1
#![cfg_attr(docsrs, feature(doc_auto_cfg, doc_cfg))]
2
#![doc = include_str!("../README.md")]
3
// @@ begin lint list maintained by maint/add_warning @@
4
#![allow(renamed_and_removed_lints)] // @@REMOVE_WHEN(ci_arti_stable)
5
#![allow(unknown_lints)] // @@REMOVE_WHEN(ci_arti_nightly)
6
#![warn(missing_docs)]
7
#![warn(noop_method_call)]
8
#![warn(unreachable_pub)]
9
#![warn(clippy::all)]
10
#![deny(clippy::await_holding_lock)]
11
#![deny(clippy::cargo_common_metadata)]
12
#![deny(clippy::cast_lossless)]
13
#![deny(clippy::checked_conversions)]
14
#![warn(clippy::cognitive_complexity)]
15
#![deny(clippy::debug_assert_with_mut_call)]
16
#![deny(clippy::exhaustive_enums)]
17
#![deny(clippy::exhaustive_structs)]
18
#![deny(clippy::expl_impl_clone_on_copy)]
19
#![deny(clippy::fallible_impl_from)]
20
#![deny(clippy::implicit_clone)]
21
#![deny(clippy::large_stack_arrays)]
22
#![warn(clippy::manual_ok_or)]
23
#![deny(clippy::missing_docs_in_private_items)]
24
#![warn(clippy::needless_borrow)]
25
#![warn(clippy::needless_pass_by_value)]
26
#![warn(clippy::option_option)]
27
#![deny(clippy::print_stderr)]
28
#![deny(clippy::print_stdout)]
29
#![warn(clippy::rc_buffer)]
30
#![deny(clippy::ref_option_ref)]
31
#![warn(clippy::semicolon_if_nothing_returned)]
32
#![warn(clippy::trait_duplication_in_bounds)]
33
#![deny(clippy::unchecked_duration_subtraction)]
34
#![deny(clippy::unnecessary_wraps)]
35
#![warn(clippy::unseparated_literal_suffix)]
36
#![deny(clippy::unwrap_used)]
37
#![deny(clippy::mod_module_files)]
38
#![allow(clippy::let_unit_value)] // This can reasonably be done for explicitness
39
#![allow(clippy::uninlined_format_args)]
40
#![allow(clippy::significant_drop_in_scrutinee)] // arti/-/merge_requests/588/#note_2812945
41
#![allow(clippy::result_large_err)] // temporary workaround for arti#587
42
#![allow(clippy::needless_raw_string_hashes)] // complained-about code is fine, often best
43
#![allow(clippy::needless_lifetimes)] // See arti#1765
44
//! <!-- @@ end lint list maintained by maint/add_warning @@ -->
45

            
46
pub mod details;
47
mod err;
48
#[cfg(feature = "hs-common")]
49
mod hsdir_params;
50
#[cfg(feature = "hs-common")]
51
mod hsdir_ring;
52
pub mod params;
53
mod weight;
54

            
55
#[cfg(any(test, feature = "testing"))]
56
pub mod testnet;
57
#[cfg(feature = "testing")]
58
pub mod testprovider;
59

            
60
use async_trait::async_trait;
61
#[cfg(feature = "hs-service")]
62
use itertools::chain;
63
use static_assertions::const_assert;
64
use tor_error::warn_report;
65
use tor_linkspec::{
66
    ChanTarget, DirectChanMethodsHelper, HasAddrs, HasRelayIds, RelayIdRef, RelayIdType,
67
};
68
use tor_llcrypto as ll;
69
use tor_llcrypto::pk::{ed25519::Ed25519Identity, rsa::RsaIdentity};
70
use tor_netdoc::doc::microdesc::{MdDigest, Microdesc};
71
use tor_netdoc::doc::netstatus::{self, MdConsensus, MdConsensusRouterStatus, RouterStatus};
72
#[cfg(feature = "hs-common")]
73
use {hsdir_ring::HsDirRing, std::iter};
74

            
75
use derive_more::{From, Into};
76
use futures::{stream::BoxStream, StreamExt};
77
use num_enum::{IntoPrimitive, TryFromPrimitive};
78
use rand::seq::{IndexedRandom as _, SliceRandom as _, WeightError};
79
use serde::Deserialize;
80
use std::collections::HashMap;
81
use std::net::IpAddr;
82
use std::ops::Deref;
83
use std::sync::Arc;
84
use std::time::SystemTime;
85
use strum::{EnumCount, EnumIter};
86
use tracing::warn;
87
use typed_index_collections::{TiSlice, TiVec};
88

            
89
#[cfg(feature = "hs-common")]
90
use {
91
    itertools::Itertools,
92
    std::collections::HashSet,
93
    tor_error::{internal, Bug},
94
    tor_hscrypto::{pk::HsBlindId, time::TimePeriod},
95
};
96

            
97
pub use err::Error;
98
pub use weight::WeightRole;
99
/// A Result using the Error type from the tor-netdir crate
100
pub type Result<T> = std::result::Result<T, Error>;
101

            
102
#[cfg(feature = "hs-common")]
103
pub use err::OnionDirLookupError;
104

            
105
use params::NetParameters;
106
#[cfg(feature = "geoip")]
107
use tor_geoip::{CountryCode, GeoipDb, HasCountryCode};
108

            
109
#[cfg(feature = "hs-common")]
110
#[cfg_attr(docsrs, doc(cfg(feature = "hs-common")))]
111
pub use hsdir_params::HsDirParams;
112

            
113
/// Index into the consensus relays
114
///
115
/// This is an index into the list of relays returned by
116
/// [`.c_relays()`](ConsensusRelays::c_relays)
117
/// (on the corresponding consensus or netdir).
118
///
119
/// This is just a `usize` inside, but using a newtype prevents getting a relay index
120
/// confused with other kinds of slice indices or counts.
121
///
122
/// If you are in a part of the code which needs to work with multiple consensuses,
123
/// the typechecking cannot tell if you try to index into the wrong consensus.
124
#[derive(Debug, From, Into, Copy, Clone, Ord, PartialOrd, Eq, PartialEq, Hash)]
125
pub(crate) struct RouterStatusIdx(usize);
126

