@push.rocks/smartnftables 1.6.0 → 1.8.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -0,0 +1,73 @@
1
+ use super::*;
2
+
3
+ impl Owner {
4
+ /// Terminal retention, never adoption/deletion. Exact retry may observe an
5
+ /// orphan after an ACK was lost, including through a fresh native process.
6
+ pub fn detach(&mut self, expected: Applied) -> Result<()> {
7
+ self.validate_receipt(&expected)?;
8
+ if self.released || self.pending.is_some() {
9
+ return Err(Error::Conflict);
10
+ }
11
+ if let Some(retention) = &self.retention {
12
+ if retention.expected != expected {
13
+ return Err(Error::Conflict);
14
+ }
15
+ } else {
16
+ self.retention = Some(Retention {
17
+ expected: expected.clone(),
18
+ confirmed: false,
19
+ });
20
+ }
21
+ let result = (|| {
22
+ self.verify_retention(&expected, false)?;
23
+ // OWNER disappears with this socket; PERSIST must retain exactly
24
+ // the same table. The fresh verification socket never adopts it.
25
+ self.socket = None;
26
+ self.verify_retention(&expected, true)?;
27
+ self.socket = None;
28
+ Ok(())
29
+ })();
30
+ self.retention.as_mut().unwrap().confirmed = result.is_ok();
31
+ match result {
32
+ Ok(()) => {
33
+ self.applied = Some(expected);
34
+ self.error = None;
35
+ Ok(())
36
+ }
37
+ Err(error) => {
38
+ self.error = Some(error);
39
+ Err(error)
40
+ }
41
+ }
42
+ }
43
+
44
+ fn verify_retention(&mut self, expected: &Applied, orphan_only: bool) -> Result<()> {
45
+ let graph = self.graph()?.ok_or(Error::Conflict)?;
46
+ self.binding(&graph, Some(expected))?;
47
+ let flags = wire::number(&graph.table, 2)?;
48
+ if (orphan_only && flags != 4)
49
+ || (flags == 6 && wire::number(&graph.table, 7)? != self.socket()?.port)
50
+ || !self.matches(&graph, &expected.prepared)?
51
+ {
52
+ return Err(Error::Conflict);
53
+ }
54
+ expected.prepared.validate_interfaces()
55
+ }
56
+
57
+ pub(super) fn inspect_retained(&mut self) -> Status {
58
+ let expected = self.retention.as_ref().unwrap().expected.clone();
59
+ if let Err(error) = self.verify_retention(&expected, true) {
60
+ self.retention.as_mut().unwrap().confirmed = false;
61
+ self.error = Some(error);
62
+ }
63
+ let enforced = self.retention.as_ref().unwrap().confirmed && self.error.is_none();
64
+ Status {
65
+ identity: self.identity.clone(),
66
+ state: if enforced { "retained" } else { "failed-owned" }.into(),
67
+ enforced,
68
+ applied: self.applied.clone(),
69
+ pending: self.pending.clone(),
70
+ error: self.error.map(|error| error.code().into()),
71
+ }
72
+ }
73
+ }
package/rust/src/owner.rs CHANGED
@@ -68,8 +68,17 @@ pub struct Owner {
68
68
  pending: Option<Transition>,
69
69
  error: Option<Error>,
70
70
  released: bool,
71
+ retention: Option<Retention>,
71
72
  }
72
73
 
74
+ struct Retention {
75
+ expected: Applied,
76
+ confirmed: bool,
77
+ }
78
+
79
+ #[path = "owner.detach.rs"]
80
+ mod detach;
81
+
73
82
  pub(crate) fn context() -> Result<(String, String, String)> {
74
83
  let boot = std::fs::read_to_string("/proc/sys/kernel/random/boot_id")
75
84
  .map_err(|_| Error::Unavailable)?
