OXYGENHUB
# cce63af9ec54 oxygenflow / crates / flow / src / home.rs
11.7 KiB · Rust · last changed flow builds for Windows, offering the WASM tier 10 days ago source blame history raw
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
104
fn yes() -> bool {
105
    true
106
}
107
108
impl RunnerConfig {
109
    /// The labels offered to the hub: the configured ones, and `egress`
110
    /// when egress is on.
111
    pub fn offered_labels(&self) -> Vec<String> {
112
        let mut l = self.labels.clone();
113
        if self.egress && !l.iter().any(|x| x == flow_core::EGRESS_LABEL) {
114
            l.push(flow_core::EGRESS_LABEL.to_string());
115
        }
116
        l
117
    }
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 {
150
    pub fn locate(explicit: Option<&Path>) -> Result<Home> {
151
        let root = match explicit {
152
            Some(p) => p.to_path_buf(),
153
            None => match std::env::var_os("FLOW_HOME") {
154
                Some(p) => PathBuf::from(p),
155
                None => PathBuf::from(std::env::var_os("HOME").context("HOME is not set")?).join(".local/share/flow"),
156
            },
157
        };
158
        let file = root.join("flow.toml");
159
        let mut config: Config = if file.is_file() {
160
            let text = std::fs::read_to_string(&file).with_context(|| format!("read {}", file.display()))?;
161
            toml::from_str(&text).with_context(|| format!("parse {}", file.display()))?
162
        } else {
163
            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.
167
        let registered = root.join(RUNNER_TOML);
168
        if config.runner.is_none() && registered.is_file() {
169
            let text = std::fs::read_to_string(&registered).with_context(|| format!("read {}", registered.display()))?;
170
            config.runner = Some(toml::from_str(&text).with_context(|| format!("parse {}", registered.display()))?);
171
        }
172
        for d in ["cas", "work", "repos"] {
173
            std::fs::create_dir_all(root.join(d)).with_context(|| format!("create {}", root.join(d).display()))?;
174
        }
175
        Ok(Home { root, config, verified: Mutex::new(HashMap::new()) })
176
    }
177
178
    pub fn resolve(&self, p: &Path) -> PathBuf {
179
        if p.is_absolute() {
180
            p.to_path_buf()
181
        } else {
182
            self.root.join(p)
183
        }
184
    }
185
186
    /// Where a rootfs image with this hash lives.
187
    pub fn rootfs(&self, hash: &str) -> PathBuf {
188
        self.root.join("cas").join(format!("{hash}.ext4"))
189
    }
190
191
    /// An image's manifest (`cas/<hash>.json`): what `flow rootfs import`
192
    /// found out about it.
193
    pub fn image_manifest(&self, hash: &str) -> PathBuf {
194
        self.root.join("cas").join(format!("{hash}.json"))
195
    }
196
197
    /// The architecture an image was imported for, when its manifest says.
198
    pub fn image_arch(&self, hash: &str) -> Option<String> {
199
        let text = std::fs::read_to_string(self.image_manifest(hash)).ok()?;
200
        let v: serde_json::Value = serde_json::from_str(&text).ok()?;
201
        v["arch"].as_str().map(String::from)
202
    }
203
204
    /// Every image with a known architecture (spec §11.2).
205
    pub fn image_archs(&self) -> std::collections::BTreeMap<String, String> {
206
        self.images().into_iter().filter_map(|h| self.image_arch(&h).map(|a| (h, a))).collect()
207
    }
208
209
    /// The rootfs images this home holds: the hashes of `cas/*.ext4`.
210
    pub fn images(&self) -> Vec<String> {
211
        let Ok(dir) = std::fs::read_dir(self.root.join("cas")) else {
212
            return Vec::new();
213
        };
214
        let mut out: Vec<String> = dir
215
            .filter_map(|e| {
216
                let name = e.ok()?.file_name().into_string().ok()?;
217
                let hash = name.strip_suffix(".ext4")?;
218
                (hash.len() == 64 && hash.chars().all(|c| c.is_ascii_hexdigit())).then(|| hash.to_string())
219
            })
220
            .collect();
221
        out.sort();
222
        out
223
    }
224
225
    /// A pinned file, checked against its BLAKE3 before use (again only
226
    /// when its size or modification time changed since).