            
127
/// Extension trait to provide index-type-safe `.c_relays()` method
128
//
129
// TODO: Really it would be better to have MdConsensns::relays() return TiSlice,
130
// but that would be an API break there.
131
pub(crate) trait ConsensusRelays {
132
    /// Obtain the list of relays in the consensus
133
    //
134
    fn c_relays(&self) -> &TiSlice<RouterStatusIdx, MdConsensusRouterStatus>;
135
}
136
impl ConsensusRelays for MdConsensus {
137
35298072
    fn c_relays(&self) -> &TiSlice<RouterStatusIdx, MdConsensusRouterStatus> {
138
35298072
        TiSlice::from_ref(MdConsensus::relays(self))
139
35298072
    }
140
}
141
impl ConsensusRelays for NetDir {
142
35222841
    fn c_relays(&self) -> &TiSlice<RouterStatusIdx, MdConsensusRouterStatus> {
143
35222841
        self.consensus.c_relays()
144
35222841
    }
145
}
146

            
147
/// Configuration for determining when two relays have addresses "too close" in
148
/// the network.
149
///
150
/// Used by [`Relay::low_level_details().in_same_subnet()`].
151
#[derive(Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
152
#[serde(deny_unknown_fields)]
153
pub struct SubnetConfig {
154
    /// Consider IPv4 nodes in the same /x to be the same family.
155
    ///
156
    /// If this value is 0, all nodes with IPv4 addresses will be in the
157
    /// same family.  If this value is above 32, then no nodes will be
158
    /// placed im the same family based on their IPv4 addresses.
159
    subnets_family_v4: u8,
160
    /// Consider IPv6 nodes in the same /x to be the same family.
161
    ///
162
    /// If this value is 0, all nodes with IPv6 addresses will be in the
163
    /// same family.  If this value is above 128, then no nodes will be
164
    /// placed im the same family based on their IPv6 addresses.
165
    subnets_family_v6: u8,
166
}
167

            
168
impl Default for SubnetConfig {
169
364099
    fn default() -> Self {
170
364099
        Self::new(16, 32)
171
364099
    }
172
}
173

            
174
impl SubnetConfig {
175
    /// Construct a new SubnetConfig from a pair of bit prefix lengths.
176
    ///
177
    /// The values are clamped to the appropriate ranges if they are
178
    /// out-of-bounds.
179
1111774
    pub fn new(subnets_family_v4: u8, subnets_family_v6: u8) -> Self {
180
1111774
        Self {
181
1111774
            subnets_family_v4,
182
1111774
            subnets_family_v6,
183
1111774
        }
184
1111774
    }
185

            
186
    /// Construct a new SubnetConfig such that addresses are not in the same
187
    /// family with anything--not even with themselves.
188
2246222
    pub fn no_addresses_match() -> SubnetConfig {
189
2246222
        SubnetConfig {
190
2246222
            subnets_family_v4: 33,
191
2246222
            subnets_family_v6: 129,
192
2246222
        }
193
2246222
    }
194

            
195
    /// Return true if the two addresses in the same subnet, according to this
196
    /// configuration.
197
31275113
    pub fn addrs_in_same_subnet(&self, a: &IpAddr, b: &IpAddr) -> bool {
198
31275113
        match (a, b) {
199
31275081
            (IpAddr::V4(a), IpAddr::V4(b)) => {
200
31275081
                let bits = self.subnets_family_v4;
201
31275081
                if bits > 32 {
202
6668
                    return false;
203
31268413
                }
204
31268413
                let a = u32::from_be_bytes(a.octets());
205
31268413
                let b = u32::from_be_bytes(b.octets());
206
31268413
                (a >> (32 - bits)) == (b >> (32 - bits))
207
            }
208
20
            (IpAddr::V6(a), IpAddr::V6(b)) => {
209
20
                let bits = self.subnets_family_v6;
210
20
                if bits > 128 {
211
4
                    return false;
212
16
                }
213
16
                let a = u128::from_be_bytes(a.octets());
214
16
                let b = u128::from_be_bytes(b.octets());
215
16
                (a >> (128 - bits)) == (b >> (128 - bits))
216
            }
217
12
            _ => false,
218
        }
219
31275113
    }
220

            
221
    /// Return true if any of the addresses in `a` shares a subnet with any of
222
    /// the addresses in `b`, according to this configuration.
223
31275063
    pub fn any_addrs_in_same_subnet<T, U>(&self, a: &T, b: &U) -> bool
224
31275063
    where
225
31275063
        T: tor_linkspec::HasAddrs,
226
31275063
        U: tor_linkspec::HasAddrs,
227
31275063
    {
228
31275069
        a.addrs().iter().any(|aa| {
229
31275069
            b.addrs()
230
31275069
                .iter()
231
31275077
                .any(|bb| self.addrs_in_same_subnet(&aa.ip(), &bb.ip()))
232
31275069
        })
233
31275063
    }
234

            
235
    /// Return a new subnet configuration that is the union of `self` and
236
    /// `other`.
237
    ///
238
    /// That is, return a subnet configuration that puts all addresses in the
239
    /// same subnet if and only if at least one of `self` and `other` would put
240
    /// them in the same subnet.
241
1123296
    pub fn union(&self, other: &Self) -> Self {
242
        use std::cmp::min;
243
1123296
        Self {
244
1123296
            subnets_family_v4: min(self.subnets_family_v4, other.subnets_family_v4),
245
1123296
            subnets_family_v6: min(self.subnets_family_v6, other.subnets_family_v6),
246
1123296
        }
247
1123296
    }
248
}
249

