src/tools/daemon.rs

   1 //! Helpers for daemons/services.
   2
   3 use std::ffi::CString;
   4 use std::future::Future;
   5 use std::io::{Read, Write};
   6 use std::os::raw::{c_char, c_uchar, c_int};
   7 use std::os::unix::io::{AsRawFd, FromRawFd, IntoRawFd, RawFd};
   8 use std::os::unix::ffi::OsStrExt;
   9 use std::panic::UnwindSafe;
  10 use std::pin::Pin;
  11 use std::task::{Context, Poll};
  12 use std::path::PathBuf;
  13
  14 use anyhow::{bail, format_err, Error};
  15 use futures::future::{self, Either};
  16
  17 use proxmox::tools::io::{ReadExt, WriteExt};
  18
  19 use crate::server;
  20 use crate::tools::{fd_change_cloexec, self};
  21
  22 #[link(name = "systemd")]
  23 extern "C" {
  24     fn sd_journal_stream_fd(identifier: *const c_uchar, priority: c_int, level_prefix: c_int) -> c_int;
  25 }
  26
  27 // Unfortunately FnBox is nightly-only and Box<FnOnce> is unusable, so just use Box<Fn>...
  28 pub type BoxedStoreFunc = Box<dyn FnMut() -> Result<String, Error> + UnwindSafe + Send>;
  29
  30 /// Helper trait to "store" something in the environment to be re-used after re-executing the
  31 /// service on a reload.
  32 pub trait Reloadable: Sized {
  33     fn restore(var: &str) -> Result<Self, Error>;
  34     fn get_store_func(&self) -> Result<BoxedStoreFunc, Error>;
  35 }
  36
  37 /// Manages things to be stored and reloaded upon reexec.
  38 /// Anything which should be restorable should be instantiated via this struct's `restore` method,
  39 #[derive(Default)]
  40 pub struct Reloader {
  41     pre_exec: Vec<PreExecEntry>,
  42     self_exe: PathBuf,
  43 }
  44
  45 // Currently we only need environment variables for storage, but in theory we could also add
  46 // variants which need temporary files or pipes...
  47 struct PreExecEntry {
  48     name: &'static str, // Feel free to change to String if necessary...
  49     store_fn: BoxedStoreFunc,
  50 }
  51
  52 impl Reloader {
  53     pub fn new() -> Result<Self, Error> {
  54         Ok(Self {
  55             pre_exec: Vec::new(),
  56
  57             // Get the path to our executable as PathBuf
  58             self_exe: std::fs::read_link("/proc/self/exe")?,
  59         })
  60     }
  61
  62     /// Restore an object from an environment variable of the given name, or, if none exists, uses
  63     /// the function provided in the `or_create` parameter to instantiate the new "first" instance.
  64     ///
  65     /// Values created via this method will be remembered for later re-execution.
  66     pub async fn restore<T, F, U>(&mut self, name: &'static str, or_create: F) -> Result<T, Error>
  67     where
  68         T: Reloadable,
  69         F: FnOnce() -> U,
  70         U: Future<Output = Result<T, Error>>,
  71     {
  72         let res = match std::env::var(name) {
  73             Ok(varstr) => T::restore(&varstr)?,
  74             Err(std::env::VarError::NotPresent) => or_create().await?,
  75             Err(_) => bail!("variable {} has invalid value", name),
  76         };
  77
  78         self.pre_exec.push(PreExecEntry {
  79             name,
  80             store_fn: res.get_store_func()?,
  81         });
  82         Ok(res)
  83     }
  84
  85     fn pre_exec(self) -> Result<(), Error> {
  86         for mut item in self.pre_exec {
  87             std::env::set_var(item.name, (item.store_fn)()?);
  88         }
  89         Ok(())
  90     }
  91
  92     pub fn fork_restart(self) -> Result<(), Error> {
  93         // Get our parameters as Vec<CString>
  94         let args = std::env::args_os();
  95         let mut new_args = Vec::with_capacity(args.len());
  96         for arg in args {
  97             new_args.push(CString::new(arg.as_bytes())?);
  98         }
  99
 100         // Synchronisation pipe:
 101         let (pold, pnew) = super::socketpair()?;
 102
 103         // Start ourselves in the background:
 104         use nix::unistd::{fork, ForkResult};
 105         match unsafe { fork() } {
 106             Ok(ForkResult::Child) => {
 107                 // Double fork so systemd can supervise us without nagging...
