.len()
+        );
+
+        let is_commit = want_commit && batch_id.is_some();
+        // Weird syntax for calling a function object that is a property.
+        let resp_or_error = self.poster.post(
+            self.queued.clone(),
+            self.last_modified,
+            batch_id,
+            is_commit,
+            self,
+        );
+
+        self.queued.truncate(0);
+
+        if want_commit || self.batch == BatchState::Unsupported {
+            self.batch_limits.clear();
+        }
+        self.post_limits.clear();
+
+        let resp = resp_or_error?;
+
+        let (status, last_modified, record) = match resp {
+            Sync15ClientResponse::Success {
+                status,
+                last_modified,
+                ref record,
+                ..
+            } => (status, last_modified, record),
+            _ => {
+                self.on_response.handle_response(resp, !want_commit)?;
+                // on_response() should always fail!
+                unreachable!();
+            }
+        };
+
+        if want_commit || self.batch == BatchState::Unsupported {
+            self.last_modified = last_modified;
+        }
+
+        if want_commit {
+            log::debug!("Committed batch {:?}", self.batch);
+            self.batch = BatchState::NoBatch;
+            self.on_response.handle_response(resp, false)?;
+            return Ok(());
+        }
+
+        if status != status_codes::ACCEPTED {
+            if self.in_batch() {
+                return Err(ErrorKind::ServerBatchProblem(
+                    "Server responded non-202 success code while a batch was in progress",
+                ));
+            }
+            self.last_modified = last_modified;
+            self.batch = BatchState::Unsupported;
+            self.batch_limits.clear();
+            self.on_response.handle_response(resp, false)?;
+            return Ok(());
+        }
+
+        let batch_id = record
+            .batch
+            .as_ref()
+            .ok_or({
+                ErrorKind::ServerBatchProblem("Invalid server response: 202 without a batch ID")
+            })?
+            .clone();
+
+        match &self.batch {
+            BatchState::Unsupported => {
+                log::warn!("Server changed its mind about supporting batching mid-batch...");
+            }
+
+            BatchState::InBatch(ref cur_id) => {
+                if cur_id != &batch_id {
+                    return Err(ErrorKind::ServerBatchProblem(
+                        "Invalid server response: 202 without a batch ID",
+                    ));
+                }
+            }
+            _ => {}
+        }
+
+        // Can't change this in match arms without NLL
+        self.batch = BatchState::InBatch(batch_id);
+        self.last_modified = last_modified;
+
+        self.on_response.handle_response(resp, true)?;
+
+        Ok(())
+    }
+}
+
+#[derive(Clone)]
+pub struct UploadInfo {
+    pub successful_ids: Vec<Guid>,
+    pub failed_ids: Vec<Guid>,
+    pub modified_timestamp: ServerTimestamp,
+}
+
+impl<Poster> PostQueue<Poster, NormalResponseHandler> {
+    // TODO: should take by move
+    pub fn completed_upload_info(&mut self) -> UploadInfo {
+        let mut result = UploadInfo {
+            successful_ids: Vec::with_capacity(self.on_response.successful_ids.len()),
+            failed_ids: Vec::with_capacity(
+                self.on_response.failed_ids.len()
+                    + self.on_response.pending_failed.len()
+                    + self.on_response.pending_success.len(),
+            ),
+            modified_timestamp: self.last_modified,
+        };
+
+        result
+            .successful_ids
+            .append(&mut self.on_response.successful_ids);
+
+        result.failed_ids.append(&mut self.on_response.failed_ids);
+        result
+            .failed_ids
+            .append(&mut self.on_response.pending_failed);
+        result
+            .failed_ids
+            .append(&mut self.on_response.pending_success);
+
+        result
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::bso::{IncomingEncryptedBso, OutgoingEncryptedBso, OutgoingEnvelope};
+    use crate::EncryptedPayload;
+    use lazy_static::lazy_static;
+    use std::cell::RefCell;
+    use std::collections::VecDeque;
+    use std::rc::Rc;
+
+    #[derive(Debug, Clone)]
+    struct PostedData {
+        body: String,
+        _xius: ServerTimestamp,
+        batch: Option<String>,
+        commit: bool,
+        payload_bytes: usize,
+        records: usize,
+    }
+
+    impl PostedData {
+        fn records_as_json(&self) -> Vec<serde_json::Value> {
+            let values =
+                serde_json::from_str::<serde_json::Value>(&self.body).expect("Posted invalid json");
+            // Check that they actually deserialize as what we want
+            let records_or_err =
+                serde_json::from_value::<Vec<IncomingEncryptedBso>>(values.clone());
+            records_or_err.expect("Failed to deserialize data");
+            serde_json::from_value(values).unwrap()
+        }
+    }
+
+    #[derive(Debug, Clone)]
+    struct BatchInfo {
+        id: Option<String>,
+        posts: Vec<PostedData>,
+        bytes: usize,
+        records: usize,
+    }
+
+    #[derive(Debug, Clone)]
+    struct TestPoster {
+        all_posts: Vec<PostedData>,
+        responses: VecDeque<PostResponse>,
+        batches: Vec<BatchInfo>,
+        cur_batch: Option<BatchInfo>,
+        cfg: InfoConfiguration,
+    }
+
+    type TestPosterRef = Rc<RefCell<TestPoster>>;
+    impl TestPoster {
+        pub fn new<T>(cfg: &InfoConfiguration, responses: T) -> TestPosterRef
+        where
+            T: Into<VecDeque<PostResponse>>,
+        {
+            Rc::new(RefCell::new(TestPoster {
+                all_posts: vec![],
+                responses: responses.into(),
+                batches: vec![],
+                cur_batch: None,
+                cfg: cfg.clone(),
+            }))
+        }
+        // Adds &mut
+        fn do_post<T, O>(
+            &mut self,
+            body: &[u8],
+            xius: ServerTimestamp,
+            batch: Option<String>,
+            commit: bool,
+            queue: &PostQueue<T, O>,
+        ) -> Sync15ClientResponse<UploadResult> {
+            let mut post = PostedData {
+                body: String::from_utf8(body.into()).expect("Posted invalid utf8..."),
+                batch: batch.clone(),
+                _xius: xius,
+                commit,
+                payload_bytes: 0,
+                records: 0,
+            };
+
+            assert!(body.len() <= self.cfg.max_request_bytes);
+
+            let (num_records, record_payload_bytes) = {
+                let recs = post.records_as_json();
+                assert!(recs.len() <= self.cfg.max_post_records);
+                assert!(recs.len() <= self.cfg.max_total_records);
+                let payload_bytes: usize = recs
+                    .iter()
+                    .map(|r| {
+                        let len = r["payload"]
+                            .as_str()
+                            .expect("Non string payload property")
+                            .len();
+                        assert!(len <= self.cfg.max_record_payload_bytes);
+                        len
+                    })
+                    .sum();
+                assert!(payload_bytes <= self.cfg.max_post_bytes);
+                assert!(payload_bytes <= self.cfg.max_total_bytes);
+
+                assert_eq!(queue.post_limits.cur_bytes, payload_bytes);
+                assert_eq!(queue.post_limits.cur_records, recs.len());
+                (recs.len(), payload_bytes)
+            };
+            post.payload_bytes = record_payload_bytes;
+            post.records = num_records;
+
+            self.all_posts.push(post.clone());
+            let response = self.responses.pop_front().unwrap();
+
+            let record = match response {
+                Sync15ClientResponse::Success { ref record, .. } => record,
+                _ => {
+                    panic!("only success codes are used in this test");
+                }
+            };
+
+            if self.cur_batch.is_none() {
+                assert!(
+                    batch.is_none() || batch == Some("true".into()),
+                    "We shouldn't be in a batch now"
+                );
+                self.cur_batch = Some(BatchInfo {
+                    id: record.batch.clone(),
+                    posts: vec![],
+                    records: 0,
+                    bytes: 0,
+                });
+            } else {
+                assert_eq!(
+                    batch,
+                    self.cur_batch.as_ref().unwrap().id,
+                    "We're in a batch but got the wrong batch id"
+                );
+            }
+
+            {
+                let batch = self.cur_batch.as_mut().unwrap();
+                batch.posts.push(post);
+                batch.records += num_records;
+                batch.bytes += record_payload_bytes;
+
+                assert!(batch.bytes <= self.cfg.max_total_bytes);
+                assert!(batch.records <= self.cfg.max_total_records);
+
+                assert_eq!(batch.records, queue.batch_limits.cur_records);
+                assert_eq!(batch.bytes, queue.batch_limits.cur_bytes);
+            }
+
+            if commit || record.batch.is_none() {
+                let batch = self.cur_batch.take().unwrap();
+                self.batches.push(batch);
+            }
+
+            response
+        }
+
+        fn do_handle_response(&mut self, _: PostResponse, mid_batch: bool) {
+            assert_eq!(mid_batch, self.cur_batch.is_some());
+        }
+    }
+    impl BatchPoster for TestPosterRef {
+        fn post<T, O>(
+            &self,
+            body: Vec<u8>,
+            xius: ServerTimestamp,
+            batch: Option<String>,
+            commit: bool,
+            queue: &PostQueue<T, O>,
+        ) -> Result<PostResponse> {
+            Ok(self.borrow_mut().do_post(&body, xius, batch, commit, queue))
+        }
+    }
+
+    impl PostResponseHandler for TestPosterRef {
+        fn handle_response(&mut self, r: PostResponse, mid_batch: bool) -> Result<()> {
+            self.borrow_mut().do_handle_response(r, mid_batch);
+            Ok(())
+        }
+    }
+
+    type MockedPostQueue = PostQueue<TestPosterRef, TestPosterRef>;
+
+    fn pq_test_setup(
+        cfg: InfoConfiguration,
+        lm: i64,
+        resps: Vec<PostResponse>,
+    ) -> (MockedPostQueue, TestPosterRef) {
+        let tester = TestPoster::new(&cfg, resps);
+        let pq = PostQueue::new(&cfg, ServerTimestamp(lm), tester.clone(), tester.clone());
+        (pq, tester)
+    }
+
+    fn fake_response<'a, T: Into<Option<&'a str>>>(status: u16, lm: i64, batch: T) -> PostResponse {
+        assert!(status_codes::is_success_code(status));
+        Sync15ClientResponse::Success {
+            status,
+            last_modified: ServerTimestamp(lm),
+            record: UploadResult {
+                batch: batch.into().map(Into::into),
+                failed: HashMap::new(),
+                success: vec![],
+            },
+            route: "test/path".into(),
+        }
+    }
+
+    lazy_static! {
+        // ~40b
+        static ref PAYLOAD_OVERHEAD: usize = {
+            let payload = EncryptedPayload {
+                iv: "".into(),
+                hmac: "".into(),
+                ciphertext: "".into()
+            };
+            serde_json::to_string(&payload).unwrap().len()
+        };
+        // ~80b
+        static ref TOTAL_RECORD_OVERHEAD: usize = {
+            let val = serde_json::to_value(OutgoingEncryptedBso::new(OutgoingEnvelope {
+                    id: "".into(),
+                    sortindex: None,
+                    ttl: None,
+                },
+                EncryptedPayload {
+                    iv: "".into(),
+                    hmac: "".into(),
+                    ciphertext: "".into()
+                },
+            )).unwrap();
+            serde_json::to_string(&val).unwrap().len()
+        };
+        // There's some subtlety in how we calulate this having to do with the fact that
+        // the quotes in the payload are escaped but the escape chars count to the request len
+        // and *not* to the payload len (the payload len check happens after json parsing the
+        // top level object).
+        static ref NON_PAYLOAD_OVERHEAD: usize = {
+            *TOTAL_RECORD_OVERHEAD - *PAYLOAD_OVERHEAD
+        };
+    }
+
+    // Actual record size (for max_request_len) will be larger by some amount
+    fn make_record(payload_size: usize) -> OutgoingEncryptedBso {
+        assert!(payload_size > *PAYLOAD_OVERHEAD);
+        let ciphertext_len = payload_size - *PAYLOAD_OVERHEAD;
+        OutgoingEncryptedBso::new(
+            OutgoingEnvelope {
+                id: "".into(),
+                sortindex: None,
+                ttl: None,
+            },
+            EncryptedPayload {
+                iv: "".into(),
+                hmac: "".into(),
+                ciphertext: "x".repeat(ciphertext_len),
+            },
+        )
+    }
+
+    fn request_bytes_for_payloads(payloads: &[usize]) -> usize {
+        1 + payloads
+            .iter()
+            .map(|&size| size + 1 + *NON_PAYLOAD_OVERHEAD)
+            .sum::<usize>()
+    }
+
+    #[test]
+    fn test_pq_basic() {
+        let cfg = InfoConfiguration {
+            max_request_bytes: 1000,
+            max_record_payload_bytes: 1000,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![fake_response(status_codes::OK, time + 100_000, None)],
+        );
+
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.flush(true).unwrap();
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 1);
+        assert_eq!(t.batches.len(), 1);
+        assert_eq!(t.batches[0].posts.len(), 1);
+        assert_eq!(t.batches[0].records, 1);
+        assert_eq!(t.batches[0].bytes, 100);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[100])
+        );
+    }
+
+    #[test]
+    fn test_pq_max_request_bytes_no_batch() {
+        let cfg = InfoConfiguration {
+            max_request_bytes: 250,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![
+                fake_response(status_codes::OK, time + 100_000, None),
+                fake_response(status_codes::OK, time + 200_000, None),
+            ],
+        );
+
+        // Note that the total record overhead is around 85 bytes
+        let payload_size = 100 - *NON_PAYLOAD_OVERHEAD;
+        pq.enqueue(&make_record(payload_size)).unwrap(); // total size == 102; [r]
+        pq.enqueue(&make_record(payload_size)).unwrap(); // total size == 203; [r,r]
+        pq.enqueue(&make_record(payload_size)).unwrap(); // too big, 2nd post.
+        pq.flush(true).unwrap();
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 2);
+        assert_eq!(t.batches.len(), 2);
+        assert_eq!(t.batches[0].posts.len(), 1);
+        assert_eq!(t.batches[0].records, 2);
+        assert_eq!(t.batches[0].bytes, payload_size * 2);
+        assert_eq!(t.batches[0].posts[0].batch, Some("true".into()));
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[payload_size, payload_size])
+        );
+
+        assert_eq!(t.batches[1].posts.len(), 1);
+        assert_eq!(t.batches[1].records, 1);
+        assert_eq!(t.batches[1].bytes, payload_size);
+        // We know at this point that the server does not support batching.
+        assert_eq!(t.batches[1].posts[0].batch, None);
+        assert!(!t.batches[1].posts[0].commit);
+        assert_eq!(
+            t.batches[1].posts[0].body.len(),
+            request_bytes_for_payloads(&[payload_size])
+        );
+    }
+
+    #[test]
+    fn test_pq_max_record_payload_bytes_no_batch() {
+        let cfg = InfoConfiguration {
+            max_record_payload_bytes: 150,
+            max_request_bytes: 350,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![
+                fake_response(status_codes::OK, time + 100_000, None),
+                fake_response(status_codes::OK, time + 200_000, None),
+            ],
+        );
+
+        // Note that the total record overhead is around 85 bytes
+        let payload_size = 100 - *NON_PAYLOAD_OVERHEAD;
+        pq.enqueue(&make_record(payload_size)).unwrap(); // total size == 102; [r]
+        let enqueued = pq.enqueue(&make_record(151)).unwrap(); // still 102
+        assert!(!enqueued, "Should not have fit");
+        pq.enqueue(&make_record(payload_size)).unwrap();
+        pq.flush(true).unwrap();
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 1);
+        assert_eq!(t.batches.len(), 1);
+        assert_eq!(t.batches[0].posts.len(), 1);
+        assert_eq!(t.batches[0].records, 2);
+        assert_eq!(t.batches[0].bytes, payload_size * 2);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[payload_size, payload_size])
+        );
+    }
+
+    #[test]
+    fn test_pq_single_batch() {
+        let cfg = InfoConfiguration::default();
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![fake_response(
+                status_codes::ACCEPTED,
+                time + 100_000,
+                Some("1234"),
+            )],
+        );
+
+        let payload_size = 100 - *NON_PAYLOAD_OVERHEAD;
+        pq.enqueue(&make_record(payload_size)).unwrap();
+        pq.enqueue(&make_record(payload_size)).unwrap();
+        pq.enqueue(&make_record(payload_size)).unwrap();
+        pq.flush(true).unwrap();
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 1);
+        assert_eq!(t.batches.len(), 1);
+        assert_eq!(t.batches[0].id.as_ref().unwrap(), "1234");
+        assert_eq!(t.batches[0].posts.len(), 1);
+        assert_eq!(t.batches[0].records, 3);
+        assert_eq!(t.batches[0].bytes, payload_size * 3);
+        assert!(t.batches[0].posts[0].commit);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[payload_size, payload_size, payload_size])
+        );
+    }
+
+    #[test]
+    fn test_pq_multi_post_batch_bytes() {
+        let cfg = InfoConfiguration {
+            max_post_bytes: 200,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![
+                fake_response(status_codes::ACCEPTED, time, Some("1234")),
+                fake_response(status_codes::ACCEPTED, time + 100_000, Some("1234")),
+            ],
+        );
+
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.flush(true).unwrap(); // COMMIT
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 2);
+        assert_eq!(t.batches.len(), 1);
+        assert_eq!(t.batches[0].posts.len(), 2);
+        assert_eq!(t.batches[0].records, 3);
+        assert_eq!(t.batches[0].bytes, 300);
+
+        assert_eq!(t.batches[0].posts[0].batch.as_ref().unwrap(), "true");
+        assert_eq!(t.batches[0].posts[0].records, 2);
+        assert_eq!(t.batches[0].posts[0].payload_bytes, 200);
+        assert!(!t.batches[0].posts[0].commit);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[100, 100])
+        );
+
+        assert_eq!(t.batches[0].posts[1].batch.as_ref().unwrap(), "1234");
+        assert_eq!(t.batches[0].posts[1].records, 1);
+        assert_eq!(t.batches[0].posts[1].payload_bytes, 100);
+        assert!(t.batches[0].posts[1].commit);
+        assert_eq!(
+            t.batches[0].posts[1].body.len(),
+            request_bytes_for_payloads(&[100])
+        );
+    }
+
+    #[test]
+    fn test_pq_multi_post_batch_records() {
+        let cfg = InfoConfiguration {
+            max_post_records: 3,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![
+                fake_response(status_codes::ACCEPTED, time, Some("1234")),
+                fake_response(status_codes::ACCEPTED, time, Some("1234")),
+                fake_response(status_codes::ACCEPTED, time + 100_000, Some("1234")),
+            ],
+        );
+
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.flush(true).unwrap(); // COMMIT
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 3);
+        assert_eq!(t.batches.len(), 1);
+        assert_eq!(t.batches[0].posts.len(), 3);
+        assert_eq!(t.batches[0].records, 7);
+        assert_eq!(t.batches[0].bytes, 700);
+
+        assert_eq!(t.batches[0].posts[0].batch.as_ref().unwrap(), "true");
+        assert_eq!(t.batches[0].posts[0].records, 3);
+        assert_eq!(t.batches[0].posts[0].payload_bytes, 300);
+        assert!(!t.batches[0].posts[0].commit);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[100, 100, 100])
+        );
+
+        assert_eq!(t.batches[0].posts[1].batch.as_ref().unwrap(), "1234");
+        assert_eq!(t.batches[0].posts[1].records, 3);
+        assert_eq!(t.batches[0].posts[1].payload_bytes, 300);
+        assert!(!t.batches[0].posts[1].commit);
+        assert_eq!(
+            t.batches[0].posts[1].body.len(),
+            request_bytes_for_payloads(&[100, 100, 100])
+        );
+
+        assert_eq!(t.batches[0].posts[2].batch.as_ref().unwrap(), "1234");
+        assert_eq!(t.batches[0].posts[2].records, 1);
+        assert_eq!(t.batches[0].posts[2].payload_bytes, 100);
+        assert!(t.batches[0].posts[2].commit);
+        assert_eq!(
+            t.batches[0].posts[2].body.len(),
+            request_bytes_for_payloads(&[100])
+        );
+    }
+
+    #[test]
+    #[allow(clippy::cognitive_complexity)]
+    fn test_pq_multi_post_multi_batch_records() {
+        let cfg = InfoConfiguration {
+            max_post_records: 3,
+            max_total_records: 5,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![
+                fake_response(status_codes::ACCEPTED, time, Some("1234")),
+                fake_response(status_codes::ACCEPTED, time + 100_000, Some("1234")),
+                fake_response(status_codes::ACCEPTED, time + 100_000, Some("abcd")),
+                fake_response(status_codes::ACCEPTED, time + 200_000, Some("abcd")),
+            ],
+        );
+
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST + COMMIT
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.flush(true).unwrap(); // COMMIT
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 4);
+        assert_eq!(t.batches.len(), 2);
+        assert_eq!(t.batches[0].posts.len(), 2);
+        assert_eq!(t.batches[1].posts.len(), 2);
+
+        assert_eq!(t.batches[0].records, 5);
+        assert_eq!(t.batches[1].records, 4);
+
+        assert_eq!(t.batches[0].bytes, 500);
+        assert_eq!(t.batches[1].bytes, 400);
+
+        assert_eq!(t.batches[0].posts[0].batch.as_ref().unwrap(), "true");
+        assert_eq!(t.batches[0].posts[0].records, 3);
+        assert_eq!(t.batches[0].posts[0].payload_bytes, 300);
+        assert!(!t.batches[0].posts[0].commit);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[100, 100, 100])
+        );
+
+        assert_eq!(t.batches[0].posts[1].batch.as_ref().unwrap(), "1234");
+        assert_eq!(t.batches[0].posts[1].records, 2);
+        assert_eq!(t.batches[0].posts[1].payload_bytes, 200);
+        assert!(t.batches[0].posts[1].commit);
+        assert_eq!(
+            t.batches[0].posts[1].body.len(),
+            request_bytes_for_payloads(&[100, 100])
+        );
+
+        assert_eq!(t.batches[1].posts[0].batch.as_ref().unwrap(), "true");
+        assert_eq!(t.batches[1].posts[0].records, 3);
+        assert_eq!(t.batches[1].posts[0].payload_bytes, 300);
+        assert!(!t.batches[1].posts[0].commit);
+        assert_eq!(
+            t.batches[1].posts[0].body.len(),
+            request_bytes_for_payloads(&[100, 100, 100])
+        );
+
+        assert_eq!(t.batches[1].posts[1].batch.as_ref().unwrap(), "abcd");
+        assert_eq!(t.batches[1].posts[1].records, 1);
+        assert_eq!(t.batches[1].posts[1].payload_bytes, 100);
+        assert!(t.batches[1].posts[1].commit);
+        assert_eq!(
+            t.batches[1].posts[1].body.len(),
+            request_bytes_for_payloads(&[100])
+        );
+    }
+
+    #[test]
+    #[allow(clippy::cognitive_complexity)]
+    fn test_pq_multi_post_multi_batch_bytes() {
+        let cfg = InfoConfiguration {
+            max_post_bytes: 300,
+            max_total_bytes: 500,
+            ..InfoConfiguration::default()
+        };
+        let time = 11_111_111_000;
+        let (mut pq, tester) = pq_test_setup(
+            cfg,
+            time,
+            vec![
+                fake_response(status_codes::ACCEPTED, time, Some("1234")),
+                fake_response(status_codes::ACCEPTED, time + 100_000, Some("1234")), // should commit
+                fake_response(status_codes::ACCEPTED, time + 100_000, Some("abcd")),
+                fake_response(status_codes::ACCEPTED, time + 200_000, Some("abcd")), // should commit
+            ],
+        );
+
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        assert_eq!(pq.last_modified.0, time);
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+        // POST + COMMIT
+        pq.enqueue(&make_record(100)).unwrap();
+        assert_eq!(pq.last_modified.0, time + 100_000);
+        pq.enqueue(&make_record(100)).unwrap();
+        pq.enqueue(&make_record(100)).unwrap();
+
+        // POST
+        pq.enqueue(&make_record(100)).unwrap();
+        assert_eq!(pq.last_modified.0, time + 100_000);
+        pq.flush(true).unwrap(); // COMMIT
+
+        assert_eq!(pq.last_modified.0, time + 200_000);
+
+        let t = tester.borrow();
+        assert!(t.cur_batch.is_none());
+        assert_eq!(t.all_posts.len(), 4);
+        assert_eq!(t.batches.len(), 2);
+        assert_eq!(t.batches[0].posts.len(), 2);
+        assert_eq!(t.batches[1].posts.len(), 2);
+
+        assert_eq!(t.batches[0].records, 5);
+        assert_eq!(t.batches[1].records, 4);
+
+        assert_eq!(t.batches[0].bytes, 500);
+        assert_eq!(t.batches[1].bytes, 400);
+
+        assert_eq!(t.batches[0].posts[0].batch.as_ref().unwrap(), "true");
+        assert_eq!(t.batches[0].posts[0].records, 3);
+        assert_eq!(t.batches[0].posts[0].payload_bytes, 300);
+        assert!(!t.batches[0].posts[0].commit);
+        assert_eq!(
+            t.batches[0].posts[0].body.len(),
+            request_bytes_for_payloads(&[100, 100, 100])
+        );
+
+        assert_eq!(t.batches[0].posts[1].batch.as_ref().unwrap(), "1234");
+        assert_eq!(t.batches[0].posts[1].records, 2);
+        assert_eq!(t.batches[0].posts[1].payload_bytes, 200);
+        assert!(t.batches[0].posts[1].commit);
+        assert_eq!(
+            t.batches[0].posts[1].body.len(),
+            request_bytes_for_payloads(&[100, 100])
+        );
+
+        assert_eq!(t.batches[1].posts[0].batch.as_ref().unwrap(), "true");
+        assert_eq!(t.batches[1].posts[0].records, 3);
+        assert_eq!(t.batches[1].posts[0].payload_bytes, 300);
+        assert!(!t.batches[1].posts[0].commit);
+        assert_eq!(
+            t.batches[1].posts[0].body.len(),
+            request_bytes_for_payloads(&[100, 100, 100])
+        );
+
+        assert_eq!(t.batches[1].posts[1].batch.as_ref().unwrap(), "abcd");
+        assert_eq!(t.batches[1].posts[1].records, 1);
+        assert_eq!(t.batches[1].posts[1].payload_bytes, 100);
+        assert!(t.batches[1].posts[1].commit);
+        assert_eq!(
+            t.batches[1].posts[1].body.len(),
+            request_bytes_for_payloads(&[100])
+        );
+    }
+
+    // TODO: Test
+    //
+    // - error cases!!! We don't test our handling of server errors at all!
+    // - mixed bytes/record limits
+    //
+    // A lot of these have good examples in test_postqueue.js on deskftop sync
+}
diff --git a/third_party/rust/sync15/src/client/state.rs b/third_party/rust/sync15/src/client/state.rs
new file mode 100644
index 0000000000..78e9a6a718
--- /dev/null
+++ b/third_party/rust/sync15/src/client/state.rs
@@ -0,0 +1,1089 @@
+/* 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/. */
+
+use std::collections::{HashMap, HashSet};
+
+use super::request::{InfoCollections, InfoConfiguration};
+use super::storage_client::{SetupStorageClient, Sync15ClientResponse};
+use super::CollectionKeys;
+use crate::bso::OutgoingEncryptedBso;
+use crate::error::{self, Error as ErrorKind, ErrorResponse};
+use crate::record_types::{MetaGlobalEngine, MetaGlobalRecord};
+use crate::EncryptedPayload;
+use crate::{Guid, KeyBundle, ServerTimestamp};
+use interrupt_support::Interruptee;
+use serde_derive::*;
+
+use self::SetupState::*;
+
+const STORAGE_VERSION: usize = 5;
+
+/// Maps names to storage versions for engines to include in a fresh
+/// `meta/global` record. We include engines that we don't implement
+/// because they'll be disabled on other clients if we omit them
+/// (bug 1479929).
+const DEFAULT_ENGINES: &[(&str, usize)] = &[
+    ("passwords", 1),
+    ("clients", 1),
+    ("addons", 1),
+    ("addresses", 1),
+    ("bookmarks", 2),
+    ("creditcards", 1),
+    ("forms", 1),
+    ("history", 1),
+    ("prefs", 2),
+    ("tabs", 1),
+];
+
+// Declined engines to include in a fresh `meta/global` record.
+const DEFAULT_DECLINED: &[&str] = &[];
+
+/// State that we require the app to persist to storage for us.
+/// It's a little unfortunate we need this, because it's only tracking
+/// "declined engines", and even then, only needed in practice when there's
+/// no meta/global so we need to create one. It's extra unfortunate because we
+/// want to move away from "globally declined" engines anyway, moving towards
+/// allowing engines to be enabled or disabled per client rather than globally.
+///
+/// Apps are expected to treat this as opaque, so we support serializing it.
+/// Note that this structure is *not* used to *change* the declined engines
+/// list - that will be done in the future, but the API exposed for that
+/// purpose will also take a mutable PersistedGlobalState.
+#[derive(Debug, Serialize, Deserialize)]
+#[serde(tag = "schema_version")]
+pub enum PersistedGlobalState {
+    /// V1 was when we persisted the entire GlobalState, keys and all!
+
+    /// V2 is just tracking the globally declined list.
+    /// None means "I've no idea" and theoretically should only happen on the
+    /// very first sync for an app.
+    V2 { declined: Option<Vec<String>> },
+}
+
+impl Default for PersistedGlobalState {
+    #[inline]
+    fn default() -> PersistedGlobalState {
+        PersistedGlobalState::V2 { declined: None }
+    }
+}
+
+#[derive(Debug, Default, Clone, PartialEq)]
+pub(crate) struct EngineChangesNeeded {
+    pub local_resets: HashSet<String>,
+    pub remote_wipes: HashSet<String>,
+}
+
+#[derive(Debug, Default, Clone, PartialEq)]
+struct RemoteEngineState {
+    info_collections: HashSet<String>,
+    declined: HashSet<String>,
+}
+
+#[derive(Debug, Default, Clone, PartialEq)]
+struct EngineStateInput {
+    local_declined: HashSet<String>,
+    remote: Option<RemoteEngineState>,
+    user_changes: HashMap<String, bool>,
+}
+
+#[derive(Debug, Default, Clone, PartialEq)]
+struct EngineStateOutput {
+    // The new declined.
+    declined: HashSet<String>,
+    // Which engines need resets or wipes.
+    changes_needed: EngineChangesNeeded,
+}
+
+fn compute_engine_states(input: EngineStateInput) -> EngineStateOutput {
+    use super::util::*;
+    log::debug!("compute_engine_states: input {:?}", input);
+    let (must_enable, must_disable) = partition_by_value(&input.user_changes);
+    let have_remote = input.remote.is_some();
+    let RemoteEngineState {
+        info_collections,
+        declined: remote_declined,
+    } = input.remote.clone().unwrap_or_default();
+
+    let both_declined_and_remote = set_intersection(&info_collections, &remote_declined);
+    if !both_declined_and_remote.is_empty() {
+        // Should we wipe these too?
+        log::warn!(
+            "Remote state contains engines which are in both info/collections and meta/global's declined: {:?}",
+            both_declined_and_remote,
+        );
+    }
+
+    let most_recent_declined_list = if have_remote {
+        &remote_declined
+    } else {
+        &input.local_declined
+    };
+
+    let result_declined = set_difference(
+        &set_union(most_recent_declined_list, &must_disable),
+        &must_enable,
+    );
+
+    let output = EngineStateOutput {
+        changes_needed: EngineChangesNeeded {
+            // Anything now declined which wasn't in our declined list before gets a reset.
+            local_resets: set_difference(&result_declined, &input.local_declined),
+            // Anything remote that we just declined gets a wipe. In the future
+            // we might want to consider wiping things in both remote declined
+            // and info/collections, but we'll let other clients pick up their
+            // own mess for now.
+            remote_wipes: set_intersection(&info_collections, &must_disable),
+        },
+        declined: result_declined,
+    };
+    // No PII here and this helps debug problems.