            
250
/// Configuration for which listed family information to use when deciding
251
/// whether relays belong to the same family.
252
///
253
/// Derived from network parameters.
254
#[derive(Clone, Copy, Debug)]
255
pub struct FamilyRules {
256
    /// If true, we use family information from lists of family members.
257
    use_family_lists: bool,
258
    /// If true, we use family information from lists of family IDs and from family certs.
259
    use_family_ids: bool,
260
}
261

            
262
impl<'a> From<&'a NetParameters> for FamilyRules {
263
373770
    fn from(params: &'a NetParameters) -> Self {
264
373770
        FamilyRules {
265
373770
            use_family_lists: bool::from(params.use_family_lists),
266
373770
            use_family_ids: bool::from(params.use_family_ids),
267
373770
        }
268
373770
    }
269
}
270

            
271
impl FamilyRules {
272
    /// Return a `FamilyRules` that will use all recognized kinds of family information.
273
587
    pub fn all_family_info() -> Self {
274
587
        Self {
275
587
            use_family_lists: true,
276
587
            use_family_ids: true,
277
587
        }
278
587
    }
279

            
280
    /// Return a `FamilyRules` that will ignore all family information declared by relays.
281
2246310
    pub fn ignore_declared_families() -> Self {
282
2246310
        Self {
283
2246310
            use_family_lists: false,
284
2246310
            use_family_ids: false,
285
2246310
        }
286
2246310
    }
287

            
288
    /// Configure this `FamilyRules` to use (or not use) family information from
289
    /// lists of family members.
290
45
    pub fn use_family_lists(&mut self, val: bool) -> &mut Self {
291
45
        self.use_family_lists = val;
292
45
        self
293
45
    }
294

            
295
    /// Configure this `FamilyRules` to use (or not use) family information from
296
    /// family IDs and family certs.
297
45
    pub fn use_family_ids(&mut self, val: bool) -> &mut Self {
298
45
        self.use_family_ids = val;
299
45
        self
300
45
    }
301

            
302
    /// Return a `FamilyRules` that will look at every source of information
303
    /// requested by `self` or by `other`.
304
1123290
    pub fn union(&self, other: &Self) -> Self {
305
1123290
        Self {
306
1123290
            use_family_lists: self.use_family_lists || other.use_family_lists,
307
1123290
            use_family_ids: self.use_family_ids || other.use_family_ids,
308
        }
309
1123290
    }
310
}
311

            
312
/// An opaque type representing the weight with which a relay or set of
313
/// relays will be selected for a given role.
314
///
315
/// Most users should ignore this type, and just use pick_relay instead.
316
#[derive(
317
    Copy,
318
    Clone,
319
    Debug,
320
    derive_more::Add,
321
    derive_more::Sum,
322
    derive_more::AddAssign,
323
    Eq,
324
    PartialEq,
325
    Ord,
326
    PartialOrd,
327
)]
328
pub struct RelayWeight(u64);
329

            
330
impl RelayWeight {
331
    /// Try to divide this weight by `rhs`.
332
    ///
333
    /// Return a ratio on success, or None on division-by-zero.
334
2344
    pub fn checked_div(&self, rhs: RelayWeight) -> Option<f64> {
335
2344
        if rhs.0 == 0 {
336
2
            None
337
        } else {
338
2342
            Some((self.0 as f64) / (rhs.0 as f64))
339
        }
340
2344
    }
341

            
342
    /// Compute a ratio `frac` of this weight.
343
    ///
344
    /// Return None if frac is less than zero, since negative weights
345
    /// are impossible.
346
24349
    pub fn ratio(&self, frac: f64) -> Option<RelayWeight> {
347
24349
        let product = (self.0 as f64) * frac;
348
24349
        if product >= 0.0 && product.is_finite() {
349
24347
            Some(RelayWeight(product as u64))
350
        } else {
351
2
            None
352
        }
353
24349
    }
354
}
355

            
356
impl From<u64> for RelayWeight {
357
4680
    fn from(val: u64) -> Self {
358
4680
        RelayWeight(val)
359
4680
    }
360
}
361

            
362
/// An operation for which we might be requesting a hidden service directory.
363
#[derive(Copy, Clone, Debug, PartialEq)]
364
// TODO: make this pub(crate) once NetDir::hs_dirs is removed
365
#[non_exhaustive]
366
pub enum HsDirOp {
367
    /// Uploading an onion service descriptor.
368
    #[cfg(feature = "hs-service")]
369
    Upload,
370
    /// Downloading an onion service descriptor.
371
    Download,
372
}
373