 108                 match unsafe { fork() } {
 109                     Ok(ForkResult::Child) => {
 110                         std::mem::drop(pold);
 111                         // At this point we call pre-exec helpers. We must be certain that if they fail for
 112                         // whatever reason we can still call `_exit()`, so use catch_unwind.
 113                         match std::panic::catch_unwind(move || {
 114                             let mut pnew = unsafe {
 115                                 std::fs::File::from_raw_fd(pnew.into_raw_fd())
 116                             };
 117                             let pid = nix::unistd::Pid::this();
 118                             if let Err(e) = unsafe { pnew.write_host_value(pid.as_raw()) } {
 119                                 log::error!("failed to send new server PID to parent: {}", e);
 120                                 unsafe {
 121                                     libc::_exit(-1);
 122                                 }
 123                             }
 124
 125                             let mut ok = [0u8];
 126                             if let Err(e) = pnew.read_exact(&mut ok) {
 127                                 log::error!("parent vanished before notifying systemd: {}", e);
 128                                 unsafe {
 129                                     libc::_exit(-1);
 130                                 }
 131                             }
 132                             assert_eq!(ok[0], 1, "reload handshake should have sent a 1 byte");
 133
 134                             std::mem::drop(pnew);
 135
 136                             // Try to reopen STDOUT/STDERR journald streams to get correct PID in logs
 137                             let ident = CString::new(self.self_exe.file_name().unwrap().as_bytes()).unwrap();
 138                             let ident = ident.as_bytes();
 139                             let fd = unsafe { sd_journal_stream_fd(ident.as_ptr(), libc::LOG_INFO, 1) };
 140                             if fd >= 0 && fd != 1 {
 141                                 let fd = proxmox::tools::fd::Fd(fd); // add drop handler
 142                                 nix::unistd::dup2(fd.as_raw_fd(), 1)?;
 143                             } else {
 144                                 log::error!("failed to update STDOUT journal redirection ({})", fd);
 145                             }
 146                             let fd = unsafe { sd_journal_stream_fd(ident.as_ptr(), libc::LOG_ERR, 1) };
 147                             if fd >= 0 && fd != 2 {
 148                                 let fd = proxmox::tools::fd::Fd(fd); // add drop handler
 149                                 nix::unistd::dup2(fd.as_raw_fd(), 2)?;
 150                             } else {
 151                                 log::error!("failed to update STDERR journal redirection ({})", fd);
 152                             }
 153
 154                             self.do_reexec(new_args)
 155                         })
 156                         {
 157                             Ok(Ok(())) => eprintln!("do_reexec returned!"),
 158                             Ok(Err(err)) => eprintln!("do_reexec failed: {}", err),
 159                             Err(_) => eprintln!("panic in re-exec"),
 160                         }
 161                     }
 162                     Ok(ForkResult::Parent { child }) => {
 163                         std::mem::drop((pold, pnew));
 164                         log::debug!("forked off a new server (second pid: {})", child);
 165                     }
 166                     Err(e) => log::error!("fork() failed, restart delayed: {}", e),
 167                 }
 168                 // No matter how we managed to get here, this is the time where we bail out quickly:
 169                 unsafe {
 170                     libc::_exit(-1)
 171                 }
 172             }
 173             Ok(ForkResult::Parent { child }) => {
 174                 log::debug!("forked off a new server (first pid: {}), waiting for 2nd pid", child);
 175                 std::mem::drop(pnew);
 176                 let mut pold = unsafe {
 177                     std::fs::File::from_raw_fd(pold.into_raw_fd())
 178                 };
 179                 let child = nix::unistd::Pid::from_raw(match unsafe { pold.read_le_value() } {
 180                     Ok(v) => v,
 181                     Err(e) => {
 182                         log::error!("failed to receive pid of double-forked child process: {}", e);
 183                         // systemd will complain but won't kill the service...