+    log::debug!("compute_engine_states: output {:?}", output);
+    output
+}
+
+impl PersistedGlobalState {
+    fn set_declined(&mut self, new_declined: Vec<String>) {
+        match self {
+            Self::V2 { ref mut declined } => *declined = Some(new_declined),
+        }
+    }
+    pub(crate) fn get_declined(&self) -> &[String] {
+        match self {
+            Self::V2 { declined: Some(d) } => d,
+            Self::V2 { declined: None } => &[],
+        }
+    }
+}
+
+/// Holds global Sync state, including server upload limits, the
+/// last-fetched collection modified times, `meta/global` record, and
+/// an encrypted copy of the crypto/keys resource (avoids keeping them
+/// in memory longer than necessary; avoids key mismatches by ensuring the same KeyBundle
+/// is used for both the keys and encrypted payloads.)
+#[derive(Debug, Clone)]
+pub struct GlobalState {
+    pub config: InfoConfiguration,
+    pub collections: InfoCollections,
+    pub global: MetaGlobalRecord,
+    pub global_timestamp: ServerTimestamp,
+    pub keys: EncryptedPayload,
+    pub keys_timestamp: ServerTimestamp,
+}
+
+/// Creates a fresh `meta/global` record, using the default engine selections,
+/// and declined engines from our PersistedGlobalState.
+fn new_global(pgs: &PersistedGlobalState) -> MetaGlobalRecord {
+    let sync_id = Guid::random();
+    let mut engines: HashMap<String, _> = HashMap::new();
+    for (name, version) in DEFAULT_ENGINES.iter() {
+        let sync_id = Guid::random();
+        engines.insert(
+            (*name).to_string(),
+            MetaGlobalEngine {
+                version: *version,
+                sync_id,
+            },
+        );
+    }
+    // We only need our PersistedGlobalState to fill out a new meta/global - if
+    // we previously saw a meta/global then we would have updated it with what
+    // it was at the time.
+    let declined = match pgs {
+        PersistedGlobalState::V2 { declined: Some(d) } => d.clone(),
+        _ => DEFAULT_DECLINED.iter().map(ToString::to_string).collect(),
+    };
+
+    MetaGlobalRecord {
+        sync_id,
+        storage_version: STORAGE_VERSION,
+        engines,
+        declined,
+    }
+}
+
+fn fixup_meta_global(global: &mut MetaGlobalRecord) -> bool {
+    let mut changed_any = false;
+    for &(name, version) in DEFAULT_ENGINES.iter() {
+        let had_engine = global.engines.contains_key(name);
+        let should_have_engine = !global.declined.iter().any(|c| c == name);
+        if had_engine != should_have_engine {
+            if should_have_engine {
+                log::debug!("SyncID for engine {:?} was missing", name);
+                global.engines.insert(
+                    name.to_string(),
+                    MetaGlobalEngine {
+                        version,
+                        sync_id: Guid::random(),
+                    },
+                );
+            } else {
+                log::debug!("SyncID for engine {:?} was present, but shouldn't be", name);
+                global.engines.remove(name);
+            }
+            changed_any = true;
+        }
+    }
+    changed_any
+}
+
+pub struct SetupStateMachine<'a> {
+    client: &'a dyn SetupStorageClient,
+    root_key: &'a KeyBundle,
+    pgs: &'a mut PersistedGlobalState,
+    // `allowed_states` is designed so that we can arrange for the concept of
+    // a "fast" sync - so we decline to advance if we need to setup from scratch.
+    // The idea is that if we need to sync before going to sleep we should do
+    // it as fast as possible. However, in practice this isn't going to do
+    // what we expect - a "fast sync" that finds lots to do is almost certainly
+    // going to take longer than a "full sync" that finds nothing to do.
+    // We should almost certainly remove this and instead allow for a "time
+    // budget", after which we get interrupted. Later...
+    allowed_states: Vec<&'static str>,
+    sequence: Vec<&'static str>,
+    engine_updates: Option<&'a HashMap<String, bool>>,
+    interruptee: &'a dyn Interruptee,
+    pub(crate) changes_needed: Option<EngineChangesNeeded>,
+}
+
+impl<'a> SetupStateMachine<'a> {
+    /// Creates a state machine for a "classic" Sync 1.5 client that supports
+    /// all states, including uploading a fresh `meta/global` and `crypto/keys`
+    /// after a node reassignment.
+    pub fn for_full_sync(
+        client: &'a dyn SetupStorageClient,
+        root_key: &'a KeyBundle,
+        pgs: &'a mut PersistedGlobalState,
+        engine_updates: Option<&'a HashMap<String, bool>>,
+        interruptee: &'a dyn Interruptee,
+    ) -> SetupStateMachine<'a> {
+        SetupStateMachine::with_allowed_states(
+            client,
+            root_key,
+            pgs,
+            interruptee,
+            engine_updates,
+            vec![
+                "Initial",
+                "InitialWithConfig",
+                "InitialWithInfo",
+                "InitialWithMetaGlobal",
+                "Ready",
+                "FreshStartRequired",
+                "WithPreviousState",
+            ],
+        )
+    }
+
+    fn with_allowed_states(
+        client: &'a dyn SetupStorageClient,
+        root_key: &'a KeyBundle,
+        pgs: &'a mut PersistedGlobalState,
+        interruptee: &'a dyn Interruptee,
+        engine_updates: Option<&'a HashMap<String, bool>>,
+        allowed_states: Vec<&'static str>,
+    ) -> SetupStateMachine<'a> {
+        SetupStateMachine {
+            client,
+            root_key,
+            pgs,
+            sequence: Vec::new(),
+            allowed_states,
+            engine_updates,
+            interruptee,
+            changes_needed: None,
+        }
+    }
+
+    fn advance(&mut self, from: SetupState) -> error::Result<SetupState> {
+        match from {
+            // Fetch `info/configuration` with current server limits, and
+            // `info/collections` with collection last modified times.
+            Initial => {
+                let config = match self.client.fetch_info_configuration()? {
+                    Sync15ClientResponse::Success { record, .. } => record,
+                    Sync15ClientResponse::Error(ErrorResponse::NotFound { .. }) => {
+                        InfoConfiguration::default()
+                    }
+                    other => return Err(other.create_storage_error()),
+                };
+                Ok(InitialWithConfig { config })
+            }
+
+            // XXX - we could consider combining these Initial* states, because we don't
+            // attempt to support filling in "missing" global state - *any* 404 in them
+            // means `FreshStart`.
+            // IOW, in all cases, they either `Err()`, move to `FreshStartRequired`, or
+            // advance to a specific next state.
+            InitialWithConfig { config } => {
+                match self.client.fetch_info_collections()? {
+                    Sync15ClientResponse::Success {
+                        record: collections,
+                        ..
+                    } => Ok(InitialWithInfo {
+                        config,
+                        collections,
+                    }),
+                    // If the server doesn't have a `crypto/keys`, start over
+                    // and reupload our `meta/global` and `crypto/keys`.
+                    Sync15ClientResponse::Error(ErrorResponse::NotFound { .. }) => {
+                        Ok(FreshStartRequired { config })
+                    }
+                    other => Err(other.create_storage_error()),
+                }
+            }
+
+            InitialWithInfo {
+                config,
+                collections,
+            } => {
+                match self.client.fetch_meta_global()? {
+                    Sync15ClientResponse::Success {
+                        record: mut global,
+                        last_modified: mut global_timestamp,
+                        ..
+                    } => {
+                        // If the server has a newer storage version, we can't
+                        // sync until our client is updated.
+                        if global.storage_version > STORAGE_VERSION {
+                            return Err(ErrorKind::ClientUpgradeRequired);
+                        }
+
+                        // If the server has an older storage version, wipe and
+                        // reupload.
+                        if global.storage_version < STORAGE_VERSION {
+                            Ok(FreshStartRequired { config })
+                        } else {
+                            log::info!("Have info/collections and meta/global. Computing new engine states");
+                            let initial_global_declined: HashSet<String> =
+                                global.declined.iter().cloned().collect();
+                            let result = compute_engine_states(EngineStateInput {
+                                local_declined: self.pgs.get_declined().iter().cloned().collect(),
+                                user_changes: self.engine_updates.cloned().unwrap_or_default(),
+                                remote: Some(RemoteEngineState {
+                                    declined: initial_global_declined.clone(),
+                                    info_collections: collections.keys().cloned().collect(),
+                                }),
+                            });
+                            // Persist the new declined.
+                            self.pgs
+                                .set_declined(result.declined.iter().cloned().collect());
+                            // If the declined engines differ from remote, fix that.
+                            let fixed_declined = if result.declined != initial_global_declined {
+                                global.declined = result.declined.iter().cloned().collect();
+                                log::info!(
+                                    "Uploading new declined {:?} to meta/global with timestamp {:?}",
+                                    global.declined,
+                                    global_timestamp,
+                                );
+                                true
+                            } else {
+                                false
+                            };
+                            // If there are missing syncIds, we need to fix those as well
+                            let fixed_ids = if fixup_meta_global(&mut global) {
+                                log::info!(
+                                    "Uploading corrected meta/global with timestamp {:?}",
+                                    global_timestamp,
+                                );
+                                true
+                            } else {
+                                false
+                            };
+
+                            if fixed_declined || fixed_ids {
+                                global_timestamp =
+                                    self.client.put_meta_global(global_timestamp, &global)?;
+                                log::debug!("new global_timestamp: {:?}", global_timestamp);
+                            }
+                            // Update the set of changes needed.
+                            if self.changes_needed.is_some() {
+                                // Should never happen (we prevent state machine
+                                // loops elsewhere) but if it did, the info is stale
+                                // anyway.
+                                log::warn!("Already have a set of changes needed, Overwriting...");
+                            }
+                            self.changes_needed = Some(result.changes_needed);
+                            Ok(InitialWithMetaGlobal {
+                                config,
+                                collections,
+                                global,
+                                global_timestamp,
+                            })
+                        }
+                    }
+                    Sync15ClientResponse::Error(ErrorResponse::NotFound { .. }) => {
+                        Ok(FreshStartRequired { config })
+                    }
+                    other => Err(other.create_storage_error()),
+                }
+            }
+
+            InitialWithMetaGlobal {
+                config,
+                collections,
+                global,
+                global_timestamp,
+            } => {
+                // Now try and get keys etc - if we fresh-start we'll re-use declined.
+                match self.client.fetch_crypto_keys()? {
+                    Sync15ClientResponse::Success {
+                        record,
+                        last_modified,
+                        ..
+                    } => {
+                        // Note that collection/keys is itself a bso, so the
+                        // json body also carries the timestamp. If they aren't
+                        // identical something has screwed up and we should die.
+                        assert_eq!(last_modified, record.envelope.modified);
+                        let state = GlobalState {
+                            config,
+                            collections,
+                            global,
+                            global_timestamp,
+                            keys: record.payload,
+                            keys_timestamp: last_modified,
+                        };
+                        Ok(Ready { state })
+                    }
+                    // If the server doesn't have a `crypto/keys`, start over
+                    // and reupload our `meta/global` and `crypto/keys`.
+                    Sync15ClientResponse::Error(ErrorResponse::NotFound { .. }) => {
+                        Ok(FreshStartRequired { config })
+                    }
+                    other => Err(other.create_storage_error()),
+                }
+            }
+
+            // We've got old state that's likely to be OK.
+            // We keep things simple here - if there's evidence of a new/missing
+            // meta/global or new/missing keys we just restart from scratch.
+            WithPreviousState { old_state } => match self.client.fetch_info_collections()? {
+                Sync15ClientResponse::Success {
+                    record: collections,
+                    ..
+                } => Ok(
+                    if self.engine_updates.is_none()
+                        && is_same_timestamp(old_state.global_timestamp, &collections, "meta")
+                        && is_same_timestamp(old_state.keys_timestamp, &collections, "crypto")
+                    {
+                        Ready {
+                            state: GlobalState {
+                                collections,
+                                ..old_state
+                            },
+                        }
+                    } else {
+                        InitialWithConfig {
+                            config: old_state.config,
+                        }
+                    },
+                ),
+                _ => Ok(InitialWithConfig {
+                    config: old_state.config,
+                }),
+            },
+
+            Ready { state } => Ok(Ready { state }),
+
+            FreshStartRequired { config } => {
+                // Wipe the server.
+                log::info!("Fresh start: wiping remote");
+                self.client.wipe_all_remote()?;
+
+                // Upload a fresh `meta/global`...
+                log::info!("Uploading meta/global");
+                let computed = compute_engine_states(EngineStateInput {
+                    local_declined: self.pgs.get_declined().iter().cloned().collect(),
+                    user_changes: self.engine_updates.cloned().unwrap_or_default(),
+                    remote: None,
+                });
+                self.pgs
+                    .set_declined(computed.declined.iter().cloned().collect());
+
+                self.changes_needed = Some(computed.changes_needed);
+
+                let new_global = new_global(self.pgs);
+
+                self.client
+                    .put_meta_global(ServerTimestamp::default(), &new_global)?;
+
+                // ...And a fresh `crypto/keys`.
+                let new_keys = CollectionKeys::new_random()?.to_encrypted_payload(self.root_key)?;
+                let bso = OutgoingEncryptedBso::new(Guid::new("keys").into(), new_keys);
+                self.client
+                    .put_crypto_keys(ServerTimestamp::default(), &bso)?;
+
+                // TODO(lina): Can we pass along server timestamps from the PUTs
+                // above, and avoid re-fetching the `m/g` and `c/k` we just
+                // uploaded?
+                // OTOH(mark): restarting the state machine keeps life simple and rare.
+                Ok(InitialWithConfig { config })
+            }
+        }
+    }
+
+    /// Runs through the state machine to the ready state.
+    pub fn run_to_ready(&mut self, state: Option<GlobalState>) -> error::Result<GlobalState> {
+        let mut s = match state {
+            Some(old_state) => WithPreviousState { old_state },
+            None => Initial,
+        };
+        loop {
+            self.interruptee.err_if_interrupted()?;
+            let label = &s.label();
+            log::trace!("global state: {:?}", label);
+            match s {
+                Ready { state } => {
+                    self.sequence.push(label);
+                    return Ok(state);
+                }
+                // If we already started over once before, we're likely in a
+                // cycle, and should try again later. Intermediate states
+                // aren't a problem, just the initial ones.
+                FreshStartRequired { .. } | WithPreviousState { .. } | Initial => {
+                    if self.sequence.contains(label) {
+                        // Is this really the correct error?
+                        return Err(ErrorKind::SetupRace);
+                    }
+                }
+                _ => {
+                    if !self.allowed_states.contains(label) {
+                        return Err(ErrorKind::SetupRequired);
+                    }
+                }
+            };
+            self.sequence.push(label);
+            s = self.advance(s)?;
+        }
+    }
+}
+
+/// States in the remote setup process.
+/// TODO(lina): Add link once #56 is merged.
+#[derive(Debug)]
+#[allow(clippy::large_enum_variant)]
+enum SetupState {
+    // These "Initial" states are only ever used when starting from scratch.
+    Initial,
+    InitialWithConfig {
+        config: InfoConfiguration,
+    },
+    InitialWithInfo {
+        config: InfoConfiguration,
+        collections: InfoCollections,
+    },
+    InitialWithMetaGlobal {
+        config: InfoConfiguration,
+        collections: InfoCollections,
+        global: MetaGlobalRecord,
+        global_timestamp: ServerTimestamp,
+    },
+    WithPreviousState {
+        old_state: GlobalState,
+    },
+    Ready {
+        state: GlobalState,
+    },
+    FreshStartRequired {
+        config: InfoConfiguration,
+    },
+}
+
+impl SetupState {
+    fn label(&self) -> &'static str {
+        match self {
+            Initial { .. } => "Initial",
+            InitialWithConfig { .. } => "InitialWithConfig",
+            InitialWithInfo { .. } => "InitialWithInfo",
+            InitialWithMetaGlobal { .. } => "InitialWithMetaGlobal",
+            Ready { .. } => "Ready",
+            WithPreviousState { .. } => "WithPreviousState",
+            FreshStartRequired { .. } => "FreshStartRequired",
+        }
+    }
+}
+
+/// Whether we should skip fetching an item. Used when we already have timestamps
+/// and want to check if we should reuse our existing state. The state's fairly
+/// cheap to recreate and very bad to use if it is wrong, so we insist on the
+/// *exact* timestamp matching and not a simple "later than" check.
+fn is_same_timestamp(local: ServerTimestamp, collections: &InfoCollections, key: &str) -> bool {
+    collections.get(key).map_or(false, |ts| local == *ts)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    use crate::bso::{IncomingEncryptedBso, IncomingEnvelope};
+    use interrupt_support::NeverInterrupts;
+
+    struct InMemoryClient {
+        info_configuration: error::Result<Sync15ClientResponse<InfoConfiguration>>,
+        info_collections: error::Result<Sync15ClientResponse<InfoCollections>>,
+        meta_global: error::Result<Sync15ClientResponse<MetaGlobalRecord>>,
+        crypto_keys: error::Result<Sync15ClientResponse<IncomingEncryptedBso>>,
+    }
+
+    impl SetupStorageClient for InMemoryClient {
+        fn fetch_info_configuration(
+            &self,
+        ) -> error::Result<Sync15ClientResponse<InfoConfiguration>> {
+            match &self.info_configuration {
+                Ok(client_response) => Ok(client_response.clone()),
+                Err(_) => Ok(Sync15ClientResponse::Error(ErrorResponse::ServerError {
+                    status: 500,
+                    route: "test/path".into(),
+                })),
+            }
+        }
+
+        fn fetch_info_collections(&self) -> error::Result<Sync15ClientResponse<InfoCollections>> {
+            match &self.info_collections {
+                Ok(collections) => Ok(collections.clone()),
+                Err(_) => Ok(Sync15ClientResponse::Error(ErrorResponse::ServerError {
+                    status: 500,
+                    route: "test/path".into(),
+                })),
+            }
+        }
+
+        fn fetch_meta_global(&self) -> error::Result<Sync15ClientResponse<MetaGlobalRecord>> {
+            match &self.meta_global {
+                Ok(global) => Ok(global.clone()),
+                // TODO(lina): Special handling for 404s, we want to ensure we
+                // handle missing keys and other server errors correctly.
+                Err(_) => Ok(Sync15ClientResponse::Error(ErrorResponse::ServerError {
+                    status: 500,
+                    route: "test/path".into(),
+                })),
+            }
+        }
+
+        fn put_meta_global(
+            &self,
+            xius: ServerTimestamp,
+            global: &MetaGlobalRecord,
+        ) -> error::Result<ServerTimestamp> {
+            // Ensure that the meta/global record we uploaded is "fixed up"
+            assert!(DEFAULT_ENGINES
+                .iter()
+                .filter(|e| e.0 != "logins")
+                .all(|&(k, _v)| global.engines.contains_key(k)));
+            assert!(!global.engines.contains_key("logins"));
+            assert_eq!(global.declined, vec!["logins".to_string()]);
+            // return a different timestamp.
+            Ok(ServerTimestamp(xius.0 + 1))
+        }
+
+        fn fetch_crypto_keys(&self) -> error::Result<Sync15ClientResponse<IncomingEncryptedBso>> {
+            match &self.crypto_keys {
+                Ok(Sync15ClientResponse::Success {
+                    status,
+                    record,
+                    last_modified,
+                    route,
+                }) => Ok(Sync15ClientResponse::Success {
+                    status: *status,
+                    record: IncomingEncryptedBso::new(
+                        record.envelope.clone(),
+                        record.payload.clone(),
+                    ),
+                    last_modified: *last_modified,
+                    route: route.clone(),
+                }),
+                // TODO(lina): Same as above, for 404s.
+                _ => Ok(Sync15ClientResponse::Error(ErrorResponse::ServerError {
+                    status: 500,
+                    route: "test/path".into(),
+                })),
+            }
+        }
+
+        fn put_crypto_keys(
+            &self,
+            xius: ServerTimestamp,
+            _keys: &OutgoingEncryptedBso,
+        ) -> error::Result<()> {
+            assert_eq!(xius, ServerTimestamp(888_800));
+            Err(ErrorKind::StorageHttpError(ErrorResponse::ServerError {
+                status: 500,
+                route: "crypto/keys".to_string(),
+            }))
+        }
+
+        fn wipe_all_remote(&self) -> error::Result<()> {
+            Ok(())
+        }
+    }
+
+    #[allow(clippy::unnecessary_wraps)]
+    fn mocked_success_ts<T>(t: T, ts: i64) -> error::Result<Sync15ClientResponse<T>> {
+        Ok(Sync15ClientResponse::Success {
+            status: 200,
+            record: t,
+            last_modified: ServerTimestamp(ts),
+            route: "test/path".into(),
+        })
+    }
+
+    fn mocked_success<T>(t: T) -> error::Result<Sync15ClientResponse<T>> {
+        mocked_success_ts(t, 0)
+    }
+
+    fn mocked_success_keys(
+        keys: CollectionKeys,
+        root_key: &KeyBundle,
+    ) -> error::Result<Sync15ClientResponse<IncomingEncryptedBso>> {
+        let timestamp = keys.timestamp;
+        let payload = keys.to_encrypted_payload(root_key).unwrap();
+        let bso = IncomingEncryptedBso::new(
+            IncomingEnvelope {
+                id: Guid::new("keys"),
+                modified: timestamp,
+                sortindex: None,
+                ttl: None,
+            },
+            payload,
+        );
+        Ok(Sync15ClientResponse::Success {
+            status: 200,
+            record: bso,
+            last_modified: timestamp,
+            route: "test/path".into(),
+        })
+    }
+
+    #[test]
+    fn test_state_machine_ready_from_empty() {
+        let _ = env_logger::try_init();
+        let root_key = KeyBundle::new_random().unwrap();
+        let keys = CollectionKeys {
+            timestamp: ServerTimestamp(123_400),
+            default: KeyBundle::new_random().unwrap(),
+            collections: HashMap::new(),
+        };
+        let mg = MetaGlobalRecord {
+            sync_id: "syncIDAAAAAA".into(),
+            storage_version: 5usize,
+            engines: vec![(
+                "bookmarks",
+                MetaGlobalEngine {
+                    version: 1usize,
+                    sync_id: "syncIDBBBBBB".into(),
+                },
+            )]
+            .into_iter()
+            .map(|(key, value)| (key.to_owned(), value))
+            .collect(),
+            // We ensure that the record we upload doesn't have a logins record.
+            declined: vec!["logins".to_string()],
+        };
+        let client = InMemoryClient {
+            info_configuration: mocked_success(InfoConfiguration::default()),
+            info_collections: mocked_success(InfoCollections::new(
+                vec![("meta", 123_456), ("crypto", 145_000)]
+                    .into_iter()
+                    .map(|(key, value)| (key.to_owned(), ServerTimestamp(value)))
+                    .collect(),
+            )),
+            meta_global: mocked_success_ts(mg, 999_000),
+            crypto_keys: mocked_success_keys(keys, &root_key),
+        };
+        let mut pgs = PersistedGlobalState::V2 { declined: None };
+
+        let mut state_machine =
+            SetupStateMachine::for_full_sync(&client, &root_key, &mut pgs, None, &NeverInterrupts);
+        assert!(
+            state_machine.run_to_ready(None).is_ok(),
+            "Should drive state machine to ready"
+        );
+        assert_eq!(
+            state_machine.sequence,
+            vec![
+                "Initial",
+                "InitialWithConfig",
+                "InitialWithInfo",
+                "InitialWithMetaGlobal",
+                "Ready",
+            ],
+            "Should cycle through all states"
+        );
+    }
+
+    #[test]
+    fn test_from_previous_state_declined() {
+        let _ = env_logger::try_init();
+        // The state-machine sequence where we didn't use the previous state
+        // (ie, where the state machine restarted)
+        let sm_seq_restarted = vec![
+            "WithPreviousState",
+            "InitialWithConfig",
+            "InitialWithInfo",
+            "InitialWithMetaGlobal",
+            "Ready",
+        ];
+        // The state-machine sequence where we used the previous state.
+        let sm_seq_used_previous = vec!["WithPreviousState", "Ready"];
+
+        // do the actual test.
+        fn do_test(
+            client: &dyn SetupStorageClient,
+            root_key: &KeyBundle,
+            pgs: &mut PersistedGlobalState,
+            engine_updates: Option<&HashMap<String, bool>>,
+            old_state: GlobalState,
+            expected_states: &[&str],
+        ) {
+            let mut state_machine = SetupStateMachine::for_full_sync(
+                client,
+                root_key,
+                pgs,
+                engine_updates,
+                &NeverInterrupts,
+            );
+            assert!(
+                state_machine.run_to_ready(Some(old_state)).is_ok(),
+                "Should drive state machine to ready"
+            );
+            assert_eq!(state_machine.sequence, expected_states);
+        }
+
+        // and all the complicated setup...
+        let ts_metaglobal = 123_456;
+        let ts_keys = 145_000;
+        let root_key = KeyBundle::new_random().unwrap();
+        let keys = CollectionKeys {
+            timestamp: ServerTimestamp(ts_keys + 1),
+            default: KeyBundle::new_random().unwrap(),
+            collections: HashMap::new(),
+        };
+        let mg = MetaGlobalRecord {
+            sync_id: "syncIDAAAAAA".into(),
+            storage_version: 5usize,
+            engines: vec![(
+                "bookmarks",
+                MetaGlobalEngine {
+                    version: 1usize,
+                    sync_id: "syncIDBBBBBB".into(),
+                },
+            )]
+            .into_iter()
+            .map(|(key, value)| (key.to_owned(), value))
+            .collect(),
+            // We ensure that the record we upload doesn't have a logins record.
+            declined: vec!["logins".to_string()],
+        };
+        let collections = InfoCollections::new(
+            vec![("meta", ts_metaglobal), ("crypto", ts_keys)]
+                .into_iter()
+                .map(|(key, value)| (key.to_owned(), ServerTimestamp(value)))
+                .collect(),
+        );
+        let client = InMemoryClient {
+            info_configuration: mocked_success(InfoConfiguration::default()),
+            info_collections: mocked_success(collections.clone()),
+            meta_global: mocked_success_ts(mg.clone(), ts_metaglobal),
+            crypto_keys: mocked_success_keys(keys.clone(), &root_key),
+        };
+
+        // First a test where the "previous" global state is OK to reuse.
+        {
+            let mut pgs = PersistedGlobalState::V2 { declined: None };
+            // A "previous" global state.
+            let old_state = GlobalState {
+                config: InfoConfiguration::default(),
+                collections: collections.clone(),
+                global: mg.clone(),
+                global_timestamp: ServerTimestamp(ts_metaglobal),
+                keys: keys
+                    .to_encrypted_payload(&root_key)
+                    .expect("should always work in this test"),
+                keys_timestamp: ServerTimestamp(ts_keys),
+            };
+            do_test(
+                &client,
+                &root_key,
+                &mut pgs,
+                None,
+                old_state,
+                &sm_seq_used_previous,
+            );
+        }
+
+        // Now where the meta/global record on the server is later.
+        {
+            let mut pgs = PersistedGlobalState::V2 { declined: None };
+            // A "previous" global state.
+            let old_state = GlobalState {
+                config: InfoConfiguration::default(),
+                collections: collections.clone(),
+                global: mg.clone(),
+                global_timestamp: ServerTimestamp(999_999),
+                keys: keys
+                    .to_encrypted_payload(&root_key)
+                    .expect("should always work in this test"),
+                keys_timestamp: ServerTimestamp(ts_keys),
+            };
+            do_test(
+                &client,
+                &root_key,
+                &mut pgs,
+                None,
+                old_state,
+                &sm_seq_restarted,
+            );
+        }
+
+        // Where keys on the server is later.
+        {
+            let mut pgs = PersistedGlobalState::V2 { declined: None };
+            // A "previous" global state.
+            let old_state = GlobalState {
+                config: InfoConfiguration::default(),
+                collections: collections.clone(),
+                global: mg.clone(),
+                global_timestamp: ServerTimestamp(ts_metaglobal),
+                keys: keys
+                    .to_encrypted_payload(&root_key)
+                    .expect("should always work in this test"),
+                keys_timestamp: ServerTimestamp(999_999),
+            };
+            do_test(
+                &client,
+                &root_key,
+                &mut pgs,
+                None,
+                old_state,
+                &sm_seq_restarted,
+            );
+        }
+
+        // Where there are engine-state changes.
+        {
+            let mut pgs = PersistedGlobalState::V2 { declined: None };
+            // A "previous" global state.
+            let old_state = GlobalState {
+                config: InfoConfiguration::default(),
+                collections,
+                global: mg,
+                global_timestamp: ServerTimestamp(ts_metaglobal),
+                keys: keys
+                    .to_encrypted_payload(&root_key)
+                    .expect("should always work in this test"),
+                keys_timestamp: ServerTimestamp(ts_keys),
+            };
+            let mut engine_updates = HashMap::<String, bool>::new();
+            engine_updates.insert("logins".to_string(), false);
+            do_test(
+                &client,
+                &root_key,
+                &mut pgs,
+                Some(&engine_updates),
+                old_state,
+                &sm_seq_restarted,
+            );
+            let declined = match pgs {
+                PersistedGlobalState::V2 { declined: d } => d,
+            };
+            // and check we now consider logins as declined.
+            assert_eq!(declined, Some(vec!["logins".to_string()]));
+        }
+    }
+
+    fn string_set(s: &[&str]) -> HashSet<String> {
+        s.iter().map(ToString::to_string).collect()
+    }
+    fn string_map<T: Clone>(s: &[(&str, T)]) -> HashMap<String, T> {
+        s.iter().map(|v| (v.0.to_string(), v.1.clone())).collect()
+    }
+    #[test]
+    fn test_engine_states() {
+        assert_eq!(
+            compute_engine_states(EngineStateInput {
+                local_declined: string_set(&["foo", "bar"]),
+                remote: None,
+                user_changes: Default::default(),
+            }),
+            EngineStateOutput {
+                declined: string_set(&["foo", "bar"]),
+                // No wipes, no resets
+                changes_needed: Default::default(),
+            }
+        );
+        assert_eq!(
+            compute_engine_states(EngineStateInput {
+                local_declined: string_set(&["foo", "bar"]),
+                remote: Some(RemoteEngineState {
+                    declined: string_set(&["foo"]),
+                    info_collections: string_set(&["bar"])
+                }),
+                user_changes: Default::default(),
+            }),
+            EngineStateOutput {
+                // Now we have `foo`.
+                declined: string_set(&["foo"]),
+                // No wipes, no resets, should just be a local update.
+                changes_needed: Default::default(),
+            }
+        );
+        assert_eq!(
+            compute_engine_states(EngineStateInput {
+                local_declined: string_set(&["foo", "bar"]),
+                remote: Some(RemoteEngineState {
+                    declined: string_set(&["foo", "bar", "quux"]),
+                    info_collections: string_set(&[])
+                }),
+                user_changes: Default::default(),
+            }),
+            EngineStateOutput {
+                // Now we have `foo`.
+                declined: string_set(&["foo", "bar", "quux"]),
+                changes_needed: EngineChangesNeeded {
+                    // Should reset `quux`.
+                    local_resets: string_set(&["quux"]),
+                    // No wipes, though.
+                    remote_wipes: string_set(&[]),
+                }
+            }
+        );
+        assert_eq!(
+            compute_engine_states(EngineStateInput {
+                local_declined: string_set(&["bar", "baz"]),
+                remote: Some(RemoteEngineState {
+                    declined: string_set(&["bar", "baz",]),
+                    info_collections: string_set(&["quux"])
+                }),
+                // Change a declined engine to undeclined.
+                user_changes: string_map(&[("bar", true)]),
+            }),
+            EngineStateOutput {
+                declined: string_set(&["baz"]),
+                // No wipes, just undecline it.
+                changes_needed: Default::default()
+            }
+        );
+        assert_eq!(
+            compute_engine_states(EngineStateInput {
+                local_declined: string_set(&["bar", "baz"]),
+                remote: Some(RemoteEngineState {
+                    declined: string_set(&["bar", "baz"]),
+                    info_collections: string_set(&["foo"])
+                }),
+                // Change an engine which exists remotely to declined.
+                user_changes: string_map(&[("foo", false)]),
+            }),
+            EngineStateOutput {
+                declined: string_set(&["baz", "bar", "foo"]),
+                // No wipes, just undecline it.
+                changes_needed: EngineChangesNeeded {
+                    // Should reset our local foo
+                    local_resets: string_set(&["foo"]),
+                    // And wipe the server.