            
374
/// A view of the Tor directory, suitable for use in building circuits.
375
///
376
/// Abstractly, a [`NetDir`] is a set of usable public [`Relay`]s, each of which
377
/// has its own properties, identity, and correct weighted probability for use
378
/// under different circumstances.
379
///
380
/// A [`NetDir`] is constructed by making a [`PartialNetDir`] from a consensus
381
/// document, and then adding enough microdescriptors to that `PartialNetDir` so
382
/// that it can be used to build paths. (Thus, if you have a NetDir, it is
383
/// definitely adequate to build paths.)
384
///
385
/// # "Usable" relays
386
///
387
/// Many methods on NetDir are defined in terms of <a name="usable">"Usable"</a> relays.  Unless
388
/// otherwise stated, a relay is "usable" if it is listed in the consensus,
389
/// if we have full directory information for that relay (including a
390
/// microdescriptor), and if that relay does not have any flags indicating that
391
/// we should never use it. (Currently, `NoEdConsensus` is the only such flag.)
392
///
393
/// # Limitations
394
///
395
/// The current NetDir implementation assumes fairly strongly that every relay
396
/// has an Ed25519 identity and an RSA identity, that the consensus is indexed
397
/// by RSA identities, and that the Ed25519 identities are stored in
398
/// microdescriptors.
399
///
400
/// If these assumptions someday change, then we'll have to revise the
401
/// implementation.
402
#[derive(Debug, Clone)]
403
pub struct NetDir {
404
    /// A microdescriptor consensus that lists the members of the network,
405
    /// and maps each one to a 'microdescriptor' that has more information
406
    /// about it
407
    consensus: Arc<MdConsensus>,
408
    /// A map from keys to integer values, distributed in the consensus,
409
    /// and clamped to certain defaults.
410
    params: NetParameters,
411
    /// Map from routerstatus index, to that routerstatus's microdescriptor (if we have one.)
412
    mds: TiVec<RouterStatusIdx, Option<Arc<Microdesc>>>,
413
    /// Map from SHA256 of _missing_ microdescriptors to the index of their
414
    /// corresponding routerstatus.
415
    rsidx_by_missing: HashMap<MdDigest, RouterStatusIdx>,
416
    /// Map from ed25519 identity to index of the routerstatus.
417
    ///
418
    /// Note that we don't know the ed25519 identity of a relay until
419
    /// we get the microdescriptor for it, so this won't be filled in
420
    /// until we get the microdescriptors.
421
    ///
422
    /// # Implementation note
423
    ///
424
    /// For this field, and for `rsidx_by_rsa`,
425
    /// it might be cool to have references instead.
426
    /// But that would make this into a self-referential structure,
427
    /// which isn't possible in safe rust.
428
    rsidx_by_ed: HashMap<Ed25519Identity, RouterStatusIdx>,
429
    /// Map from RSA identity to index of the routerstatus.
430
    ///
431
    /// This is constructed at the same time as the NetDir object, so it
432
    /// can be immutable.
433
    rsidx_by_rsa: Arc<HashMap<RsaIdentity, RouterStatusIdx>>,
434

            
435
    /// Hash ring(s) describing the onion service directory.
436
    ///
437
    /// This is empty in a PartialNetDir, and is filled in before the NetDir is
438
    /// built.
439
    //
440
    // TODO hs: It is ugly to have this exist in a partially constructed state
441
    // in a PartialNetDir.
442
    // Ideally, a PartialNetDir would contain only an HsDirs<HsDirParams>,
443
    // or perhaps nothing at all, here.
444
    #[cfg(feature = "hs-common")]
445
    hsdir_rings: Arc<HsDirs<HsDirRing>>,
446

            
447
    /// Weight values to apply to a given relay when deciding how frequently
448
    /// to choose it for a given role.
449
    weights: weight::WeightSet,
450

            
451
    #[cfg(feature = "geoip")]
452
    /// Country codes for each router in our consensus.
453
    ///
454
    /// This is indexed by the `RouterStatusIdx` (i.e. a router idx of zero has
455
    /// the country code at position zero in this array).
456
    country_codes: Vec<Option<CountryCode>>,
457
}
458

            
459
/// Collection of hidden service directories (or parameters for them)
460
///
461
/// In [`NetDir`] this is used to store the actual hash rings.
462
/// (But, in a NetDir in a [`PartialNetDir`], it contains [`HsDirRing`]s
463
/// where only the `params` are populated, and the `ring` is empty.)
464
///
465
/// This same generic type is used as the return type from
466
/// [`HsDirParams::compute`](HsDirParams::compute),
467
/// where it contains the *parameters* for the primary and secondary rings.
468
#[derive(Debug, Clone)]
469
#[cfg(feature = "hs-common")]
470
pub(crate) struct HsDirs<D> {
471
    /// The current ring
472
    ///
473
    /// It corresponds to the time period containing the `valid-after` time in
474
    /// the consensus. Its SRV is whatever SRV was most current at the time when
475
    /// that time period began.
476
    ///
477
    /// This is the hash ring that we should use whenever we are fetching an
478
    /// onion service descriptor.
479
    current: D,
480

            
481
    /// Secondary rings (based on the parameters for the previous and next time periods)
482
    ///
483
    /// Onion services upload to positions on these ring as well, based on how
484
    /// far into the current time period this directory is, so that
485
    /// not-synchronized clients can still find their descriptor.
486
    ///
487
    /// Note that with the current (2023) network parameters, with
488
    /// `hsdir_interval = SRV lifetime = 24 hours` at most one of these
489
    /// secondary rings will be active at a time.  We have two here in order
490
    /// to conform with a more flexible regime in proposal 342.
491
    //
492
    // TODO: hs clients never need this; so I've made it not-present for thm.
493
    // But does that risk too much with respect to side channels?
494
    //
495
    // TODO: Perhaps we should refactor this so that it is clear that these
496
    // are immutable?  On the other hand, the documentation for this type
497
    // declares that it is immutable, so we are likely okay.
498
    //
499
    // TODO: this `Vec` is only ever 0,1,2 elements.
500
    // Maybe it should be an ArrayVec or something.
501
    #[cfg(feature = "hs-service")]
502
    secondary: Vec<D>,
503
}
504

            
505
#[cfg(feature = "hs-common")]
506
impl<D> HsDirs<D> {
507
    /// Convert an `HsDirs<D>` to `HsDirs<D2>` by mapping each contained `D`
508
14916
    pub(crate) fn map<D2>(self, mut f: impl FnMut(D) -> D2) -> HsDirs<D2> {
509
14916
        HsDirs {
510
14916
            current: f(self.current),
511
14916
            #[cfg(feature = "hs-service")]
512
14916
            secondary: self.secondary.into_iter().map(f).collect(),
513
14916
        }
514
14916
    }
515

            
516
    /// Iterate over some of the contained hsdirs, according to `secondary`
517
    ///
518
    /// The current ring is always included.
519
    /// Secondary rings are included iff `secondary` and the `hs-service` feature is enabled.
520
724
    fn iter_filter_secondary(&self, secondary: bool) -> impl Iterator<Item = &D> {
521
724
        let i = iter::once(&self.current);
522
724

            
523
724
        // With "hs-service" disabled, there are no secondary rings,
524
724
        // so we don't care.
525
724
        let _ = secondary;
526
724

            
527
724
        #[cfg(feature = "hs-service")]
528
724
        let i = chain!(i, self.secondary.iter().filter(move |_| secondary));
529
724

