OXYGENHUB

Coverage / crates/flow/src/home.rs 80.6%

133 of 165 lines run, 22 of 59 functions · run #181 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
}
91
92
impl RunnerConfig {
93
    /// The labels offered to the hub: the configured ones, and `egress`
94
    /// when egress is on.
950×
    pub fn offered_labels(&self) -> Vec<String> {
960×
        let mut l = self.labels.clone();
970×
        if self.egress && !l.iter().any(|x| x == flow_core::EGRESS_LABEL) {
980×
            l.push(flow_core::EGRESS_LABEL.to_string());
990×
        }
1000×
        l
1010×
    }
102
}
103
104
/// `[runner.swarm]` in flow.toml, or `[swarm]` in runner.toml.
105
#[derive(Clone, Debug, Deserialize)]
106
pub struct SwarmConfig {
107
    /// Where peers reach this runner, e.g. `0.0.0.0:7401`.
108
    pub listen: String,
109
    /// The address peers are given (default: `listen`).
110
    #[serde(default)]
111
    pub advertise: Option<String>,
112
    /// Peers are only ever in the same site.
113
    #[serde(default)]
114
    pub site: String,
115
    /// The most this runner sends to peers, MiB a second (default 50).
116
    #[serde(default)]
117
    pub upload_mib_s: Option<u32>,
118
}
119
120
#[derive(Clone, Debug, Default, Deserialize)]
121
pub struct Config {
122
    pub microvm: Option<MicrovmConfig>,
123
    pub runner: Option<RunnerConfig>,
124
}
125
126
pub struct Home {
127
    pub root: PathBuf,
128
    pub config: Config,
129
    /// Files already verified, by path, with the (len, mtime) they had then.
130
    verified: Mutex<HashMap<PathBuf, (u64, std::time::SystemTime)>>,
131
}
132
133
impl Home {
13418×
    pub fn locate(explicit: Option<&Path>) -> Result<Home> {
13518×
        let root = match explicit {
13618×
            Some(p) => p.to_path_buf(),
1370×
            None => match std::env::var_os("FLOW_HOME") {
1380×
                Some(p) => PathBuf::from(p),
1390×
                None => PathBuf::from(std::env::var_os("HOME").context("HOME is not set")?).join(".local/share/flow"),
140
            },
141
        };
14218×
        let file = root.join("flow.toml");
14318×
        let mut config: Config = if file.is_file() {
1440×
            let text = std::fs::read_to_string(&file).with_context(|| format!("read {}", file.display()))?;
1450×
            toml::from_str(&text).with_context(|| format!("parse {}", file.display()))?
146
        } else {
14718×
            Config::default()
148
        };
149
        // `flow runner register` writes the runner's hub and credential to
150
        // runner.toml, which stands in for a [runner] section flow.toml lacks.
15118×
        let registered = root.join(RUNNER_TOML);
15218×
        if config.runner.is_none() && registered.is_file() {
1539×
            let text = std::fs::read_to_string(&registered).with_context(|| format!("read {}", registered.display()))?;
1549×
            config.runner = Some(toml::from_str(&text).with_context(|| format!("parse {}", registered.display()))?);
1559×
        }
15654×
        for d in ["cas", "work", "repos"] {
15754×
            std::fs::create_dir_all(root.join(d)).with_context(|| format!("create {}", root.join(d).display()))?;
158
        }
15918×
        Ok(Home { root, config, verified: Mutex::new(HashMap::new()) })
16018×
    }
161
1620×
    pub fn resolve(&self, p: &Path) -> PathBuf {
1630×
        if p.is_absolute() {
1640×
            p.to_path_buf()
165
        } else {
1660×
            self.root.join(p)
167
        }
1680×
    }
169
170
    /// Where a rootfs image with this hash lives.
17135×
    pub fn rootfs(&self, hash: &str) -> PathBuf {
17235×
        self.root.join("cas").join(format!("{hash}.ext4"))
17335×
    }
174
175
    /// An image's manifest (`cas/<hash>.json`): what `flow rootfs import`
176
    /// found out about it.