+                    remote_wipes: string_set(&["foo"]),
+                }
+            }
+        );
+    }
+}
diff --git a/third_party/rust/sync15/src/client/status.rs b/third_party/rust/sync15/src/client/status.rs
new file mode 100644
index 0000000000..407efeec12
--- /dev/null
+++ b/third_party/rust/sync15/src/client/status.rs
@@ -0,0 +1,106 @@
+/* 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/. */
+
+use crate::error::{Error, ErrorResponse};
+use crate::telemetry::SyncTelemetryPing;
+use std::collections::HashMap;
+use std::time::{Duration, SystemTime};
+
+/// The general status of sync - should probably be moved to the "sync manager"
+/// once we have one!
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum ServiceStatus {
+    /// Everything is fine.
+    Ok,
+    /// Some general network issue.
+    NetworkError,
+    /// Some apparent issue with the servers.
+    ServiceError,
+    /// Some external FxA action needs to be taken.
+    AuthenticationError,
+    /// We declined to do anything for backoff or rate-limiting reasons.
+    BackedOff,
+    /// We were interrupted.
+    Interrupted,
+    /// Something else - you need to check the logs for more details. May
+    /// or may not be transient, we really don't know.
+    OtherError,
+}
+
+impl ServiceStatus {
+    // This is a bit naive and probably will not survive in this form in the
+    // SyncManager - eg, we'll want to handle backoff etc.
+    pub fn from_err(err: &Error) -> ServiceStatus {
+        match err {
+            // HTTP based errors.
+            Error::TokenserverHttpError(status) => {
+                // bit of a shame the tokenserver is different to storage...
+                if *status == 401 {
+                    ServiceStatus::AuthenticationError
+                } else {
+                    ServiceStatus::ServiceError
+                }
+            }
+            // BackoffError is also from the tokenserver.
+            Error::BackoffError(_) => ServiceStatus::ServiceError,
+            Error::StorageHttpError(ref e) => match e {
+                ErrorResponse::Unauthorized { .. } => ServiceStatus::AuthenticationError,
+                _ => ServiceStatus::ServiceError,
+            },
+
+            // Network errors.
+            Error::RequestError(_) | Error::UnexpectedStatus(_) | Error::HawkError(_) => {
+                ServiceStatus::NetworkError
+            }
+
+            Error::Interrupted(_) => ServiceStatus::Interrupted,
+            _ => ServiceStatus::OtherError,
+        }
+    }
+}
+
+/// The result of a sync request. This too is from the "sync manager", but only
+/// has a fraction of the things it will have when we actually build that.
+#[derive(Debug)]
+pub struct SyncResult {
+    /// The general health.
+    pub service_status: ServiceStatus,
+
+    /// The set of declined engines, if we know them.
+    pub declined: Option<Vec<String>>,
+
+    /// The result of the sync.
+    pub result: Result<(), Error>,
+
+    /// The result for each engine.
+    /// Note that we expect the `String` to be replaced with an enum later.
+    pub engine_results: HashMap<String, Result<(), Error>>,
+
+    pub telemetry: SyncTelemetryPing,
+
+    pub next_sync_after: Option<std::time::SystemTime>,
+}
+
+// If `r` has a BackoffError, then returns the later backoff value.
+fn advance_backoff(cur_best: SystemTime, r: &Result<(), Error>) -> SystemTime {
+    if let Err(e) = r {
+        if let Some(time) = e.get_backoff() {
+            return std::cmp::max(time, cur_best);
+        }
+    }
+    cur_best
+}
+
+impl SyncResult {
+    pub(crate) fn set_sync_after(&mut self, backoff_duration: Duration) {
+        let now = SystemTime::now();
+        let toplevel = advance_backoff(now + backoff_duration, &self.result);
+        let sync_after = self.engine_results.values().fold(toplevel, advance_backoff);
+        if sync_after <= now {
+            self.next_sync_after = None;
+        } else {
+            self.next_sync_after = Some(sync_after);
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/client/storage_client.rs b/third_party/rust/sync15/src/client/storage_client.rs
new file mode 100644
index 0000000000..7b8f650d24
--- /dev/null
+++ b/third_party/rust/sync15/src/client/storage_client.rs
@@ -0,0 +1,605 @@
+/* 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/. */
+
+use super::request::{
+    BatchPoster, InfoCollections, InfoConfiguration, PostQueue, PostResponse, PostResponseHandler,
+};
+use super::token;
+use crate::bso::{IncomingBso, IncomingEncryptedBso, OutgoingBso, OutgoingEncryptedBso};
+use crate::engine::{CollectionPost, CollectionRequest};
+use crate::error::{self, Error, ErrorResponse};
+use crate::record_types::MetaGlobalRecord;
+use crate::{CollectionName, Guid, ServerTimestamp};
+use serde_json::Value;
+use std::str::FromStr;
+use std::sync::atomic::{AtomicU32, Ordering};
+use url::Url;
+use viaduct::{
+    header_names::{self, AUTHORIZATION},
+    Method, Request, Response,
+};
+
+/// A response from a GET request on a Sync15StorageClient, encapsulating all
+/// the variants users of this client needs to care about.
+#[derive(Debug, Clone)]
+pub enum Sync15ClientResponse<T> {
+    Success {
+        status: u16,
+        record: T,
+        last_modified: ServerTimestamp,
+        route: String,
+    },
+    Error(ErrorResponse),
+}
+
+fn parse_seconds(seconds_str: &str) -> Option<u32> {
+    let secs = seconds_str.parse::<f64>().ok()?.ceil();
+    // Note: u32 doesn't impl TryFrom<f64> :(
+    if !secs.is_finite() || secs < 0.0 || secs > f64::from(u32::max_value()) {
+        log::warn!("invalid backoff value: {}", secs);
+        None
+    } else {
+        Some(secs as u32)
+    }
+}
+
+impl<T> Sync15ClientResponse<T> {
+    pub fn from_response(resp: Response, backoff_listener: &BackoffListener) -> error::Result<Self>
+    where
+        for<'a> T: serde::de::Deserialize<'a>,
+    {
+        let route: String = resp.url.path().into();
+        // Android seems to respect retry_after even on success requests, so we
+        // will too if it's present. This also lets us handle both backoff-like
+        // properties in the same place.
+        let retry_after = resp
+            .headers
+            .get(header_names::RETRY_AFTER)
+            .and_then(parse_seconds);
+
+        let backoff = resp
+            .headers
+            .get(header_names::X_WEAVE_BACKOFF)
+            .and_then(parse_seconds);
+
+        if let Some(b) = backoff {
+            backoff_listener.note_backoff(b);
+        }
+        if let Some(ra) = retry_after {
+            backoff_listener.note_retry_after(ra);
+        }
+
+        Ok(if resp.is_success() {
+            let record: T = resp.json()?;
+            let last_modified = resp
+                .headers
+                .get(header_names::X_LAST_MODIFIED)
+                .and_then(|s| ServerTimestamp::from_str(s).ok())
+                .ok_or(Error::MissingServerTimestamp)?;
+            log::info!(
+                "Successful request to \"{}\", incoming x-last-modified={:?}",
+                route,
+                last_modified
+            );
+
+            Sync15ClientResponse::Success {
+                status: resp.status,
+                record,
+                last_modified,
+                route,
+            }
+        } else {
+            let status = resp.status;
+            log::info!("Request \"{}\" was an error (status={})", route, status);
+            match status {
+                404 => Sync15ClientResponse::Error(ErrorResponse::NotFound { route }),
+                401 => Sync15ClientResponse::Error(ErrorResponse::Unauthorized { route }),
+                412 => Sync15ClientResponse::Error(ErrorResponse::PreconditionFailed { route }),
+                500..=600 => {
+                    Sync15ClientResponse::Error(ErrorResponse::ServerError { route, status })
+                }
+                _ => Sync15ClientResponse::Error(ErrorResponse::RequestFailed { route, status }),
+            }
+        })
+    }
+
+    pub fn create_storage_error(self) -> Error {
+        let inner = match self {
+            Sync15ClientResponse::Success { status, route, .. } => {
+                // This should never happen as callers are expected to have
+                // already special-cased this response, so warn if it does.
+                // (or maybe we could panic?)
+                log::warn!("Converting success response into an error");
+                ErrorResponse::RequestFailed { status, route }
+            }
+            Sync15ClientResponse::Error(e) => e,
+        };
+        Error::StorageHttpError(inner)
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct Sync15StorageClientInit {
+    pub key_id: String,
+    pub access_token: String,
+    pub tokenserver_url: Url,
+}
+
+/// A trait containing the methods required to run through the setup state
+/// machine. This is factored out into a separate trait to make mocking
+/// easier.
+pub trait SetupStorageClient {
+    fn fetch_info_configuration(&self) -> error::Result<Sync15ClientResponse<InfoConfiguration>>;
+    fn fetch_info_collections(&self) -> error::Result<Sync15ClientResponse<InfoCollections>>;
+    fn fetch_meta_global(&self) -> error::Result<Sync15ClientResponse<MetaGlobalRecord>>;
+    fn fetch_crypto_keys(&self) -> error::Result<Sync15ClientResponse<IncomingEncryptedBso>>;
+
+    fn put_meta_global(
+        &self,
+        xius: ServerTimestamp,
+        global: &MetaGlobalRecord,
+    ) -> error::Result<ServerTimestamp>;
+    fn put_crypto_keys(
+        &self,
+        xius: ServerTimestamp,
+        keys: &OutgoingEncryptedBso,
+    ) -> error::Result<()>;
+    fn wipe_all_remote(&self) -> error::Result<()>;
+}
+
+#[derive(Debug, Default)]
+pub struct BackoffState {
+    pub backoff_secs: AtomicU32,
+    pub retry_after_secs: AtomicU32,
+}
+
+pub(crate) type BackoffListener = std::sync::Arc<BackoffState>;
+
+pub(crate) fn new_backoff_listener() -> BackoffListener {
+    std::sync::Arc::new(BackoffState::default())
+}
+
+impl BackoffState {
+    pub fn note_backoff(&self, noted: u32) {
+        super::util::atomic_update_max(&self.backoff_secs, noted)
+    }
+
+    pub fn note_retry_after(&self, noted: u32) {
+        super::util::atomic_update_max(&self.retry_after_secs, noted)
+    }
+
+    pub fn get_backoff_secs(&self) -> u32 {
+        self.backoff_secs.load(Ordering::SeqCst)
+    }
+
+    pub fn get_retry_after_secs(&self) -> u32 {
+        self.retry_after_secs.load(Ordering::SeqCst)
+    }
+
+    pub fn get_required_wait(&self, ignore_soft_backoff: bool) -> Option<std::time::Duration> {
+        let bo = self.get_backoff_secs();
+        let ra = self.get_retry_after_secs();
+        let secs = u64::from(if ignore_soft_backoff { ra } else { bo.max(ra) });
+        if secs > 0 {
+            Some(std::time::Duration::from_secs(secs))
+        } else {
+            None
+        }
+    }
+
+    pub fn reset(&self) {
+        self.backoff_secs.store(0, Ordering::SeqCst);
+        self.retry_after_secs.store(0, Ordering::SeqCst);
+    }
+}
+
+// meta/global is a clear-text Bso (ie, there's a String `payload` which has a MetaGlobalRecord)
+// We don't use the 'content' helpers here because we want json errors to be fatal here
+// (ie, we don't need tombstones and can't just skip a malformed record)
+type IncMetaGlobalBso = IncomingBso;
+type OutMetaGlobalBso = OutgoingBso;
+
+#[derive(Debug)]
+pub struct Sync15StorageClient {
+    tsc: token::TokenProvider,
+    pub(crate) backoff: BackoffListener,
+}
+
+impl SetupStorageClient for Sync15StorageClient {
+    fn fetch_info_configuration(&self) -> error::Result<Sync15ClientResponse<InfoConfiguration>> {
+        self.relative_storage_request(Method::Get, "info/configuration")
+    }
+
+    fn fetch_info_collections(&self) -> error::Result<Sync15ClientResponse<InfoCollections>> {
+        self.relative_storage_request(Method::Get, "info/collections")
+    }
+
+    fn fetch_meta_global(&self) -> error::Result<Sync15ClientResponse<MetaGlobalRecord>> {
+        let got: Sync15ClientResponse<IncMetaGlobalBso> =
+            self.relative_storage_request(Method::Get, "storage/meta/global")?;
+        Ok(match got {
+            Sync15ClientResponse::Success {
+                record,
+                last_modified,
+                route,
+                status,
+            } => {
+                log::debug!(
+                    "Got meta global with modified = {}; last-modified = {}",
+                    record.envelope.modified,
+                    last_modified
+                );
+                Sync15ClientResponse::Success {
+                    record: serde_json::from_str(&record.payload)?,
+                    last_modified,
+                    route,
+                    status,
+                }
+            }
+            Sync15ClientResponse::Error(e) => Sync15ClientResponse::Error(e),
+        })
+    }
+
+    fn fetch_crypto_keys(&self) -> error::Result<Sync15ClientResponse<IncomingEncryptedBso>> {
+        self.relative_storage_request(Method::Get, "storage/crypto/keys")
+    }
+
+    fn put_meta_global(
+        &self,
+        xius: ServerTimestamp,
+        global: &MetaGlobalRecord,
+    ) -> error::Result<ServerTimestamp> {
+        let bso = OutMetaGlobalBso::new(Guid::new("global").into(), global)?;
+        self.put("storage/meta/global", xius, &bso)
+    }
+
+    fn put_crypto_keys(
+        &self,
+        xius: ServerTimestamp,
+        keys: &OutgoingEncryptedBso,
+    ) -> error::Result<()> {
+        self.put("storage/crypto/keys", xius, keys)?;
+        Ok(())
+    }
+
+    fn wipe_all_remote(&self) -> error::Result<()> {
+        let s = self.tsc.api_endpoint()?;
+        let url = Url::parse(&s)?;
+
+        let req = self.build_request(Method::Delete, url)?;
+        match self.exec_request::<Value>(req, false) {
+            Ok(Sync15ClientResponse::Error(ErrorResponse::NotFound { .. }))
+            | Ok(Sync15ClientResponse::Success { .. }) => Ok(()),
+            Ok(resp) => Err(resp.create_storage_error()),
+            Err(e) => Err(e),
+        }
+    }
+}
+
+impl Sync15StorageClient {
+    pub fn new(init_params: Sync15StorageClientInit) -> error::Result<Sync15StorageClient> {
+        rc_crypto::ensure_initialized();
+        let tsc = token::TokenProvider::new(
+            init_params.tokenserver_url,
+            init_params.access_token,
+            init_params.key_id,
+        )?;
+        Ok(Sync15StorageClient {
+            tsc,
+            backoff: new_backoff_listener(),
+        })
+    }
+
+    pub fn get_encrypted_records(
+        &self,
+        collection_request: CollectionRequest,
+    ) -> error::Result<Sync15ClientResponse<Vec<IncomingEncryptedBso>>> {
+        self.collection_request(Method::Get, collection_request)
+    }
+
+    #[inline]
+    fn authorized(&self, req: Request) -> error::Result<Request> {
+        let hawk_header_value = self.tsc.authorization(&req)?;
+        Ok(req.header(AUTHORIZATION, hawk_header_value)?)
+    }
+
+    // TODO: probably want a builder-like API to do collection requests (e.g. something
+    // that occupies roughly the same conceptual role as the Collection class in desktop)
+    fn build_request(&self, method: Method, url: Url) -> error::Result<Request> {
+        self.authorized(Request::new(method, url).header(header_names::ACCEPT, "application/json")?)
+    }
+
+    fn relative_storage_request<P, T>(
+        &self,
+        method: Method,
+        relative_path: P,
+    ) -> error::Result<Sync15ClientResponse<T>>
+    where
+        P: AsRef<str>,
+        for<'a> T: serde::de::Deserialize<'a>,
+    {
+        let s = self.tsc.api_endpoint()? + "/";
+        let url = Url::parse(&s)?.join(relative_path.as_ref())?;
+        self.exec_request(self.build_request(method, url)?, false)
+    }
+
+    fn exec_request<T>(
+        &self,
+        req: Request,
+        require_success: bool,
+    ) -> error::Result<Sync15ClientResponse<T>>
+    where
+        for<'a> T: serde::de::Deserialize<'a>,
+    {
+        log::trace!(
+            "request: {} {} ({:?})",
+            req.method,
+            req.url.path(),
+            req.url.query()
+        );
+        let resp = req.send()?;
+
+        let result = Sync15ClientResponse::from_response(resp, &self.backoff)?;
+        match result {
+            Sync15ClientResponse::Success { .. } => Ok(result),
+            _ => {
+                if require_success {
+                    Err(result.create_storage_error())
+                } else {
+                    Ok(result)
+                }
+            }
+        }
+    }
+
+    fn collection_request<T>(
+        &self,
+        method: Method,
+        r: CollectionRequest,
+    ) -> error::Result<Sync15ClientResponse<T>>
+    where
+        for<'a> T: serde::de::Deserialize<'a>,
+    {
+        let url = build_collection_request_url(Url::parse(&self.tsc.api_endpoint()?)?, r)?;
+        self.exec_request(self.build_request(method, url)?, false)
+    }
+
+    pub fn new_post_queue<'a, F: PostResponseHandler>(
+        &'a self,
+        coll: &'a CollectionName,
+        config: &InfoConfiguration,
+        ts: ServerTimestamp,
+        on_response: F,
+    ) -> error::Result<PostQueue<PostWrapper<'a>, F>> {
+        let pw = PostWrapper { client: self, coll };
+        Ok(PostQueue::new(config, ts, pw, on_response))
+    }
+
+    fn put<P, B>(
+        &self,
+        relative_path: P,
+        xius: ServerTimestamp,
+        body: &B,
+    ) -> error::Result<ServerTimestamp>
+    where
+        P: AsRef<str>,
+        B: serde::ser::Serialize,
+    {
+        let s = self.tsc.api_endpoint()? + "/";
+        let url = Url::parse(&s)?.join(relative_path.as_ref())?;
+
+        let req = self
+            .build_request(Method::Put, url)?
+            .json(body)
+            .header(header_names::X_IF_UNMODIFIED_SINCE, format!("{}", xius))?;
+
+        let resp = self.exec_request::<Value>(req, true)?;
+        // Note: we pass `true` for require_success, so this panic never happens.
+        if let Sync15ClientResponse::Success { last_modified, .. } = resp {
+            Ok(last_modified)
+        } else {
+            unreachable!("Error returned exec_request when `require_success` was true");
+        }
+    }
+
+    pub fn hashed_uid(&self) -> error::Result<String> {
+        self.tsc.hashed_uid()
+    }
+
+    pub(crate) fn wipe_remote_engine(&self, engine: &str) -> error::Result<()> {
+        let s = self.tsc.api_endpoint()? + "/";
+        let url = Url::parse(&s)?.join(&format!("storage/{}", engine))?;
+        log::debug!("Wiping: {:?}", url);
+        let req = self.build_request(Method::Delete, url)?;
+        match self.exec_request::<Value>(req, false) {
+            Ok(Sync15ClientResponse::Error(ErrorResponse::NotFound { .. }))
+            | Ok(Sync15ClientResponse::Success { .. }) => Ok(()),
+            Ok(resp) => Err(resp.create_storage_error()),
+            Err(e) => Err(e),
+        }
+    }
+}
+
+pub struct PostWrapper<'a> {
+    client: &'a Sync15StorageClient,
+    coll: &'a CollectionName,
+}
+
+impl<'a> BatchPoster for PostWrapper<'a> {
+    fn post<T, O>(
+        &self,
+        bytes: Vec<u8>,
+        xius: ServerTimestamp,
+        batch: Option<String>,
+        commit: bool,
+        _: &PostQueue<T, O>,
+    ) -> error::Result<PostResponse> {
+        let r = CollectionPost::new(self.coll.clone())
+            .batch(batch)
+            .commit(commit);
+        let url = build_collection_post_url(Url::parse(&self.client.tsc.api_endpoint()?)?, r)?;
+
+        let req = self
+            .client
+            .build_request(Method::Post, url)?
+            .header(header_names::CONTENT_TYPE, "application/json")?
+            .header(header_names::X_IF_UNMODIFIED_SINCE, format!("{}", xius))?
+            .body(bytes);
+        self.client.exec_request(req, false)
+    }
+}
+
+fn build_collection_url(mut base_url: Url, collection: CollectionName) -> error::Result<Url> {
+    base_url
+        .path_segments_mut()
+        .map_err(|_| Error::UnacceptableUrl("Storage server URL is not a base".to_string()))?
+        .extend(&["storage", &collection]);
+
+    // This is strange but just accessing query_pairs_mut makes you have
+    // a trailing question mark on your url. I don't think anything bad
+    // would happen here, but I don't know, and also, it looks dumb so
+    // I'd rather not have it.
+    if base_url.query() == Some("") {
+        base_url.set_query(None);
+    }
+    Ok(base_url)
+}
+
+fn build_collection_request_url(mut base_url: Url, r: CollectionRequest) -> error::Result<Url> {
+    let mut pairs = base_url.query_pairs_mut();
+    if r.full {
+        pairs.append_pair("full", "1");
+    }
+    if let Some(ids) = &r.ids {
+        // Most ids are 12 characters, and we comma separate them, so 13.
+        let mut buf = String::with_capacity(ids.len() * 13);
+        for (i, id) in ids.iter().enumerate() {
+            if i > 0 {
+                buf.push(',');
+            }
+            buf.push_str(id.as_str());
+        }
+        pairs.append_pair("ids", &buf);
+    }
+    if let Some(ts) = r.older {
+        pairs.append_pair("older", &ts.to_string());
+    }
+    if let Some(ts) = r.newer {
+        pairs.append_pair("newer", &ts.to_string());
+    }
+    if let Some(l) = r.limit {
+        pairs.append_pair("sort", l.order.as_str());
+        pairs.append_pair("limit", &l.num.to_string());
+    }
+    pairs.finish();
+    drop(pairs);
+    build_collection_url(base_url, r.collection)
+}
+
+#[cfg(feature = "sync-client")]
+fn build_collection_post_url(mut base_url: Url, r: CollectionPost) -> error::Result<Url> {
+    let mut pairs = base_url.query_pairs_mut();
+    if let Some(batch) = &r.batch {
+        pairs.append_pair("batch", batch);
+    }
+    if r.commit {
+        pairs.append_pair("commit", "true");
+    }
+    pairs.finish();
+    drop(pairs);
+    build_collection_url(base_url, r.collection)
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    #[test]
+    fn test_send() {
+        fn ensure_send<T: Send>() {}
+        // Compile will fail if not send.
+        ensure_send::<Sync15StorageClient>();
+    }
+
+    #[test]
+    fn test_parse_seconds() {
+        assert_eq!(parse_seconds("1"), Some(1));
+        assert_eq!(parse_seconds("1.4"), Some(2));
+        assert_eq!(parse_seconds("1.5"), Some(2));
+        assert_eq!(parse_seconds("3600.0"), Some(3600));
+        assert_eq!(parse_seconds("3600"), Some(3600));
+        assert_eq!(parse_seconds("-1"), None);
+        assert_eq!(parse_seconds("inf"), None);
+        assert_eq!(parse_seconds("-inf"), None);
+        assert_eq!(parse_seconds("one-thousand"), None);
+        assert_eq!(parse_seconds("4294967295"), Some(4294967295));
+        assert_eq!(parse_seconds("4294967296"), None);
+    }
+
+    #[test]
+    fn test_query_building() {
+        use crate::engine::RequestOrder;
+        let base = Url::parse("https://example.com/sync").unwrap();
+
+        let empty =
+            build_collection_request_url(base.clone(), CollectionRequest::new("foo".into()))
+                .unwrap();
+        assert_eq!(empty.as_str(), "https://example.com/sync/storage/foo");
+
+        let idreq = build_collection_request_url(
+            base.clone(),
+            CollectionRequest::new("wutang".into())
+                .full()
+                .ids(&["rza", "gza"]),
+        )
+        .unwrap();
+        assert_eq!(
+            idreq.as_str(),
+            "https://example.com/sync/storage/wutang?full=1&ids=rza%2Cgza"
+        );
+
+        let complex = build_collection_request_url(
+            base,
+            CollectionRequest::new("specific".into())
+                .full()
+                .limit(10, RequestOrder::Oldest)
+                .older_than(ServerTimestamp(9_876_540))
+                .newer_than(ServerTimestamp(1_234_560)),
+        )
+        .unwrap();
+        assert_eq!(complex.as_str(),
+            "https://example.com/sync/storage/specific?full=1&older=9876.54&newer=1234.56&sort=oldest&limit=10");
+    }
+
+    #[cfg(feature = "sync-client")]
+    #[test]
+    fn test_post_query_building() {
+        let base = Url::parse("https://example.com/sync").unwrap();
+
+        let empty =
+            build_collection_post_url(base.clone(), CollectionPost::new("foo".into())).unwrap();
+        assert_eq!(empty.as_str(), "https://example.com/sync/storage/foo");
+        let batch_start = build_collection_post_url(
+            base.clone(),
+            CollectionPost::new("bar".into())
+                .batch(Some("true".into()))
+                .commit(false),
+        )
+        .unwrap();
+        assert_eq!(
+            batch_start.as_str(),
+            "https://example.com/sync/storage/bar?batch=true"
+        );
+        let batch_commit = build_collection_post_url(
+            base,
+            CollectionPost::new("asdf".into())
+                .batch(Some("1234abc".into()))
+                .commit(true),
+        )
+        .unwrap();
+        assert_eq!(
+            batch_commit.as_str(),
+            "https://example.com/sync/storage/asdf?batch=1234abc&commit=true"
+        );
+    }
+}
diff --git a/third_party/rust/sync15/src/client/sync.rs b/third_party/rust/sync15/src/client/sync.rs
new file mode 100644
index 0000000000..55bcb491e9
--- /dev/null
+++ b/third_party/rust/sync15/src/client/sync.rs
@@ -0,0 +1,94 @@
+/* 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/. */
+
+use super::{CollectionUpdate, GlobalState, LocalCollStateMachine, Sync15StorageClient};
+use crate::clients_engine;
+use crate::engine::{IncomingChangeset, SyncEngine};
+use crate::error::Error;
+use crate::telemetry;
+use crate::KeyBundle;
+use interrupt_support::Interruptee;
+
+#[allow(clippy::too_many_arguments)]
+pub fn synchronize_with_clients_engine(
+    client: &Sync15StorageClient,
+    global_state: &GlobalState,
+    root_sync_key: &KeyBundle,
+    clients: Option<&clients_engine::Engine<'_>>,
+    engine: &dyn SyncEngine,
+    fully_atomic: bool,
+    telem_engine: &mut telemetry::Engine,
+    interruptee: &dyn Interruptee,
+) -> Result<(), Error> {
+    let collection = engine.collection_name();
+    log::info!("Syncing collection {}", collection);
+
+    // our global state machine is ready - get the collection machine going.
+    let coll_state = match LocalCollStateMachine::get_state(engine, global_state, root_sync_key)? {
+        Some(coll_state) => coll_state,
+        None => {
+            // XXX - this is either "error" or "declined".
+            log::warn!(
+                "can't setup for the {} collection - hopefully it works later",
+                collection
+            );
+            return Ok(());
+        }
+    };
+
+    if let Some(clients) = clients {
+        engine.prepare_for_sync(&|| clients.get_client_data())?;
+    }
+
+    let collection_requests = engine.get_collection_requests(coll_state.last_modified)?;
+    let incoming = if collection_requests.is_empty() {
+        log::info!("skipping incoming for {} - not needed.", collection);
+        vec![IncomingChangeset::new(collection, coll_state.last_modified)]
+    } else {
+        assert_eq!(collection_requests.last().unwrap().collection, collection);
+
+        let count = collection_requests.len();
+        collection_requests
+            .into_iter()
+            .enumerate()
+            .map(|(idx, collection_request)| {
+                interruptee.err_if_interrupted()?;
+                let incoming_changes =
+                    super::fetch_incoming(client, &coll_state, collection_request)?;
+
+                log::info!(
+                    "Downloaded {} remote changes (request {} of {})",
+                    incoming_changes.changes.len(),
+                    idx,
+                    count,
+                );
+                Ok(incoming_changes)
+            })
+            .collect::<Result<Vec<_>, Error>>()?
+    };
+
+    let outgoing = engine.apply_incoming(incoming, telem_engine)?;
+    interruptee.err_if_interrupted()?;
+    log::info!("Uploading {} outgoing changes", outgoing.changes.len());
+    let upload_info =
+        CollectionUpdate::new_from_changeset(client, &coll_state, outgoing, fully_atomic)?
+            .upload()?;
+    log::info!(
+        "Upload success ({} records success, {} records failed)",
+        upload_info.successful_ids.len(),
+        upload_info.failed_ids.len()
+    );
+
+    // ideally we'd report this per-batch, but for now, let's just report it
+    // as a total.
+    let mut telem_outgoing = telemetry::EngineOutgoing::new();
+    telem_outgoing.sent(upload_info.successful_ids.len() + upload_info.failed_ids.len());
+    telem_outgoing.failed(upload_info.failed_ids.len());
+    telem_engine.outgoing(telem_outgoing);
+
+    engine.sync_finished(upload_info.modified_timestamp, upload_info.successful_ids)?;
+
+    log::info!("Sync finished!");
+    Ok(())
+}
diff --git a/third_party/rust/sync15/src/client/sync_multiple.rs b/third_party/rust/sync15/src/client/sync_multiple.rs
new file mode 100644
index 0000000000..79ddceff3c
--- /dev/null
+++ b/third_party/rust/sync15/src/client/sync_multiple.rs
@@ -0,0 +1,493 @@
+/* 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/. */
+
+// This helps you perform a sync of multiple engines and helps you manage
+// global and local state between syncs.
+
+use super::state::{EngineChangesNeeded, GlobalState, PersistedGlobalState, SetupStateMachine};
+use super::status::{ServiceStatus, SyncResult};
+use super::storage_client::{BackoffListener, Sync15StorageClient, Sync15StorageClientInit};
+use crate::clients_engine::{self, CommandProcessor, CLIENTS_TTL_REFRESH};
+use crate::engine::{EngineSyncAssociation, SyncEngine};
+use crate::error::Error;
+use crate::telemetry;
+use crate::KeyBundle;
+use interrupt_support::Interruptee;
+use std::collections::HashMap;
+use std::mem;
+use std::result;
+use std::time::{Duration, SystemTime};
+
+/// Info about the client to use. We reuse the client unless
+/// we discover the client_init has changed, in which case we re-create one.
+#[derive(Debug)]
+struct ClientInfo {
+    // the client_init used to create `client`.
+    client_init: Sync15StorageClientInit,
+    // the client (our tokenserver state machine state, and our http library's state)
+    client: Sync15StorageClient,
+}
+
+impl ClientInfo {
+    fn new(ci: &Sync15StorageClientInit) -> Result<Self, Error> {
+        Ok(Self {
+            client_init: ci.clone(),
+            client: Sync15StorageClient::new(ci.clone())?,
+        })
+    }
+}
+
+/// Info we want callers to engine *in memory* for us so that subsequent
+/// syncs are faster. This should never be persisted to storage as it holds
+/// sensitive information, such as the sync decryption keys.
+#[derive(Debug, Default)]
+pub struct MemoryCachedState {
+    last_client_info: Option<ClientInfo>,
+    last_global_state: Option<GlobalState>,
+    // These are just engined in memory, as persisting an invalid value far in the
+    // future has the potential to break sync for good.
+    next_sync_after: Option<SystemTime>,
+    next_client_refresh_after: Option<SystemTime>,
+}
+
+impl MemoryCachedState {
+    // Called we notice the cached state is stale.
+    pub fn clear_sensitive_info(&mut self) {
+        self.last_client_info = None;
+        self.last_global_state = None;
+        // Leave the backoff time, as there's no reason to think it's not still
+        // true.