            
530
724
        i
531
724
    }
532

            
533
    /// Iterate over all the contained hsdirs
534
720
    pub(crate) fn iter(&self) -> impl Iterator<Item = &D> {
535
720
        self.iter_filter_secondary(true)
536
720
    }
537

            
538
    /// Iterate over the hsdirs relevant for `op`
539
4
    pub(crate) fn iter_for_op(&self, op: HsDirOp) -> impl Iterator<Item = &D> {
540
4
        self.iter_filter_secondary(match op {
541
            #[cfg(feature = "hs-service")]
542
2
            HsDirOp::Upload => true,
543
2
            HsDirOp::Download => false,
544
        })
545
4
    }
546
}
547

            
548
/// An event that a [`NetDirProvider`] can broadcast to indicate that a change in
549
/// the status of its directory.
550
#[derive(
551
3238
    Debug, Clone, Copy, PartialEq, Eq, EnumIter, EnumCount, IntoPrimitive, TryFromPrimitive,
552
)]
553
#[non_exhaustive]
554
#[repr(u16)]
555
pub enum DirEvent {
556
    /// A new consensus has been received, and has enough information to be
557
    /// used.
558
    ///
559
    /// This event is also broadcast when a new set of consensus parameters is
560
    /// available, even if that set of parameters comes from a configuration
561
    /// change rather than from the latest consensus.
562
    NewConsensus,
563

            
564
    /// New descriptors have been received for the current consensus.
565
    ///
566
    /// (This event is _not_ broadcast when receiving new descriptors for a
567
    /// consensus which is not yet ready to replace the current consensus.)
568
    NewDescriptors,
569

            
570
    /// We have received updated recommendations and requirements
571
    /// for which subprotocols we should have to use the network.
572
    NewProtocolRecommendation,
573
}
574

            
575
/// The network directory provider is shutting down without giving us the
576
/// netdir we asked for.
577
#[derive(Clone, Copy, Debug, thiserror::Error)]
578
#[error("Network directory provider is shutting down")]
579
#[non_exhaustive]
580
pub struct NetdirProviderShutdown;
581

            
582
impl tor_error::HasKind for NetdirProviderShutdown {
583
    fn kind(&self) -> tor_error::ErrorKind {
584
        tor_error::ErrorKind::ArtiShuttingDown
585
    }
586
}
587

            
588
/// How "timely" must a network directory be?
589
///
590
/// This enum is used as an argument when requesting a [`NetDir`] object from
591
/// [`NetDirProvider`] and other APIs, to specify how recent the information
592
/// must be in order to be useful.
593
#[derive(Copy, Clone, Eq, PartialEq, Debug)]
594
#[allow(clippy::exhaustive_enums)]
595
pub enum Timeliness {
596
    /// The network directory must be strictly timely.
597
    ///
598
    /// That is, it must be based on a consensus that valid right now, with no
599
    /// tolerance for skew or consensus problems.
600
    ///
601
    /// Avoid using this option if you could use [`Timeliness::Timely`] instead.
602
    Strict,
603
    /// The network directory must be roughly timely.
604
    ///
605
    /// This is, it must be be based on a consensus that is not _too_ far in the
606
    /// future, and not _too_ far in the past.
607
    ///
608
    /// (The tolerances for "too far" will depend on configuration.)
609
    ///
610
    /// This is almost always the option that you want to use.
611
    Timely,
612
    /// Any network directory is permissible, regardless of how untimely.
613
    ///
614
    /// Avoid using this option if you could use [`Timeliness::Timely`] instead.
615
    Unchecked,
616
}
617

            
618
/// An object that can provide [`NetDir`]s, as well as inform consumers when
619
/// they might have changed.
620
///
621
/// It is the responsibility of the implementor of `NetDirProvider`
622
/// to try to obtain an up-to-date `NetDir`,
623
/// and continuously to maintain and update it.
624
///
625
/// In usual configurations, Arti uses `tor_dirmgr::DirMgr`
626
/// as its `NetDirProvider`.
627
#[async_trait]
628
pub trait NetDirProvider: UpcastArcNetDirProvider + Send + Sync {
629
    /// Return a network directory that's live according to the provided
630
    /// `timeliness`.
631
    fn netdir(&self, timeliness: Timeliness) -> Result<Arc<NetDir>>;
632

            
633
    /// Return a reasonable netdir for general usage.
634
    ///
635
    /// This is an alias for
636
    /// [`NetDirProvider::netdir`]`(`[`Timeliness::Timely`]`)`.
637
686
    fn timely_netdir(&self) -> Result<Arc<NetDir>> {
638
686
        self.netdir(Timeliness::Timely)
639
686
    }
640

            
641
    /// Return a new asynchronous stream that will receive notification
642
    /// whenever the consensus has changed.
643
    ///
644
    /// Multiple events may be batched up into a single item: each time
645
    /// this stream yields an event, all you can assume is that the event has
646
    /// occurred at least once.
647
    fn events(&self) -> BoxStream<'static, DirEvent>;
648

            
649
    /// Return the latest network parameters.
650
    ///
651
    /// If we have no directory, return a reasonable set of defaults.
652
    fn params(&self) -> Arc<dyn AsRef<NetParameters>>;
653

            
654
    /// Get a NetDir from `provider`, waiting until one exists.
655
    async fn wait_for_netdir(
656
        &self,
657
        timeliness: Timeliness,
658
8
    ) -> std::result::Result<Arc<NetDir>, NetdirProviderShutdown> {
659
8
        if let Ok(nd) = self.netdir(timeliness) {
660
8
            return Ok(nd);
661
        }
662