 184                         return Ok(());
 185                     }
 186                 });
 187
 188                 if let Err(e) = systemd_notify(SystemdNotify::MainPid(child)) {
 189                     log::error!("failed to notify systemd about the new main pid: {}", e);
 190                 }
 191
 192                 // notify child that it is now the new main process:
 193                 if let Err(e) = pold.write_all(&[1u8]) {
 194                     log::error!("child vanished during reload: {}", e);
 195                 }
 196
 197                 Ok(())
 198             }
 199             Err(e) => {
 200                 log::error!("fork() failed, restart delayed: {}", e);
 201                 Ok(())
 202             }
 203         }
 204     }
 205
 206     fn do_reexec(self, args: Vec<CString>) -> Result<(), Error> {
 207         let exe = CString::new(self.self_exe.as_os_str().as_bytes())?;
 208         self.pre_exec()?;
 209         nix::unistd::setsid()?;
 210         let args: Vec<&std::ffi::CStr> = args.iter().map(|s| s.as_ref()).collect();
 211         nix::unistd::execvp(&exe, &args)?;
 212         panic!("exec misbehaved");
 213     }
 214 }
 215
 216 // For now all we need to do is store and reuse a tcp listening socket:
 217 impl Reloadable for tokio::net::TcpListener {
 218     // NOTE: The socket must not be closed when the store-function is called:
 219     // FIXME: We could become "independent" of the TcpListener and its reference to the file
 220     // descriptor by `dup()`ing it (and check if the listener still exists via kcmp()?)
 221     fn get_store_func(&self) -> Result<BoxedStoreFunc, Error> {
 222         let mut fd_opt = Some(tools::Fd(
 223             nix::fcntl::fcntl(self.as_raw_fd(), nix::fcntl::FcntlArg::F_DUPFD_CLOEXEC(0))?
 224         ));
 225         Ok(Box::new(move || {
 226             let fd = fd_opt.take().unwrap();
 227             fd_change_cloexec(fd.as_raw_fd(), false)?;
 228             Ok(fd.into_raw_fd().to_string())
 229         }))
 230     }
 231
 232     fn restore(var: &str) -> Result<Self, Error> {
 233         let fd = var.parse::<u32>()
 234             .map_err(|e| format_err!("invalid file descriptor: {}", e))?
 235             as RawFd;
 236         fd_change_cloexec(fd, true)?;
 237         Ok(Self::from_std(
 238             unsafe { std::net::TcpListener::from_raw_fd(fd) },
 239         )?)
 240     }
 241 }
 242
 243 pub struct NotifyReady;
 244
 245 impl Future for NotifyReady {
 246     type Output = Result<(), Error>;
 247
 248     fn poll(self: Pin<&mut Self>, _cx: &mut Context) -> Poll<Result<(), Error>> {
 249         systemd_notify(SystemdNotify::Ready)?;
 250         Poll::Ready(Ok(()))
 251     }
 252 }
 253
 254 /// This creates a future representing a daemon which reloads itself when receiving a SIGHUP.
 255 /// If this is started regularly, a listening socket is created. In this case, the file descriptor
 256 /// number will be remembered in `PROXMOX_BACKUP_LISTEN_FD`.
 257 /// If the variable already exists, its contents will instead be used to restore the listening
 258 /// socket.  The finished listening socket is then passed to the `create_service` function which
 259 /// can be used to setup the TLS and the HTTP daemon.