227
    pub fn verified(&self, path: &Path, blake3_hex: &str) -> Result<PathBuf> {
228
        let meta = std::fs::metadata(path).with_context(|| format!("{} is missing", path.display()))?;
229
        let stamp = (meta.len(), meta.modified()?);
230
        if self.verified.lock().unwrap().get(path) == Some(&stamp) {
231
            return Ok(path.to_path_buf());
232
        }
233
        let got = hash_file(path)?;
234
        if got != blake3_hex {
235
            bail!("{} has BLAKE3 {got}, but {blake3_hex} is pinned", path.display());
236
        }
237
        self.verified.lock().unwrap().insert(path.to_path_buf(), stamp);
238
        Ok(path.to_path_buf())
239
    }
240
}
241
242
pub fn hash_file(path: &Path) -> Result<String> {
243
    let mut h = blake3::Hasher::new();
244
    h.update_reader(std::fs::File::open(path).with_context(|| format!("open {}", path.display()))?)?;
245
    Ok(h.finalize().to_hex().to_string())
246
}
247
248
/// The key a runner signs receipts with (Oxygen spec §14.6).
249
pub struct RunnerKey {
250
    pub identity: oxy_core::identity::Identity,
251
    pub secret: oxy_core::identity::SecretKey,
252
}
253
254
impl Home {
255
    /// The runner's key, made on first use as `runner.key` (owner-only)
256
    /// and `runner.identity`, named `name`.
257
    pub fn runner_key(&self, name: &str) -> Result<RunnerKey> {
258
        use oxy_core::identity::{Identity, SecretKey};
259
        let key_path = self.root.join("runner.key");
260
        let id_path = self.root.join("runner.identity");
261
        if key_path.exists() {
262
            let (key, secret) = SecretKey::decode(&std::fs::read(&key_path)?).context("runner.key")?;
263
            let identity = Identity::decode(&std::fs::read(&id_path)?).context("runner.identity")?;
264
            if identity.hash() != key {
265
                bail!("runner.identity does not belong to runner.key");
266
            }
267
            return Ok(RunnerKey { identity, secret });
268
        }
269
        let secret = SecretKey::generate();
270
        let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).map(|d| d.as_millis() as u64).unwrap_or(0);
271
        let identity = secret.identity(&format!("{name} (OxygenFlow runner)"), "", now);
272
        std::fs::write(&id_path, identity.encode())?;
273
        let tmp = self.root.join("runner.key.tmp");
274
        {
275
            use std::io::Write;
276
            let mut f = crate::portable::private(std::fs::OpenOptions::new().write(true).create(true).truncate(true)).open(&tmp)?;
277
            f.write_all(&secret.encode(&identity.hash()))?;
278
            f.sync_all()?;
279
        }
280
        std::fs::rename(&tmp, &key_path)?;
281
        Ok(RunnerKey { identity, secret })
282
    }
283
}
284
285
#[cfg(test)]
286
mod tests {
287
    use super::*;
288
289
    #[test]
290
    fn the_runner_key_is_made_once_and_kept_private() {
291
        use std::os::unix::fs::PermissionsExt;
292
        let dir = tempfile::tempdir().unwrap();
293
        let home = Home::locate(Some(dir.path())).unwrap();
294
        let a = home.runner_key("earth-1").unwrap();
295
        assert_eq!(a.identity.name, "earth-1 (OxygenFlow runner)");
296
        let mode = std::fs::metadata(dir.path().join("runner.key")).unwrap().permissions().mode();
297
        assert_eq!(mode & 0o777, 0o600);
298
        let b = home.runner_key("renamed").unwrap();
299
        assert_eq!(a.identity, b.identity, "the same key after a restart");
300
        let sig = b.secret.sign(oxy_core::identity::Domain::Attest, &b.identity.hash(), &oxy_core::hash::ObjectRef([1; 32]));
301
        oxy_core::identity::verify(&a.identity, &sig, oxy_core::identity::Domain::Attest, &oxy_core::hash::ObjectRef([1; 32])).unwrap();
302
    }
303
}