+    }
+    pub fn get_next_sync_after(&self) -> Option<SystemTime> {
+        self.next_sync_after
+    }
+    pub fn should_refresh_client(&self) -> bool {
+        match self.next_client_refresh_after {
+            Some(t) => SystemTime::now() > t,
+            None => true,
+        }
+    }
+    pub fn note_client_refresh(&mut self) {
+        self.next_client_refresh_after =
+            Some(SystemTime::now() + Duration::from_secs(CLIENTS_TTL_REFRESH));
+    }
+}
+
+/// Sync multiple engines
+/// * `engines` - The engines to sync
+/// * `persisted_global_state` - The global state to use, or None if never
+///   before provided. At the end of the sync, and even when the sync fails,
+///   the value in this cell should be persisted to permanent storage and
+///   provided next time the sync is called.
+/// * `last_client_info` - The client state to use, or None if never before
+///   provided. At the end of the sync, the value should be persisted
+///   *in memory only* - it should not be persisted to disk.
+/// * `storage_init` - Information about how the sync http client should be
+///   configured.
+/// * `root_sync_key` - The KeyBundle used for encryption.
+///
+/// Returns a map, keyed by name and holding an error value - if any engine
+/// fails, the sync will continue on to other engines, but the error will be
+/// places in this map. The absence of a name in the map implies the engine
+/// succeeded.
+pub fn sync_multiple(
+    engines: &[&dyn SyncEngine],
+    persisted_global_state: &mut Option<String>,
+    mem_cached_state: &mut MemoryCachedState,
+    storage_init: &Sync15StorageClientInit,
+    root_sync_key: &KeyBundle,
+    interruptee: &dyn Interruptee,
+    req_info: Option<SyncRequestInfo<'_>>,
+) -> SyncResult {
+    sync_multiple_with_command_processor(
+        None,
+        engines,
+        persisted_global_state,
+        mem_cached_state,
+        storage_init,
+        root_sync_key,
+        interruptee,
+        req_info,
+    )
+}
+
+/// Like `sync_multiple`, but specifies an optional command processor to handle
+/// commands from the clients collection. This function is called by the sync
+/// manager, which provides its own processor.
+#[allow(clippy::too_many_arguments)]
+pub fn sync_multiple_with_command_processor(
+    command_processor: Option<&dyn CommandProcessor>,
+    engines: &[&dyn SyncEngine],
+    persisted_global_state: &mut Option<String>,
+    mem_cached_state: &mut MemoryCachedState,
+    storage_init: &Sync15StorageClientInit,
+    root_sync_key: &KeyBundle,
+    interruptee: &dyn Interruptee,
+    req_info: Option<SyncRequestInfo<'_>>,
+) -> SyncResult {
+    log::info!("Syncing {} engines", engines.len());
+    let mut sync_result = SyncResult {
+        service_status: ServiceStatus::OtherError,
+        result: Ok(()),
+        declined: None,
+        next_sync_after: None,
+        engine_results: HashMap::with_capacity(engines.len()),
+        telemetry: telemetry::SyncTelemetryPing::new(),
+    };
+    let backoff = super::storage_client::new_backoff_listener();
+    let req_info = req_info.unwrap_or_default();
+    let driver = SyncMultipleDriver {
+        command_processor,
+        engines,
+        storage_init,
+        interruptee,
+        engines_to_state_change: req_info.engines_to_state_change,
+        backoff: backoff.clone(),
+        root_sync_key,
+        result: &mut sync_result,
+        persisted_global_state,
+        mem_cached_state,
+        saw_auth_error: false,
+        ignore_soft_backoff: req_info.is_user_action,
+    };
+    match driver.sync() {
+        Ok(()) => {
+            log::debug!(
+                "sync was successful, final status={:?}",
+                sync_result.service_status
+            );
+        }
+        Err(e) => {
+            log::warn!(
+                "sync failed: {}, final status={:?}",
+                e,
+                sync_result.service_status,
+            );
+            sync_result.result = Err(e);
+        }
+    }
+    // Respect `backoff` value when computing the next sync time even if we were
+    // ignoring it during the sync
+    sync_result.set_sync_after(backoff.get_required_wait(false).unwrap_or_default());
+    mem_cached_state.next_sync_after = sync_result.next_sync_after;
+    log::trace!("Sync result: {:?}", sync_result);
+    sync_result
+}
+
+/// This is essentially a bag of information that the sync manager knows, but
+/// otherwise we won't. It should probably be rethought if it gains many more
+/// fields.
+#[derive(Debug, Default)]
+pub struct SyncRequestInfo<'a> {
+    pub engines_to_state_change: Option<&'a HashMap<String, bool>>,
+    pub is_user_action: bool,
+}
+
+// The sync multiple driver
+struct SyncMultipleDriver<'info, 'res, 'pgs, 'mcs> {
+    command_processor: Option<&'info dyn CommandProcessor>,
+    engines: &'info [&'info dyn SyncEngine],
+    storage_init: &'info Sync15StorageClientInit,
+    root_sync_key: &'info KeyBundle,
+    interruptee: &'info dyn Interruptee,
+    backoff: BackoffListener,
+    engines_to_state_change: Option<&'info HashMap<String, bool>>,
+    result: &'res mut SyncResult,
+    persisted_global_state: &'pgs mut Option<String>,
+    mem_cached_state: &'mcs mut MemoryCachedState,
+    ignore_soft_backoff: bool,
+    saw_auth_error: bool,
+}
+
+impl<'info, 'res, 'pgs, 'mcs> SyncMultipleDriver<'info, 'res, 'pgs, 'mcs> {
+    /// The actual worker for sync_multiple.
+    fn sync(mut self) -> result::Result<(), Error> {
+        log::info!("Loading/initializing persisted state");
+        let mut pgs = self.prepare_persisted_state();
+
+        log::info!("Preparing client info");
+        let client_info = self.prepare_client_info()?;
+
+        if self.was_interrupted() {
+            return Ok(());
+        }
+
+        log::info!("Entering sync state machine");
+        // Advance the state machine to the point where it can perform a full
+        // sync. This may involve uploading meta/global, crypto/keys etc.
+        let mut global_state = self.run_state_machine(&client_info, &mut pgs)?;
+
+        if self.was_interrupted() {
+            return Ok(());
+        }
+
+        // Set the service status to OK here - we may adjust it based on an individual
+        // engine failing.
+        self.result.service_status = ServiceStatus::Ok;
+
+        let clients_engine = if let Some(command_processor) = self.command_processor {
+            log::info!("Synchronizing clients engine");
+            let should_refresh = self.mem_cached_state.should_refresh_client();
+            let mut engine = clients_engine::Engine::new(command_processor, self.interruptee);
+            if let Err(e) = engine.sync(
+                &client_info.client,
+                &global_state,
+                self.root_sync_key,
+                should_refresh,
+            ) {
+                // Record telemetry with the error just in case...
+                let mut telem_sync = telemetry::SyncTelemetry::new();
+                let mut telem_engine = telemetry::Engine::new("clients");
+                telem_engine.failure(&e);
+                telem_sync.engine(telem_engine);
+                self.result.service_status = ServiceStatus::from_err(&e);
+
+                // ...And bail, because a clients engine sync failure is fatal.
+                return Err(e);
+            }
+            // We don't record telemetry for successful clients engine
+            // syncs, since we only keep client records in memory, we
+            // expect the counts to be the same most times, and a
+            // failure aborts the entire sync.
+            if self.was_interrupted() {
+                return Ok(());
+            }
+            self.mem_cached_state.note_client_refresh();
+            Some(engine)
+        } else {
+            None
+        };
+
+        log::info!("Synchronizing engines");
+
+        let telem_sync =
+            self.sync_engines(&client_info, &mut global_state, clients_engine.as_ref());
+        self.result.telemetry.sync(telem_sync);
+
+        log::info!("Finished syncing engines.");
+
+        if !self.saw_auth_error {
+            log::trace!("Updating persisted global state");
+            self.mem_cached_state.last_client_info = Some(client_info);
+            self.mem_cached_state.last_global_state = Some(global_state);
+        }
+
+        Ok(())
+    }
+
+    fn was_interrupted(&mut self) -> bool {
+        if self.interruptee.was_interrupted() {
+            log::info!("Interrupted, bailing out");
+            self.result.service_status = ServiceStatus::Interrupted;
+            true
+        } else {
+            false
+        }
+    }
+
+    fn sync_engines(
+        &mut self,
+        client_info: &ClientInfo,
+        global_state: &mut GlobalState,
+        clients: Option<&clients_engine::Engine<'_>>,
+    ) -> telemetry::SyncTelemetry {
+        let mut telem_sync = telemetry::SyncTelemetry::new();
+        for engine in self.engines {
+            let name = engine.collection_name();
+            if self
+                .backoff
+                .get_required_wait(self.ignore_soft_backoff)
+                .is_some()
+            {
+                log::warn!("Got backoff, bailing out of sync early");
+                break;
+            }
+            if global_state.global.declined.iter().any(|e| e == &*name) {
+                log::info!("The {} engine is declined. Skipping", name);
+                continue;
+            }
+            log::info!("Syncing {} engine!", name);
+
+            let mut telem_engine = telemetry::Engine::new(&*name);
+            let result = super::sync::synchronize_with_clients_engine(
+                &client_info.client,
+                global_state,
+                self.root_sync_key,
+                clients,
+                *engine,
+                true,
+                &mut telem_engine,
+                self.interruptee,
+            );
+
+            match result {
+                Ok(()) => log::info!("Sync of {} was successful!", name),
+                Err(ref e) => {
+                    log::warn!("Sync of {} failed! {:?}", name, e);
+                    let this_status = ServiceStatus::from_err(e);
+                    // The only error which forces us to discard our state is an
+                    // auth error.
+                    self.saw_auth_error =
+                        self.saw_auth_error || this_status == ServiceStatus::AuthenticationError;
+                    telem_engine.failure(e);
+                    // If the failure from the engine looks like anything other than
+                    // a "engine error" we don't bother trying the others.
+                    if this_status != ServiceStatus::OtherError {
+                        telem_sync.engine(telem_engine);
+                        self.result.engine_results.insert(name.into(), result);
+                        self.result.service_status = this_status;
+                        break;
+                    }
+                }
+            }
+            telem_sync.engine(telem_engine);
+            self.result.engine_results.insert(name.into(), result);
+            if self.was_interrupted() {
+                break;
+            }
+        }
+        telem_sync
+    }
+
+    fn run_state_machine(
+        &mut self,
+        client_info: &ClientInfo,
+        pgs: &mut PersistedGlobalState,
+    ) -> result::Result<GlobalState, Error> {
+        let last_state = mem::replace(&mut self.mem_cached_state.last_global_state, None);
+
+        let mut state_machine = SetupStateMachine::for_full_sync(
+            &client_info.client,
+            self.root_sync_key,
+            pgs,
+            self.engines_to_state_change,
+            self.interruptee,
+        );
+
+        log::info!("Advancing state machine to ready (full)");
+        let res = state_machine.run_to_ready(last_state);
+        // Grab this now even though we don't need it until later to avoid a
+        // lifetime issue
+        let changes = state_machine.changes_needed.take();
+        // The state machine might have updated our persisted_global_state, so
+        // update the caller's repr of it.
+        *self.persisted_global_state = Some(serde_json::to_string(&pgs)?);
+
+        // Now that we've gone through the state machine, engine the declined list in
+        // the sync_result
+        self.result.declined = Some(pgs.get_declined().to_vec());
+        log::debug!(
+            "Declined engines list after state machine set to: {:?}",
+            self.result.declined,
+        );
+
+        if let Some(c) = changes {
+            self.wipe_or_reset_engines(c, &client_info.client)?;
+        }
+        let state = match res {
+            Err(e) => {
+                self.result.service_status = ServiceStatus::from_err(&e);
+                return Err(e);
+            }
+            Ok(state) => state,
+        };
+        self.result.telemetry.uid(client_info.client.hashed_uid()?);
+        // As for client_info, put None back now so we start from scratch on error.
+        self.mem_cached_state.last_global_state = None;
+        Ok(state)
+    }
+
+    fn wipe_or_reset_engines(
+        &mut self,
+        changes: EngineChangesNeeded,
+        client: &Sync15StorageClient,
+    ) -> result::Result<(), Error> {
+        if changes.local_resets.is_empty() && changes.remote_wipes.is_empty() {
+            return Ok(());
+        }
+        for e in &changes.remote_wipes {
+            log::info!("Engine {:?} just got disabled locally, wiping server", e);
+            client.wipe_remote_engine(e)?;
+        }
+
+        for s in self.engines {
+            let name = s.collection_name();
+            if changes.local_resets.contains(&*name) {
+                log::info!("Resetting engine {}, as it was declined remotely", name);
+                s.reset(&EngineSyncAssociation::Disconnected)?;
+            }
+        }
+
+        Ok(())
+    }
+
+    fn prepare_client_info(&mut self) -> result::Result<ClientInfo, Error> {
+        let mut client_info = match mem::replace(&mut self.mem_cached_state.last_client_info, None)
+        {
+            Some(client_info) => {
+                // if our storage_init has changed it probably means the user has
+                // changed, courtesy of the 'kid' in the structure. Thus, we can't
+                // reuse the client or the memory cached state. We do keep the disk
+                // state as currently that's only the declined list.
+                if client_info.client_init != *self.storage_init {
+                    log::info!("Discarding all state as the account might have changed");
+                    *self.mem_cached_state = MemoryCachedState::default();
+                    ClientInfo::new(self.storage_init)?
+                } else {
+                    log::debug!("Reusing memory-cached client_info");
+                    // we can reuse it (which should be the common path)
+                    client_info
+                }
+            }
+            None => {
+                log::debug!("mem_cached_state was stale or missing, need setup");
+                // We almost certainly have no other state here, but to be safe, we
+                // throw away any memory state we do have.
+                self.mem_cached_state.clear_sensitive_info();
+                ClientInfo::new(self.storage_init)?
+            }
+        };
+        // Ensure we use the correct listener here rather than on all the branches
+        // above, since it seems less error prone.
+        client_info.client.backoff = self.backoff.clone();
+        Ok(client_info)
+    }
+
+    fn prepare_persisted_state(&mut self) -> PersistedGlobalState {
+        // Note that any failure to use a persisted state means we also decline
+        // to use our memory cached state, so that we fully rebuild that
+        // persisted state for next time.
+        match self.persisted_global_state {
+            Some(persisted_string) if !persisted_string.is_empty() => {
+                match serde_json::from_str::<PersistedGlobalState>(persisted_string) {
+                    Ok(state) => {
+                        log::trace!("Read persisted state: {:?}", state);
+                        // Note that we don't set `result.declined` from the
+                        // data in state - it remains None, which explicitly
+                        // indicates "we don't have updated info".
+                        state
+                    }
+                    _ => {
+                        // Don't log the error since it might contain sensitive
+                        // info (although currently it only contains the declined engines list)
+                        error_support::report_error!(
+                            "sync15-prepare-persisted-state",
+                            "Failed to parse PersistedGlobalState from JSON! Falling back to default"
+                        );
+                        *self.mem_cached_state = MemoryCachedState::default();
+                        PersistedGlobalState::default()
+                    }
+                }
+            }
+            _ => {
+                log::info!(
+                    "The application didn't give us persisted state - \
+                     this is only expected on the very first run for a given user."
+                );
+                *self.mem_cached_state = MemoryCachedState::default();
+                PersistedGlobalState::default()
+            }
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/client/token.rs b/third_party/rust/sync15/src/client/token.rs
new file mode 100644
index 0000000000..b416c0c12a
--- /dev/null
+++ b/third_party/rust/sync15/src/client/token.rs
@@ -0,0 +1,602 @@
+/* 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/. */
+
+use crate::error::{self, Error as ErrorKind, Result};
+use crate::ServerTimestamp;
+use rc_crypto::hawk;
+use serde_derive::*;
+use std::borrow::{Borrow, Cow};
+use std::cell::RefCell;
+use std::fmt;
+use std::time::{Duration, SystemTime};
+use url::Url;
+use viaduct::{header_names, Request};
+
+const RETRY_AFTER_DEFAULT_MS: u64 = 10000;
+
+// The TokenserverToken is the token as received directly from the token server
+// and deserialized from JSON.
+#[derive(Deserialize, Clone, PartialEq, Eq)]
+struct TokenserverToken {
+    id: String,
+    key: String,
+    api_endpoint: String,
+    uid: u64,
+    duration: u64,
+    hashed_fxa_uid: String,
+}
+
+impl std::fmt::Debug for TokenserverToken {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("TokenserverToken")
+            .field("api_endpoint", &self.api_endpoint)
+            .field("uid", &self.uid)
+            .field("duration", &self.duration)
+            .field("hashed_fxa_uid", &self.hashed_fxa_uid)
+            .finish()
+    }
+}
+
+// The struct returned by the TokenFetcher - the token itself and the
+// server timestamp.
+struct TokenFetchResult {
+    token: TokenserverToken,
+    server_timestamp: ServerTimestamp,
+}
+
+// The trait for fetching tokens - we'll provide a "real" implementation but
+// tests will re-implement it.
+trait TokenFetcher {
+    fn fetch_token(&self) -> crate::Result<TokenFetchResult>;
+    // We allow the trait to tell us what the time is so tests can get funky.
+    fn now(&self) -> SystemTime;
+}
+
+// Our "real" token fetcher, implementing the TokenFetcher trait, which hits
+// the token server
+#[derive(Debug)]
+struct TokenServerFetcher {
+    // The stuff needed to fetch a token.
+    server_url: Url,
+    access_token: String,
+    key_id: String,
+}
+
+fn fixup_server_url(mut url: Url) -> url::Url {
+    // The given `url` is the end-point as returned by .well-known/fxa-client-configuration,
+    // or as directly specified by self-hosters. As a result, it may or may not have
+    // the sync 1.5 suffix of "/1.0/sync/1.5", so add it on here if it does not.
+    if url.as_str().ends_with("1.0/sync/1.5") {
+        // ok!
+    } else if url.as_str().ends_with("1.0/sync/1.5/") {
+        // Shouldn't ever be Err() here, but the result is `Result<PathSegmentsMut, ()>`
+        // and I don't want to unwrap or add a new error type just for PathSegmentsMut failing.
+        if let Ok(mut path) = url.path_segments_mut() {
+            path.pop();
+        }
+    } else {
+        // We deliberately don't use `.join()` here in order to preserve all path components.
+        // For example, "http://example.com/token" should produce "http://example.com/token/1.0/sync/1.5"
+        // but using `.join()` would produce "http://example.com/1.0/sync/1.5".
+        if let Ok(mut path) = url.path_segments_mut() {
+            path.pop_if_empty();
+            path.extend(&["1.0", "sync", "1.5"]);
+        }
+    };
+    url
+}
+
+impl TokenServerFetcher {
+    fn new(base_url: Url, access_token: String, key_id: String) -> TokenServerFetcher {
+        TokenServerFetcher {
+            server_url: fixup_server_url(base_url),
+            access_token,
+            key_id,
+        }
+    }
+}
+
+impl TokenFetcher for TokenServerFetcher {
+    fn fetch_token(&self) -> Result<TokenFetchResult> {
+        log::debug!("Fetching token from {}", self.server_url);
+        let resp = Request::get(self.server_url.clone())
+            .header(
+                header_names::AUTHORIZATION,
+                format!("Bearer {}", self.access_token),
+            )?
+            .header(header_names::X_KEYID, self.key_id.clone())?
+            .send()?;
+
+        if !resp.is_success() {
+            log::warn!("Non-success status when fetching token: {}", resp.status);
+            // TODO: the body should be JSON and contain a status parameter we might need?
+            log::trace!("  Response body {}", resp.text());
+            // XXX - shouldn't we "chain" these errors - ie, a BackoffError could
+            // have a TokenserverHttpError as its cause?
+            if let Some(res) = resp.headers.get_as::<f64, _>(header_names::RETRY_AFTER) {
+                let ms = res
+                    .ok()
+                    .map_or(RETRY_AFTER_DEFAULT_MS, |f| (f * 1000f64) as u64);
+                let when = self.now() + Duration::from_millis(ms);
+                return Err(ErrorKind::BackoffError(when));
+            }
+            let status = resp.status;
+            return Err(ErrorKind::TokenserverHttpError(status));
+        }
+
+        let token: TokenserverToken = resp.json()?;
+        let server_timestamp = resp
+            .headers
+            .try_get::<ServerTimestamp, _>(header_names::X_TIMESTAMP)
+            .ok_or(ErrorKind::MissingServerTimestamp)?;
+        Ok(TokenFetchResult {
+            token,
+            server_timestamp,
+        })
+    }
+
+    fn now(&self) -> SystemTime {
+        SystemTime::now()
+    }
+}
+
+// The context stored by our TokenProvider when it has a TokenState::Token
+// state.
+struct TokenContext {
+    token: TokenserverToken,
+    credentials: hawk::Credentials,
+    server_timestamp: ServerTimestamp,
+    valid_until: SystemTime,
+}
+
+// hawk::Credentials doesn't implement debug -_-
+impl fmt::Debug for TokenContext {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> ::std::result::Result<(), fmt::Error> {
+        f.debug_struct("TokenContext")
+            .field("token", &self.token)
+            .field("credentials", &"(omitted)")
+            .field("server_timestamp", &self.server_timestamp)
+            .field("valid_until", &self.valid_until)
+            .finish()
+    }
+}
+
+impl TokenContext {
+    fn new(
+        token: TokenserverToken,
+        credentials: hawk::Credentials,
+        server_timestamp: ServerTimestamp,
+        valid_until: SystemTime,
+    ) -> Self {
+        Self {
+            token,
+            credentials,
+            server_timestamp,
+            valid_until,
+        }
+    }
+
+    fn is_valid(&self, now: SystemTime) -> bool {
+        // We could consider making the duration a little shorter - if it
+        // only has 1 second validity there seems a reasonable chance it will
+        // have expired by the time it gets presented to the remote that wants
+        // it.
+        // Either way though, we will eventually need to handle a token being
+        // rejected as a non-fatal error and recover, so maybe we don't care?
+        now < self.valid_until
+    }
+
+    fn authorization(&self, req: &Request) -> Result<String> {
+        let url = &req.url;
+
+        let path_and_query = match url.query() {
+            None => Cow::from(url.path()),
+            Some(qs) => Cow::from(format!("{}?{}", url.path(), qs)),
+        };
+
+        let host = url
+            .host_str()
+            .ok_or_else(|| ErrorKind::UnacceptableUrl("Storage URL has no host".into()))?;
+
+        // Known defaults exist for https? (among others), so this should be impossible
+        let port = url.port_or_known_default().ok_or_else(|| {
+            ErrorKind::UnacceptableUrl(
+                "Storage URL has no port and no default port is known for the protocol".into(),
+            )
+        })?;
+
+        let header =
+            hawk::RequestBuilder::new(req.method.as_str(), host, port, path_and_query.borrow())
+                .request()
+                .make_header(&self.credentials)?;
+
+        Ok(format!("Hawk {}", header))
+    }
+}
+
+// The state our TokenProvider holds to reflect the state of the token.
+#[derive(Debug)]
+enum TokenState {
+    // We've never fetched a token.
+    NoToken,
+    // Have a token and last we checked it remained valid.
+    Token(TokenContext),
+    // We failed to fetch a token. First elt is the error, second elt is
+    // the api_endpoint we had before we failed to fetch a new token (or
+    // None if the very first attempt at fetching a token failed)
+    Failed(Option<error::Error>, Option<String>),
+    // Previously failed and told to back-off for SystemTime duration. Second
+    // elt is the api_endpoint we had before we hit the backoff error.
+    // XXX - should we roll Backoff and Failed together?
+    Backoff(SystemTime, Option<String>),
+    // api_endpoint changed - we are never going to get a token nor move out
+    // of this state.
+    NodeReassigned,
+}
+
+/// The generic TokenProvider implementation - long lived and fetches tokens
+/// on demand (eg, when first needed, or when an existing one expires.)
+#[derive(Debug)]
+struct TokenProviderImpl<TF: TokenFetcher> {
+    fetcher: TF,
+    // Our token state (ie, whether we have a token, and if not, why not)
+    current_state: RefCell<TokenState>,
+}
+
+impl<TF: TokenFetcher> TokenProviderImpl<TF> {
+    fn new(fetcher: TF) -> Self {
+        // We check this at the real entrypoint of the application, but tests
+        // can/do bypass that, so check this here too.
+        rc_crypto::ensure_initialized();
+        TokenProviderImpl {
+            fetcher,
+            current_state: RefCell::new(TokenState::NoToken),
+        }
+    }
+
+    // Uses our fetcher to grab a new token and if successfull, derives other
+    // info from that token into a usable TokenContext.
+    fn fetch_context(&self) -> Result<TokenContext> {
+        let result = self.fetcher.fetch_token()?;
+        let token = result.token;
+        let valid_until = SystemTime::now() + Duration::from_secs(token.duration);
+
+        let credentials = hawk::Credentials {
+            id: token.id.clone(),
+            key: hawk::Key::new(token.key.as_bytes(), hawk::SHA256)?,
+        };
+
+        Ok(TokenContext::new(
+            token,
+            credentials,
+            result.server_timestamp,
+            valid_until,
+        ))
+    }
+
+    // Attempt to fetch a new token and return a new state reflecting that
+    // operation. If it worked a TokenState will be returned, but errors may
+    // cause other states.
+    fn fetch_token(&self, previous_endpoint: Option<&str>) -> TokenState {
+        match self.fetch_context() {
+            Ok(tc) => {
+                // We got a new token - check that the endpoint is the same
+                // as a previous endpoint we saw (if any)
+                match previous_endpoint {
+                    Some(prev) => {
+                        if prev == tc.token.api_endpoint {
+                            TokenState::Token(tc)
+                        } else {
+                            log::warn!(
+                                "api_endpoint changed from {} to {}",
+                                prev,
+                                tc.token.api_endpoint
+                            );
+                            TokenState::NodeReassigned
+                        }
+                    }
+                    None => {
+                        // Never had an api_endpoint in the past, so this is OK.
+                        TokenState::Token(tc)
+                    }
+                }
+            }
+            Err(e) => {
+                // Early to avoid nll issues...
+                if let ErrorKind::BackoffError(be) = e {
+                    return TokenState::Backoff(be, previous_endpoint.map(ToString::to_string));
+                }
+                TokenState::Failed(Some(e), previous_endpoint.map(ToString::to_string))
+            }
+        }
+    }
+
+    // Given the state we are currently in, return a new current state.
+    // Returns None if the current state should be used (eg, if we are
+    // holding a token that remains valid) or Some() if the state has changed
+    // (which may have changed to a state with a token or an error state)
+    fn advance_state(&self, state: &TokenState) -> Option<TokenState> {
+        match state {
+            TokenState::NoToken => Some(self.fetch_token(None)),
+            TokenState::Failed(_, existing_endpoint) => {
+                Some(self.fetch_token(existing_endpoint.as_ref().map(String::as_str)))
+            }
+            TokenState::Token(existing_context) => {
+                if existing_context.is_valid(self.fetcher.now()) {
+                    None
+                } else {
+                    Some(self.fetch_token(Some(existing_context.token.api_endpoint.as_str())))
+                }
+            }
+            TokenState::Backoff(ref until, ref existing_endpoint) => {
+                if let Ok(remaining) = until.duration_since(self.fetcher.now()) {
+                    log::debug!("enforcing existing backoff - {:?} remains", remaining);
+                    None
+                } else {
+                    // backoff period is over
+                    Some(self.fetch_token(existing_endpoint.as_ref().map(String::as_str)))
+                }
+            }
+            TokenState::NodeReassigned => {
+                // We never leave this state.
+                None
+            }
+        }
+    }
+
+    fn with_token<T, F>(&self, func: F) -> Result<T>
+    where
+        F: FnOnce(&TokenContext) -> Result<T>,
+    {
+        // first get a mutable ref to our existing state, advance to the
+        // state we will use, then re-stash that state for next time.
+        let state: &mut TokenState = &mut self.current_state.borrow_mut();
+        if let Some(new_state) = self.advance_state(state) {
+            *state = new_state;
+        }
+
+        // Now re-fetch the state we should use for this call - if it's
+        // anything other than TokenState::Token we will fail.
+        match state {
+            TokenState::NoToken => {
+                // it should be impossible to get here.
+                panic!("Can't be in NoToken state after advancing");
+            }
+            TokenState::Token(ref token_context) => {
+                // make the call.
+                func(token_context)
+            }
+            TokenState::Failed(e, _) => {
+                // We swap the error out of the state enum and return it.
+                Err(e.take().unwrap())
+            }
+            TokenState::NodeReassigned => {
+                // this is unrecoverable.
+                Err(ErrorKind::StorageResetError)
+            }
+            TokenState::Backoff(ref remaining, _) => Err(ErrorKind::BackoffError(*remaining)),
+        }
+    }
+
+    fn hashed_uid(&self) -> Result<String> {
+        self.with_token(|ctx| Ok(ctx.token.hashed_fxa_uid.clone()))
+    }
+
+    fn authorization(&self, req: &Request) -> Result<String> {
+        self.with_token(|ctx| ctx.authorization(req))
+    }
+
+    fn api_endpoint(&self) -> Result<String> {
+        self.with_token(|ctx| Ok(ctx.token.api_endpoint.clone()))
+    }
+    // TODO: we probably want a "drop_token/context" type method so that when
+    // using a token with some validity fails the caller can force a new one
+    // (in which case the new token request will probably fail with a 401)
+}
+
+// The public concrete object exposed by this module
+#[derive(Debug)]
+pub struct TokenProvider {
+    imp: TokenProviderImpl<TokenServerFetcher>,
+}
+
+impl TokenProvider {
+    pub fn new(url: Url, access_token: String, key_id: String) -> Result<Self> {
+        let fetcher = TokenServerFetcher::new(url, access_token, key_id);
+        Ok(Self {
+            imp: TokenProviderImpl::new(fetcher),
+        })
+    }
+
+    pub fn hashed_uid(&self) -> Result<String> {
+        self.imp.hashed_uid()
+    }
+
+    pub fn authorization(&self, req: &Request) -> Result<String> {
+        self.imp.authorization(req)
+    }
+
+    pub fn api_endpoint(&self) -> Result<String> {
+        self.imp.api_endpoint()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::cell::Cell;
+
+    struct TestFetcher<FF, FN>
+    where
+        FF: Fn() -> Result<TokenFetchResult>,
+        FN: Fn() -> SystemTime,
+    {
+        fetch: FF,
+        now: FN,
+    }
+    impl<FF, FN> TokenFetcher for TestFetcher<FF, FN>
+    where
+        FF: Fn() -> Result<TokenFetchResult>,
+        FN: Fn() -> SystemTime,
+    {
+        fn fetch_token(&self) -> Result<TokenFetchResult> {
+            (self.fetch)()
+        }
+        fn now(&self) -> SystemTime {
+            (self.now)()
+        }
+    }
+
+    fn make_tsc<FF, FN>(fetch: FF, now: FN) -> TokenProviderImpl<TestFetcher<FF, FN>>
+    where
+        FF: Fn() -> Result<TokenFetchResult>,
+        FN: Fn() -> SystemTime,
+    {
+        let fetcher: TestFetcher<FF, FN> = TestFetcher { fetch, now };
+        TokenProviderImpl::new(fetcher)
+    }
+
+    #[test]
+    fn test_endpoint() {
+        // Use a cell to avoid the closure having a mutable ref to this scope.
+        let counter: Cell<u32> = Cell::new(0);
+        let fetch = || {
+            counter.set(counter.get() + 1);
+            Ok(TokenFetchResult {
+                token: TokenserverToken {
+                    id: "id".to_string(),
+                    key: "key".to_string(),
+                    api_endpoint: "api_endpoint".to_string(),
+                    uid: 1,
+                    duration: 1000,
+                    hashed_fxa_uid: "hash".to_string(),
+                },
+                server_timestamp: ServerTimestamp(0i64),
+            })
+        };
+
+        let tsc = make_tsc(fetch, SystemTime::now);
+
+        let e = tsc.api_endpoint().expect("should work");
+        assert_eq!(e, "api_endpoint".to_string());
+        assert_eq!(counter.get(), 1);
+
+        let e2 = tsc.api_endpoint().expect("should work");
+        assert_eq!(e2, "api_endpoint".to_string());
+        // should not have re-fetched.