            
663
        let mut stream = self.events();
664
        loop {
665
            // We need to retry `self.netdir()` before waiting for any stream events, to
666
            // avoid deadlock.
667
            //
668
            // We ignore all errors here: they can all potentially be fixed by
669
            // getting a fresh consensus, and they will all get warned about
670
            // by the NetDirProvider itself.
671
            if let Ok(nd) = self.netdir(timeliness) {
672
                return Ok(nd);
673
            }
674
            match stream.next().await {
675
                Some(_) => {}
676
                None => {
677
                    return Err(NetdirProviderShutdown);
678
                }
679
            }
680
        }
681
16
    }
682

            
683
    /// Wait until `provider` lists `target`.
684
    ///
685
    /// NOTE: This might potentially wait indefinitely, if `target` is never actually
686
    /// becomes listed in the directory.  It will exit if the `NetDirProvider` shuts down.
687
    async fn wait_for_netdir_to_list(
688
        &self,
689
        target: &tor_linkspec::RelayIds,
690
        timeliness: Timeliness,
691
    ) -> std::result::Result<(), NetdirProviderShutdown> {
692
        let mut events = self.events();
693
        loop {
694
            // See if the desired relay is in the netdir.
695
            //
696
            // We do this before waiting for any events, to avoid race conditions.
697
            {
698
                let netdir = self.wait_for_netdir(timeliness).await?;
699
                if netdir.ids_listed(target) == Some(true) {
700
                    return Ok(());
701
                }
702
                // If we reach this point, then ids_listed returned `Some(false)`,
703
                // meaning "This relay is definitely not in the current directory";
704
                // or it returned `None`, meaning "waiting for more information
705
                // about this network directory.
706
                // In both cases, it's reasonable to just wait for another netdir
707
                // event and try again.
708
            }
709
            // We didn't find the relay; wait for the provider to have a new netdir
710
            // or more netdir information.
711
            if events.next().await.is_none() {
712
                // The event stream is closed; the provider has shut down.
713
                return Err(NetdirProviderShutdown);
714
            }
715
        }
716
    }
717

            
718
    /// Return the latest set of recommended and required protocols, if there is one.
719
    ///
720
    /// This may be more recent (or more available) than this provider's associated NetDir.
721
    fn protocol_statuses(&self) -> Option<(SystemTime, Arc<netstatus::ProtoStatuses>)>;
722
}
723

            
724
impl<T> NetDirProvider for Arc<T>
725
where
726
    T: NetDirProvider,
727
{
728
50
    fn netdir(&self, timeliness: Timeliness) -> Result<Arc<NetDir>> {
729
50
        self.deref().netdir(timeliness)
730
50
    }
731

            
732
4
    fn timely_netdir(&self) -> Result<Arc<NetDir>> {
733
4
        self.deref().timely_netdir()
734
4
    }
735

            
736
72
    fn events(&self) -> BoxStream<'static, DirEvent> {
737
72
        self.deref().events()
738
72
    }
739

            
740
12
    fn params(&self) -> Arc<dyn AsRef<NetParameters>> {
741
12
        self.deref().params()
742
12
    }
743

            
744
8
    fn protocol_statuses(&self) -> Option<(SystemTime, Arc<netstatus::ProtoStatuses>)> {
745
8
        self.deref().protocol_statuses()
746
8
    }
747
}
748

            
749
/// Helper trait: allows any `Arc<X>` to be upcast to a `Arc<dyn
750
/// NetDirProvider>` if X is an implementation or supertrait of NetDirProvider.
751
///
752
/// This trait exists to work around a limitation in rust: when trait upcasting
753
/// coercion is stable, this will be unnecessary.
754
///
755
/// The Rust tracking issue is <https://github.com/rust-lang/rust/issues/65991>.
756
pub trait UpcastArcNetDirProvider {
757
    /// Return a view of this object as an `Arc<dyn NetDirProvider>`
758
    fn upcast_arc<'a>(self: Arc<Self>) -> Arc<dyn NetDirProvider + 'a>
759
    where
760
        Self: 'a;
761
}
762

            
763
impl<T> UpcastArcNetDirProvider for T
764
where
765
    T: NetDirProvider + Sized,
766
{
767
48
    fn upcast_arc<'a>(self: Arc<Self>) -> Arc<dyn NetDirProvider + 'a>
768
48
    where
769
48
        Self: 'a,
770
48
    {
771
48
        self
772
48
    }
773
}
774

            
775
impl AsRef<NetParameters> for NetDir {
776
855
    fn as_ref(&self) -> &NetParameters {
777
855
        self.params()
778
855
    }
779
}
780

            
781
/// A partially build NetDir -- it can't be unwrapped until it has
782
/// enough information to build safe paths.
783
#[derive(Debug, Clone)]
784
pub struct PartialNetDir {
785
    /// The netdir that's under construction.
786
    netdir: NetDir,
787

            
788
    /// The previous netdir, if we had one
789
    ///
790
    /// Used as a cache, so we can reuse information
791
    #[cfg(feature = "hs-common")]
792
    prev_netdir: Option<Arc<NetDir>>,
793
}
794

            
795
/// A view of a relay on the Tor network, suitable for building circuits.
796
// TODO: This should probably be a more specific struct, with a trait
797
// that implements it.
798
#[derive(Clone)]
799
pub struct Relay<'a> {
800
    /// A router descriptor for this relay.
801
    rs: &'a netstatus::MdConsensusRouterStatus,
802
    /// A microdescriptor for this relay.
803
    md: &'a Microdesc,
804
    /// The country code this relay is in, if we know one.
805
    #[cfg(feature = "geoip")]
806
    cc: Option<CountryCode>,
807
}
808

            
809
/// A relay that we haven't checked for validity or usability in
810
/// routing.
811
#[derive(Debug)]
812
pub struct UncheckedRelay<'a> {
813
    /// A router descriptor for this relay.
814
    rs: &'a netstatus::MdConsensusRouterStatus,
815
    /// A microdescriptor for this relay, if there is one.
816
    md: Option<&'a Microdesc>,
817
    /// The country code this relay is in, if we know one.
818
    #[cfg(feature = "geoip")]
819
    cc: Option<CountryCode>,
820
}
821