17714×
    pub fn image_manifest(&self, hash: &str) -> PathBuf {
17814×
        self.root.join("cas").join(format!("{hash}.json"))
17914×
    }
180
181
    /// The architecture an image is for: as its manifest records, else
182
    /// read from the image's own shell (spec §11.2) and recorded, so an
183
    /// image fetched or imported without one is read once and a runner
184
    /// declines what it cannot run instead of failing it. `None` when the
185
    /// image is not here or names no architecture.
1869×
    pub fn image_arch(&self, hash: &str) -> Option<String> {
1879×
        let path = self.image_manifest(hash);
1889×
        let mut v: serde_json::Value =
1899×
            std::fs::read_to_string(&path).ok().and_then(|t| serde_json::from_str(&t).ok()).unwrap_or_else(|| serde_json::json!({}));
1909×
        if let Some(a) = v["arch"].as_str() {
1913×
            return Some(a.to_string());
1926×
        }
1936×
        let image = self.rootfs(hash);
1946×
        if v.get("arch_read_ms").is_some() || !image.is_file() {
1951×
            return None;
1965×
        }
1975×
        let arch = jet::rootfs::arch_of_file(&image);
1985×
        let o = v.as_object_mut()?;
1995×
        if let Some(a) = arch {
2001×
            o.insert("arch".into(), a.into());
2014×
        }
2025×
        let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).map(|d| d.as_millis() as u64).unwrap_or(0);
2035×
        o.insert("arch_read_ms".into(), now.into());
2045×
        let tmp = path.with_extension("json.part");
2055×
        if std::fs::write(&tmp, v.to_string()).is_ok() {
2065×
            let _ = std::fs::rename(&tmp, &path);
2075×
        }
2085×
        arch.map(String::from)
2099×
    }
210
211
    /// Every image with a known architecture (spec §11.2).
2121×
    pub fn image_archs(&self) -> std::collections::BTreeMap<String, String> {
2131×
        self.images().into_iter().filter_map(|h| self.image_arch(&h).map(|a| (h, a))).collect()
2141×
    }
215
216
    /// The rootfs images this home holds: the hashes of `cas/*.ext4`.
21710×
    pub fn images(&self) -> Vec<String> {
21810×
        let Ok(dir) = std::fs::read_dir(self.root.join("cas")) else {
2190×
            return Vec::new();
220
        };
22110×
        let mut out: Vec<String> = dir
22210×
            .filter_map(|e| {
2235×
                let name = e.ok()?.file_name().into_string().ok()?;
2245×
                let hash = name.strip_suffix(".ext4")?;
225256×
                (hash.len() == 64 && hash.chars().all(|c| c.is_ascii_hexdigit())).then(|| hash.to_string())
2265×
            })
22710×
            .collect();
22810×
        out.sort();
22910×
        out
23010×
    }
231
232
    /// A pinned file, checked against its BLAKE3 before use (again only
233
    /// when its size or modification time changed since).
2344×
    pub fn verified(&self, path: &Path, blake3_hex: &str) -> Result<PathBuf> {
2354×
        let meta = std::fs::metadata(path).with_context(|| format!("{} is missing", path.display()))?;
2364×
        let stamp = (meta.len(), meta.modified()?);
2374×
        if self.verified.lock().unwrap().get(path) == Some(&stamp) {
2381×
            return Ok(path.to_path_buf());
2393×
        }
2403×
        let mut got = hash_file(path)?;
2413×
        if got != blake3_hex {
242
            // The read may have come from bad cached pages rather than the
243
            // disk (oxygenflow#22: a RAM fault on Earth failed every job for
244
            // hours while the file was intact): drop the file's pages and
245
            // read it again from the disk before giving up.
2461×
            drop_cached(path);
2471×
            let again = hash_file(path)?;
2481×
            if again == blake3_hex {
2490×
                eprintln!(
2500×
                    "warning: {} hashed {got} from the page cache but {again} from the disk: the cached copy was bad (check this machine's memory)",
2510×
                    path.display()
2520×
                );
2531×
            }
2541×
            got = again;
2552×
        }
2563×
        if got != blake3_hex {
2571×
            bail!("{} has BLAKE3 {got}, but {blake3_hex} is pinned", path.display());
2582×
        }
2592×
        self.verified.lock().unwrap().insert(path.to_path_buf(), stamp);
2602×
        Ok(path.to_path_buf())
2614×
    }
262
}
263
264
/// Ask the kernel to forget a file's cached pages, so the next read comes
265
/// from the disk. Best effort: dirty pages and other platforms stay.