+        assert_eq!(counter.get(), 1);
+    }
+
+    #[test]
+    fn test_backoff() {
+        let counter: Cell<u32> = Cell::new(0);
+        let fetch = || {
+            counter.set(counter.get() + 1);
+            let when = SystemTime::now() + Duration::from_millis(10000);
+            Err(ErrorKind::BackoffError(when))
+        };
+        let now: Cell<SystemTime> = Cell::new(SystemTime::now());
+        let tsc = make_tsc(fetch, || now.get());
+
+        tsc.api_endpoint().expect_err("should bail");
+        // XXX - check error type.
+        assert_eq!(counter.get(), 1);
+        // try and get another token - should not re-fetch as backoff is still
+        // in progress.
+        tsc.api_endpoint().expect_err("should bail");
+        assert_eq!(counter.get(), 1);
+
+        // Advance the clock.
+        now.set(now.get() + Duration::new(20, 0));
+
+        // Our token fetch mock is still returning a backoff error, so we
+        // still fail, but should have re-hit the fetch function.
+        tsc.api_endpoint().expect_err("should bail");
+        assert_eq!(counter.get(), 2);
+    }
+
+    #[test]
+    fn test_validity() {
+        let counter: Cell<u32> = Cell::new(0);
+        let fetch = || {
+            counter.set(counter.get() + 1);
+            Ok(TokenFetchResult {
+                token: TokenserverToken {
+                    id: "id".to_string(),
+                    key: "key".to_string(),
+                    api_endpoint: "api_endpoint".to_string(),
+                    uid: 1,
+                    duration: 10,
+                    hashed_fxa_uid: "hash".to_string(),
+                },
+                server_timestamp: ServerTimestamp(0i64),
+            })
+        };
+        let now: Cell<SystemTime> = Cell::new(SystemTime::now());
+        let tsc = make_tsc(fetch, || now.get());
+
+        tsc.api_endpoint().expect("should get a valid token");
+        assert_eq!(counter.get(), 1);
+
+        // try and get another token - should not re-fetch as the old one
+        // remains valid.
+        tsc.api_endpoint().expect("should reuse existing token");
+        assert_eq!(counter.get(), 1);
+
+        // Advance the clock.
+        now.set(now.get() + Duration::new(20, 0));
+
+        // We should discard our token and fetch a new one.
+        tsc.api_endpoint().expect("should re-fetch");
+        assert_eq!(counter.get(), 2);
+    }
+
+    #[test]
+    fn test_server_url() {
+        assert_eq!(
+            fixup_server_url(
+                Url::parse("https://token.services.mozilla.com/1.0/sync/1.5").unwrap()
+            )
+            .as_str(),
+            "https://token.services.mozilla.com/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(
+                Url::parse("https://token.services.mozilla.com/1.0/sync/1.5/").unwrap()
+            )
+            .as_str(),
+            "https://token.services.mozilla.com/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(Url::parse("https://token.services.mozilla.com").unwrap()).as_str(),
+            "https://token.services.mozilla.com/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(Url::parse("https://token.services.mozilla.com/").unwrap()).as_str(),
+            "https://token.services.mozilla.com/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(
+                Url::parse("https://selfhosted.example.com/token/1.0/sync/1.5").unwrap()
+            )
+            .as_str(),
+            "https://selfhosted.example.com/token/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(
+                Url::parse("https://selfhosted.example.com/token/1.0/sync/1.5/").unwrap()
+            )
+            .as_str(),
+            "https://selfhosted.example.com/token/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(Url::parse("https://selfhosted.example.com/token/").unwrap()).as_str(),
+            "https://selfhosted.example.com/token/1.0/sync/1.5"
+        );
+        assert_eq!(
+            fixup_server_url(Url::parse("https://selfhosted.example.com/token").unwrap()).as_str(),
+            "https://selfhosted.example.com/token/1.0/sync/1.5"
+        );
+    }
+}
diff --git a/third_party/rust/sync15/src/client/util.rs b/third_party/rust/sync15/src/client/util.rs
new file mode 100644
index 0000000000..01fff77afa
--- /dev/null
+++ b/third_party/rust/sync15/src/client/util.rs
@@ -0,0 +1,102 @@
+/* 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/. */
+
+use std::collections::{HashMap, HashSet};
+use std::sync::atomic::{AtomicU32, Ordering};
+
+/// Finds the maximum of the current value and the argument `val`, and sets the
+/// new value to the result.
+///
+/// Note: `AtomicFoo::fetch_max` is unstable, and can't really be implemented as
+/// a single atomic operation from outside the stdlib ;-;
+pub(crate) fn atomic_update_max(v: &AtomicU32, new: u32) {
+    // For loads (and the compare_exchange_weak second ordering argument) this
+    // is too strong, we could probably get away with Acquire (or maybe Relaxed
+    // because we don't need the result?). In either case, this fn isn't called
+    // from a hot spot so whatever.
+    let mut cur = v.load(Ordering::SeqCst);
+    while cur < new {
+        // we're already handling the failure case so there's no reason not to
+        // use _weak here.
+        match v.compare_exchange_weak(cur, new, Ordering::SeqCst, Ordering::SeqCst) {
+            Ok(_) => {
+                // Success.
+                break;
+            }
+            Err(new_cur) => {
+                // Interrupted, keep trying.
+                cur = new_cur
+            }
+        }
+    }
+}
+
+// Slight wrappers around the builtin methods for doing this.
+pub(crate) fn set_union(a: &HashSet<String>, b: &HashSet<String>) -> HashSet<String> {
+    a.union(b).cloned().collect()
+}
+
+pub(crate) fn set_difference(a: &HashSet<String>, b: &HashSet<String>) -> HashSet<String> {
+    a.difference(b).cloned().collect()
+}
+
+pub(crate) fn set_intersection(a: &HashSet<String>, b: &HashSet<String>) -> HashSet<String> {
+    a.intersection(b).cloned().collect()
+}
+
+pub(crate) fn partition_by_value(v: &HashMap<String, bool>) -> (HashSet<String>, HashSet<String>) {
+    let mut true_: HashSet<String> = HashSet::new();
+    let mut false_: HashSet<String> = HashSet::new();
+    for (s, val) in v {
+        if *val {
+            true_.insert(s.clone());
+        } else {
+            false_.insert(s.clone());
+        }
+    }
+    (true_, false_)
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_set_ops() {
+        fn hash_set(s: &[&str]) -> HashSet<String> {
+            s.iter()
+                .copied()
+                .map(ToOwned::to_owned)
+                .collect::<HashSet<_>>()
+        }
+
+        assert_eq!(
+            set_union(&hash_set(&["a", "b", "c"]), &hash_set(&["b", "d"])),
+            hash_set(&["a", "b", "c", "d"]),
+        );
+
+        assert_eq!(
+            set_difference(&hash_set(&["a", "b", "c"]), &hash_set(&["b", "d"])),
+            hash_set(&["a", "c"]),
+        );
+        assert_eq!(
+            set_intersection(&hash_set(&["a", "b", "c"]), &hash_set(&["b", "d"])),
+            hash_set(&["b"]),
+        );
+        let m: HashMap<String, bool> = [
+            ("foo", true),
+            ("bar", true),
+            ("baz", false),
+            ("quux", false),
+        ]
+        .iter()
+        .copied()
+        .map(|(a, b)| (a.to_owned(), b))
+        .collect();
+        assert_eq!(
+            partition_by_value(&m),
+            (hash_set(&["foo", "bar"]), hash_set(&["baz", "quux"])),
+        );
+    }
+}
diff --git a/third_party/rust/sync15/src/client_types.rs b/third_party/rust/sync15/src/client_types.rs
new file mode 100644
index 0000000000..2966a190e2
--- /dev/null
+++ b/third_party/rust/sync15/src/client_types.rs
@@ -0,0 +1,129 @@
+/* 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/. */
+
+//! This module has to be here because of some hard-to-avoid hacks done for the
+//! tabs engine... See issue #2590
+
+use crate::DeviceType;
+use serde::{Deserialize, Serialize};
+use std::collections::HashMap;
+
+/// Argument to Store::prepare_for_sync. See comment there for more info. Only
+/// really intended to be used by tabs engine.
+#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
+pub struct ClientData {
+    pub local_client_id: String,
+    /// A hashmap of records in the `clients` collection. Key is the id of the record in
+    /// that collection, which may or may not be the device's fxa_device_id.
+    pub recent_clients: HashMap<String, RemoteClient>,
+}
+
+/// Information about a remote client in the clients collection.
+#[derive(Clone, Debug, Eq, Hash, PartialEq, Serialize, Deserialize)]
+pub struct RemoteClient {
+    pub fxa_device_id: Option<String>,
+    pub device_name: String,
+    #[serde(default)]
+    pub device_type: DeviceType,
+}
+
+#[cfg(test)]
+mod client_types_tests {
+    use super::*;
+
+    #[test]
+    fn test_remote_client() {
+        // Missing `device_type` gets DeviceType::Unknown.
+        let dt = serde_json::from_str::<RemoteClient>("{\"device_name\": \"foo\"}").unwrap();
+        assert_eq!(dt.device_type, DeviceType::Unknown);
+        // But reserializes as null.
+        assert_eq!(
+            serde_json::to_string(&dt).unwrap(),
+            "{\"fxa_device_id\":null,\"device_name\":\"foo\",\"device_type\":null}"
+        );
+
+        // explicit null is also unknown.
+        assert_eq!(
+            serde_json::from_str::<RemoteClient>(
+                "{\"device_name\": \"foo\", \"device_type\": null}",
+            )
+            .unwrap()
+            .device_type,
+            DeviceType::Unknown
+        );
+
+        // Unknown device_type string deserializes as DeviceType::Unknown.
+        let dt = serde_json::from_str::<RemoteClient>(
+            "{\"device_name\": \"foo\", \"device_type\": \"foo\"}",
+        )
+        .unwrap();
+        assert_eq!(dt.device_type, DeviceType::Unknown);
+        // The None gets re-serialized as null.
+        assert_eq!(
+            serde_json::to_string(&dt).unwrap(),
+            "{\"fxa_device_id\":null,\"device_name\":\"foo\",\"device_type\":null}"
+        );
+
+        // DeviceType::Unknown gets serialized as null.
+        let dt = RemoteClient {
+            device_name: "bar".to_string(),
+            fxa_device_id: None,
+            device_type: DeviceType::Unknown,
+        };
+        assert_eq!(
+            serde_json::to_string(&dt).unwrap(),
+            "{\"fxa_device_id\":null,\"device_name\":\"bar\",\"device_type\":null}"
+        );
+
+        // DeviceType::Desktop gets serialized as "desktop".
+        let dt = RemoteClient {
+            device_name: "bar".to_string(),
+            fxa_device_id: Some("fxa".to_string()),
+            device_type: DeviceType::Desktop,
+        };
+        assert_eq!(
+            serde_json::to_string(&dt).unwrap(),
+            "{\"fxa_device_id\":\"fxa\",\"device_name\":\"bar\",\"device_type\":\"desktop\"}"
+        );
+    }
+
+    #[test]
+    fn test_client_data() {
+        let client_data = ClientData {
+            local_client_id: "my-device".to_string(),
+            recent_clients: HashMap::from([
+                (
+                    "my-device".to_string(),
+                    RemoteClient {
+                        fxa_device_id: None,
+                        device_name: "my device".to_string(),
+                        device_type: DeviceType::Unknown,
+                    },
+                ),
+                (
+                    "device-no-tabs".to_string(),
+                    RemoteClient {
+                        fxa_device_id: None,
+                        device_name: "device with no tabs".to_string(),
+                        device_type: DeviceType::Unknown,
+                    },
+                ),
+                (
+                    "device-with-a-tab".to_string(),
+                    RemoteClient {
+                        fxa_device_id: None,
+                        device_name: "device with a tab".to_string(),
+                        device_type: DeviceType::Desktop,
+                    },
+                ),
+            ]),
+        };
+        //serialize
+        let client_data_ser = serde_json::to_string(&client_data).unwrap();
+        println!("SER: {}", client_data_ser);
+        // deserialize
+        let client_data_des: ClientData = serde_json::from_str(&client_data_ser).unwrap();
+        assert_eq!(client_data_des, client_data);
+    }
+}
diff --git a/third_party/rust/sync15/src/clients_engine/engine.rs b/third_party/rust/sync15/src/clients_engine/engine.rs
new file mode 100644
index 0000000000..eb746d5516
--- /dev/null
+++ b/third_party/rust/sync15/src/clients_engine/engine.rs
@@ -0,0 +1,804 @@
+/* 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/. */
+
+use std::collections::{HashMap, HashSet};
+
+use crate::bso::{IncomingKind, OutgoingBso, OutgoingEnvelope};
+use crate::client::{
+    CollState, CollectionKeys, CollectionUpdate, GlobalState, InfoConfiguration,
+    Sync15StorageClient,
+};
+use crate::client_types::{ClientData, RemoteClient};
+use crate::engine::{CollectionRequest, IncomingChangeset, OutgoingChangeset};
+use crate::{error::Result, Guid, KeyBundle};
+use interrupt_support::Interruptee;
+
+use super::{
+    record::{ClientRecord, CommandRecord},
+    ser::shrink_to_fit,
+    Command, CommandProcessor, CommandStatus, CLIENTS_TTL,
+};
+
+const COLLECTION_NAME: &str = "clients";
+
+/// The driver for the clients engine. Internal; split out from the `Engine`
+/// struct to make testing easier.
+struct Driver<'a> {
+    command_processor: &'a dyn CommandProcessor,
+    interruptee: &'a dyn Interruptee,
+    config: &'a InfoConfiguration,
+    recent_clients: HashMap<String, RemoteClient>,
+}
+
+impl<'a> Driver<'a> {
+    fn new(
+        command_processor: &'a dyn CommandProcessor,
+        interruptee: &'a dyn Interruptee,
+        config: &'a InfoConfiguration,
+    ) -> Driver<'a> {
+        Driver {
+            command_processor,
+            interruptee,
+            config,
+            recent_clients: HashMap::new(),
+        }
+    }
+
+    fn note_recent_client(&mut self, client: &ClientRecord) {
+        self.recent_clients.insert(client.id.clone(), client.into());
+    }
+
+    fn sync(
+        &mut self,
+        inbound: IncomingChangeset,
+        should_refresh_client: bool,
+    ) -> Result<OutgoingChangeset> {
+        self.interruptee.err_if_interrupted()?;
+        let outgoing_commands = self.command_processor.fetch_outgoing_commands()?;
+
+        let mut has_own_client_record = false;
+        let mut changes = Vec::new();
+
+        for bso in inbound.changes {
+            self.interruptee.err_if_interrupted()?;
+
+            let content = bso.into_content();
+
+            let client: ClientRecord = match content.kind {
+                IncomingKind::Malformed => {
+                    log::debug!("Error unpacking record");
+                    continue;
+                }
+                IncomingKind::Tombstone => {
+                    log::debug!("Record has been deleted; skipping...");
+                    continue;
+                }
+                IncomingKind::Content(client) => client,
+            };
+
+            if client.id == self.command_processor.settings().fxa_device_id {
+                log::debug!("Found my record on the server");
+                // If we see our own client record, apply any incoming commands,
+                // remove them from the list, and reupload the record. Any
+                // commands that we don't understand also go back in the list.
+                // https://github.com/mozilla/application-services/issues/1800
+                // tracks if that's the right thing to do.
+                has_own_client_record = true;
+                let mut current_client_record = self.current_client_record();
+                for c in &client.commands {
+                    let status = match c.as_command() {
+                        Some(command) => self.command_processor.apply_incoming_command(command)?,
+                        None => CommandStatus::Unsupported,
+                    };
+                    match status {
+                        CommandStatus::Applied => {}
+                        CommandStatus::Ignored => {
+                            log::debug!("Ignored command {:?}", c);
+                        }
+                        CommandStatus::Unsupported => {
+                            log::warn!("Don't know how to apply command {:?}", c);
+                            current_client_record.commands.push(c.clone());
+                        }
+                    }
+                }
+
+                // The clients collection has a hard limit on the payload size,
+                // after which the server starts rejecting our records. Large
+                // command lists can cause us to exceed this, so we truncate
+                // the list.
+                shrink_to_fit(
+                    &mut current_client_record.commands,
+                    self.memcache_max_record_payload_size(),
+                )?;
+
+                // Add the new client record to our map of recently synced
+                // clients, so that downstream consumers like synced tabs can
+                // access them.
+                self.note_recent_client(&current_client_record);
+
+                // We periodically upload our own client record, even if it
+                // doesn't change, to keep it fresh.
+                if should_refresh_client || client != current_client_record {
+                    log::debug!("Will update our client record on the server");
+                    let envelope = OutgoingEnvelope {
+                        id: content.envelope.id,
+                        ttl: Some(CLIENTS_TTL),
+                        ..Default::default()
+                    };
+                    changes.push(OutgoingBso::from_content(envelope, current_client_record)?);
+                }
+            } else {
+                // Add the other client to our map of recently synced clients.
+                self.note_recent_client(&client);
+
+                // Bail if we don't have any outgoing commands to write into
+                // the other client's record.
+                if outgoing_commands.is_empty() {
+                    continue;
+                }
+
+                // Determine if we have new commands, that aren't already in the
+                // client's command list.
+                let current_commands: HashSet<Command> = client
+                    .commands
+                    .iter()
+                    .filter_map(|c| c.as_command())
+                    .collect();
+                let mut new_outgoing_commands = outgoing_commands
+                    .difference(&current_commands)
+                    .cloned()
+                    .collect::<Vec<_>>();
+                // Sort, to ensure deterministic ordering for tests.
+                new_outgoing_commands.sort();
+                let mut new_client = client.clone();
+                new_client
+                    .commands
+                    .extend(new_outgoing_commands.into_iter().map(CommandRecord::from));
+                if new_client.commands.len() == client.commands.len() {
+                    continue;
+                }
+
+                // Hooray, we added new commands! Make sure the record still
+                // fits in the maximum record size, or the server will reject
+                // our upload.
+                shrink_to_fit(
+                    &mut new_client.commands,
+                    self.memcache_max_record_payload_size(),
+                )?;
+
+                let envelope = OutgoingEnvelope {
+                    id: content.envelope.id,
+                    ttl: Some(CLIENTS_TTL),
+                    ..Default::default()
+                };
+                changes.push(OutgoingBso::from_content(envelope, new_client)?);
+            }
+        }
+
+        // Upload a record for our own client, if we didn't replace it already.
+        if !has_own_client_record {
+            let current_client_record = self.current_client_record();
+            self.note_recent_client(&current_client_record);
+            let envelope = OutgoingEnvelope {
+                id: Guid::new(&current_client_record.id),
+                ttl: Some(CLIENTS_TTL),
+                ..Default::default()
+            };
+            changes.push(OutgoingBso::from_content(envelope, current_client_record)?);
+        }
+
+        Ok(OutgoingChangeset::new(COLLECTION_NAME.into(), changes))
+    }
+
+    /// Builds a fresh client record for this device.
+    fn current_client_record(&self) -> ClientRecord {
+        let settings = self.command_processor.settings();
+        ClientRecord {
+            id: settings.fxa_device_id.clone(),
+            name: settings.device_name.clone(),
+            typ: settings.device_type,
+            commands: Vec::new(),
+            fxa_device_id: Some(settings.fxa_device_id.clone()),
+            version: None,
+            protocols: vec!["1.5".into()],
+            form_factor: None,
+            os: None,
+            app_package: None,
+            application: None,
+            device: None,
+        }
+    }
+
+    fn max_record_payload_size(&self) -> usize {
+        let payload_max = self.config.max_record_payload_bytes;
+        if payload_max <= self.config.max_post_bytes {
+            self.config.max_post_bytes.saturating_sub(4096)
+        } else {
+            payload_max
+        }
+    }
+
+    /// Collections stored in memcached ("tabs", "clients" or "meta") have a
+    /// different max size than ones stored in the normal storage server db.
+    /// In practice, the real limit here is 1M (bug 1300451 comment 40), but
+    /// there's overhead involved that is hard to calculate on the client, so we
+    /// use 512k to be safe (at the recommendation of the server team). Note
+    /// that if the server reports a lower limit (via info/configuration), we
+    /// respect that limit instead. See also bug 1403052.
+    /// XXX - the above comment is stale and refers to the world before the
+    /// move to spanner and the rust sync server.
+    fn memcache_max_record_payload_size(&self) -> usize {
+        self.max_record_payload_size().min(512 * 1024)
+    }
+}
+
+pub struct Engine<'a> {
+    pub command_processor: &'a dyn CommandProcessor,
+    pub interruptee: &'a dyn Interruptee,
+    pub recent_clients: HashMap<String, RemoteClient>,
+}
+
+impl<'a> Engine<'a> {
+    /// Creates a new clients engine that delegates to the given command
+    /// processor to apply incoming commands.
+    pub fn new<'b>(
+        command_processor: &'b dyn CommandProcessor,
+        interruptee: &'b dyn Interruptee,
+    ) -> Engine<'b> {
+        Engine {
+            command_processor,
+            interruptee,
+            recent_clients: HashMap::new(),
+        }
+    }
+
+    /// Syncs the clients collection. This works a little differently than
+    /// other collections:
+    ///
+    ///   1. It can't be disabled or declined.
+    ///   2. The sync ID and last sync time aren't meaningful, since we always
+    ///      fetch all client records on every sync. As such, the
+    ///      `LocalCollStateMachine` that we use for other engines doesn't
+    ///      apply to it.
+    ///   3. It doesn't persist state directly, but relies on the sync manager
+    ///      to persist device settings, and process commands.
+    ///   4. Failing to sync the clients collection is fatal, and aborts the
+    ///      sync.
+    ///
+    /// For these reasons, we implement this engine directly in the `sync15`
+    /// crate, and provide a specialized `sync` method instead of implementing
+    /// `sync15::Store`.
+    pub fn sync(
+        &mut self,
+        storage_client: &Sync15StorageClient,
+        global_state: &GlobalState,
+        root_sync_key: &KeyBundle,
+        should_refresh_client: bool,
+    ) -> Result<()> {
+        log::info!("Syncing collection clients");
+
+        let coll_keys = CollectionKeys::from_encrypted_payload(
+            global_state.keys.clone(),
+            global_state.keys_timestamp,
+            root_sync_key,
+        )?;
+        let coll_state = CollState {
+            config: global_state.config.clone(),
+            last_modified: global_state
+                .collections
+                .get(COLLECTION_NAME)
+                .cloned()
+                .unwrap_or_default(),
+            key: coll_keys.key_for_collection(COLLECTION_NAME).clone(),
+        };
+
+        let inbound = self.fetch_incoming(storage_client, &coll_state)?;
+
+        let mut driver = Driver::new(
+            self.command_processor,
+            self.interruptee,
+            &global_state.config,
+        );
+
+        let outgoing = driver.sync(inbound, should_refresh_client)?;
+        self.recent_clients = driver.recent_clients;
+
+        self.interruptee.err_if_interrupted()?;
+        let upload_info =
+            CollectionUpdate::new_from_changeset(storage_client, &coll_state, outgoing, true)?
+                .upload()?;
+
+        log::info!(
+            "Upload success ({} records success, {} records failed)",
+            upload_info.successful_ids.len(),
+            upload_info.failed_ids.len()
+        );
+
+        log::info!("Finished syncing clients");
+        Ok(())
+    }
+
+    fn fetch_incoming(
+        &self,
+        storage_client: &Sync15StorageClient,
+        coll_state: &CollState,
+    ) -> Result<IncomingChangeset> {
+        // Note that, unlike other stores, we always fetch the full collection
+        // on every sync, so `inbound` will return all clients, not just the
+        // ones that changed since the last sync.
+        let coll_request = CollectionRequest::new(COLLECTION_NAME.into()).full();
+
+        self.interruptee.err_if_interrupted()?;
+        let inbound = crate::client::fetch_incoming(storage_client, coll_state, coll_request)?;
+
+        Ok(inbound)
+    }
+
+    pub fn local_client_id(&self) -> String {
+        // Bit dirty but it's the easiest way to reach to our own
+        // device ID without refactoring the whole sync manager crate.
+        self.command_processor.settings().fxa_device_id.clone()
+    }
+
+    pub fn get_client_data(&self) -> ClientData {
+        ClientData {
+            local_client_id: self.local_client_id(),
+            recent_clients: self.recent_clients.clone(),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::super::{CommandStatus, DeviceType, Settings};
+    use super::*;
+    use crate::bso::IncomingBso;
+    use crate::ServerTimestamp;
+    use anyhow::Result;
+    use interrupt_support::NeverInterrupts;
+    use serde_json::{json, Value};
+    use std::iter::zip;
+
+    struct TestProcessor {
+        settings: Settings,
+        outgoing_commands: HashSet<Command>,
+    }
+
+    impl CommandProcessor for TestProcessor {
+        fn settings(&self) -> &Settings {
+            &self.settings
+        }
+
+        fn apply_incoming_command(&self, command: Command) -> Result<CommandStatus> {
+            Ok(if let Command::Reset(name) = command {
+                if name == "forms" {
+                    CommandStatus::Unsupported
+                } else {
+                    CommandStatus::Applied
+                }
+            } else {
+                CommandStatus::Ignored
+            })
+        }
+
+        fn fetch_outgoing_commands(&self) -> Result<HashSet<Command>> {
+            Ok(self.outgoing_commands.clone())
+        }
+    }
+
+    fn inbound_from_clients(clients: Value) -> IncomingChangeset {
+        if let Value::Array(clients) = clients {
+            let changes = clients
+                .into_iter()
+                .map(IncomingBso::from_test_content)
+                .collect();
+            IncomingChangeset {
+                changes,
+                timestamp: ServerTimestamp(0),
+                collection: COLLECTION_NAME.into(),
+            }
+        } else {
+            unreachable!("`clients` must be an array of client records")
+        }
+    }
+
+    #[test]
+    fn test_clients_sync() {
+        let processor = TestProcessor {
+            settings: Settings {
+                fxa_device_id: "deviceAAAAAA".into(),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            outgoing_commands: [
+                Command::Wipe("bookmarks".into()),
+                Command::Reset("history".into()),
+            ]
+            .iter()
+            .cloned()
+            .collect(),
+        };
+
+        let config = InfoConfiguration::default();
+
+        let mut driver = Driver::new(&processor, &NeverInterrupts, &config);
+
+        let inbound = inbound_from_clients(json!([{
+            "id": "deviceBBBBBB",
+            "name": "iPhone",
+            "type": "mobile",
+            "commands": [{
+                "command": "resetEngine",
+                "args": ["history"],
+            }],
+            "fxaDeviceId": "iPhooooooone",
+            "protocols": ["1.5"],
+            "device": "iPhone",
+        }, {
+            "id": "deviceCCCCCC",
+            "name": "Fenix",
+            "type": "mobile",
+            "commands": [],
+            "fxaDeviceId": "deviceCCCCCC",
+        }, {
+            "id": "deviceAAAAAA",
+            "name": "Laptop with a different name",
+            "type": "desktop",
+            "commands": [{
+                "command": "wipeEngine",
+                "args": ["logins"]
+            }, {
+                "command": "displayURI",
+                "args": ["http://example.com", "Fennec", "Example page"],
+                "flowID": "flooooooooow",
+            }, {
+                "command": "resetEngine",
+                "args": ["forms"],
+            }, {
+                "command": "logout",
+                "args": [],
+            }],
+            "fxaDeviceId": "deviceAAAAAA",
+        }]));
+
+        // Passing false for `should_refresh_client` - it should be ignored
+        // because we've changed the commands.
+        let mut outgoing = driver.sync(inbound, false).expect("Should sync clients");
+        outgoing
+            .changes
+            .sort_by(|a, b| a.envelope.id.cmp(&b.envelope.id));
+
+        // Make sure the list of recently synced remote clients is correct.
+        let expected_ids = &["deviceAAAAAA", "deviceBBBBBB", "deviceCCCCCC"];
+        let mut actual_ids = driver.recent_clients.keys().collect::<Vec<&String>>();
+        actual_ids.sort();
+        assert_eq!(actual_ids, expected_ids);
+
+        let expected_remote_clients = &[
+            RemoteClient {
+                fxa_device_id: Some("deviceAAAAAA".to_string()),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            RemoteClient {
+                fxa_device_id: Some("iPhooooooone".to_string()),
+                device_name: "iPhone".into(),
+                device_type: DeviceType::Mobile,
+            },
+            RemoteClient {
+                fxa_device_id: Some("deviceCCCCCC".to_string()),
+                device_name: "Fenix".into(),
+                device_type: DeviceType::Mobile,
+            },
+        ];
+        let actual_remote_clients = expected_ids
+            .iter()
+            .filter_map(|&id| driver.recent_clients.get(id))
+            .cloned()
+            .collect::<Vec<RemoteClient>>();
+        assert_eq!(actual_remote_clients, expected_remote_clients);
+
+        let expected = json!([{
+            "id": "deviceAAAAAA",
+            "name": "Laptop",
+            "type": "desktop",
+            "commands": [{
+                "command": "displayURI",
+                "args": ["http://example.com", "Fennec", "Example page"],
+                "flowID": "flooooooooow",
+            }, {
+                "command": "resetEngine",
+                "args": ["forms"],
+            }, {
+                "command": "logout",
+                "args": [],
+            }],
+            "fxaDeviceId": "deviceAAAAAA",
+            "protocols": ["1.5"],
+        }, {
+            "id": "deviceBBBBBB",
+            "name": "iPhone",
+            "type": "mobile",
+            "commands": [{
+                "command": "resetEngine",
+                "args": ["history"],
+            }, {
+                "command": "wipeEngine",
+                "args": ["bookmarks"],
+            }],
+            "fxaDeviceId": "iPhooooooone",
+            "protocols": ["1.5"],
+            "device": "iPhone",
+        }, {
+            "id": "deviceCCCCCC",
+            "name": "Fenix",
+            "type": "mobile",
+            "commands": [{
+                "command": "wipeEngine",
+                "args": ["bookmarks"],
+            }, {
+                "command": "resetEngine",
+                "args": ["history"],
+            }],
+            "fxaDeviceId": "deviceCCCCCC",
+        }]);
+        // turn outgoing into an incoming payload.
+        let incoming = IncomingChangeset {
+            changes: outgoing
+                .changes
+                .into_iter()
+                .map(|c| OutgoingBso::to_test_incoming(&c))
+                .collect(),
+            timestamp: ServerTimestamp::default(),
+            collection: outgoing.collection,
+        };
+        if let Value::Array(expected) = expected {
+            for (incoming_cleartext, exp_client) in zip(incoming.changes, expected) {
+                let incoming_client: ClientRecord =
+                    incoming_cleartext.into_content().content().unwrap();
+                assert_eq!(incoming_client, serde_json::from_value(exp_client).unwrap());
+            }
+        } else {
+            unreachable!("`expected_clients` must be an array of client records")
+        }
+    }
+
+    #[test]
+    fn test_clients_sync_bad_incoming_record_skipped() {
+        let processor = TestProcessor {
+            settings: Settings {
+                fxa_device_id: "deviceAAAAAA".into(),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            outgoing_commands: [].iter().cloned().collect(),
+        };
+
+        let config = InfoConfiguration::default();
+
+        let mut driver = Driver::new(&processor, &NeverInterrupts, &config);
+
+        let inbound = inbound_from_clients(json!([{
+            "id": "deviceBBBBBB",
+            "name": "iPhone",
+            "type": "mobile",
+            "commands": [{
+                "command": "resetEngine",
+                "args": ["history"],
+            }],
+            "fxaDeviceId": "iPhooooooone",
+            "protocols": ["1.5"],
+            "device": "iPhone",
+        }, {
+            "id": "garbage",
+            "garbage": "value",
+        }, {
+            "id": "deviceCCCCCC",
+            "deleted": true,
+            "name": "Fenix",
+            "type": "mobile",
+            "commands": [],
+            "fxaDeviceId": "deviceCCCCCC",
+        }]));
+
+        driver.sync(inbound, false).expect("Should sync clients");
+
+        // Make sure the list of recently synced remote clients is correct.