            
822
/// A partial or full network directory that we can download
823
/// microdescriptors for.
824
pub trait MdReceiver {
825
    /// Return an iterator over the digests for all of the microdescriptors
826
    /// that this netdir is missing.
827
    fn missing_microdescs(&self) -> Box<dyn Iterator<Item = &MdDigest> + '_>;
828
    /// Add a microdescriptor to this netdir, if it was wanted.
829
    ///
830
    /// Return true if it was indeed wanted.
831
    fn add_microdesc(&mut self, md: Microdesc) -> bool;
832
    /// Return the number of missing microdescriptors.
833
    fn n_missing(&self) -> usize;
834
}
835

            
836
impl PartialNetDir {
837
    /// Create a new PartialNetDir with a given consensus, and no
838
    /// microdescriptors loaded.
839
    ///
840
    /// If `replacement_params` is provided, override network parameters from
841
    /// the consensus with those from `replacement_params`.
842
7371
    pub fn new(
843
7371
        consensus: MdConsensus,
844
7371
        replacement_params: Option<&netstatus::NetParams<i32>>,
845
7371
    ) -> Self {
846
7371
        Self::new_inner(
847
7371
            consensus,
848
7371
            replacement_params,
849
7371
            #[cfg(feature = "geoip")]
850
7371
            None,
851
7371
        )
852
7371
    }
853

            
854
    /// Create a new PartialNetDir with GeoIP support.
855
    ///
856
    /// This does the same thing as `new()`, except the provided GeoIP database is used to add
857
    /// country codes to relays.
858
    #[cfg(feature = "geoip")]
859
    #[cfg_attr(docsrs, doc(cfg(feature = "geoip")))]
860
137
    pub fn new_with_geoip(
861
137
        consensus: MdConsensus,
862
137
        replacement_params: Option<&netstatus::NetParams<i32>>,
863
137
        geoip_db: &GeoipDb,
864
137
    ) -> Self {
865
137
        Self::new_inner(consensus, replacement_params, Some(geoip_db))
866
137
    }
867

            
868
    /// Implementation of the `new()` functions.
869
7508
    fn new_inner(
870
7508
        consensus: MdConsensus,
871
7508
        replacement_params: Option<&netstatus::NetParams<i32>>,
872
7508
        #[cfg(feature = "geoip")] geoip_db: Option<&GeoipDb>,
873
7508
    ) -> Self {
874
7508
        let mut params = NetParameters::default();
875
7508

            
876
7508
        // (We ignore unrecognized options here, since they come from
877
7508
        // the consensus, and we don't expect to recognize everything
878
7508
        // there.)
879
7508
        let _ = params.saturating_update(consensus.params().iter());
880

            
881
        // Now see if the user has any parameters to override.
882
        // (We have to do this now, or else changes won't be reflected in our
883
        // weights.)
884
7508
        if let Some(replacement) = replacement_params {
885
7496
            for u in params.saturating_update(replacement.iter()) {
886
2
                warn!("Unrecognized option: override_net_params.{}", u);
887
            }
888
12
        }
889

            
890
        // Compute the weights we'll want to use for these relays.
891
7508
        let weights = weight::WeightSet::from_consensus(&consensus, &params);
892
7508

            
893
7508
        let n_relays = consensus.c_relays().len();
894
7508

            
895
7508
        let rsidx_by_missing = consensus
896
7508
            .c_relays()
897
7508
            .iter_enumerated()
898
280030
            .map(|(rsidx, rs)| (*rs.md_digest(), rsidx))
899
7508
            .collect();
900
7508

            
901
7508
        let rsidx_by_rsa = consensus
902
7508
            .c_relays()
903
7508
            .iter_enumerated()
904
280030
            .map(|(rsidx, rs)| (*rs.rsa_identity(), rsidx))
905
7508
            .collect();
906

            
907
        #[cfg(feature = "geoip")]
908
7508
        let country_codes = if let Some(db) = geoip_db {
909
137
            consensus
910
137
                .c_relays()
911
137
                .iter()
912
714
                .map(|rs| {
913
710
                    let ret = db
914
990
                        .lookup_country_code_multi(rs.addrs().iter().map(|x| x.ip()))
915
710
                        .cloned();
916
710
                    ret
917
714
                })
918
137
                .collect()
919
        } else {
920
7371
            Default::default()
921
        };
922

            
923
        #[cfg(feature = "hs-common")]
924
7508
        let hsdir_rings = Arc::new({
925
7508
            let params = HsDirParams::compute(&consensus, &params).expect("Invalid consensus!");
926
7508
            // TODO: It's a bit ugly to use expect above, but this function does
927
7508
            // not return a Result. On the other hand, the error conditions under which
928
7508
            // HsDirParams::compute can return Err are _very_ narrow and hard to
929
7508
            // hit; see documentation in that function.  As such, we probably
930
7508
            // don't need to have this return a Result.
931
7508

            
932
7508
            params.map(HsDirRing::empty_from_params)
933
7508
        });
934
7508

            
935
7508
        let netdir = NetDir {
936
7508
            consensus: Arc::new(consensus),
937
7508
            params,
938
7508
            mds: vec![None; n_relays].into(),
939
7508
            rsidx_by_missing,
940
7508
            rsidx_by_rsa: Arc::new(rsidx_by_rsa),
941
7508
            rsidx_by_ed: HashMap::with_capacity(n_relays),
942
7508
            #[cfg(feature = "hs-common")]
943
7508
            hsdir_rings,
944
7508
            weights,
945
7508
            #[cfg(feature = "geoip")]
946
7508
            country_codes,
947
7508
        };
948
7508

            
949
7508
        PartialNetDir {
950
7508
            netdir,
951
7508
            #[cfg(feature = "hs-common")]
952
7508
            prev_netdir: None,
953
7508
        }
954
7508
    }
955

            
956
    /// Return the declared lifetime of this PartialNetDir.
957
2
    pub fn lifetime(&self) -> &netstatus::Lifetime {
958
2
        self.netdir.lifetime()
959
2
    }
960

