feat: protect queued changes at shutdown and sign-out (B66)

While the queue holds changes the drain can send, the daemon holds a logind
shutdown inhibitor through systemd-inhibit, so desktops show why a power-off
is held up. Parked transient files and changes that need the user never hold
it. pdfs sync flush retries failed changes and waits for the queue, exiting 3
when what is left needs the user and 4 on timeout. Signing out, in the app or
with pdfs logout, says how many changes are not on Proton Drive yet.
This commit is contained in:
nvrl committed 2026-11-21 23:42:54 +01:00
1 parent 2c4c95812f
commit 2b5d6f2436
18 files changed
+3644 -3252

No files matched your search

+102 -5
View File
@@ -158,8 +158,12 @@ enum Command {
#[arg(long)]
forget: bool,
},
/// Forget the stored session.
Logout,
/// Forget the stored session. Asks first while changes are still queued.
Logout {
/// Do not ask first.
#[arg(long)]
yes: bool,
},
/// Show account login state, and mount status if a daemon is running.
Status,
/// List every local Proton Drive location.
@@ -857,6 +861,13 @@ enum SyncCmd {
},
/// List uploads and changes not yet on Proton Drive.
Queue,
/// Wait until every queued change is on Proton Drive. Exits 0 once nothing
/// is left, 3 when what is left needs you (see `sync issues`), 4 on timeout.
Flush {
/// Give up after this many seconds.
#[arg(long)]
timeout: Option<u64>,
},
/// List queued changes Proton Drive refused or that keep failing, and what to do.
Issues,
/// Save the content of a queued upload (by id from `sync queue`) to a file.
@@ -1047,7 +1058,7 @@ fn main() -> Result<()> {
cli.command,
Command::Login { .. }
| Command::Unlock { .. }
| Command::Logout
| Command::Logout { .. }
| Command::Mount { .. }
| Command::Daemon { .. }
) {
@@ -1057,7 +1068,7 @@ fn main() -> Result<()> {
Command::Login { username, remember } => cmd_login(username, remember),
Command::Unlock { remember } => cmd_unlock(remember),
Command::Lock { forget } => cmd_lock(forget),
Command::Logout => cmd_logout(),
Command::Logout { yes } => cmd_logout(yes),
Command::Status => cmd_status(),
Command::Locations => cmd_locations(),
Command::Mount { mountpoint } => cmd_mount(mountpoint),
@@ -1362,6 +1373,7 @@ fn cmd_sync(action: SyncCmd) -> Result<()> {
})?)?
}
SyncCmd::Queue => return cmd_sync_queue(),
SyncCmd::Flush { timeout } => return cmd_sync_flush(timeout),
SyncCmd::Issues => return cmd_sync_issues(),
SyncCmd::Export { id, dest } => return cmd_sync_export(id, &dest),
SyncCmd::Limit { up, down } => return cmd_sync_limit(up, down),
@@ -1465,6 +1477,71 @@ fn cmd_sync_queue() -> Result<()> {
Ok(())
}
/// `pdfs sync flush`: wait for the queue to drain.
///
/// Failed changes are retried at once rather than after their backoff. The
/// wait ends early, with exit code 3, when everything left is waiting on the
/// user: refused by Drive, failing again and again, or a temporary file
/// waiting for its final name. Paused or offline, it keeps waiting and says
/// why.
fn cmd_sync_flush(timeout: Option<u64>) -> Result<()> {
const POLL: std::time::Duration = std::time::Duration::from_secs(1);
let deadline =
timeout.map(|secs| std::time::Instant::now() + std::time::Duration::from_secs(secs));
ok_or_bail(control_request(CtlRequest::RetryPendingOp { id: None })?)?;
let mut last_line = String::new();
loop {
let items = match control_request(CtlRequest::ListPendingOps)? {
CtlResponse::PendingOps { items } => items,
CtlResponse::Error { message, kind } => bail!("{}", cli_error(kind, &message)),
other => bail!("unexpected response: {other:?}"),
};
let waiting = items
.iter()
.filter(|op| !op.parked && !op.failing && op.issue.is_none())
.count();
if items.is_empty() {
finish_line(&last_line);
println!("Everything is on Proton Drive.");
return Ok(());
}
if waiting == 0 {
finish_line(&last_line);
eprintln!(
"{} queued change(s) need you before they can be sent. See `pdfs sync issues`.",
items.len()
);
std::process::exit(3);
}
let reason = match control_request(CtlRequest::Status)? {
CtlResponse::Status { paused: true, .. } => " (sync is paused; `pdfs sync resume`)",
CtlResponse::Status { online: false, .. } => " (offline; waiting for Proton Drive)",
_ => "",
};
let line = format!("{waiting} change(s) left to send{reason}");
if line != last_line {
if !last_line.is_empty() {
print!("\r\x1b[2K");
}
print!("{line}");
last_line = line;
}
if deadline.is_some_and(|deadline| std::time::Instant::now() >= deadline) {
finish_line(&last_line);
eprintln!("Timed out with {waiting} change(s) still queued.");
std::process::exit(4);
}
std::thread::sleep(POLL);
}
}
/// End a progress line so the next message starts on its own.