 260 pub async fn create_daemon<F, S>(
 261     address: std::net::SocketAddr,
 262     create_service: F,
 263     service_name: &str,
 264 ) -> Result<(), Error>
 265 where
 266     F: FnOnce(tokio::net::TcpListener, NotifyReady) -> Result<S, Error>,
 267     S: Future<Output = ()> + Unpin,
 268 {
 269     let mut reloader = Reloader::new()?;
 270
 271     let listener: tokio::net::TcpListener = reloader.restore(
 272         "PROXMOX_BACKUP_LISTEN_FD",
 273         move || async move { Ok(tokio::net::TcpListener::bind(&address).await?) },
 274     ).await?;
 275
 276     let server_future = create_service(listener, NotifyReady)?;
 277     let shutdown_future = server::shutdown_future();
 278
 279     let finish_future = match future::select(server_future, shutdown_future).await {
 280         Either::Left((_, _)) => {
 281             crate::tools::request_shutdown(); // make sure we are in shutdown mode
 282             None
 283         }
 284         Either::Right((_, server_future)) => Some(server_future),
 285     };
 286
 287     let mut reloader = Some(reloader);
 288
 289     if server::is_reload_request() {
 290         log::info!("daemon reload...");
 291         if let Err(e) = systemd_notify(SystemdNotify::Reloading) {
 292             log::error!("failed to notify systemd about the state change: {}", e);
 293         }
 294         if let Err(e) = reloader.take().unwrap().fork_restart() {
 295             log::error!("error during reload: {}", e);
 296             let _ = systemd_notify(SystemdNotify::Status("error during reload".to_string()));
 297         }
 298     } else {
 299         log::info!("daemon shutting down...");
 300     }
 301
 302     if let Some(future) = finish_future {
 303         future.await;
 304     }
 305
 306     // FIXME: this is a hack, replace with sd_notify_barrier when available
 307     if server::is_reload_request() {
 308         check_service_is_active(service_name).await?;
 309     }
 310
 311     log::info!("daemon shut down...");
 312     Ok(())
 313 }
 314
 315 pub async fn check_service_is_active(service: &str) -> Result<(), Error> {
 316     for _ in 0..5 {
 317         tokio::time::delay_for(std::time::Duration::new(5, 0)).await;
 318         if let Ok(output) = tokio::process::Command::new("systemctl")
 319             .args(&["is-active", service])
 320             .output()
 321             .await
 322         {
 323             if let Ok(text) = String::from_utf8(output.stdout) {
 324                 if text.trim().trim_start() != "reloading" {
 325                     return Ok(());
 326                 }
 327             }
 328         }
 329     }
 330
 331     Ok(())
 332 }
 333
 334 #[link(name = "systemd")]
 335 extern "C" {
 336     fn sd_notify(unset_environment: c_int, state: *const c_char) -> c_int;
 337 }
 338
 339 pub enum SystemdNotify {
 340     Ready,
 341     Reloading,
 342     Stopping,
 343     Status(String),
 344     MainPid(nix::unistd::Pid),
 345 }
 346
 347 pub fn systemd_notify(state: SystemdNotify) -> Result<(), Error> {
 348     let message = match state {
 349         SystemdNotify::Ready => CString::new("READY=1"),
 350         SystemdNotify::Reloading => CString::new("RELOADING=1"),
 351         SystemdNotify::Stopping => CString::new("STOPPING=1"),
 352         SystemdNotify::Status(msg) => CString::new(format!("STATUS={}", msg)),
 353         SystemdNotify::MainPid(pid) => CString::new(format!("MAINPID={}", pid)),
 354     }?;
 355     let rc = unsafe { sd_notify(0, message.as_ptr()) };
 356     if rc < 0 {
 357         bail!(
 358             "systemd_notify failed: {}",
 359             std::io::Error::from_raw_os_error(-rc),
 360         );
 361     }
 362     Ok(())
 363 }