OXYGENHUB

Coverage / crates/flow/src/home.rs 79.2%

133 of 168 lines run, 22 of 61 functions · run #196 on main · run missed

1
//! `$FLOW_HOME` (default `~/.local/share/flow`): the pinned parts of each
2
//! isolation tier, rootfs images by hash, scratch space and the runner's
3
//! configuration, all described by `flow.toml`.
4
//!
5
//! ```toml
6
//! [microvm]
7
//! firecracker = { path = "bin/firecracker", blake3 = "…" }
8
//! kernel = { path = "kernel/vmlinux", blake3 = "…" }
9
//! init = { path = "bin/flow-init", blake3 = "…" }
10
//!
11
//! [runner]                       # only for `flow runner`
12
//! hub = "https://hub.example:7343"
13
//! token = "…"                    # a runner token from the hub's [flow] runner_tokens
14
//! ca = "/path/to/hub-ca.pem"     # when the hub's certificate is self-signed
15
//! name = "earth-1"
16
//! ```
17
//!
18
//! A runner signs its receipts with a composite key kept beside `flow.toml`
19
//! (`runner.key`, `runner.identity`), made on first start.
20
21
use anyhow::{bail, Context, Result};
22
use serde::Deserialize;
23
use std::collections::HashMap;
24
use std::path::{Path, PathBuf};
25
use std::sync::Mutex;
26
27
#[derive(Clone, Debug, Deserialize)]
28
pub struct Pinned {
29
    pub path: PathBuf,
30
    pub blake3: String,
31
}
32
33
#[derive(Clone, Debug, Deserialize)]
34
pub struct MicrovmConfig {
35
    pub firecracker: Pinned,
36
    pub kernel: Pinned,
37
    pub init: Pinned,
38
}
39
40
/// Where `flow runner register` keeps the runner's hub and credential.
41
pub const RUNNER_TOML: &str = "runner.toml";
42
43
#[derive(Clone, Debug, Deserialize)]
44
pub struct RunnerConfig {
45
    pub hub: String,
46
    pub token: String,
47
    pub ca: Option<PathBuf>,
48
    pub name: Option<String>,
49
    /// Jobs run at once (default 1).
50
    #[serde(default)]
51
    pub max_jobs: Option<usize>,
52
    /// Labels the runner offers, e.g. `gpu`, `big-disk`.
53
    #[serde(default)]
54
    pub labels: Vec<String>,
55
    /// Memory (MiB) and vCPUs the jobs running at once may use together.
56
    #[serde(default)]
57
    pub memory_mib: Option<u32>,
58
    #[serde(default)]
59
    pub cpus: Option<u32>,
60
    /// Sharing images with other runners (docs/jetstream.md §8.1): absent,
61
    /// the runner neither serves nor asks peers.
62
    #[serde(default)]
63
    pub swarm: Option<SwarmConfig>,
64
    /// The build cache's size limit in GiB (#28; default 64).
65
    #[serde(default)]
66
    pub cache_gb: Option<u64>,
67
    /// The share of its disk, in percent, a kept build cache may fill
68
    /// (oxygenflow#72; default 60): over it, the files the job did not use
69
    /// are pruned, oldest first, and a cache still over it is not kept.
70
    #[serde(default)]
71
    pub cache_prune_percent: Option<u8>,
72
    /// Let microVM jobs reach the hosts their sandbox allows, through this
73
    /// runner (oxygenflow#55); off unless set. The runner then offers the
74
    /// `egress` label, which such jobs need.
75
    #[serde(default)]
76
    pub egress: bool,
77
    /// Sign each job's receipt with a fresh key the hub certifies for that
78
    /// job, just before signing (keyless, oxygenhub#241), rather than with
79
    /// the runner's own key; off unless set. When the hub cannot certify,
80
    /// the runner's key signs, and the job's log says why.
81
    #[serde(default)]
82
    pub keyless: bool,
83
    /// Serve this runner's metrics at this address, e.g. `127.0.0.1:9464`
84
    /// (oxygenflow#35); off unless set.
85
    #[serde(default)]
86
    pub metrics: Option<String>,
87
    /// A bearer token a scrape of the metrics must send.
88
    #[serde(default)]
89
    pub metrics_token: Option<String>,
90
    /// `[runner.update]`: whether this runner takes updates (#85).