+        let expected_ids = &["deviceAAAAAA", "deviceBBBBBB"];
+        let mut actual_ids = driver.recent_clients.keys().collect::<Vec<&String>>();
+        actual_ids.sort();
+        assert_eq!(actual_ids, expected_ids);
+
+        let expected_remote_clients = &[
+            RemoteClient {
+                fxa_device_id: Some("deviceAAAAAA".to_string()),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            RemoteClient {
+                fxa_device_id: Some("iPhooooooone".to_string()),
+                device_name: "iPhone".into(),
+                device_type: DeviceType::Mobile,
+            },
+        ];
+        let actual_remote_clients = expected_ids
+            .iter()
+            .filter_map(|&id| driver.recent_clients.get(id))
+            .cloned()
+            .collect::<Vec<RemoteClient>>();
+        assert_eq!(actual_remote_clients, expected_remote_clients);
+    }
+
+    #[test]
+    fn test_clients_sync_explicit_refresh() {
+        let processor = TestProcessor {
+            settings: Settings {
+                fxa_device_id: "deviceAAAAAA".into(),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            outgoing_commands: [].iter().cloned().collect(),
+        };
+
+        let config = InfoConfiguration::default();
+
+        let mut driver = Driver::new(&processor, &NeverInterrupts, &config);
+
+        let test_clients = json!([{
+            "id": "deviceBBBBBB",
+            "name": "iPhone",
+            "type": "mobile",
+            "commands": [{
+                "command": "resetEngine",
+                "args": ["history"],
+            }],
+            "fxaDeviceId": "iPhooooooone",
+            "protocols": ["1.5"],
+            "device": "iPhone",
+        }, {
+            "id": "deviceAAAAAA",
+            "name": "Laptop",
+            "type": "desktop",
+            "commands": [],
+            "fxaDeviceId": "deviceAAAAAA",
+            "protocols": ["1.5"],
+        }]);
+
+        let outgoing = driver
+            .sync(inbound_from_clients(test_clients.clone()), false)
+            .expect("Should sync clients");
+        // should be no outgoing changes.
+        assert_eq!(outgoing.changes.len(), 0);
+
+        // Make sure the list of recently synced remote clients is correct and
+        // still includes our record we didn't update.
+        let expected_ids = &["deviceAAAAAA", "deviceBBBBBB"];
+        let mut actual_ids = driver.recent_clients.keys().collect::<Vec<&String>>();
+        actual_ids.sort();
+        assert_eq!(actual_ids, expected_ids);
+
+        // Do it again - still no changes, but force a refresh.
+        let outgoing = driver
+            .sync(inbound_from_clients(test_clients), true)
+            .expect("Should sync clients");
+        assert_eq!(outgoing.changes.len(), 1);
+
+        // Do it again - but this time with our own client record needing
+        // some change.
+        let inbound = inbound_from_clients(json!([{
+            "id": "deviceAAAAAA",
+            "name": "Laptop with New Name",
+            "type": "desktop",
+            "commands": [],
+            "fxaDeviceId": "deviceAAAAAA",
+            "protocols": ["1.5"],
+        }]));
+        let outgoing = driver.sync(inbound, false).expect("Should sync clients");
+        // should still be outgoing because the name changed.
+        assert_eq!(outgoing.changes.len(), 1);
+    }
+
+    #[test]
+    fn test_fresh_client_record() {
+        let processor = TestProcessor {
+            settings: Settings {
+                fxa_device_id: "deviceAAAAAA".into(),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            outgoing_commands: HashSet::new(),
+        };
+
+        let config = InfoConfiguration::default();
+
+        let mut driver = Driver::new(&processor, &NeverInterrupts, &config);
+
+        let clients = json!([{
+            "id": "deviceBBBBBB",
+            "name": "iPhone",
+            "type": "mobile",
+            "commands": [{
+                "command": "resetEngine",
+                "args": ["history"],
+            }],
+            "fxaDeviceId": "iPhooooooone",
+            "protocols": ["1.5"],
+            "device": "iPhone",
+        }]);
+
+        let inbound = if let Value::Array(clients) = clients {
+            let changes = clients
+                .into_iter()
+                .map(IncomingBso::from_test_content)
+                .collect();
+            IncomingChangeset {
+                changes,
+                timestamp: ServerTimestamp(0),
+                collection: COLLECTION_NAME.into(),
+            }
+        } else {
+            unreachable!("`clients` must be an array of client records")
+        };
+
+        // Passing false here for should_refresh_client, but it should be
+        // ignored as we don't have an existing record yet.
+        let mut outgoing = driver.sync(inbound, false).expect("Should sync clients");
+        outgoing
+            .changes
+            .sort_by(|a, b| a.envelope.id.cmp(&b.envelope.id));
+
+        // Make sure the list of recently synced remote clients is correct.
+        let expected_ids = &["deviceAAAAAA", "deviceBBBBBB"];
+        let mut actual_ids = driver.recent_clients.keys().collect::<Vec<&String>>();
+        actual_ids.sort();
+        assert_eq!(actual_ids, expected_ids);
+
+        let expected_remote_clients = &[
+            RemoteClient {
+                fxa_device_id: Some("deviceAAAAAA".to_string()),
+                device_name: "Laptop".into(),
+                device_type: DeviceType::Desktop,
+            },
+            RemoteClient {
+                fxa_device_id: Some("iPhooooooone".to_string()),
+                device_name: "iPhone".into(),
+                device_type: DeviceType::Mobile,
+            },
+        ];
+        let actual_remote_clients = expected_ids
+            .iter()
+            .filter_map(|&id| driver.recent_clients.get(id))
+            .cloned()
+            .collect::<Vec<RemoteClient>>();
+        assert_eq!(actual_remote_clients, expected_remote_clients);
+
+        let expected = json!([{
+            "id": "deviceAAAAAA",
+            "name": "Laptop",
+            "type": "desktop",
+            "fxaDeviceId": "deviceAAAAAA",
+            "protocols": ["1.5"],
+            "ttl": CLIENTS_TTL,
+        }]);
+        if let Value::Array(expected) = expected {
+            // turn outgoing into an incoming payload.
+            let incoming = IncomingChangeset {
+                changes: outgoing
+                    .changes
+                    .into_iter()
+                    .map(|c| OutgoingBso::to_test_incoming(&c))
+                    .collect(),
+                timestamp: ServerTimestamp::default(),
+                collection: outgoing.collection,
+            };
+            for (incoming_cleartext, record) in zip(incoming.changes, expected) {
+                let incoming_client: ClientRecord =
+                    incoming_cleartext.into_content().content().unwrap();
+                assert_eq!(incoming_client, serde_json::from_value(record).unwrap());
+            }
+        } else {
+            unreachable!("`expected_clients` must be an array of client records")
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/clients_engine/mod.rs b/third_party/rust/sync15/src/clients_engine/mod.rs
new file mode 100644
index 0000000000..7346712dc7
--- /dev/null
+++ b/third_party/rust/sync15/src/clients_engine/mod.rs
@@ -0,0 +1,93 @@
+/* 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/. */
+
+//! The client engine is a [crate::engine](Sync Engine) used to manage the
+//! "clients" collection. The clients engine manages the client record for
+//! "this device, and also manages "commands".
+//! In short, commands target one or more engines and instruct them to
+//! perform various operations - such as wiping all local data.
+//! These commands are used very rarely - currently the only command used
+//! in practice is for bookmarks to wipe all their data, which is sent when
+//! a desktop device restores all bookmarks from a backup. In this scenario,
+//! desktop will delete all local bookmarks then replace them with the backed
+//! up set, which without a "wipe" command would almost certainly cause other
+//! connected devices to "resurrect" the deleted bookmarks.
+use std::collections::HashSet;
+
+mod engine;
+mod record;
+mod ser;
+
+use crate::DeviceType;
+use anyhow::Result;
+pub use engine::Engine;
+
+// These are what desktop uses.
+const CLIENTS_TTL: u32 = 15_552_000; // 180 days
+pub(crate) const CLIENTS_TTL_REFRESH: u64 = 604_800; // 7 days
+
+/// A command processor applies incoming commands like wipes and resets for all
+/// stores, and returns commands to send to other clients. It also manages
+/// settings like the device name and type, which is stored in the special
+/// `clients` collection.
+///
+/// In practice, this trait only has one implementation, in the sync manager.
+/// It's split this way because the clients engine depends on internal `sync15`
+/// structures, and can't be implemented as a syncable store...but `sync15`
+/// doesn't know anything about multiple engines. This lets the sync manager
+/// provide its own implementation for handling wipe and reset commands for all
+/// the engines that it manages.
+pub trait CommandProcessor {
+    fn settings(&self) -> &Settings;
+
+    /// Fetches commands to send to other clients. An error return value means
+    /// commands couldn't be fetched, and halts the sync.
+    fn fetch_outgoing_commands(&self) -> Result<HashSet<Command>>;
+
+    /// Applies a command sent to this client from another client. This method
+    /// should return a `CommandStatus` indicating whether the command was
+    /// processed.
+    ///
+    /// An error return value means the sync manager encountered an error
+    /// applying the command, and halts the sync to prevent unexpected behavior
+    /// (for example, merging local and remote bookmarks, when we were told to
+    /// wipe our local bookmarks).
+    fn apply_incoming_command(&self, command: Command) -> Result<CommandStatus>;
+}
+
+/// Indicates if a command was applied successfully, ignored, or not supported.
+/// Applied and ignored commands are removed from our client record, and never
+/// retried. Unsupported commands are put back into our record, and retried on
+/// subsequent syncs. This is to handle clients adding support for new data
+/// types.
+#[derive(Clone, Copy, Debug, Eq, Hash, PartialEq)]
+pub enum CommandStatus {
+    Applied,
+    Ignored,
+    Unsupported,
+}
+
+/// Information about this device to include in its client record. This should
+/// be persisted across syncs, as part of the sync manager state.
+#[derive(Clone, Debug, Eq, Hash, PartialEq)]
+pub struct Settings {
+    /// The FxA device ID of this client, also used as this client's record ID
+    /// in the clients collection.
+    pub fxa_device_id: String,
+    /// The name of this client. This should match the client's name in the
+    /// FxA device manager.
+    pub device_name: String,
+    /// The type of this client: mobile, tablet, desktop, or other.
+    pub device_type: DeviceType,
+}
+
+#[derive(Clone, Debug, Eq, Hash, Ord, PartialEq, PartialOrd)]
+pub enum Command {
+    /// Erases all local data for a specific engine.
+    Wipe(String),
+    /// Resets local sync state for all engines.
+    ResetAll,
+    /// Resets local sync state for a specific engine.
+    Reset(String),
+}
diff --git a/third_party/rust/sync15/src/clients_engine/record.rs b/third_party/rust/sync15/src/clients_engine/record.rs
new file mode 100644
index 0000000000..fe17c12861
--- /dev/null
+++ b/third_party/rust/sync15/src/clients_engine/record.rs
@@ -0,0 +1,124 @@
+/* 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/. */
+
+use serde_derive::*;
+
+use super::Command;
+
+/// The serialized form of a client record.
+#[derive(Clone, Debug, Eq, Deserialize, Hash, PartialEq, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ClientRecord {
+    #[serde(rename = "id")]
+    pub id: String,
+
+    pub name: String,
+
+    #[serde(rename = "type")]
+    pub typ: crate::DeviceType,
+
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub commands: Vec<CommandRecord>,
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub fxa_device_id: Option<String>,
+
+    /// `version`, `protocols`, `formfactor`, `os`, `appPackage`, `application`,
+    /// and `device` are unused and optional in all implementations (Desktop,
+    /// iOS, and Fennec), but we round-trip them.
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub version: Option<String>,
+
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub protocols: Vec<String>,
+
+    #[serde(
+        default,
+        rename = "formfactor",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub form_factor: Option<String>,
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub os: Option<String>,
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub app_package: Option<String>,
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub application: Option<String>,
+
+    /// The model of the device, like "iPhone" or "iPod touch" on iOS. Note
+    /// that this is _not_ the client ID (`id`) or the FxA device ID
+    /// (`fxa_device_id`).
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub device: Option<String>,
+}
+
+impl From<&ClientRecord> for crate::RemoteClient {
+    fn from(record: &ClientRecord) -> crate::RemoteClient {
+        crate::RemoteClient {
+            fxa_device_id: record.fxa_device_id.clone(),
+            device_name: record.name.clone(),
+            device_type: record.typ,
+        }
+    }
+}
+
+/// The serialized form of a client command.
+#[derive(Clone, Debug, Eq, Deserialize, Hash, PartialEq, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct CommandRecord {
+    /// The command name. This is a string, not an enum, because we want to
+    /// round-trip commands that we don't support yet.
+    #[serde(rename = "command")]
+    pub name: String,
+
+    /// Extra, command-specific arguments. Note that we must send an empty
+    /// array if the command expects no arguments.
+    #[serde(default)]
+    pub args: Vec<String>,
+
+    /// Some commands, like repair, send a "flow ID" that other cliennts can
+    /// record in their telemetry. We don't currently send commands with
+    /// flow IDs, but we round-trip them.
+    #[serde(default, rename = "flowID", skip_serializing_if = "Option::is_none")]
+    pub flow_id: Option<String>,
+}
+
+impl CommandRecord {
+    /// Converts a serialized command into one that we can apply. Returns `None`
+    /// if we don't support the command.
+    pub fn as_command(&self) -> Option<Command> {
+        match self.name.as_str() {
+            "wipeEngine" => self.args.get(0).map(|e| Command::Wipe(e.into())),
+            "resetEngine" => self.args.get(0).map(|e| Command::Reset(e.into())),
+            "resetAll" => Some(Command::ResetAll),
+            _ => None,
+        }
+    }
+}
+
+impl From<Command> for CommandRecord {
+    fn from(command: Command) -> CommandRecord {
+        match command {
+            Command::Wipe(engine) => CommandRecord {
+                name: "wipeEngine".into(),
+                args: vec![engine],
+                flow_id: None,
+            },
+            Command::Reset(engine) => CommandRecord {
+                name: "resetEngine".into(),
+                args: vec![engine],
+                flow_id: None,
+            },
+            Command::ResetAll => CommandRecord {
+                name: "resetAll".into(),
+                args: Vec::new(),
+                flow_id: None,
+            },
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/clients_engine/ser.rs b/third_party/rust/sync15/src/clients_engine/ser.rs
new file mode 100644
index 0000000000..2e7b0817b8
--- /dev/null
+++ b/third_party/rust/sync15/src/clients_engine/ser.rs
@@ -0,0 +1,125 @@
+/* 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/. */
+
+use crate::error::Result;
+use serde::Serialize;
+use std::io::{self, Write};
+
+/// A writer that counts the number of bytes it's asked to write, and discards
+/// the data. Used to calculate the serialized size of the commands list.
+#[derive(Clone, Copy, Default)]
+pub struct WriteCount(usize);
+
+impl WriteCount {
+    #[inline]
+    pub fn len(self) -> usize {
+        self.0
+    }
+}
+
+impl Write for WriteCount {
+    #[inline]
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        self.0 += buf.len();
+        Ok(buf.len())
+    }
+
+    #[inline]
+    fn flush(&mut self) -> io::Result<()> {
+        Ok(())
+    }
+}
+
+/// Returns the size of the given value, in bytes, when serialized to JSON.
+fn compute_serialized_size<T: Serialize>(value: &T) -> Result<usize> {
+    let mut w = WriteCount::default();
+    serde_json::to_writer(&mut w, value)?;
+    Ok(w.len())
+}
+
+/// Truncates `list` to fit within `payload_size_max_bytes` when serialized to
+/// JSON.
+pub fn shrink_to_fit<T: Serialize>(list: &mut Vec<T>, payload_size_max_bytes: usize) -> Result<()> {
+    let size = compute_serialized_size(&list)?;
+    // See bug 535326 comment 8 for an explanation of the estimation
+    match ((payload_size_max_bytes / 4) * 3).checked_sub(1500) {
+        Some(max_serialized_size) => {
+            if size > max_serialized_size {
+                // Estimate a little more than the direct fraction to maximize packing
+                let cutoff = (list.len() * max_serialized_size - 1) / size + 1;
+                list.truncate(cutoff + 1);
+                // Keep dropping off the last entry until the data fits.
+                while compute_serialized_size(&list)? > max_serialized_size {
+                    if list.pop().is_none() {
+                        break;
+                    }
+                }
+            }
+            Ok(())
+        }
+        None => {
+            list.clear();
+            Ok(())
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::super::record::CommandRecord;
+    use super::*;
+
+    #[test]
+    fn test_compute_serialized_size() {
+        assert_eq!(compute_serialized_size(&1).unwrap(), 1);
+        assert_eq!(compute_serialized_size(&"hi").unwrap(), 4);
+        assert_eq!(
+            compute_serialized_size(&["hi", "hello", "bye"]).unwrap(),
+            20
+        );
+    }
+
+    #[test]
+    fn test_shrink_to_fit() {
+        let mut commands = vec![
+            CommandRecord {
+                name: "wipeEngine".into(),
+                args: vec!["bookmarks".into()],
+                flow_id: Some("flow".into()),
+            },
+            CommandRecord {
+                name: "resetEngine".into(),
+                args: vec!["history".into()],
+                flow_id: Some("flow".into()),
+            },
+            CommandRecord {
+                name: "logout".into(),
+                args: Vec::new(),
+                flow_id: None,
+            },
+        ];
+
+        // 4096 bytes is enough to fit all three commands.
+        shrink_to_fit(&mut commands, 4096).unwrap();
+        assert_eq!(commands.len(), 3);
+
+        let sizes = commands
+            .iter()
+            .map(|c| compute_serialized_size(c).unwrap())
+            .collect::<Vec<_>>();
+        assert_eq!(sizes, &[61, 60, 30]);
+
+        // `logout` won't fit within 2168 bytes.
+        shrink_to_fit(&mut commands, 2168).unwrap();
+        assert_eq!(commands.len(), 2);
+
+        // `resetEngine` won't fit within 2084 bytes.
+        shrink_to_fit(&mut commands, 2084).unwrap();
+        assert_eq!(commands.len(), 1);
+
+        // `wipeEngine` won't fit at all.
+        shrink_to_fit(&mut commands, 1024).unwrap();
+        assert!(commands.is_empty());
+    }
+}
diff --git a/third_party/rust/sync15/src/device_type.rs b/third_party/rust/sync15/src/device_type.rs
new file mode 100644
index 0000000000..8afea0dd00
--- /dev/null
+++ b/third_party/rust/sync15/src/device_type.rs
@@ -0,0 +1,152 @@
+/* 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/. */
+
+//! This type is strictly owned by FxA, but is defined in this crate because of
+//! some hard-to-avoid hacks done for the tabs engine... See issue #2590.
+//!
+//! Thus, fxa-client ends up taking a dep on this crate, which is roughly
+//! the opposite of reality.
+
+use serde::{Deserialize, Deserializer, Serialize, Serializer};
+
+/// Enumeration for the different types of device.
+///
+/// Firefox Accounts and the broader Sync universe separates devices into broad categories for
+/// various purposes, such as distinguishing a desktop PC from a mobile phone.
+///
+/// A special variant in this enum, `DeviceType::Unknown` is used to capture
+/// the string values we don't recognise. It also has a custom serde serializer and deserializer
+/// which implements the following semantics:
+/// * deserializing a `DeviceType` which uses a string value we don't recognise or null will return
+///   `DeviceType::Unknown` rather than returning an error.
+/// * serializing `DeviceType::Unknown` will serialize `null`.
+///
+/// This has a few important implications:
+/// * In general, `Option<DeviceType>` should be avoided, and a plain `DeviceType` used instead,
+///   because in that case, `None` would be semantically identical to `DeviceType::Unknown` and
+///   as mentioned above, `null` already deserializes as `DeviceType::Unknown`.
+/// * Any unknown device types can not be round-tripped via this enum - eg, if you deserialize
+///   a struct holding a `DeviceType` string value we don't recognize, then re-serialize it, the
+///   original string value is lost. We don't consider this a problem because in practice, we only
+///   upload records with *this* device's type, not the type of other devices, and it's reasonable
+///   to assume that this module knows about all valid device types for the device type it is
+///   deployed on.
+#[derive(Copy, Clone, Debug, PartialEq, Eq, Hash)]
+pub enum DeviceType {
+    Desktop,
+    Mobile,
+    Tablet,
+    VR,
+    TV,
+    // See docstrings above re how Unknown is serialized and deserialized.
+    Unknown,
+}
+
+impl Default for DeviceType {
+    fn default() -> Self {
+        DeviceType::Unknown
+    }
+}
+
+impl<'de> Deserialize<'de> for DeviceType {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        Ok(match String::deserialize(deserializer) {
+            Ok(s) => match s.as_str() {
+                "desktop" => DeviceType::Desktop,
+                "mobile" => DeviceType::Mobile,
+                "tablet" => DeviceType::Tablet,
+                "vr" => DeviceType::VR,
+                "tv" => DeviceType::TV,
+                // There's a vague possibility that desktop might serialize "phone" for mobile
+                // devices - https://searchfox.org/mozilla-central/rev/a156a65ced2dae5913ae35a68e9445b8ee7ca457/services/sync/modules/engines/clients.js#292
+                "phone" => DeviceType::Mobile,
+                // Everything else is Unknown.
+                _ => DeviceType::Unknown,
+            },
+            // Anything other than a string is "unknown" - this isn't ideal - we really only want
+            // to handle null and, eg, a number probably should be an error, but meh.
+            Err(_) => DeviceType::Unknown,
+        })
+    }
+}
+impl Serialize for DeviceType {
+    fn serialize<S>(&self, s: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        match self {
+            // It's unfortunate we need to duplicate the strings here...
+            DeviceType::Desktop => s.serialize_unit_variant("DeviceType", 0, "desktop"),
+            DeviceType::Mobile => s.serialize_unit_variant("DeviceType", 1, "mobile"),
+            DeviceType::Tablet => s.serialize_unit_variant("DeviceType", 2, "tablet"),
+            DeviceType::VR => s.serialize_unit_variant("DeviceType", 3, "vr"),
+            DeviceType::TV => s.serialize_unit_variant("DeviceType", 4, "tv"),
+            // This is the important bit - Unknown -> None
+            DeviceType::Unknown => s.serialize_none(),
+        }
+    }
+}
+
+#[cfg(test)]
+mod device_type_tests {
+    use super::*;
+
+    #[test]
+    fn test_serde_ser() {
+        assert_eq!(
+            serde_json::to_string(&DeviceType::Desktop).unwrap(),
+            "\"desktop\""
+        );
+        assert_eq!(
+            serde_json::to_string(&DeviceType::Mobile).unwrap(),
+            "\"mobile\""
+        );
+        assert_eq!(
+            serde_json::to_string(&DeviceType::Tablet).unwrap(),
+            "\"tablet\""
+        );
+        assert_eq!(serde_json::to_string(&DeviceType::VR).unwrap(), "\"vr\"");
+        assert_eq!(serde_json::to_string(&DeviceType::TV).unwrap(), "\"tv\"");
+        assert_eq!(serde_json::to_string(&DeviceType::Unknown).unwrap(), "null");
+    }
+
+    #[test]
+    fn test_serde_de() {
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("\"desktop\"").unwrap(),
+            DeviceType::Desktop
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("\"mobile\"").unwrap(),
+            DeviceType::Mobile
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("\"tablet\"").unwrap(),
+            DeviceType::Tablet
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("\"vr\"").unwrap(),
+            DeviceType::VR
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("\"tv\"").unwrap(),
+            DeviceType::TV
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("\"something-else\"").unwrap(),
+            DeviceType::Unknown,
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("null").unwrap(),
+            DeviceType::Unknown,
+        ));
+        assert!(matches!(
+            serde_json::from_str::<DeviceType>("99").unwrap(),
+            DeviceType::Unknown,
+        ));
+    }
+}
diff --git a/third_party/rust/sync15/src/enc_payload.rs b/third_party/rust/sync15/src/enc_payload.rs
new file mode 100644
index 0000000000..2adc031f70
--- /dev/null
+++ b/third_party/rust/sync15/src/enc_payload.rs
@@ -0,0 +1,110 @@
+/* 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/. */
+
+use crate::error;
+use crate::key_bundle::KeyBundle;
+use lazy_static::lazy_static;
+use serde::{Deserialize, Serialize};
+
+/// A representation of an encrypted payload. Used as the payload in EncryptedBso and
+/// also anywhere else the sync keys might be used to encrypt/decrypt, such as send-tab payloads.
+#[derive(Deserialize, Serialize, Clone, Debug)]
+pub struct EncryptedPayload {
+    #[serde(rename = "IV")]
+    pub iv: String,
+    pub hmac: String,
+    pub ciphertext: String,
+}
+
+impl EncryptedPayload {
+    #[inline]
+    pub fn serialized_len(&self) -> usize {
+        (*EMPTY_ENCRYPTED_PAYLOAD_SIZE) + self.ciphertext.len() + self.hmac.len() + self.iv.len()
+    }
+
+    pub fn decrypt(&self, key: &KeyBundle) -> error::Result<String> {
+        key.decrypt(&self.ciphertext, &self.iv, &self.hmac)
+    }
+
+    pub fn decrypt_into<T>(&self, key: &KeyBundle) -> error::Result<T>
+    where
+        for<'a> T: Deserialize<'a>,
+    {
+        Ok(serde_json::from_str(&self.decrypt(key)?)?)
+    }
+
+    pub fn from_cleartext(key: &KeyBundle, cleartext: String) -> error::Result<Self> {
+        let (enc_base64, iv_base64, hmac_base16) =
+            key.encrypt_bytes_rand_iv(cleartext.as_bytes())?;
+        Ok(EncryptedPayload {
+            iv: iv_base64,
+            hmac: hmac_base16,
+            ciphertext: enc_base64,
+        })
+    }
+
+    pub fn from_cleartext_payload<T: Serialize>(
+        key: &KeyBundle,
+        cleartext_payload: &T,
+    ) -> error::Result<Self> {
+        Self::from_cleartext(key, serde_json::to_string(cleartext_payload)?)
+    }
+}
+
+// Our "postqueue", which chunks records for upload, needs to know this value.
+// It's tricky to determine at compile time, so do it once at at runtime.
+lazy_static! {
+    // The number of bytes taken up by padding in a EncryptedPayload.
+    static ref EMPTY_ENCRYPTED_PAYLOAD_SIZE: usize = serde_json::to_string(
+        &EncryptedPayload { iv: "".into(), hmac: "".into(), ciphertext: "".into() }
+    ).unwrap().len();
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use serde_json::json;
+
+    #[derive(Serialize, Deserialize, Debug)]
+    struct TestStruct {
+        id: String,
+        age: u32,
+        meta: String,
+    }
+
+    #[test]
+    fn test_roundtrip_crypt_record() {
+        let key = KeyBundle::new_random().unwrap();
+        let payload_json = json!({ "id": "aaaaaaaaaaaa", "age": 105, "meta": "data" });
+        let payload =
+            EncryptedPayload::from_cleartext(&key, serde_json::to_string(&payload_json).unwrap())
+                .unwrap();
+
+        let record = payload.decrypt_into::<TestStruct>(&key).unwrap();
+        assert_eq!(record.id, "aaaaaaaaaaaa");
+        assert_eq!(record.age, 105);
+        assert_eq!(record.meta, "data");
+
+        // While we're here, check on EncryptedPayload::serialized_len
+        let val_rec = serde_json::to_string(&serde_json::to_value(&payload).unwrap()).unwrap();
+        assert_eq!(payload.serialized_len(), val_rec.len());
+    }
+
+    #[test]
+    fn test_record_bad_hmac() {
+        let key1 = KeyBundle::new_random().unwrap();
+        let json = json!({ "id": "aaaaaaaaaaaa", "deleted": true, });
+
+        let payload =
+            EncryptedPayload::from_cleartext(&key1, serde_json::to_string(&json).unwrap()).unwrap();
+
+        let key2 = KeyBundle::new_random().unwrap();
+        let e = payload
+            .decrypt(&key2)
+            .expect_err("Should fail because wrong keybundle");
+
+        // Note: ErrorKind isn't PartialEq, so.
+        assert!(matches!(e, error::Error::CryptoError(_)));
+    }
+}
diff --git a/third_party/rust/sync15/src/engine/bridged_engine.rs b/third_party/rust/sync15/src/engine/bridged_engine.rs
new file mode 100644
index 0000000000..c12780ba04
--- /dev/null
+++ b/third_party/rust/sync15/src/engine/bridged_engine.rs
@@ -0,0 +1,119 @@
+/* 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/. */
+
+use anyhow::Result;
+
+use crate::bso::{IncomingBso, OutgoingBso};
+use crate::Guid;
+
+/// A BridgedEngine acts as a bridge between application-services, rust
+/// implemented sync engines and sync engines as defined by Desktop Firefox.
+///
+/// [Desktop Firefox has an abstract implementation of a Sync
+/// Engine](https://searchfox.org/mozilla-central/source/services/sync/modules/engines.js)
+/// with a number of functions each engine is expected to override. Engines
+/// implemented in Rust use a different shape (specifically, the
+/// [SyncEngine](crate::SyncEngine) trait), so this BridgedEngine trait adapts
+/// between the 2.
+pub trait BridgedEngine: Send + Sync {
+    /// Returns the last sync time, in milliseconds, for this engine's
+    /// collection. This is called before each sync, to determine the lower
+    /// bound for new records to fetch from the server.
+    fn last_sync(&self) -> Result<i64>;
+
+    /// Sets the last sync time, in milliseconds. This is called throughout
+    /// the sync, to fast-forward the stored last sync time to match the
+    /// timestamp on the uploaded records.
+    fn set_last_sync(&self, last_sync_millis: i64) -> Result<()>;
+
+    /// Returns the sync ID for this engine's collection. This is only used in
+    /// tests.
+    fn sync_id(&self) -> Result<Option<String>>;
+
+    /// Resets the sync ID for this engine's collection, returning the new ID.
+    /// As a side effect, implementations should reset all local Sync state,
+    /// as in `reset`.
+    fn reset_sync_id(&self) -> Result<String>;
+
+    /// Ensures that the locally stored sync ID for this engine's collection
+    /// matches the `new_sync_id` from the server. If the two don't match,
+    /// implementations should reset all local Sync state, as in `reset`.
+    /// This method returns the assigned sync ID, which can be either the
+    /// `new_sync_id`, or a different one if the engine wants to force other
+    /// devices to reset their Sync state for this collection the next time they
+    /// sync.
+    fn ensure_current_sync_id(&self, new_sync_id: &str) -> Result<String>;
+
+    /// Tells the tabs engine about recent FxA devices. A bit of a leaky abstration as it only
+    /// makes sense for tabs.
+    /// The arg is a json serialized `ClientData` struct.
+    fn prepare_for_sync(&self, _client_data: &str) -> Result<()> {
+        Ok(())
+    }
+
+    /// Indicates that the engine is about to start syncing. This is called
+    /// once per sync, and always before `store_incoming`.
+    fn sync_started(&self) -> Result<()>;
+
+    /// Stages a batch of incoming Sync records. This is called multiple
+    /// times per sync, once for each batch. Implementations can use the
+    /// signal to check if the operation was aborted, and cancel any
+    /// pending work.
+    fn store_incoming(&self, incoming_records: Vec<IncomingBso>) -> Result<()>;
+
+    /// Applies all staged records, reconciling changes on both sides and
+    /// resolving conflicts. Returns a list of records to upload.
+    fn apply(&self) -> Result<ApplyResults>;
+
+    /// Indicates that the given record IDs were uploaded successfully to the
+    /// server. This is called multiple times per sync, once for each batch
+    /// upload.
+    fn set_uploaded(&self, server_modified_millis: i64, ids: &[Guid]) -> Result<()>;
+
+    /// Indicates that all records have been uploaded. At this point, any record
+    /// IDs marked for upload that haven't been passed to `set_uploaded`, can be
+    /// assumed to have failed: for example, because the server rejected a record
+    /// with an invalid TTL or sort index.
+    fn sync_finished(&self) -> Result<()>;
+
+    /// Resets all local Sync state, including any change flags, mirrors, and
+    /// the last sync time, such that the next sync is treated as a first sync
+    /// with all new local data. Does not erase any local user data.
+    fn reset(&self) -> Result<()>;
+
+    /// Erases all local user data for this collection, and any Sync metadata.
+    /// This method is destructive, and unused for most collections.
+    fn wipe(&self) -> Result<()>;
+}
+
+// TODO: We should replace this with OutgoingChangeset to reduce the number
+// of types engines need to deal with.