@@ -154,6 +163,7 @@ impl Owner {
154
163
  pending: None,
155
164
  error: None,
156
165
  released: false,
166
+ retention: None,
157
167
  })
158
168
  }
159
169
  fn socket(&mut self) -> Result<&mut Socket> {
@@ -350,7 +360,7 @@ impl Owner {
350
360
  })
351
361
  }
352
362
  pub fn reconcile(&mut self, transition: Transition) -> Result<Applied> {
353
- if self.released {
363
+ if self.released || self.retention.is_some() {
354
364
  return Err(Error::Conflict);
355
365
  }
356
366
  transition.target.validate()?;
@@ -453,6 +463,9 @@ impl Owner {
453
463
  self.applied(&graph, transition.target.clone())
454
464
  }
455
465
  pub fn inspect(&mut self) -> Status {
466
+ if self.retention.is_some() {
467
+ return self.inspect_retained();
468
+ }
456
469
  let mut enforced = false;
457
470
  if let Some(applied) = self.applied.clone().filter(|_| !self.released) {
458
471
  match self.graph().and_then(|graph| {
@@ -490,6 +503,9 @@ impl Owner {
490
503
  }
491
504
  }
492
505
  pub fn release(&mut self, expected: Applied) -> Result<()> {
506
+ if self.retention.is_some() {
507
+ return Err(Error::Conflict);
508
+ }
493
509
  self.validate_receipt(&expected)?;
494
510
  if self.pending.is_some() {
495
511
  return Err(Error::Conflict);
@@ -545,6 +561,9 @@ impl Owner {
545
561
  }
546
562
  }
547
563
  pub fn close(&mut self) -> Result<()> {
564
+ if self.retention.is_some() {
565
+ return Err(Error::Conflict);
566
+ }
548
567
  if self.released {
549
568
  return Ok(());
550
569
  }
@@ -0,0 +1,252 @@
1
+ use super::namespace_tests::NativeOwner;
2
+ use super::*;
3
+ use crate::egress;
4
+ use serde_json::json;
5
+ use std::io::Write;
6
+ use std::time::{Duration, Instant};
7
+
8
+ fn pool_target(revision: u64) -> Prepared {
9
+ let mut policy = egress::tests::pool_guard();
10
+ policy["revision"] = json!(revision);
11
+ serde_json::from_value::<egress::Policy>(policy)
12
+ .unwrap()
13
+ .prepare()
14
+ .unwrap()
15
+ .into()
16
+ }
17
+ fn apply(name: &str) -> (Owner, Applied) {
18
+ let mut owner = Owner::new(options(name)).unwrap();
19
+ let applied = owner
20
+ .reconcile(Transition {
21
+ previous: None,
22
+ target: pool_target(1),
23
+ })
24
+ .unwrap();
25
+ (owner, applied)
26
+ }
27
+ fn retained(owner: &mut Owner, applied: &Applied) {
28
+ let status = owner.inspect();
29
+ assert_eq!(status.state, "retained");
30
+ assert!(status.enforced);
31
+ assert_eq!(status.applied, Some(applied.clone()));
32
+ let graph = owner.graph().unwrap().unwrap();
33
+ assert_eq!(wire::number(&graph.table, 2).unwrap(), 4);
34
+ assert_eq!(
35
+ wire::handle(&graph.table, 4).unwrap().to_string(),
36
+ applied.receipt.table_handle
37
+ );
38
+ assert!(owner.matches(&graph, &applied.prepared).unwrap());
39
+ }
40
+
41
+ #[test]
42
+ #[ignore = "requires disposable isolated native qualification guest"]
43
+ fn detach_retains_exact_graph_is_terminal_and_replays_without_adoption() {
44
+ isolated();
45
+ let (mut owner, applied) = apply("detach_replay");
46
+ owner.detach(applied.clone()).unwrap();
47
+ retained(&mut owner, &applied);
48
+ owner.detach(applied.clone()).unwrap();
49
+ assert_eq!(owner.close(), Err(Error::Conflict));
50
+ assert_eq!(owner.release(applied.clone()), Err(Error::Conflict));
51
+ assert_eq!(
52
+ owner.reconcile(Transition {
53
+ previous: Some(applied.clone()),
54
+ target: pool_target(2)
55
+ }),
56
+ Err(Error::Conflict)
57
+ );
58
+ let mut different = applied.clone();
59
+ different.prepared = pool_target(2);
60
+ different.receipt.revision = 2;
61
+ different.receipt.digest = different.prepared.digest.clone();
62
+ assert_eq!(owner.detach(different), Err(Error::Conflict));
63
+ retained(&mut owner, &applied);
64
+ drop(owner);
65
+ let mut fresh = Owner::new(options("detach_replay")).unwrap();
66
+ fresh.detach(applied.clone()).unwrap();
67
+ retained(&mut fresh, &applied);
68
+ // A separate active owner invalidates retention inspection. It is never
69
+ // detached by this reader, even with the same full logical identity.