fn finish_line(last_line: &str) {
if !last_line.is_empty() {
println!();
}
}
/// `pdfs sync issues`: the queued changes that need the user, and what to do.
fn cmd_sync_issues() -> Result<()> {
let response = control_request(CtlRequest::ListPendingOps)?;
@@ -2298,7 +2375,27 @@ fn key_state_name() -> &'static str {
}
}
fn cmd_logout() -> Result<()> {
fn cmd_logout(yes: bool) -> Result<()> {
// Queued changes stay staged on this computer and are sent after the next
// sign-in, but until then they exist nowhere else. Without a running
// daemon there is no count to show.
if !yes
&& let Ok(CtlResponse::Status {
pending_uploads,
pending_changes,
parked_uploads,
..
}) = control_request(CtlRequest::Status)
{
let queued = pending_uploads + pending_changes + parked_uploads;
if queued > 0 {
println!(
"{queued} change(s) are not on Proton Drive yet. They stay on this computer and are \
sent after you sign in again. Run `pdfs sync flush` first to send them now."
);
confirm("Sign out anyway?")?;
}
}
if auth::logout()? {
println!("Signed out: the session is ended on Proton and cleared here.");
} else {
+129
View File
@@ -0,0 +1,129 @@
//! Hold off a shutdown while queued changes exist only on this computer.
//!
//! A change made through the mount is staged here and sent by the drain. Until
//! it lands, this computer has the only copy, so powering off then risks it
//! more than the user expects (`docs/BUGS.md` B66). While the queue holds work
//! the drain can still send, the daemon takes a logind shutdown inhibitor
//! through `systemd-inhibit`, which desktops show in their power-off dialog
//! with the reason. It is let go as soon as the queue empties.
//!
//! Work that cannot drain on its own — a temporary file waiting for its final
//! name, or a change Drive refused or that keeps failing — never holds the
//! lock, or a shutdown would be held up for something that is not coming.
//! Without `systemd-inhibit` nothing happens.
use std::process::{Child, Command, Stdio};
use std::time::Duration;
use pdfs_core::db::PendingCounts;
use tracing::{debug, warn};
use super::Core;
/// How often the queue is looked at.
const CHECK_EVERY: Duration = Duration::from_secs(5);
/// Changes the drain can still send by itself.
fn sendable(counts: &PendingCounts) -> i64 {
(counts.uploads + counts.changes - counts.parked - counts.failing).max(0)
}
/// Run until the daemon stops, holding the inhibitor while changes wait.
pub(crate) fn run(core: Core) {
let mut held: Option<Child> = None;
loop {
let waiting = core
.db
.pending_op_counts()
.map(|counts| sendable(&counts))
.unwrap_or(0);
// The helper can go away (logind restarted, or it was killed); the
// lock is then taken again on the next round.