91
    #[serde(default)]
92
    pub update: Option<UpdateConfig>,
93
}
94
95
/// `[runner.update]` in flow.toml (docs/jetstream.md §8.2): the runner's
96
/// last word on updates, whatever the hub's policy says.
97
#[derive(Clone, Debug, Deserialize)]
98
pub struct UpdateConfig {
99
    /// `false` refuses every update, so a site can hold a machine back.
100
    #[serde(default = "yes")]
101
    pub allow: bool,
102
}
103
1040×
fn yes() -> bool {
1050×
    true
1060×
}
107
108
impl RunnerConfig {
109
    /// The labels offered to the hub: the configured ones, and `egress`
110
    /// when egress is on.
1110×
    pub fn offered_labels(&self) -> Vec<String> {
1120×
        let mut l = self.labels.clone();
1130×
        if self.egress && !l.iter().any(|x| x == flow_core::EGRESS_LABEL) {
1140×
            l.push(flow_core::EGRESS_LABEL.to_string());
1150×
        }
1160×
        l
1170×
    }
118
}
119
120
/// `[runner.swarm]` in flow.toml, or `[swarm]` in runner.toml.
121
#[derive(Clone, Debug, Deserialize)]
122
pub struct SwarmConfig {
123
    /// Where peers reach this runner, e.g. `0.0.0.0:7401`.
124
    pub listen: String,
125
    /// The address peers are given (default: `listen`).
126
    #[serde(default)]
127
    pub advertise: Option<String>,
128
    /// Peers are only ever in the same site.
129
    #[serde(default)]
130
    pub site: String,
131
    /// The most this runner sends to peers, MiB a second (default 50).
132
    #[serde(default)]
133
    pub upload_mib_s: Option<u32>,
134
}
135
136
#[derive(Clone, Debug, Default, Deserialize)]
137
pub struct Config {
138
    pub microvm: Option<MicrovmConfig>,
139
    pub runner: Option<RunnerConfig>,
140
}
141
142
pub struct Home {
143
    pub root: PathBuf,
144
    pub config: Config,
145
    /// Files already verified, by path, with the (len, mtime) they had then.
146
    verified: Mutex<HashMap<PathBuf, (u64, std::time::SystemTime)>>,
147
}
148
149
impl Home {
15018×
    pub fn locate(explicit: Option<&Path>) -> Result<Home> {
15118×
        let root = match explicit {
15218×
            Some(p) => p.to_path_buf(),
1530×
            None => match std::env::var_os("FLOW_HOME") {
1540×
                Some(p) => PathBuf::from(p),
1550×
                None => PathBuf::from(std::env::var_os("HOME").context("HOME is not set")?).join(".local/share/flow"),
156
            },
157
        };
15818×
        let file = root.join("flow.toml");
15918×
        let mut config: Config = if file.is_file() {
1600×
            let text = std::fs::read_to_string(&file).with_context(|| format!("read {}", file.display()))?;
1610×
            toml::from_str(&text).with_context(|| format!("parse {}", file.display()))?
162
        } else {
16318×
            Config::default()
164
        };
165
        // `flow runner register` writes the runner's hub and credential to
166
        // runner.toml, which stands in for a [runner] section flow.toml lacks.
16718×
        let registered = root.join(RUNNER_TOML);
16818×
        if config.runner.is_none() && registered.is_file() {
1699×
            let text = std::fs::read_to_string(&registered).with_context(|| format!("read {}", registered.display()))?;
1709×
            config.runner = Some(toml::from_str(&text).with_context(|| format!("parse {}", registered.display()))?);
1719×
        }
17254×
        for d in ["cas", "work", "repos"] {
17354×
            std::fs::create_dir_all(root.join(d)).with_context(|| format!("create {}", root.join(d).display()))?;
174
        }
17518×
        Ok(Home { root, config, verified: Mutex::new(HashMap::new()) })
17618×
    }
177
1780×
    pub fn resolve(&self, p: &Path) -> PathBuf {
1790×
        if p.is_absolute() {
1800×
            p.to_path_buf()
181
        } else {
1820×
            self.root.join(p)
183
        }
1840×
    }
185
186
    /// Where a rootfs image with this hash lives.
18735×
    pub fn rootfs(&self, hash: &str) -> PathBuf {
18835×
        self.root.join("cas").join(format!("{hash}.ext4"))
18935×
    }
190
191
    /// An image's manifest (`cas/<hash>.json`): what `flow rootfs import`
192
    /// found out about it.