70
+ let mut active = Owner::new(options("detach_replay")).unwrap();
71
+ active
72
+ .reconcile(Transition {
73
+ previous: None,
74
+ target: applied.prepared.clone(),
75
+ })
76
+ .unwrap();
77
+ assert_eq!(fresh.detach(applied.clone()), Err(Error::Conflict));
78
+ assert!(!fresh.inspect().enforced);
79
+ assert!(active.inspect().enforced);
80
+ assert_eq!(fresh.close(), Err(Error::Conflict));
81
+ drop(active);
82
+ fresh.detach(applied.clone()).unwrap();
83
+ retained(&mut fresh, &applied);
84
+ drop(fresh);
85
+ Owner::new(options("detach_replay"))
86
+ .unwrap()
87
+ .release(applied)
88
+ .unwrap();
89
+ }
90
+
91
+ #[test]
92
+ #[ignore = "requires disposable isolated native qualification guest"]
93
+ fn detach_rejects_pending_wrong_receipts_foreign_mutation_and_absence() {
94
+ isolated();
95
+ let (mut owner, applied) = apply("detach_faults");
96
+ let pending = Transition {
97
+ previous: Some(applied.clone()),
98
+ target: pool_target(2),
99
+ };
100
+ owner.pending = Some(pending.clone());
101
+ assert_eq!(owner.detach(applied.clone()), Err(Error::Conflict));
102
+ assert!(owner.retention.is_none());
103
+ let applied = owner.reconcile(pending).unwrap();
104
+ for field in ["boot", "namespace", "handle", "digest"] {
105
+ let mut invalid = applied.clone();
106
+ match field {
107
+ "boot" => {
108
+ invalid.receipt.identity.boot_id = "00000000-0000-0000-0000-000000000000".into()
109
+ }
110
+ "namespace" => invalid.receipt.identity.namespace_inode = "0".into(),
111
+ "handle" => invalid.receipt.table_handle = "18446744073709551615".into(),
112
+ _ => invalid.receipt.digest = format!("sha256:{}", "0".repeat(64)),
113
+ }
114
+ let mut reader = Owner::new(options("detach_faults")).unwrap();
115
+ assert_eq!(reader.detach(invalid), Err(Error::Conflict));
116
+ assert!(owner.inspect().enforced);
117
+ }
118
+ owner.detach(applied.clone()).unwrap();
119
+ // Mutation after detachment is observed on exact retry and permanently
120
+ // retains the failed owner. No failure cleanup adopts or deletes the graph.
121
+ let mut foreign = Socket::open().unwrap();
122
+ let generation = foreign.generation().unwrap();
123
+ foreign
124
+ .batch(
125
+ generation,
126
+ vec![(
127
+ 3,
128
+ 0x600,
129
+ vec![
130
+ Attr::string(1, "snft_detach_faults"),
131
+ Attr::string(3, "foreign"),
132
+ ],
133
+ )],
134
+ )
135
+ .unwrap();
136
+ assert_eq!(owner.detach(applied.clone()), Err(Error::Conflict));
137
+ assert!(!owner.inspect().enforced);
138
+ assert_eq!(owner.close(), Err(Error::Conflict));
139
+ assert_eq!(owner.release(applied.clone()), Err(Error::Conflict));
140
+ assert!(owner
141
+ .graph()
142
+ .unwrap()
143
+ .unwrap()
144
+ .chains
145
+ .iter()
146
+ .any(|chain| wire::text(chain, 3).unwrap() == "foreign"));
147
+ drop(owner);
148
+ // This disposable fixture's foreign owner deletes only its exact orphan.