if let Some(child) = &mut held
&& !matches!(child.try_wait(), Ok(None))
{
held = None;
}
match (&mut held, waiting > 0) {
(None, true) => held = take(waiting),
(Some(child), false) => {
release(child);
held = None;
}
_ => {}
}
if !core.shutdown.sleep(CHECK_EVERY) {
break;
}
}
if let Some(mut child) = held {
release(&mut child);
}
}
fn take(waiting: i64) -> Option<Child> {
let why = if waiting == 1 {
"Sending 1 change to Proton Drive".to_string()
} else {
format!("Sending {waiting} changes to Proton Drive")
};
match Command::new("systemd-inhibit")
.args([
"--what=shutdown",
"--who=Proton Drive",
&format!("--why={why}"),
"--mode=block",
"sleep",
"infinity",
])
.stdin(Stdio::null())
.stdout(Stdio::null())
.stderr(Stdio::null())
.spawn()
{
Ok(child) => {
debug!(waiting, "holding off shutdown while changes are queued");
Some(child)
}
Err(error) if error.kind() == std::io::ErrorKind::NotFound => None,
Err(error) => {
warn!(%error, "taking the shutdown inhibitor failed");
None
}
}
}
fn release(child: &mut Child) {
let _ = child.kill();
let _ = child.wait();
debug!("queue drained; shutdown no longer held off");
}
#[cfg(test)]
mod tests {
use super::*;
fn counts(uploads: i64, changes: i64, parked: i64, failing: i64) -> PendingCounts {
PendingCounts {
uploads,
changes,
parked,
failing,
last_error: None,
}
}
#[test]
fn only_work_the_drain_can_send_holds_off_a_shutdown() {
assert_eq!(sendable(&counts(0, 0, 0, 0)), 0);
assert_eq!(sendable(&counts(2, 1, 0, 0)), 3);
assert_eq!(sendable(&counts(1, 0, 1, 0)), 0);
assert_eq!(sendable(&counts(2, 0, 0, 2)), 0);
assert_eq!(sendable(&counts(3, 2, 1, 1)), 3);
}
#[test]
fn counts_that_overlap_never_go_negative() {
assert_eq!(sendable(&counts(1, 0, 1, 1)), 0);
}
}
+1
View File
@@ -92,6 +92,7 @@ use drive::{DriveApi, RevisionRead};
mod events;
mod filesystem;
mod gallery;
mod inhibit;
mod keeper;
mod link;
pub use filesystem::ProtonFs;
+11
View File
@@ -793,6 +793,17 @@ pub(crate) fn mount_with(
warn!(%error, "start My Files share-id repair failed");
}
}
// Hold off a shutdown while changes exist only here (B66). Only the real
// daemon: a simulation's queue is not the user's data.
if matches!(host, Host::Daemon) {
let inhibit_core = core.clone();
if let Err(error) = std::thread::Builder::new()
.name("pdfs-inhibit".into())
.spawn(move || crate::inhibit::run(inhibit_core))
{
warn!(%error, "start the shutdown inhibitor failed");
}
}
// Stop signals (SIGTERM from `systemctl --user stop`, SIGINT from Ctrl-C)
// are delivered onto the async runtime; bridge them onto a sync channel so
+33 -3
View File
@@ -675,15 +675,45 @@ fn verify_in_browser(
/// Sign out, after a confirmation: disable+stop the mount service (so the
/// daemon isn't left running without credentials), forget the stored session,
/// and drop back to the login page.
///
/// Changes still queued are named first: they stay staged here and are sent
/// after the next sign-in, but until then they exist nowhere else.
pub(crate) fn sign_out(ui: &Rc<Ui>) {
let rx = spawn_request(ui.dirs.control_socket(), Request::Status);
let ui = ui.clone();
glib::spawn_future_local(async move {
let queued = match rx.recv().await {
Ok(Ok(Response::Status {
pending_uploads,
pending_changes,
parked_uploads,
..
})) => pending_uploads + pending_changes + parked_uploads,
_ => 0,
};
confirm_sign_out(&ui, queued);
});
}
fn confirm_sign_out(ui: &Rc<Ui>, queued: u64) {
let Some(window) = ui_window(ui) else { return };
let ui = ui.clone();
let mut body = gettext(
"Proton Drive will disconnect and stop syncing until you sign in again. Files already on this computer stay where they are.",
);
if queued > 0 {
let warning = ngettext_f(
"{n} change is not on Proton Drive yet. It stays on this computer and is sent after you sign in again.",
"{n} changes are not on Proton Drive yet. They stay on this computer and are sent after you sign in again.",
queued,
&[],
);
body = format!("{warning}\n\n{body}");
}
confirm_destructive(
&window,
&gettext("Sign Out?"),
&gettext(
"Proton Drive will disconnect and stop syncing until you sign in again. Files already on this computer stay where they are.",
),
&body,
&gettext("Sign Out"),
move || {
let galleries = ui.drive_galleries.borrow().upgrade();