2661×
fn drop_cached(path: &Path) {
267
    #[cfg(target_os = "linux")]
2681×
    if let Ok(f) = std::fs::File::open(path) {
269
        use std::os::fd::AsRawFd;
270
        // SAFETY: a valid descriptor; the advice only affects caching.
2711×
        unsafe {
2721×
            libc::posix_fadvise(f.as_raw_fd(), 0, 0, libc::POSIX_FADV_DONTNEED);
2731×
        }
2740×
    }
275
    #[cfg(not(target_os = "linux"))]
276
    let _ = path;
2771×
}
278
27921×
pub fn hash_file(path: &Path) -> Result<String> {
28021×
    let mut h = blake3::Hasher::new();
28121×
    h.update_reader(std::fs::File::open(path).with_context(|| format!("open {}", path.display()))?)?;
28221×
    Ok(h.finalize().to_hex().to_string())
28321×
}
284
285
/// The key a runner signs receipts with (Oxygen spec §14.6).
286
pub struct RunnerKey {
287
    pub identity: oxy_core::identity::Identity,
288
    pub secret: oxy_core::identity::SecretKey,
289
}
290
291
impl Home {
292
    /// The runner's key, made on first use as `runner.key` (owner-only)
293
    /// and `runner.identity`, named `name`.
2944×
    pub fn runner_key(&self, name: &str) -> Result<RunnerKey> {
295
        use oxy_core::identity::{Identity, SecretKey};
2964×
        let key_path = self.root.join("runner.key");
2974×
        let id_path = self.root.join("runner.identity");
2984×
        if key_path.exists() {
2992×
            let (key, secret) = SecretKey::decode(&std::fs::read(&key_path)?).context("runner.key")?;
3002×
            let identity = Identity::decode(&std::fs::read(&id_path)?).context("runner.identity")?;
3012×
            if identity.hash() != key {
3020×
                bail!("runner.identity does not belong to runner.key");
3032×
            }
3042×
            return Ok(RunnerKey { identity, secret });
3052×
        }
3062×
        let secret = SecretKey::generate();
3072×
        let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).map(|d| d.as_millis() as u64).unwrap_or(0);
3082×
        let identity = secret.identity(&format!("{name} (OxygenFlow runner)"), "", now);
3092×
        std::fs::write(&id_path, identity.encode())?;
3102×
        let tmp = self.root.join("runner.key.tmp");
311
        {
312
            use std::io::Write;
313
            use std::os::unix::fs::OpenOptionsExt;
3142×
            let mut f = std::fs::OpenOptions::new().write(true).create(true).truncate(true).mode(0o600).open(&tmp)?;
3152×
            f.write_all(&secret.encode(&identity.hash()))?;
3162×
            f.sync_all()?;
317
        }
3182×
        std::fs::rename(&tmp, &key_path)?;
3192×
        Ok(RunnerKey { identity, secret })
3204×
    }
321
}
322
323
#[cfg(test)]
324
mod tests {
325
    use super::*;
326
327
    #[test]
3281×
    fn the_runner_key_is_made_once_and_kept_private() {
329
        use std::os::unix::fs::PermissionsExt;
3301×
        let dir = tempfile::tempdir().unwrap();
3311×
        let home = Home::locate(Some(dir.path())).unwrap();
3321×
        let a = home.runner_key("earth-1").unwrap();
3331×
        assert_eq!(a.identity.name, "earth-1 (OxygenFlow runner)");
3341×
        let mode = std::fs::metadata(dir.path().join("runner.key")).unwrap().permissions().mode();
3351×
        assert_eq!(mode & 0o777, 0o600);
3361×
        let b = home.runner_key("renamed").unwrap();
3371×
        assert_eq!(a.identity, b.identity, "the same key after a restart");
3381×
        let sig = b.secret.sign(oxy_core::identity::Domain::Attest, &b.identity.hash(), &oxy_core::hash::ObjectRef([1; 32]));
3391×
        oxy_core::identity::verify(&a.identity, &sig, oxy_core::identity::Domain::Attest, &oxy_core::hash::ObjectRef([1; 32])).unwrap();
3401×
    }
341
}