149
+ let generation = foreign.generation().unwrap();
150
+ foreign
151
+ .batch(
152
+ generation,
153
+ vec![(2, 0, vec![Attr::string(1, "snft_detach_faults")])],
154
+ )
155
+ .unwrap();
156
+ let mut absent = Owner::new(options("detach_faults")).unwrap();
157
+ assert_eq!(absent.detach(applied), Err(Error::Conflict));
158
+ assert!(!absent.inspect().enforced);
159
+ assert_eq!(absent.close(), Err(Error::Conflict));
160
+ assert!(absent.graph().unwrap().is_none());
161
+ }
162
+
163
+ #[test]
164
+ #[ignore = "requires disposable isolated native qualification guest"]
165
+ fn detach_rechecks_interface_bindings_without_releasing_on_drift() {
166
+ isolated();
167
+ let endpoint = endpoint("a", 71);
168
+ let policy = crate::policy::Policy {
169
+ schema_version: 1,
170
+ revision: 1,
171
+ endpoints: vec![endpoint],
172
+ rules: vec![],
173
+ };
174
+ let mut owner = Owner::new(options("detach_links")).unwrap();
175
+ let applied = owner
176
+ .reconcile(Transition {
177
+ previous: None,
178
+ target: policy.prepare().unwrap().into(),
179
+ })
180
+ .unwrap();
181
+ ip(&["link", "set", "worka", "name", "renameda"]);
182
+ assert_eq!(owner.detach(applied.clone()), Err(Error::Conflict));
183
+ assert_eq!(owner.close(), Err(Error::Conflict));
184
+ let graph = owner.graph().unwrap().unwrap();
185
+ assert_eq!(
186
+ wire::handle(&graph.table, 4).unwrap().to_string(),
187
+ applied.receipt.table_handle
188
+ );
189
+ ip(&["link", "set", "renameda", "name", "worka"]);
190
+ owner.detach(applied.clone()).unwrap();
191
+ retained(&mut owner, &applied);
192
+ drop(owner);
193
+ Owner::new(options("detach_links"))
194
+ .unwrap()
195
+ .release(applied)
196
+ .unwrap();
197
+ }
198
+
199
+ #[test]
200
+ #[ignore = "requires disposable isolated native qualification guest"]
201
+ fn detach_recovers_lost_response_and_undelivered_request_across_real_native_processes() {
202
+ isolated();
203
+ let namespace = std::fs::File::open("/proc/thread-self/ns/net").unwrap();
204
+ for delivered in [true, false] {
205
+ let name = if delivered {
206
+ "detach_ack"
207
+ } else {
208
+ "detach_unsent"
209
+ };
210
+ let mut native = NativeOwner::start(&namespace);
211
+ native.call("openOwner", serde_json::to_value(options(name)).unwrap());
212
+ let applied: Applied = serde_json::from_value(native.call(
213
+ "reconcilePolicy",
214
+ json!({"previous":null,"target":pool_target(1)}),
215
+ ))
216
+ .unwrap();
217
+ let mut reader = Owner::new(options(name)).unwrap();
218
+ reader.validate_receipt(&applied).unwrap();
219
+ if delivered {
220
+ writeln!(
221
+ native.child.stdin.as_mut().unwrap(),
222
+ "{}",
223
+ json!({"id":"lost-detach","method":"detachPolicy","params":applied})
224
+ )
225
+ .unwrap();
226
+ let until = Instant::now() + Duration::from_secs(3);
227
+ loop {
228
+ let graph = reader.graph().unwrap().unwrap();
229
+ if wire::number(&graph.table, 2).unwrap() == 4 {
230
+ break;
231
+ }
232
+ assert!(Instant::now() < until);
233
+ std::thread::sleep(Duration::from_millis(5));
234
+ }
235
+ // Do not read the detach response: terminate the actual process
236
+ // after observing its kernel effect but before recording an ACK.