19314×
    pub fn image_manifest(&self, hash: &str) -> PathBuf {
19414×
        self.root.join("cas").join(format!("{hash}.json"))
19514×
    }
196
197
    /// The architecture an image is for: as its manifest records, else
198
    /// read from the image's own shell (spec §11.2) and recorded, so an
199
    /// image fetched or imported without one is read once and a runner
200
    /// declines what it cannot run instead of failing it. `None` when the
201
    /// image is not here or names no architecture.
2029×
    pub fn image_arch(&self, hash: &str) -> Option<String> {
2039×
        let path = self.image_manifest(hash);
2049×
        let mut v: serde_json::Value =
2059×
            std::fs::read_to_string(&path).ok().and_then(|t| serde_json::from_str(&t).ok()).unwrap_or_else(|| serde_json::json!({}));
2069×
        if let Some(a) = v["arch"].as_str() {
2073×
            return Some(a.to_string());
2086×
        }
2096×
        let image = self.rootfs(hash);
2106×
        if v.get("arch_read_ms").is_some() || !image.is_file() {
2111×
            return None;
2125×
        }
2135×
        let arch = jet::rootfs::arch_of_file(&image);
2145×
        let o = v.as_object_mut()?;
2155×
        if let Some(a) = arch {
2161×
            o.insert("arch".into(), a.into());
2174×
        }
2185×
        let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).map(|d| d.as_millis() as u64).unwrap_or(0);
2195×
        o.insert("arch_read_ms".into(), now.into());
2205×
        let tmp = path.with_extension("json.part");
2215×
        if std::fs::write(&tmp, v.to_string()).is_ok() {
2225×
            let _ = std::fs::rename(&tmp, &path);
2235×
        }
2245×
        arch.map(String::from)
2259×
    }
226
227
    /// Every image with a known architecture (spec §11.2).
2281×
    pub fn image_archs(&self) -> std::collections::BTreeMap<String, String> {
2291×
        self.images().into_iter().filter_map(|h| self.image_arch(&h).map(|a| (h, a))).collect()
2301×
    }
231
232
    /// The rootfs images this home holds: the hashes of `cas/*.ext4`.
23310×
    pub fn images(&self) -> Vec<String> {
23410×
        let Ok(dir) = std::fs::read_dir(self.root.join("cas")) else {
2350×
            return Vec::new();
236
        };
23710×
        let mut out: Vec<String> = dir
23810×
            .filter_map(|e| {
2395×
                let name = e.ok()?.file_name().into_string().ok()?;
2405×
                let hash = name.strip_suffix(".ext4")?;
241256×
                (hash.len() == 64 && hash.chars().all(|c| c.is_ascii_hexdigit())).then(|| hash.to_string())
2425×
            })
24310×
            .collect();
24410×
        out.sort();
24510×
        out
24610×
    }
247
248
    /// A pinned file, checked against its BLAKE3 before use (again only
249
    /// when its size or modification time changed since).
2504×
    pub fn verified(&self, path: &Path, blake3_hex: &str) -> Result<PathBuf> {
2514×
        let meta = std::fs::metadata(path).with_context(|| format!("{} is missing", path.display()))?;
2524×
        let stamp = (meta.len(), meta.modified()?);
2534×
        if self.verified.lock().unwrap().get(path) == Some(&stamp) {
2541×
            return Ok(path.to_path_buf());
2553×
        }
2563×
        let mut got = hash_file(path)?;
2573×
        if got != blake3_hex {
258
            // The read may have come from bad cached pages rather than the
259
            // disk (oxygenflow#22: a RAM fault on Earth failed every job for
260
            // hours while the file was intact): drop the file's pages and
261
            // read it again from the disk before giving up.
2621×
            drop_cached(path);
2631×
            let again = hash_file(path)?;
2641×
            if again == blake3_hex {
2650×
                eprintln!(
2660×
                    "warning: {} hashed {got} from the page cache but {again} from the disk: the cached copy was bad (check this machine's memory)",
2670×
                    path.display()
2680×
                );
2691×
            }
2701×
            got = again;
2712×
        }
2723×
        if got != blake3_hex {
2731×
            bail!("{} has BLAKE3 {got}, but {blake3_hex} is pinned", path.display());
2742×
        }
2752×
        self.verified.lock().unwrap().insert(path.to_path_buf(), stamp);
2762×
        Ok(path.to_path_buf())
2774×
    }
278
}
279
280
/// Ask the kernel to forget a file's cached pages, so the next read comes
281
/// from the disk. Best effort: dirty pages and other platforms stay.