+#[derive(Debug, Default)]
+pub struct ApplyResults {
+    /// List of records
+    pub records: Vec<OutgoingBso>,
+    /// The number of incoming records whose contents were merged because they
+    /// changed on both sides. None indicates we aren't reporting this
+    /// information.
+    pub num_reconciled: Option<usize>,
+}
+
+impl ApplyResults {
+    pub fn new(records: Vec<OutgoingBso>, num_reconciled: impl Into<Option<usize>>) -> Self {
+        Self {
+            records,
+            num_reconciled: num_reconciled.into(),
+        }
+    }
+}
+
+// Shorthand for engines that don't care.
+impl From<Vec<OutgoingBso>> for ApplyResults {
+    fn from(records: Vec<OutgoingBso>) -> Self {
+        Self {
+            records,
+            num_reconciled: None,
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/engine/changeset.rs b/third_party/rust/sync15/src/engine/changeset.rs
new file mode 100644
index 0000000000..c5a4e65fcd
--- /dev/null
+++ b/third_party/rust/sync15/src/engine/changeset.rs
@@ -0,0 +1,52 @@
+/* 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/. */
+
+use crate::bso::{IncomingBso, OutgoingBso};
+use crate::{CollectionName, ServerTimestamp};
+
+// Incoming and Outgoing changesets are almost identical except for the timestamp.
+// Separate types still helps avoid confusion with that timestamp, so they're split.
+#[derive(Debug)]
+pub struct IncomingChangeset {
+    pub changes: Vec<IncomingBso>,
+    /// The server timestamp of the collection.
+    pub timestamp: ServerTimestamp,
+    pub collection: CollectionName,
+}
+
+impl IncomingChangeset {
+    #[inline]
+    pub fn new(collection: CollectionName, timestamp: ServerTimestamp) -> Self {
+        Self::new_with_changes(collection, timestamp, Vec::new())
+    }
+
+    #[inline]
+    pub fn new_with_changes(
+        collection: CollectionName,
+        timestamp: ServerTimestamp,
+        changes: Vec<IncomingBso>,
+    ) -> Self {
+        Self {
+            changes,
+            timestamp,
+            collection,
+        }
+    }
+}
+
+#[derive(Debug)]
+pub struct OutgoingChangeset {
+    pub changes: Vec<OutgoingBso>,
+    pub collection: CollectionName,
+}
+
+impl OutgoingChangeset {
+    #[inline]
+    pub fn new(collection: CollectionName, changes: Vec<OutgoingBso>) -> Self {
+        Self {
+            collection,
+            changes,
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/engine/mod.rs b/third_party/rust/sync15/src/engine/mod.rs
new file mode 100644
index 0000000000..427b779c9b
--- /dev/null
+++ b/third_party/rust/sync15/src/engine/mod.rs
@@ -0,0 +1,38 @@
+/* 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/. */
+
+//! This module is used by crates which need to implement a "sync engine".
+//! At a high-level, a "sync engine" is code which knows how to take records
+//! from a sync server, apply and reconcile them with the local data, then
+//! provide records which should be uploaded to the server.
+//!
+//! Note that the "sync engine" does not itself talk to the server, nor does
+//! it manage the state of the remote server, nor does it do any of the
+//! encryption/decryption - that is the responsbility of the "sync client", as
+//! implemented in the [client] module (or in some cases, implemented externally)
+//!
+//! There are currently 2 types of engine:
+//! * Code which implements the [crate::engine::sync_engine::SyncEngine]
+//!   trait. These are the "original" Rust engines, designed to be used with
+//!   the [crate::client](sync client)
+//! * Code which implements the [crate::engine::bridged_engine::BridgedEngine]
+//!   trait. These engines are a "bridge" between the Desktop JS Sync world and
+//!   this rust code.
+//! While these engines end up doing the same thing, the difference is due to
+//! implementation differences between the Desktop Sync client and the Rust
+//! client.
+//! We intend merging these engines - the first step will be to merge the
+//! types and payload management used by these traits, then to combine the
+//! requirements into a single trait that captures both use-cases.
+mod bridged_engine;
+mod changeset;
+mod request;
+mod sync_engine;
+
+pub use bridged_engine::{ApplyResults, BridgedEngine};
+pub use changeset::{IncomingChangeset, OutgoingChangeset};
+#[cfg(feature = "sync-client")]
+pub(crate) use request::CollectionPost;
+pub use request::{CollectionRequest, RequestOrder};
+pub use sync_engine::{CollSyncIds, EngineSyncAssociation, SyncEngine, SyncEngineId};
diff --git a/third_party/rust/sync15/src/engine/request.rs b/third_party/rust/sync15/src/engine/request.rs
new file mode 100644
index 0000000000..7d634bb5e7
--- /dev/null
+++ b/third_party/rust/sync15/src/engine/request.rs
@@ -0,0 +1,125 @@
+/* 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/. */
+use crate::{CollectionName, Guid, ServerTimestamp};
+#[derive(Debug, Default, Clone, PartialEq, Eq)]
+pub struct CollectionRequest {
+    pub collection: CollectionName,
+    pub full: bool,
+    pub ids: Option<Vec<Guid>>,
+
+    pub limit: Option<RequestLimit>,
+    pub older: Option<ServerTimestamp>,
+    pub newer: Option<ServerTimestamp>,
+}
+
+impl CollectionRequest {
+    #[inline]
+    pub fn new(collection: CollectionName) -> CollectionRequest {
+        CollectionRequest {
+            collection,
+            ..Default::default()
+        }
+    }
+
+    #[inline]
+    pub fn ids<V>(mut self, v: V) -> CollectionRequest
+    where
+        V: IntoIterator,
+        V::Item: Into<Guid>,
+    {
+        self.ids = Some(v.into_iter().map(|id| id.into()).collect());
+        self
+    }
+
+    #[inline]
+    pub fn full(mut self) -> CollectionRequest {
+        self.full = true;
+        self
+    }
+
+    #[inline]
+    pub fn older_than(mut self, ts: ServerTimestamp) -> CollectionRequest {
+        self.older = Some(ts);
+        self
+    }
+
+    #[inline]
+    pub fn newer_than(mut self, ts: ServerTimestamp) -> CollectionRequest {
+        self.newer = Some(ts);
+        self
+    }
+
+    #[inline]
+    pub fn limit(mut self, num: usize, order: RequestOrder) -> CollectionRequest {
+        self.limit = Some(RequestLimit { num, order });
+        self
+    }
+}
+
+// This is just used interally - consumers just provide the content, not request params.
+#[cfg(feature = "sync-client")]
+#[derive(Debug, Default, Clone, PartialEq, Eq)]
+pub(crate) struct CollectionPost {
+    pub collection: CollectionName,
+    pub commit: bool,
+    pub batch: Option<String>,
+}
+
+#[cfg(feature = "sync-client")]
+impl CollectionPost {
+    #[inline]
+    pub fn new(collection: CollectionName) -> Self {
+        Self {
+            collection,
+            ..Default::default()
+        }
+    }
+
+    #[inline]
+    pub fn batch(mut self, batch: Option<String>) -> Self {
+        self.batch = batch;
+        self
+    }
+
+    #[inline]
+    pub fn commit(mut self, v: bool) -> Self {
+        self.commit = v;
+        self
+    }
+}
+
+// Asking for the order of records only makes sense if you are limiting them
+// in some way - consumers don't care about the order otherwise as everything
+// is processed as a set.
+#[derive(Debug, Clone, Copy, Eq, PartialEq, Ord, PartialOrd, Hash)]
+pub enum RequestOrder {
+    Oldest,
+    Newest,
+    Index,
+}
+
+impl RequestOrder {
+    #[inline]
+    pub fn as_str(self) -> &'static str {
+        match self {
+            RequestOrder::Oldest => "oldest",
+            RequestOrder::Newest => "newest",
+            RequestOrder::Index => "index",
+        }
+    }
+}
+
+impl std::fmt::Display for RequestOrder {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.write_str(self.as_str())
+    }
+}
+
+// If you specify a numerical limit you must provide the order so backfilling
+// is possible (ie, so you know which ones you got!)
+#[derive(Debug, Clone, Copy, Eq, PartialEq, Ord, PartialOrd, Hash)]
+pub struct RequestLimit {
+    pub(crate) num: usize,
+    pub(crate) order: RequestOrder,
+}
diff --git a/third_party/rust/sync15/src/engine/sync_engine.rs b/third_party/rust/sync15/src/engine/sync_engine.rs
new file mode 100644
index 0000000000..79b9b1a522
--- /dev/null
+++ b/third_party/rust/sync15/src/engine/sync_engine.rs
@@ -0,0 +1,235 @@
+/* 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/. */
+
+use super::{CollectionRequest, IncomingChangeset, OutgoingChangeset};
+use crate::client_types::ClientData;
+use crate::{telemetry, CollectionName, Guid, ServerTimestamp};
+use anyhow::Result;
+use std::fmt;
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct CollSyncIds {
+    pub global: Guid,
+    pub coll: Guid,
+}
+
+/// Defines how an engine is associated with a particular set of records
+/// on a sync storage server. It's either disconnected, or believes it is
+/// connected with a specific set of GUIDs. If the server and the engine don't
+/// agree on the exact GUIDs, the engine will assume something radical happened
+/// so it can't believe anything it thinks it knows about the state of the
+/// server (ie, it will "reset" then do a full reconcile)
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum EngineSyncAssociation {
+    /// This store is disconnected (although it may be connected in the future).
+    Disconnected,
+    /// Sync is connected, and has the following sync IDs.
+    Connected(CollSyncIds),
+}
+
+/// The concrete `SyncEngine` implementations
+#[derive(Clone, Debug, Hash, PartialEq, Eq, PartialOrd, Ord)]
+pub enum SyncEngineId {
+    // Note that we've derived PartialOrd etc, which uses lexicographic ordering
+    // of the variants. We leverage that such that the higher priority engines
+    // are listed first.
+    // This order matches desktop.
+    Passwords,
+    Tabs,
+    Bookmarks,
+    Addresses,
+    CreditCards,
+    History,
+}
+
+impl SyncEngineId {
+    // Iterate over all possible engines. Note that we've made a policy decision
+    // that this should enumerate in "order" as defined by PartialCmp, and tests
+    // enforce this.
+    pub fn iter() -> impl Iterator<Item = SyncEngineId> {
+        [
+            Self::Passwords,
+            Self::Tabs,
+            Self::Bookmarks,
+            Self::Addresses,
+            Self::CreditCards,
+            Self::History,
+        ]
+        .into_iter()
+    }
+
+    // Get the string identifier for this engine.  This must match the strings in SyncEngineSelection.
+    pub fn name(&self) -> &'static str {
+        match self {
+            Self::Passwords => "passwords",
+            Self::History => "history",
+            Self::Bookmarks => "bookmarks",
+            Self::Tabs => "tabs",
+            Self::Addresses => "addresses",
+            Self::CreditCards => "creditcards",
+        }
+    }
+}
+
+impl fmt::Display for SyncEngineId {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{}", self.name())
+    }
+}
+
+impl TryFrom<&str> for SyncEngineId {
+    type Error = String;
+
+    fn try_from(value: &str) -> std::result::Result<Self, Self::Error> {
+        match value {
+            "passwords" => Ok(Self::Passwords),
+            "history" => Ok(Self::History),
+            "bookmarks" => Ok(Self::Bookmarks),
+            "tabs" => Ok(Self::Tabs),
+            "addresses" => Ok(Self::Addresses),
+            "creditcards" => Ok(Self::CreditCards),
+            _ => Err(value.into()),
+        }
+    }
+}
+
+/// A "sync engine" is a thing that knows how to sync. It's often implemented
+/// by a "store" (which is the generic term responsible for all storage
+/// associated with a component, including storage required for sync.)
+///
+/// Low-level engine functionality. Engines that need custom reconciliation
+/// logic should use this.
+///
+/// Different engines will produce errors of different types.  To accommodate
+/// this, we force them all to return anyhow::Error.
+pub trait SyncEngine {
+    fn collection_name(&self) -> CollectionName;
+
+    /// Prepares the engine for syncing. The tabs engine currently uses this to
+    /// store the current list of clients, which it uses to look up device names
+    /// and types.
+    ///
+    /// Note that this method is only called by `sync_multiple`, and only if a
+    /// command processor is registered. In particular, `prepare_for_sync` will
+    /// not be called if the store is synced using `sync::synchronize` or
+    /// `sync_multiple::sync_multiple`. It _will_ be called if the store is
+    /// synced via the Sync Manager.
+    ///
+    /// TODO(issue #2590): This is pretty cludgey and will be hard to extend for
+    /// any case other than the tabs case. We should find another way to support
+    /// tabs...
+    fn prepare_for_sync(&self, _get_client_data: &dyn Fn() -> ClientData) -> Result<()> {
+        Ok(())
+    }
+
+    /// Tells the engine what the local encryption key is for the data managed
+    /// by the engine. This is only used by collections that store data
+    /// encrypted locally and is unrelated to the encryption used by Sync.
+    /// The intent is that for such collections, this key can be used to
+    /// decrypt local data before it is re-encrypted by Sync and sent to the
+    /// storage servers, and similarly, data from the storage servers will be
+    /// decrypted by Sync, then encrypted by the local encryption key before
+    /// being added to the local database.
+    ///
+    /// The expectation is that the key value is being maintained by the
+    /// embedding application in some secure way suitable for the environment
+    /// in which the app is running - eg, the OS "keychain". The value of the
+    /// key is implementation dependent - it is expected that the engine and
+    /// embedding application already have some external agreement about how
+    /// to generate keys and in what form they are exchanged. Finally, there's
+    /// an assumption that sync engines are short-lived and only live for a
+    /// single sync - this means that sync doesn't hold on to the key for an
+    /// extended period.
+    ///
+    /// This will panic if called by an engine that doesn't have explicit
+    /// support for local encryption keys as that implies a degree of confusion
+    /// which shouldn't be possible to ignore.
+    fn set_local_encryption_key(&mut self, _key: &str) -> Result<()> {
+        unimplemented!("This engine does not support local encryption");
+    }
+
+    /// `inbound` is a vector to support the case where
+    /// `get_collection_requests` returned multiple requests. The changesets are
+    /// in the same order as the requests were -- e.g. if `vec![req_a, req_b]`
+    /// was returned from `get_collection_requests`, `inbound` will have the
+    /// results from `req_a` as its first index, and those from `req_b` as it's
+    /// second.
+    fn apply_incoming(
+        &self,
+        inbound: Vec<IncomingChangeset>,
+        telem: &mut telemetry::Engine,
+    ) -> Result<OutgoingChangeset>;
+
+    fn sync_finished(
+        &self,
+        new_timestamp: ServerTimestamp,
+        records_synced: Vec<Guid>,
+    ) -> Result<()>;
+
+    /// The engine is responsible for building collection requests. Engines
+    /// typically will store a lastModified timestamp and use that to build a
+    /// request saying "give me full records since that date" - however, other
+    /// engines might do something fancier. This could even later be extended to
+    /// handle "backfills" etc
+    ///
+    /// To support more advanced use cases,  multiple requests can be returned
+    /// here - either from the same or different collections. The vast majority
+    /// of engines will just want to return zero or one item in their vector
+    /// (zero is a valid optimization when the server timestamp is the same as
+    /// the engine last saw, one when it is not)
+    ///
+    /// Important: In the case when more than one collection is requested, it's
+    /// assumed the last one is the "canonical" one. (That is, it must be for
+    /// "this" collection, its timestamp is used to represent the sync, etc).
+    /// (Note that multiple collection request support is currently unused, so
+    /// it might make sense to delete it - if we need it later, we may find a
+    /// better shape for our use-case)
+    fn get_collection_requests(
+        &self,
+        server_timestamp: ServerTimestamp,
+    ) -> Result<Vec<CollectionRequest>>;
+
+    /// Get persisted sync IDs. If they don't match the global state we'll be
+    /// `reset()` with the new IDs.
+    fn get_sync_assoc(&self) -> Result<EngineSyncAssociation>;
+
+    /// Reset the engine (and associated store) without wiping local data,
+    /// ready for a "first sync".
+    /// `assoc` defines how this store is to be associated with sync.
+    fn reset(&self, assoc: &EngineSyncAssociation) -> Result<()>;
+
+    fn wipe(&self) -> Result<()>;
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use std::iter::zip;
+
+    #[test]
+    fn test_engine_priority() {
+        fn sorted(mut engines: Vec<SyncEngineId>) -> Vec<SyncEngineId> {
+            engines.sort();
+            engines
+        }
+        assert_eq!(
+            vec![SyncEngineId::Passwords, SyncEngineId::Tabs],
+            sorted(vec![SyncEngineId::Passwords, SyncEngineId::Tabs])
+        );
+        assert_eq!(
+            vec![SyncEngineId::Passwords, SyncEngineId::Tabs],
+            sorted(vec![SyncEngineId::Tabs, SyncEngineId::Passwords])
+        );
+    }
+
+    #[test]
+    fn test_engine_enum_order() {
+        let unsorted = SyncEngineId::iter().collect::<Vec<SyncEngineId>>();
+        let mut sorted = SyncEngineId::iter().collect::<Vec<SyncEngineId>>();
+        sorted.sort();
+
+        // iterating should supply identical elements in each.
+        assert!(zip(unsorted, sorted).fold(true, |acc, (a, b)| acc && (a == b)))
+    }
+}
diff --git a/third_party/rust/sync15/src/error.rs b/third_party/rust/sync15/src/error.rs
new file mode 100644
index 0000000000..e4a043b8ac
--- /dev/null
+++ b/third_party/rust/sync15/src/error.rs
@@ -0,0 +1,138 @@
+/* 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/. */
+
+use interrupt_support::Interrupted;
+
+/// This enum is to discriminate `StorageHttpError`, and not used as an error.
+#[cfg(feature = "sync-client")]
+#[derive(Debug, Clone)]
+pub enum ErrorResponse {
+    NotFound { route: String },
+    // 401
+    Unauthorized { route: String },
+    // 412
+    PreconditionFailed { route: String },
+    // 5XX
+    ServerError { route: String, status: u16 }, // TODO: info for "retry-after" and backoff handling etc here.
+    // Other HTTP responses.
+    RequestFailed { route: String, status: u16 },
+}
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[cfg(feature = "crypto")]
+    #[error("Key {0} had wrong length, got {1}, expected {2}")]
+    BadKeyLength(&'static str, usize, usize),
+
+    #[cfg(feature = "crypto")]
+    #[error("SHA256 HMAC Mismatch error")]
+    HmacMismatch,
+
+    #[cfg(feature = "crypto")]
+    #[error("Crypto/NSS error: {0}")]
+    CryptoError(#[from] rc_crypto::Error),
+
+    #[cfg(feature = "crypto")]
+    #[error("Base64 decode error: {0}")]
+    Base64Decode(#[from] base64::DecodeError),
+
+    #[error("JSON error: {0}")]
+    JsonError(#[from] serde_json::Error),
+
+    #[error("Bad cleartext UTF8: {0}")]
+    BadCleartextUtf8(#[from] std::string::FromUtf8Error),
+
+    #[cfg(feature = "crypto")]
+    #[error("HAWK error: {0}")]
+    HawkError(#[from] rc_crypto::hawk::Error),
+
+    //
+    // Errors specific to this module.
+    //
+    #[cfg(feature = "sync-client")]
+    #[error("HTTP status {0} when requesting a token from the tokenserver")]
+    TokenserverHttpError(u16),
+
+    #[cfg(feature = "sync-client")]
+    #[error("HTTP storage error: {0:?}")]
+    StorageHttpError(ErrorResponse),
+
+    #[cfg(feature = "sync-client")]
+    #[error("Server requested backoff. Retry after {0:?}")]
+    BackoffError(std::time::SystemTime),
+
+    #[cfg(feature = "sync-client")]
+    #[error("Outgoing record is too large to upload")]
+    RecordTooLargeError,
+
+    // Do we want to record the concrete problems?
+    #[cfg(feature = "sync-client")]
+    #[error("Not all records were successfully uploaded")]
+    RecordUploadFailed,
+
+    /// Used for things like a node reassignment or an unexpected syncId
+    /// implying the app needs to "reset" its understanding of remote storage.
+    #[cfg(feature = "sync-client")]
+    #[error("The server has reset the storage for this account")]
+    StorageResetError,
+
+    #[cfg(feature = "sync-client")]
+    #[error("Unacceptable URL: {0}")]
+    UnacceptableUrl(String),
+
+    #[cfg(feature = "sync-client")]
+    #[error("Missing server timestamp header in request")]
+    MissingServerTimestamp,
+
+    #[cfg(feature = "sync-client")]
+    #[error("Unexpected server behavior during batch upload: {0}")]
+    ServerBatchProblem(&'static str),
+
+    #[cfg(feature = "sync-client")]
+    #[error("It appears some other client is also trying to setup storage; try again later")]
+    SetupRace,
+
+    #[cfg(feature = "sync-client")]
+    #[error("Client upgrade required; server storage version too new")]
+    ClientUpgradeRequired,
+
+    // This means that our global state machine needs to enter a state (such as
+    // "FreshStartNeeded", but the allowed_states don't include that state.)
+    // It typically means we are trying to do a "fast" or "read-only" sync.
+    #[cfg(feature = "sync-client")]
+    #[error("Our storage needs setting up and we can't currently do it")]
+    SetupRequired,
+
+    #[cfg(feature = "sync-client")]
+    #[error("Store error: {0}")]
+    StoreError(#[from] anyhow::Error),
+
+    #[cfg(feature = "sync-client")]
+    #[error("Network error: {0}")]
+    RequestError(#[from] viaduct::Error),
+
+    #[cfg(feature = "sync-client")]
+    #[error("Unexpected HTTP status: {0}")]
+    UnexpectedStatus(#[from] viaduct::UnexpectedStatus),
+
+    #[cfg(feature = "sync-client")]
+    #[error("URL parse error: {0}")]
+    MalformedUrl(#[from] url::ParseError),
+
+    #[error("The operation was interrupted.")]
+    Interrupted(#[from] Interrupted),
+}
+
+#[cfg(feature = "sync-client")]
+impl Error {
+    pub(crate) fn get_backoff(&self) -> Option<std::time::SystemTime> {
+        if let Error::BackoffError(time) = self {
+            Some(*time)
+        } else {
+            None
+        }
+    }
+}
diff --git a/third_party/rust/sync15/src/key_bundle.rs b/third_party/rust/sync15/src/key_bundle.rs
new file mode 100644
index 0000000000..25d8ccc5ca
--- /dev/null
+++ b/third_party/rust/sync15/src/key_bundle.rs
@@ -0,0 +1,224 @@
+/* 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/. */
+
+use crate::error::{Error, Result};
+use rc_crypto::{
+    aead::{self, OpeningKey, SealingKey},
+    rand,
+};
+
+#[derive(Clone, PartialEq, Eq, Hash)]
+pub struct KeyBundle {
+    enc_key: Vec<u8>,
+    mac_key: Vec<u8>,
+}
+
+impl std::fmt::Debug for KeyBundle {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("KeyBundle").finish()
+    }
+}
+
+impl KeyBundle {
+    /// Construct a key bundle from the already-decoded encrypt and hmac keys.
+    /// Panics (asserts) if they aren't both 32 bytes.
+    pub fn new(enc: Vec<u8>, mac: Vec<u8>) -> Result<KeyBundle> {
+        if enc.len() != 32 {
+            error_support::report_error!(
+                "sync15-key-bundle",
+                "Bad key length (enc_key): {} != 32",
+                enc.len()
+            );
+            return Err(Error::BadKeyLength("enc_key", enc.len(), 32));
+        }
+        if mac.len() != 32 {
+            error_support::report_error!(
+                "sync15-key-bundle",
+                "Bad key length (mac_key): {} != 32",
+                mac.len()
+            );
+            return Err(Error::BadKeyLength("mac_key", mac.len(), 32));
+        }
+        Ok(KeyBundle {
+            enc_key: enc,
+            mac_key: mac,
+        })
+    }
+
+    pub fn new_random() -> Result<KeyBundle> {
+        let mut buffer = [0u8; 64];
+        rand::fill(&mut buffer)?;
+        KeyBundle::from_ksync_bytes(&buffer)
+    }
+
+    pub fn from_ksync_bytes(ksync: &[u8]) -> Result<KeyBundle> {
+        if ksync.len() != 64 {
+            error_support::report_error!(
+                "sync15-key-bundle",
+                "Bad key length (kSync): {} != 64",
+                ksync.len()
+            );
+            return Err(Error::BadKeyLength("kSync", ksync.len(), 64));
+        }
+        Ok(KeyBundle {
+            enc_key: ksync[0..32].into(),
+            mac_key: ksync[32..64].into(),
+        })
+    }
+
+    pub fn from_ksync_base64(ksync: &str) -> Result<KeyBundle> {
+        let bytes = base64::decode_config(ksync, base64::URL_SAFE_NO_PAD)?;
+        KeyBundle::from_ksync_bytes(&bytes)
+    }
+
+    pub fn from_base64(enc: &str, mac: &str) -> Result<KeyBundle> {
+        let enc_bytes = base64::decode(enc)?;
+        let mac_bytes = base64::decode(mac)?;
+        KeyBundle::new(enc_bytes, mac_bytes)
+    }
+
+    #[inline]
+    pub fn encryption_key(&self) -> &[u8] {
+        &self.enc_key
+    }
+
+    #[inline]
+    pub fn hmac_key(&self) -> &[u8] {
+        &self.mac_key
+    }
+
+    #[inline]
+    pub fn to_b64_array(&self) -> [String; 2] {
+        [base64::encode(&self.enc_key), base64::encode(&self.mac_key)]
+    }
+
+    /// Decrypt the provided ciphertext with the given iv, and decodes the
+    /// result as a utf8 string.
+    pub fn decrypt(&self, enc_base64: &str, iv_base64: &str, hmac_base16: &str) -> Result<String> {
+        // Decode the expected_hmac into bytes to avoid issues if a client happens to encode
+        // this as uppercase. This shouldn't happen in practice, but doing it this way is more
+        // robust and avoids an allocation.
+        let mut decoded_hmac = vec![0u8; 32];
+        if base16::decode_slice(hmac_base16, &mut decoded_hmac).is_err() {
+            log::warn!("Garbage HMAC verification string: contained non base16 characters");
+            return Err(Error::HmacMismatch);
+        }
+        let iv = base64::decode(iv_base64)?;
+        let ciphertext_bytes = base64::decode(enc_base64)?;
+        let key_bytes = [self.encryption_key(), self.hmac_key()].concat();
+        let key = OpeningKey::new(&aead::LEGACY_SYNC_AES_256_CBC_HMAC_SHA256, &key_bytes)?;
+        let nonce = aead::Nonce::try_assume_unique_for_key(
+            &aead::LEGACY_SYNC_AES_256_CBC_HMAC_SHA256,
+            &iv,
+        )?;
+        let ciphertext_and_hmac = [ciphertext_bytes, decoded_hmac].concat();
+        let cleartext_bytes = aead::open(&key, nonce, aead::Aad::empty(), &ciphertext_and_hmac)?;
+        let cleartext = String::from_utf8(cleartext_bytes)?;
+        Ok(cleartext)
+    }
+
+    /// Encrypt using the provided IV.
+    pub fn encrypt_bytes_with_iv(
+        &self,
+        cleartext_bytes: &[u8],
+        iv: &[u8],
+    ) -> Result<(String, String)> {
+        let key_bytes = [self.encryption_key(), self.hmac_key()].concat();
+        let key = SealingKey::new(&aead::LEGACY_SYNC_AES_256_CBC_HMAC_SHA256, &key_bytes)?;
+        let nonce =
+            aead::Nonce::try_assume_unique_for_key(&aead::LEGACY_SYNC_AES_256_CBC_HMAC_SHA256, iv)?;
+        let ciphertext_and_hmac = aead::seal(&key, nonce, aead::Aad::empty(), cleartext_bytes)?;
+        let ciphertext_len = ciphertext_and_hmac.len() - key.algorithm().tag_len();
+        // Do the string conversions here so we don't have to split and copy to 2 vectors.
+        let (ciphertext, hmac_signature) = ciphertext_and_hmac.split_at(ciphertext_len);
+        let enc_base64 = base64::encode(ciphertext);
+        let hmac_base16 = base16::encode_lower(&hmac_signature);
+        Ok((enc_base64, hmac_base16))
+    }
+
+    /// Generate a random iv and encrypt with it. Return both the encrypted bytes
+    /// and the generated iv.
+    pub fn encrypt_bytes_rand_iv(
+        &self,
+        cleartext_bytes: &[u8],
+    ) -> Result<(String, String, String)> {
+        let mut iv = [0u8; 16];
+        rand::fill(&mut iv)?;
+        let (enc_base64, hmac_base16) = self.encrypt_bytes_with_iv(cleartext_bytes, &iv)?;
+        let iv_base64 = base64::encode(iv);
+        Ok((enc_base64, iv_base64, hmac_base16))
+    }
+
+    pub fn encrypt_with_iv(&self, cleartext: &str, iv: &[u8]) -> Result<(String, String)> {
+        self.encrypt_bytes_with_iv(cleartext.as_bytes(), iv)
+    }
+
+    pub fn encrypt_rand_iv(&self, cleartext: &str) -> Result<(String, String, String)> {
+        self.encrypt_bytes_rand_iv(cleartext.as_bytes())
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    const HMAC_B16: &str = "b1e6c18ac30deb70236bc0d65a46f7a4dce3b8b0e02cf92182b914e3afa5eebc";
+    const IV_B64: &str = "GX8L37AAb2FZJMzIoXlX8w==";
+    const HMAC_KEY_B64: &str = "MMntEfutgLTc8FlTLQFms8/xMPmCldqPlq/QQXEjx70=";
+    const ENC_KEY_B64: &str = "9K/wLdXdw+nrTtXo4ZpECyHFNr4d7aYHqeg3KW9+m6Q=";
+
+    const CIPHERTEXT_B64_PIECES: &[&str] = &[
+        "NMsdnRulLwQsVcwxKW9XwaUe7ouJk5Wn80QhbD80l0HEcZGCynh45qIbeYBik0lgcHbK",
+        "mlIxTJNwU+OeqipN+/j7MqhjKOGIlvbpiPQQLC6/ffF2vbzL0nzMUuSyvaQzyGGkSYM2",
+        "xUFt06aNivoQTvU2GgGmUK6MvadoY38hhW2LCMkoZcNfgCqJ26lO1O0sEO6zHsk3IVz6",
+        "vsKiJ2Hq6VCo7hu123wNegmujHWQSGyf8JeudZjKzfi0OFRRvvm4QAKyBWf0MgrW1F8S",
+        "FDnVfkq8amCB7NhdwhgLWbN+21NitNwWYknoEWe1m6hmGZDgDT32uxzWxCV8QqqrpH/Z",
+        "ggViEr9uMgoy4lYaWqP7G5WKvvechc62aqnsNEYhH26A5QgzmlNyvB+KPFvPsYzxDnSC",
+        "jOoRSLx7GG86wT59QZw=",
+    ];
+
+    const CLEARTEXT_B64_PIECES: &[&str] = &[
+        "eyJpZCI6IjVxUnNnWFdSSlpYciIsImhpc3RVcmkiOiJmaWxlOi8vL1VzZXJzL2phc29u",
+        "L0xpYnJhcnkvQXBwbGljYXRpb24lMjBTdXBwb3J0L0ZpcmVmb3gvUHJvZmlsZXMva3Nn",
+        "ZDd3cGsuTG9jYWxTeW5jU2VydmVyL3dlYXZlL2xvZ3MvIiwidGl0bGUiOiJJbmRleCBv",
+        "ZiBmaWxlOi8vL1VzZXJzL2phc29uL0xpYnJhcnkvQXBwbGljYXRpb24gU3VwcG9ydC9G",
+        "aXJlZm94L1Byb2ZpbGVzL2tzZ2Q3d3BrLkxvY2FsU3luY1NlcnZlci93ZWF2ZS9sb2dz",
+        "LyIsInZpc2l0cyI6W3siZGF0ZSI6MTMxOTE0OTAxMjM3MjQyNSwidHlwZSI6MX1dfQ==",
+    ];
+
+    #[test]
+    fn test_decrypt() {
+        let key_bundle = KeyBundle::from_base64(ENC_KEY_B64, HMAC_KEY_B64).unwrap();
+        let ciphertext = CIPHERTEXT_B64_PIECES.join("");
+        let s = key_bundle.decrypt(&ciphertext, IV_B64, HMAC_B16).unwrap();
+
+        let cleartext =
+            String::from_utf8(base64::decode(CLEARTEXT_B64_PIECES.join("")).unwrap()).unwrap();
+        assert_eq!(&cleartext, &s);
+    }
+
+    #[test]
+    fn test_encrypt() {
+        let key_bundle = KeyBundle::from_base64(ENC_KEY_B64, HMAC_KEY_B64).unwrap();
+        let iv = base64::decode(IV_B64).unwrap();
+
+        let cleartext_bytes = base64::decode(CLEARTEXT_B64_PIECES.join("")).unwrap();
+        let (enc_base64, _hmac_base16) = key_bundle
+            .encrypt_bytes_with_iv(&cleartext_bytes, &iv)
+            .unwrap();
+
+        let expect_ciphertext = CIPHERTEXT_B64_PIECES.join("");
+
+        assert_eq!(&enc_base64, &expect_ciphertext);
+
+        let (enc_base64_2, iv_base64_2, hmac_base16_2) =
+            key_bundle.encrypt_bytes_rand_iv(&cleartext_bytes).unwrap();
+        assert_ne!(&enc_base64_2, &expect_ciphertext);
+
+        let s = key_bundle
+            .decrypt(&enc_base64_2, &iv_base64_2, &hmac_base16_2)
+            .unwrap();
+        assert_eq!(&cleartext_bytes, &s.as_bytes());
+    }
+}
diff --git a/third_party/rust/sync15/src/lib.rs b/third_party/rust/sync15/src/lib.rs
new file mode 100644
index 0000000000..21fcf60cb5
--- /dev/null
+++ b/third_party/rust/sync15/src/lib.rs
@@ -0,0 +1,48 @@
+/* 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/. */
+
+#![allow(unknown_lints, clippy::implicit_hasher)]
+#![warn(rust_2018_idioms)]
+
+pub mod bso;
+#[cfg(feature = "sync-client")]
+pub mod client;
+// Type to describe device types
+mod device_type;
+// Types to describe client records
+mod client_types;
+// Note that `clients_engine` should probably be in `sync_client`, but let's not make
+// things too nested at this stage...