237
+ }
238
+ drop(native);
239
+ let mut fresh = NativeOwner::start(&namespace);
240
+ fresh.call("openOwner", serde_json::to_value(options(name)).unwrap());
241
+ assert_eq!(
242
+ fresh.call("detachPolicy", serde_json::to_value(&applied).unwrap()),
243
+ json!({"detached":true})
244
+ );
245
+ assert_eq!(fresh.call("inspectPolicy", json!({}))["state"], "retained");
246
+ drop(fresh);
247
+ let graph = reader.graph().unwrap().unwrap();
248
+ assert_eq!(wire::number(&graph.table, 2).unwrap(), 4);
249
+ assert!(reader.matches(&graph, &applied.prepared).unwrap());
250
+ reader.release(applied).unwrap();
251
+ }
252
+ }
@@ -62,7 +62,7 @@ fn host() -> egress::Policy {
62
62
  value["scope"]["uplink"] = binding("uplink", "192.0.2.2", "veth");
63
63
  serde_json::from_value(value).unwrap()
64
64
  }
65
- fn recover(name: &str, policy: egress::Policy) {
65
+ pub(super) fn recover(name: &str, policy: egress::Policy) {
66
66
  let prepared: Prepared = policy.clone().prepare().unwrap().into();
67
67
  prepared.validate_interfaces().expect("fixture bindings");
68
68
  let mut owner = Owner::new(options(name)).unwrap();
@@ -205,6 +205,18 @@ fn v2_four_namespace_private_tun_local_dns_egress_and_revocation() {
205
205
  ns_ip(ROUTER, &["address", "add", "10.242.0.1/24", "dev", "vpn"]);
206
206
  ns_ip(ROUTER, &["link", "set", "vpn", "up"]);
207
207
  let mut device = in_namespace(ROUTER, tun);
208
+ let (pool_owner, pool_applied) = in_namespace(HOST, || {
209
+ let mut owner = Owner::new(options("pool_coexistence")).unwrap();
210
+ let policy: egress::Policy = serde_json::from_value(egress::tests::pool_guard()).unwrap();
211
+ let applied = owner
212
+ .reconcile(Transition {
213
+ previous: None,
214
+ target: policy.prepare().unwrap().into(),
215
+ })
216
+ .unwrap();
217
+ assert!(owner.inspect().enforced);
218
+ (owner, applied)
219
+ });
208
220
  let (host_owner, host_applied) = in_namespace(HOST, || {
209
221
  let mut owner = Owner::new(options("host_packets")).unwrap();
210
222
  let applied = owner
@@ -260,11 +272,8 @@ fn v2_four_namespace_private_tun_local_dns_egress_and_revocation() {
260
272
  });
261
273
  let listener = in_namespace(PEER, || TcpListener::bind("192.0.2.3:3000").unwrap());
262
274
  let mut client = in_namespace(WORK, || {
263
- TcpStream::connect_timeout(
264
- &"192.0.2.3:3000".parse().unwrap(),
265
- Duration::from_secs(1),
266
- )
267
- .unwrap()
275
+ TcpStream::connect_timeout(&"192.0.2.3:3000".parse().unwrap(), Duration::from_secs(1))
276
+ .unwrap()
268
277
  });
269
278
  let (mut server, peer) = listener.accept().unwrap();
270
279
  assert_eq!(peer.ip().to_string(), "192.0.2.2");
@@ -290,11 +299,8 @@ fn v2_four_namespace_private_tun_local_dns_egress_and_revocation() {
290
299
  assert_eq!(syn[33] & 0xc0, 0xc0);
291
300
  assert!((10016..=10031).contains(&u16::from_be_bytes(syn[20..22].try_into().unwrap())));
292
301
  let mut local_tcp = in_namespace(ROUTER, || {
293
- TcpStream::connect_timeout(
294
- &"192.0.2.3:3000".parse().unwrap(),
295
- Duration::from_secs(1),
296
- )
297
- .unwrap()
302
+ TcpStream::connect_timeout(&"192.0.2.3:3000".parse().unwrap(), Duration::from_secs(1))
303
+ .unwrap()
298
304
  });
299
305
  let (mut local_peer, local_outer) = listener.accept().unwrap();
300
306
  assert!((10016..=10031).contains(&local_outer.port()));
@@ -466,5 +472,10 @@ fn v2_four_namespace_private_tun_local_dns_egress_and_revocation() {