2821×
fn drop_cached(path: &Path) {
283
    #[cfg(target_os = "linux")]
2841×
    if let Ok(f) = std::fs::File::open(path) {
285
        use std::os::fd::AsRawFd;
286
        // SAFETY: a valid descriptor; the advice only affects caching.
2871×
        unsafe {
2881×
            libc::posix_fadvise(f.as_raw_fd(), 0, 0, libc::POSIX_FADV_DONTNEED);
2891×
        }
2900×
    }
291
    #[cfg(not(target_os = "linux"))]
292
    let _ = path;
2931×
}
294
29521×
pub fn hash_file(path: &Path) -> Result<String> {
29621×
    let mut h = blake3::Hasher::new();
29721×
    h.update_reader(std::fs::File::open(path).with_context(|| format!("open {}", path.display()))?)?;
29821×
    Ok(h.finalize().to_hex().to_string())
29921×
}
300
301
/// The key a runner signs receipts with (Oxygen spec §14.6).
302
pub struct RunnerKey {
303
    pub identity: oxy_core::identity::Identity,
304
    pub secret: oxy_core::identity::SecretKey,
305
}
306
307
impl Home {
308
    /// The runner's key, made on first use as `runner.key` (owner-only)
309
    /// and `runner.identity`, named `name`.
3104×
    pub fn runner_key(&self, name: &str) -> Result<RunnerKey> {
311
        use oxy_core::identity::{Identity, SecretKey};
3124×
        let key_path = self.root.join("runner.key");
3134×
        let id_path = self.root.join("runner.identity");
3144×
        if key_path.exists() {
3152×
            let (key, secret) = SecretKey::decode(&std::fs::read(&key_path)?).context("runner.key")?;
3162×
            let identity = Identity::decode(&std::fs::read(&id_path)?).context("runner.identity")?;
3172×
            if identity.hash() != key {
3180×
                bail!("runner.identity does not belong to runner.key");
3192×
            }
3202×
            return Ok(RunnerKey { identity, secret });
3212×
        }
3222×
        let secret = SecretKey::generate();
3232×
        let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).map(|d| d.as_millis() as u64).unwrap_or(0);
3242×
        let identity = secret.identity(&format!("{name} (OxygenFlow runner)"), "", now);
3252×
        std::fs::write(&id_path, identity.encode())?;
3262×
        let tmp = self.root.join("runner.key.tmp");
327
        {
328
            use std::io::Write;
3292×
            let mut f = crate::portable::private(std::fs::OpenOptions::new().write(true).create(true).truncate(true)).open(&tmp)?;
3302×
            f.write_all(&secret.encode(&identity.hash()))?;
3312×
            f.sync_all()?;
332
        }
3332×
        std::fs::rename(&tmp, &key_path)?;
3342×
        Ok(RunnerKey { identity, secret })
3354×
    }
336
}
337
338
#[cfg(test)]
339
mod tests {
340
    use super::*;
341
342
    #[test]
3431×
    fn the_runner_key_is_made_once_and_kept_private() {
344
        use std::os::unix::fs::PermissionsExt;
3451×
        let dir = tempfile::tempdir().unwrap();
3461×
        let home = Home::locate(Some(dir.path())).unwrap();
3471×
        let a = home.runner_key("earth-1").unwrap();
3481×
        assert_eq!(a.identity.name, "earth-1 (OxygenFlow runner)");
3491×
        let mode = std::fs::metadata(dir.path().join("runner.key")).unwrap().permissions().mode();
3501×
        assert_eq!(mode & 0o777, 0o600);
3511×
        let b = home.runner_key("renamed").unwrap();
3521×
        assert_eq!(a.identity, b.identity, "the same key after a restart");
3531×
        let sig = b.secret.sign(oxy_core::identity::Domain::Attest, &b.identity.hash(), &oxy_core::hash::ObjectRef([1; 32]));
3541×
        oxy_core::identity::verify(&a.identity, &sig, oxy_core::identity::Domain::Attest, &oxy_core::hash::ObjectRef([1; 32])).unwrap();
3551×
    }
356
}