+#[cfg(feature = "sync-client")]
+pub mod clients_engine;
+#[cfg(feature = "crypto")]
+mod enc_payload;
+#[cfg(feature = "sync-engine")]
+pub mod engine;
+mod error;
+#[cfg(feature = "crypto")]
+mod key_bundle;
+mod record_types;
+mod server_timestamp;
+pub mod telemetry;
+
+pub use crate::client_types::{ClientData, RemoteClient};
+pub use crate::device_type::DeviceType;
+pub use crate::error::{Error, Result};
+#[cfg(feature = "crypto")]
+pub use enc_payload::EncryptedPayload;
+#[cfg(feature = "crypto")]
+pub use key_bundle::KeyBundle;
+pub use server_timestamp::ServerTimestamp;
+pub use sync_guid::Guid;
+
+// Collection names are almost always `static, so we use a `Cow`:
+// * Either a `String` or a `'static &str` can be `.into()`'d into one of these.
+// * Cloning one with a `'static &str` is extremely cheap and doesn't allocate memory.
+pub type CollectionName = std::borrow::Cow<'static, str>;
+
+// For skip_serializing_if
+fn skip_if_default<T: PartialEq + Default>(v: &T) -> bool {
+    *v == T::default()
+}
diff --git a/third_party/rust/sync15/src/record_types.rs b/third_party/rust/sync15/src/record_types.rs
new file mode 100644
index 0000000000..f8f1449fdc
--- /dev/null
+++ b/third_party/rust/sync15/src/record_types.rs
@@ -0,0 +1,51 @@
+/* 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/. */
+
+use serde_derive::*;
+use std::collections::HashMap;
+use sync_guid::Guid;
+
+// Known record formats.
+
+#[derive(Serialize, Deserialize, Debug, Clone)]
+pub struct MetaGlobalEngine {
+    pub version: usize,
+    #[serde(rename = "syncID")]
+    pub sync_id: Guid,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone)]
+pub struct MetaGlobalRecord {
+    #[serde(rename = "syncID")]
+    pub sync_id: Guid,
+    #[serde(rename = "storageVersion")]
+    pub storage_version: usize,
+    #[serde(default)]
+    pub engines: HashMap<String, MetaGlobalEngine>,
+    #[serde(default)]
+    pub declined: Vec<String>,
+}
+
+#[derive(Deserialize, Serialize, Clone, Debug, Eq, PartialEq)]
+pub struct CryptoKeysRecord {
+    pub id: Guid,
+    pub collection: String,
+    pub default: [String; 2],
+    pub collections: HashMap<String, [String; 2]>,
+}
+
+#[cfg(test)]
+#[test]
+fn test_deserialize_meta_global() {
+    let record = serde_json::json!({
+        "syncID": "abcd1234abcd",
+        "storageVersion": 1,
+    })
+    .to_string();
+    let r = serde_json::from_str::<MetaGlobalRecord>(&record).unwrap();
+    assert_eq!(r.sync_id, "abcd1234abcd");
+    assert_eq!(r.storage_version, 1);
+    assert_eq!(r.engines.len(), 0);
+    assert_eq!(r.declined.len(), 0);
+}
diff --git a/third_party/rust/sync15/src/server_timestamp.rs b/third_party/rust/sync15/src/server_timestamp.rs
new file mode 100644
index 0000000000..7dd7a7fc42
--- /dev/null
+++ b/third_party/rust/sync15/src/server_timestamp.rs
@@ -0,0 +1,122 @@
+/* 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/. */
+use std::time::Duration;
+
+/// Typesafe way to manage server timestamps without accidentally mixing them up with
+/// local ones.
+#[derive(Debug, Copy, Clone, PartialEq, Eq, PartialOrd, Default)]
+pub struct ServerTimestamp(pub i64);
+
+impl ServerTimestamp {
+    pub fn from_float_seconds(ts: f64) -> Self {
+        let rf = (ts * 1000.0).round();
+        if !rf.is_finite() || rf < 0.0 || rf >= i64::max_value() as f64 {
+            error_support::report_error!("sync15-illegal-timestamp", "Illegal timestamp: {}", ts);
+            ServerTimestamp(0)
+        } else {
+            ServerTimestamp(rf as i64)
+        }
+    }
+
+    pub fn from_millis(ts: i64) -> Self {
+        // Catch it in tests, but just complain and replace with 0 otherwise.
+        debug_assert!(ts >= 0, "Bad timestamp: {}", ts);
+        if ts >= 0 {
+            Self(ts)
+        } else {
+            error_support::report_error!(
+                "sync15-illegal-timestamp",
+                "Illegal timestamp, substituting 0: {}",
+                ts
+            );
+            Self(0)
+        }
+    }
+}
+
+// This lets us use these in hyper header! blocks.
+impl std::str::FromStr for ServerTimestamp {
+    type Err = std::num::ParseFloatError;
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let val = f64::from_str(s)?;
+        Ok(Self::from_float_seconds(val))
+    }
+}
+
+impl std::fmt::Display for ServerTimestamp {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.0 as f64 / 1000.0)
+    }
+}
+
+impl ServerTimestamp {
+    pub const EPOCH: ServerTimestamp = ServerTimestamp(0);
+
+    /// Returns None if `other` is later than `self` (Duration may not represent
+    /// negative timespans in rust).
+    #[inline]
+    pub fn duration_since(self, other: ServerTimestamp) -> Option<Duration> {
+        let delta = self.0 - other.0;
+        if delta < 0 {
+            None
+        } else {
+            Some(Duration::from_millis(delta as u64))
+        }
+    }
+
+    /// Get the milliseconds for the timestamp.
+    #[inline]
+    pub fn as_millis(self) -> i64 {
+        self.0
+    }
+}
+
+impl serde::ser::Serialize for ServerTimestamp {
+    fn serialize<S: serde::ser::Serializer>(&self, serializer: S) -> Result<S::Ok, S::Error> {
+        serializer.serialize_f64(self.0 as f64 / 1000.0)
+    }
+}
+
+impl<'de> serde::de::Deserialize<'de> for ServerTimestamp {
+    fn deserialize<D: serde::de::Deserializer<'de>>(d: D) -> Result<Self, D::Error> {
+        f64::deserialize(d).map(Self::from_float_seconds)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_server_timestamp() {
+        let t0 = ServerTimestamp(10_300_150);
+        let t1 = ServerTimestamp(10_100_050);
+        assert!(t1.duration_since(t0).is_none());
+        assert!(t0.duration_since(t1).is_some());
+        let dur = t0.duration_since(t1).unwrap();
+        assert_eq!(dur.as_secs(), 200);
+        assert_eq!(dur.subsec_nanos(), 100_000_000);
+    }
+
+    #[test]
+    fn test_serde() {
+        let ts = ServerTimestamp(123_456);
+
+        // test serialize
+        let ser = serde_json::to_string(&ts).unwrap();
+        assert_eq!("123.456".to_string(), ser);
+
+        // test deserialize of float
+        let ts: ServerTimestamp = serde_json::from_str(&ser).unwrap();
+        assert_eq!(ServerTimestamp(123_456), ts);
+
+        // test deserialize of whole number
+        let ts: ServerTimestamp = serde_json::from_str("123").unwrap();
+        assert_eq!(ServerTimestamp(123_000), ts);
+
+        // test deserialize of negative number
+        let ts: ServerTimestamp = serde_json::from_str("-123").unwrap();
+        assert_eq!(ServerTimestamp(0), ts);
+    }
+}
diff --git a/third_party/rust/sync15/src/telemetry.rs b/third_party/rust/sync15/src/telemetry.rs
new file mode 100644
index 0000000000..1d5519c937
--- /dev/null
+++ b/third_party/rust/sync15/src/telemetry.rs
@@ -0,0 +1,824 @@
+/* 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/. */
+
+//! Manage recording sync telemetry. Assumes some external telemetry
+//! library/code which manages submitting.
+
+use crate::error::Error;
+#[cfg(feature = "sync-client")]
+use crate::error::ErrorResponse;
+
+use std::collections::HashMap;
+use std::time;
+
+use serde::{ser, Serialize, Serializer};
+
+// A test helper, used by the many test modules below.
+#[cfg(test)]
+fn assert_json<T: ?Sized>(v: &T, expected: serde_json::Value)
+where
+    T: serde::Serialize,
+{
+    assert_eq!(
+        serde_json::to_value(v).expect("should get a value"),
+        expected
+    );
+}
+
+/// What we record for 'when' and 'took' in a telemetry record.
+#[derive(Debug, Serialize)]
+struct WhenTook {
+    when: f64,
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    took: u64,
+}
+
+/// What we track while recording 'when' and 'took. It serializes as a WhenTook,
+/// except when .finished() hasn't been called, in which case it panics.
+#[derive(Debug)]
+enum Stopwatch {
+    Started(time::SystemTime, time::Instant),
+    Finished(WhenTook),
+}
+
+impl Default for Stopwatch {
+    fn default() -> Self {
+        Stopwatch::new()
+    }
+}
+
+impl Stopwatch {
+    fn new() -> Self {
+        Stopwatch::Started(time::SystemTime::now(), time::Instant::now())
+    }
+
+    // For tests we don't want real timestamps because we test against literals.
+    #[cfg(test)]
+    fn finished(&self) -> Self {
+        Stopwatch::Finished(WhenTook { when: 0.0, took: 0 })
+    }
+
+    #[cfg(not(test))]
+    fn finished(&self) -> Self {
+        match self {
+            Stopwatch::Started(st, si) => {
+                let std = st.duration_since(time::UNIX_EPOCH).unwrap_or_default();
+                let when = std.as_secs() as f64; // we don't want sub-sec accuracy. Do we need to write a float?
+
+                let sid = si.elapsed();
+                let took = sid.as_secs() * 1000 + (u64::from(sid.subsec_nanos()) / 1_000_000);
+                Stopwatch::Finished(WhenTook { when, took })
+            }
+            _ => {
+                unreachable!("can't finish twice");
+            }
+        }
+    }
+}
+
+impl Serialize for Stopwatch {
+    fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        match self {
+            Stopwatch::Started(_, _) => Err(ser::Error::custom("StopWatch has not been finished")),
+            Stopwatch::Finished(c) => c.serialize(serializer),
+        }
+    }
+}
+
+#[cfg(test)]
+mod stopwatch_tests {
+    use super::*;
+
+    // A wrapper struct because we flatten - this struct should serialize with
+    // 'when' and 'took' keys (but with no 'sw'.)
+    #[derive(Debug, Serialize)]
+    struct WT {
+        #[serde(flatten)]
+        sw: Stopwatch,
+    }
+
+    #[test]
+    fn test_not_finished() {
+        let wt = WT {
+            sw: Stopwatch::new(),
+        };
+        serde_json::to_string(&wt).expect_err("unfinished stopwatch should fail");
+    }
+
+    #[test]
+    fn test() {
+        assert_json(
+            &WT {
+                sw: Stopwatch::Finished(WhenTook { when: 1.0, took: 1 }),
+            },
+            serde_json::json!({"when": 1.0, "took": 1}),
+        );
+        assert_json(
+            &WT {
+                sw: Stopwatch::Finished(WhenTook { when: 1.0, took: 0 }),
+            },
+            serde_json::json!({"when": 1.0}),
+        );
+    }
+}
+
+/// A generic "Event" - suitable for all kinds of pings (although this module
+/// only cares about the sync ping)
+#[derive(Debug, Serialize)]
+pub struct Event {
+    // We use static str references as we expect values to be literals.
+    object: &'static str,
+
+    method: &'static str,
+
+    // Maybe "value" should be a string?
+    #[serde(skip_serializing_if = "Option::is_none")]
+    value: Option<&'static str>,
+
+    // we expect the keys to be literals but values are real strings.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    extra: Option<HashMap<&'static str, String>>,
+}
+
+impl Event {
+    pub fn new(object: &'static str, method: &'static str) -> Self {
+        assert!(object.len() <= 20);
+        assert!(method.len() <= 20);
+        Self {
+            object,
+            method,
+            value: None,
+            extra: None,
+        }
+    }
+
+    pub fn value(mut self, v: &'static str) -> Self {
+        assert!(v.len() <= 80);
+        self.value = Some(v);
+        self
+    }
+
+    pub fn extra(mut self, key: &'static str, val: String) -> Self {
+        assert!(key.len() <= 15);
+        assert!(val.len() <= 85);
+        match self.extra {
+            None => self.extra = Some(HashMap::new()),
+            Some(ref e) => assert!(e.len() < 10),
+        }
+        self.extra.as_mut().unwrap().insert(key, val);
+        self
+    }
+}
+
+#[cfg(test)]
+mod test_events {
+    use super::*;
+
+    #[test]
+    #[should_panic]
+    fn test_invalid_length_ctor() {
+        Event::new("A very long object value", "Method");
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_invalid_length_extra_key() {
+        Event::new("O", "M").extra("A very long key value", "v".to_string());
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_invalid_length_extra_val() {
+        let l = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ
+                abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+        Event::new("O", "M").extra("k", l.to_string());
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_too_many_extras() {
+        let l = "abcdefghijk";
+        let mut e = Event::new("Object", "Method");
+        for i in 0..l.len() {
+            e = e.extra(&l[i..=i], "v".to_string());
+        }
+    }
+
+    #[test]
+    fn test_json() {
+        assert_json(
+            &Event::new("Object", "Method").value("Value"),
+            serde_json::json!({"object": "Object", "method": "Method", "value": "Value"}),
+        );
+
+        assert_json(
+            &Event::new("Object", "Method").extra("one", "one".to_string()),
+            serde_json::json!({"object": "Object",
+             "method": "Method",
+             "extra": {"one": "one"}
+            }),
+        )
+    }
+}
+
+/// A Sync failure.
+#[derive(Debug, Serialize)]
+#[serde(tag = "name")]
+pub enum SyncFailure {
+    #[serde(rename = "shutdownerror")]
+    Shutdown,
+
+    #[serde(rename = "othererror")]
+    Other { error: String },
+
+    #[serde(rename = "unexpectederror")]
+    Unexpected { error: String },
+
+    #[serde(rename = "autherror")]
+    Auth { from: &'static str },
+
+    #[serde(rename = "httperror")]
+    Http { code: u16 },
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn reprs() {
+        assert_json(
+            &SyncFailure::Shutdown,
+            serde_json::json!({"name": "shutdownerror"}),
+        );
+
+        assert_json(
+            &SyncFailure::Other {
+                error: "dunno".to_string(),
+            },
+            serde_json::json!({"name": "othererror", "error": "dunno"}),
+        );
+
+        assert_json(
+            &SyncFailure::Unexpected {
+                error: "dunno".to_string(),
+            },
+            serde_json::json!({"name": "unexpectederror", "error": "dunno"}),
+        );
+
+        assert_json(
+            &SyncFailure::Auth { from: "FxA" },
+            serde_json::json!({"name": "autherror", "from": "FxA"}),
+        );
+
+        assert_json(
+            &SyncFailure::Http { code: 500 },
+            serde_json::json!({"name": "httperror", "code": 500}),
+        );
+    }
+}
+
+/// Incoming record for an engine's sync
+#[derive(Debug, Default, Serialize)]
+pub struct EngineIncoming {
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    applied: u32,
+
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    failed: u32,
+
+    #[serde(rename = "newFailed")]
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    new_failed: u32,
+
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    reconciled: u32,
+}
+
+impl EngineIncoming {
+    pub fn new() -> Self {
+        Self {
+            ..Default::default()
+        }
+    }
+
+    // A helper used via skip_serializing_if
+    fn is_empty(inc: &Option<Self>) -> bool {
+        match inc {
+            Some(a) => a.applied == 0 && a.failed == 0 && a.new_failed == 0 && a.reconciled == 0,
+            None => true,
+        }
+    }
+
+    /// Increment the value of `applied` by `n`.
+    #[inline]
+    pub fn applied(&mut self, n: u32) {
+        self.applied += n;
+    }
+
+    /// Increment the value of `failed` by `n`.
+    #[inline]
+    pub fn failed(&mut self, n: u32) {
+        self.failed += n;
+    }
+
+    /// Increment the value of `new_failed` by `n`.
+    #[inline]
+    pub fn new_failed(&mut self, n: u32) {
+        self.new_failed += n;
+    }
+
+    /// Increment the value of `reconciled` by `n`.
+    #[inline]
+    pub fn reconciled(&mut self, n: u32) {
+        self.reconciled += n;
+    }
+
+    /// Get the value of `applied`. Mostly useful for testing.
+    #[inline]
+    pub fn get_applied(&self) -> u32 {
+        self.applied
+    }
+
+    /// Get the value of `failed`. Mostly useful for testing.
+    #[inline]
+    pub fn get_failed(&self) -> u32 {
+        self.failed
+    }
+
+    /// Get the value of `new_failed`. Mostly useful for testing.
+    #[inline]
+    pub fn get_new_failed(&self) -> u32 {
+        self.new_failed
+    }
+
+    /// Get the value of `reconciled`. Mostly useful for testing.
+    #[inline]
+    pub fn get_reconciled(&self) -> u32 {
+        self.reconciled
+    }
+}
+
+/// Outgoing record for an engine's sync
+#[derive(Debug, Default, Serialize)]
+pub struct EngineOutgoing {
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    sent: usize,
+
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    failed: usize,
+}
+
+impl EngineOutgoing {
+    pub fn new() -> Self {
+        EngineOutgoing {
+            ..Default::default()
+        }
+    }
+
+    #[inline]
+    pub fn sent(&mut self, n: usize) {
+        self.sent += n;
+    }
+
+    #[inline]
+    pub fn failed(&mut self, n: usize) {
+        self.failed += n;
+    }
+}
+
+/// One engine's sync.
+#[derive(Debug, Serialize)]
+pub struct Engine {
+    name: String,
+
+    #[serde(flatten)]
+    when_took: Stopwatch,
+
+    #[serde(skip_serializing_if = "EngineIncoming::is_empty")]
+    incoming: Option<EngineIncoming>,
+
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    outgoing: Vec<EngineOutgoing>, // one for each batch posted.
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(rename = "failureReason")]
+    failure: Option<SyncFailure>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    validation: Option<Validation>,
+}
+
+impl Engine {
+    pub fn new(name: impl Into<String>) -> Self {
+        Self {
+            name: name.into(),
+            when_took: Stopwatch::new(),
+            incoming: None,
+            outgoing: Vec::new(),
+            failure: None,
+            validation: None,
+        }
+    }
+
+    pub fn incoming(&mut self, inc: EngineIncoming) {
+        assert!(self.incoming.is_none());
+        self.incoming = Some(inc);
+    }
+
+    pub fn outgoing(&mut self, out: EngineOutgoing) {
+        self.outgoing.push(out);
+    }
+
+    pub fn failure(&mut self, err: impl Into<SyncFailure>) {
+        // Currently we take the first error, under the assumption that the
+        // first is the most important and all others stem from that.
+        let failure = err.into();
+        if self.failure.is_none() {
+            self.failure = Some(failure);
+        } else {
+            log::warn!(
+                "engine already has recorded a failure of {:?} - ignoring {:?}",
+                &self.failure,
+                &failure
+            );
+        }
+    }
+
+    pub fn validation(&mut self, v: Validation) {
+        assert!(self.validation.is_none());
+        self.validation = Some(v);
+    }
+
+    fn finished(&mut self) {
+        self.when_took = self.when_took.finished();
+    }
+}
+
+#[derive(Debug, Default, Serialize)]
+pub struct Validation {
+    version: u32,
+
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    problems: Vec<Problem>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(rename = "failureReason")]
+    failure: Option<SyncFailure>,
+}
+
+impl Validation {
+    pub fn with_version(version: u32) -> Validation {
+        Validation {
+            version,
+            ..Validation::default()
+        }
+    }
+
+    pub fn problem(&mut self, name: &'static str, count: usize) -> &mut Self {
+        if count > 0 {
+            self.problems.push(Problem { name, count });
+        }
+        self
+    }
+}
+
+#[derive(Debug, Default, Serialize)]
+pub struct Problem {
+    name: &'static str,
+    #[serde(skip_serializing_if = "crate::skip_if_default")]
+    count: usize,
+}
+
+#[cfg(test)]
+mod engine_tests {
+    use super::*;
+
+    #[test]
+    fn test_engine() {
+        let mut e = Engine::new("test_engine");
+        e.finished();
+        assert_json(&e, serde_json::json!({"name": "test_engine", "when": 0.0}));
+    }
+
+    #[test]
+    fn test_engine_not_finished() {
+        let e = Engine::new("test_engine");
+        serde_json::to_value(e).expect_err("unfinished stopwatch should fail");
+    }
+
+    #[test]
+    fn test_incoming() {
+        let mut i = EngineIncoming::new();
+        i.applied(1);
+        i.failed(2);
+        let mut e = Engine::new("TestEngine");
+        e.incoming(i);
+        e.finished();
+        assert_json(
+            &e,
+            serde_json::json!({"name": "TestEngine", "when": 0.0, "incoming": {"applied": 1, "failed": 2}}),
+        );
+    }
+
+    #[test]
+    fn test_outgoing() {
+        let mut o = EngineOutgoing::new();
+        o.sent(2);
+        o.failed(1);
+        let mut e = Engine::new("TestEngine");
+        e.outgoing(o);
+        e.finished();
+        assert_json(
+            &e,
+            serde_json::json!({"name": "TestEngine", "when": 0.0, "outgoing": [{"sent": 2, "failed": 1}]}),
+        );
+    }
+
+    #[test]
+    fn test_failure() {
+        let mut e = Engine::new("TestEngine");
+        e.failure(SyncFailure::Http { code: 500 });
+        e.finished();
+        assert_json(
+            &e,
+            serde_json::json!({"name": "TestEngine",
+             "when": 0.0,
+             "failureReason": {"name": "httperror", "code": 500}
+            }),
+        );
+    }
+
+    #[test]
+    fn test_raw() {
+        let mut e = Engine::new("TestEngine");
+        let mut inc = EngineIncoming::new();
+        inc.applied(10);
+        e.incoming(inc);
+        let mut out = EngineOutgoing::new();
+        out.sent(1);
+        e.outgoing(out);
+        e.failure(SyncFailure::Http { code: 500 });
+        e.finished();
+
+        assert_eq!(e.outgoing.len(), 1);
+        assert_eq!(e.incoming.as_ref().unwrap().applied, 10);
+        assert_eq!(e.outgoing[0].sent, 1);
+        assert!(e.failure.is_some());
+        serde_json::to_string(&e).expect("should get json");
+    }
+}
+
+/// A single sync. May have many engines, may have its own failure.
+#[derive(Debug, Serialize, Default)]
+pub struct SyncTelemetry {
+    #[serde(flatten)]
+    when_took: Stopwatch,
+
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    engines: Vec<Engine>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(rename = "failureReason")]
+    failure: Option<SyncFailure>,
+}
+
+impl SyncTelemetry {
+    pub fn new() -> Self {
+        Default::default()
+    }
+
+    pub fn engine(&mut self, mut e: Engine) {
+        e.finished();
+        self.engines.push(e);
+    }
+
+    pub fn failure(&mut self, failure: SyncFailure) {
+        assert!(self.failure.is_none());
+        self.failure = Some(failure);
+    }
+
+    // Note that unlike other 'finished' methods, this isn't private - someone
+    // needs to explicitly call this before handling the json payload to
+    // whatever ends up submitting it.
+    pub fn finished(&mut self) {
+        self.when_took = self.when_took.finished();
+    }
+}
+
+#[cfg(test)]
+mod sync_tests {
+    use super::*;
+
+    #[test]
+    fn test_accum() {
+        let mut s = SyncTelemetry::new();
+        let mut inc = EngineIncoming::new();
+        inc.applied(10);
+        let mut e = Engine::new("test_engine");
+        e.incoming(inc);
+        e.failure(SyncFailure::Http { code: 500 });
+        e.finished();
+        s.engine(e);
+        s.finished();
+
+        assert_json(
+            &s,
+            serde_json::json!({
+                "when": 0.0,
+                "engines": [{
+                    "name":"test_engine",
+                    "when":0.0,
+                    "incoming": {
+                        "applied": 10
+                    },
+                    "failureReason": {
+                        "name": "httperror",
+                        "code": 500
+                    }
+                }]
+            }),
+        );
+    }
+
+    #[test]
+    fn test_multi_engine() {
+        let mut inc_e1 = EngineIncoming::new();
+        inc_e1.applied(1);
+        let mut e1 = Engine::new("test_engine");
+        e1.incoming(inc_e1);
+
+        let mut inc_e2 = EngineIncoming::new();
+        inc_e2.failed(1);
+        let mut e2 = Engine::new("test_engine_2");
+        e2.incoming(inc_e2);
+        let mut out_e2 = EngineOutgoing::new();
+        out_e2.sent(1);
+        e2.outgoing(out_e2);
+
+        let mut s = SyncTelemetry::new();
+        s.engine(e1);
+        s.engine(e2);
+        s.failure(SyncFailure::Http { code: 500 });
+        s.finished();
+        assert_json(
+            &s,
+            serde_json::json!({
+                "when": 0.0,
+                "engines": [{
+                    "name": "test_engine",
+                    "when": 0.0,
+                    "incoming": {
+                        "applied": 1
+                    }
+                },{
+                    "name": "test_engine_2",
+                    "when": 0.0,
+                    "incoming": {
+                        "failed": 1
+                    },
+                    "outgoing": [{
+                        "sent": 1
+                    }]
+                }],
+                "failureReason": {
+                    "name": "httperror",
+                    "code": 500
+                }
+            }),
+        );
+    }
+}
+
+/// The Sync ping payload, as documented at
+/// https://firefox-source-docs.mozilla.org/toolkit/components/telemetry/telemetry/data/sync-ping.html.
+/// May have many syncs, may have many events. However, due to the architecture
+/// of apps which use these components, this payload is almost certainly not
+/// suitable for submitting directly. For example, we will always return a
+/// payload with exactly 1 sync, and it will not know certain other fields
+/// in the payload, such as the *hashed* FxA device ID (see
+/// https://searchfox.org/mozilla-central/rev/c3ebaf6de2d481c262c04bb9657eaf76bf47e2ac/services/sync/modules/browserid_identity.js#185
+/// for an example of how the device ID is constructed). The intention is that
+/// consumers of this will use this to create a "real" payload - eg, accumulating
+/// until some threshold number of syncs is reached, and contributing
+/// additional data which only the consumer knows.
+#[derive(Debug, Serialize, Default)]
+pub struct SyncTelemetryPing {
+    version: u32,
+
+    uid: Option<String>,
+
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    events: Vec<Event>,
+
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    syncs: Vec<SyncTelemetry>,
+}
+
+impl SyncTelemetryPing {
+    pub fn new() -> Self {
+        Self {
+            version: 1,
+            ..Default::default()
+        }
+    }
+
+    pub fn uid(&mut self, uid: String) {
+        if let Some(ref existing) = self.uid {
+            if *existing != uid {
+                log::warn!("existing uid ${} being replaced by {}", existing, uid);
+            }
+        }
+        self.uid = Some(uid);
+    }
+
+    pub fn sync(&mut self, mut s: SyncTelemetry) {
+        s.finished();
+        self.syncs.push(s);
+    }
+
+    pub fn event(&mut self, e: Event) {
+        self.events.push(e);
+    }
+}
+
+ffi_support::implement_into_ffi_by_json!(SyncTelemetryPing);
+
+#[cfg(test)]
+mod ping_tests {
+    use super::*;
+    #[test]
+    fn test_ping() {
+        let engine = Engine::new("test");
+        let mut s = SyncTelemetry::new();
+        s.engine(engine);
+        let mut p = SyncTelemetryPing::new();
+        p.uid("user-id".into());
+        p.sync(s);
+        let event = Event::new("foo", "bar");
+        p.event(event);
+        assert_json(
+            &p,
+            serde_json::json!({
+                "events": [{
+                    "method": "bar", "object": "foo"
+                }],
+                "syncs": [{
+                    "engines": [{
+                        "name": "test", "when": 0.0
+                    }],
+                    "when": 0.0
+                }],
+                "uid": "user-id",
+                "version": 1
+            }),
+        );
+    }
+}
+
+impl<'a> From<&'a Error> for SyncFailure {
+    fn from(e: &Error) -> SyncFailure {
+        match e {
+            #[cfg(feature = "sync-client")]
+            Error::TokenserverHttpError(status) => {
+                if *status == 401 {
+                    SyncFailure::Auth {
+                        from: "tokenserver",
+                    }
+                } else {
+                    SyncFailure::Http { code: *status }
+                }
+            }
+            #[cfg(feature = "sync-client")]
+            Error::BackoffError(_) => SyncFailure::Http { code: 503 },
+            #[cfg(feature = "sync-client")]
+            Error::StorageHttpError(ref e) => match e {
+                ErrorResponse::NotFound { .. } => SyncFailure::Http { code: 404 },
+                ErrorResponse::Unauthorized { .. } => SyncFailure::Auth { from: "storage" },
+                ErrorResponse::PreconditionFailed { .. } => SyncFailure::Http { code: 412 },
+                ErrorResponse::ServerError { status, .. } => SyncFailure::Http { code: *status },
+                ErrorResponse::RequestFailed { status, .. } => SyncFailure::Http { code: *status },
+            },
+            #[cfg(feature = "crypto")]
+            Error::CryptoError(ref e) => SyncFailure::Unexpected {
+                error: e.to_string(),
+            },
+            #[cfg(feature = "sync-client")]
+            Error::RequestError(ref e) => SyncFailure::Unexpected {
+                error: e.to_string(),
+            },
+            #[cfg(feature = "sync-client")]
+            Error::UnexpectedStatus(ref e) => SyncFailure::Http { code: e.status },
+            Error::Interrupted(ref e) => SyncFailure::Unexpected {
+                error: e.to_string(),
+            },
+            e => SyncFailure::Other {
+                error: e.to_string(),
+            },
+        }
+    }
+}