466
472
  let mut owner = host_owner;
467
473
  owner.release(host_applied).unwrap();
468
474
  });
475
+ in_namespace(HOST, move || {
476
+ let mut owner = pool_owner;
477
+ assert!(owner.inspect().enforced);
478
+ owner.release(pool_applied).unwrap();
479
+ });
469
480
  println!("V2_PACKET_PROOF private_dns=true tun_private=true tun_public_denied=true workload_public=true local_dns_and_tcp=true handoff_udp_tcp_ranges=true ecn_syn=true opening_ack_rst_denied=true fragments_checked=true range_exhaustion_denied=true protected_grant_denied=true ungranted_local_denied=true retired_replies_denied=true revoked_egress_denied=true disjoint_generation_restored=true flows_drained=false reusable=false");
470
481
  }
@@ -7,7 +7,7 @@ const SOURCE: &str = "hs_source";
7
7
  const HOST: &str = "hs_host";
8
8
  const PEER: &str = "hs_peer";
9
9
 
10
- fn payload(offset: u32, bytes: Vec<u8>, transport: bool) -> Vec<Attr> {
10
+ pub(super) fn payload(offset: u32, bytes: Vec<u8>, transport: bool) -> Vec<Attr> {
11
11
  vec![
12
12
  expr(
13
13
  "payload",
@@ -21,7 +21,7 @@ fn payload(offset: u32, bytes: Vec<u8>, transport: bool) -> Vec<Attr> {
21
21
  compare(bytes),
22
22
  ]
23
23
  }
24
- fn immediate(value: Vec<u8>) -> Attr {
24
+ pub(super) fn immediate(value: Vec<u8>) -> Attr {
25
25
  expr(
26
26
  "immediate",
27
27
  vec![
@@ -196,10 +196,7 @@ fn v2_host_rejects_foreign_dnat_zone_local_diversion_and_ipv6() {
196
196
  HOST,
197
197
  &["route", "add", "203.0.113.2/32", "via", "192.0.2.3"],
198
198
  );
199
- ns_ip(
200
- HOST,
201
- &["route", "add", "10.250.0.2/32", "via", "192.0.2.3"],
202
- );
199
+ ns_ip(HOST, &["route", "add", "10.250.0.2/32", "via", "192.0.2.3"]);
203
200
  for address in ["203.0.113.2/32", "10.250.0.2/32"] {
204
201
  ns_ip(PEER, &["address", "add", address, "dev", "lo"]);
205
202
  }
@@ -285,10 +282,7 @@ fn v2_host_rejects_foreign_dnat_zone_local_diversion_and_ipv6() {
285
282
  control(source, destination, target, false);
286
283
  }
287
284
  let allowed = socket(SOURCE, "10.240.0.2:10000");
288
- assert_eq!(
289
- roundtrip(&allowed, &public).ip().to_string(),
290
- "192.0.2.2"
291
- );
285
+ assert_eq!(roundtrip(&allowed, &public).ip().to_string(), "192.0.2.2");
292
286
  let platform = socket(PEER, "192.0.2.3:5300");
293
287
  roundtrip(&allowed, &platform);
294
288
  let unleased = socket(SOURCE, "10.240.0.2:9999");
@@ -30,13 +30,13 @@ fn command(namespace: &File) -> Command {
30
30
  command
31
31
  }
32
32
 
33
- struct NativeOwner {
34
- child: Child,
33
+ pub(super) struct NativeOwner {
34
+ pub(super) child: Child,
35
35
  messages: Receiver<Value>,
36
36
  reader: Option<JoinHandle<()>>,
37
37
  }
38
38
  impl NativeOwner {
39
- fn start(namespace: &File) -> Self {
39
+ pub(super) fn start(namespace: &File) -> Self {
40
40
  let mut child = command(namespace).spawn().unwrap();
41
41
  let output = child.stdout.take().unwrap();
42
42
  let (send, messages) = mpsc::channel();
@@ -76,7 +76,7 @@ impl NativeOwner {
76
76
  assert_eq!(response["id"], method);
77
77
  response
78
78
  }
79
- fn call(&mut self, method: &str, params: Value) -> Value {
79
+ pub(super) fn call(&mut self, method: &str, params: Value) -> Value {
80
80
  let response = self.response(method, params);
81
81
  assert_eq!(response["success"], true, "{response}");
82
82
  response["result"].clone()