            
961
    /// Record a previous netdir, which can be used for reusing cached information
962
    //
963
    // Fills in as many missing microdescriptors as possible in this
964
    // netdir, using the microdescriptors from the previous netdir.
965
    //
966
    // With HS enabled, stores the netdir for reuse of relay hash ring index values.
967
    #[allow(clippy::needless_pass_by_value)] // prev might, or might not, be stored
968
2
    pub fn fill_from_previous_netdir(&mut self, prev: Arc<NetDir>) {
969
76
        for md in prev.mds.iter().flatten() {
970
76
            self.netdir.add_arc_microdesc(md.clone());
971
76
        }
972

            
973
        #[cfg(feature = "hs-common")]
974
2
        {
975
2
            self.prev_netdir = Some(prev);
976
2
        }
977
2
    }
978

            
979
    /// Compute the hash ring(s) for this NetDir
980
    #[cfg(feature = "hs-common")]
981
7408
    fn compute_rings(&mut self) {
982
7408
        let params = HsDirParams::compute(&self.netdir.consensus, &self.netdir.params)
983
7408
            .expect("Invalid consensus");
984
7408
        // TODO: see TODO by similar expect in new()
985
7408

            
986
7408
        self.netdir.hsdir_rings =
987
7946
            Arc::new(params.map(|params| {
988
7768
                HsDirRing::compute(params, &self.netdir, self.prev_netdir.as_deref())
989
7946
            }));
990
7408
    }
991

            
992
    /// Return true if this are enough information in this directory
993
    /// to build multihop paths.
994
4
    pub fn have_enough_paths(&self) -> bool {
995
4
        self.netdir.have_enough_paths()
996
4
    }
997
    /// If this directory has enough information to build multihop
998
    /// circuits, return it.
999
7592
    pub fn unwrap_if_sufficient(
7592
        #[allow(unused_mut)] mut self,
7592
    ) -> std::result::Result<NetDir, PartialNetDir> {
7592
        if self.netdir.have_enough_paths() {
            #[cfg(feature = "hs-common")]
7408
            self.compute_rings();
7408
            Ok(self.netdir)
        } else {
184
            Err(self)
        }
7592
    }
}
impl MdReceiver for PartialNetDir {
98
    fn missing_microdescs(&self) -> Box<dyn Iterator<Item = &MdDigest> + '_> {
98
        self.netdir.missing_microdescs()
98
    }
294738
    fn add_microdesc(&mut self, md: Microdesc) -> bool {
294738
        self.netdir.add_microdesc(md)
294738
    }
270
    fn n_missing(&self) -> usize {
270
        self.netdir.n_missing()
270
    }
}
impl NetDir {
    /// Return the declared lifetime of this NetDir.
317
    pub fn lifetime(&self) -> &netstatus::Lifetime {
317
        self.consensus.lifetime()
317
    }
    /// Add `md` to this NetDir.
    ///
    /// Return true if we wanted it, and false otherwise.
294814
    fn add_arc_microdesc(&mut self, md: Arc<Microdesc>) -> bool {
294814
        if let Some(rsidx) = self.rsidx_by_missing.remove(md.digest()) {
279109
            assert_eq!(self.c_relays()[rsidx].md_digest(), md.digest());
            // There should never be two approved MDs in the same
            // consensus listing the same ID... but if there is,
            // we'll let the most recent one win.
279109
            self.rsidx_by_ed.insert(*md.ed25519_id(), rsidx);
279109

            
279109
            // Happy path: we did indeed want this one.
279109
            self.mds[rsidx] = Some(md);
279109

            
279109
            // Save some space in the missing-descriptor list.
279109
            if self.rsidx_by_missing.len() < self.rsidx_by_missing.capacity() / 4 {
14059
                self.rsidx_by_missing.shrink_to_fit();
265050
            }
279109
            return true;
15705
        }
15705

            
15705
        // Either we already had it, or we never wanted it at all.
15705
        false
294814
    }
    /// Construct a (possibly invalid) Relay object from a routerstatus and its
    /// index within the consensus.
33439808
    fn relay_from_rs_and_rsidx<'a>(
33439808
        &'a self,
33439808
        rs: &'a netstatus::MdConsensusRouterStatus,
33439808
        rsidx: RouterStatusIdx,
33439808
    ) -> UncheckedRelay<'a> {
33439808
        debug_assert_eq!(self.c_relays()[rsidx].rsa_identity(), rs.rsa_identity());
33439808
        let md = self.mds[rsidx].as_deref();
33439808
        if let Some(md) = md {
33436121
            debug_assert_eq!(rs.md_digest(), md.digest());
3687
        }
33439808
        UncheckedRelay {
33439808
            rs,
33439808
            md,
33439808
            #[cfg(feature = "geoip")]
33439808
            cc: self.country_codes.get(rsidx.0).copied().flatten(),
33439808
        }
33439808
    }
    /// Return the value of the hsdir_n_replicas param.
    #[cfg(feature = "hs-common")]
405
    fn n_replicas(&self) -> u8 {
405
        self.params
405
            .hsdir_n_replicas
405
            .get()
405
            .try_into()
405
            .expect("BoundedInt did not enforce bounds")
405
    }
    /// Return the spread parameter for the specified `op`.
    #[cfg(feature = "hs-common")]
405
    fn spread(&self, op: HsDirOp) -> usize {
405
        let spread = match op {
45
            HsDirOp::Download => self.params.hsdir_spread_fetch,
            #[cfg(feature = "hs-service")]
360
            HsDirOp::Upload => self.params.hsdir_spread_store,
        };
405
        spread
405
            .get()
405
            .try_into()
405
            .expect("BoundedInt did not enforce bounds!")
405
    }
    /// Select `spread` hsdir relays for the specified `hsid` from a given `ring`.
    ///
    /// Algorithm:
    ///
    /// for idx in 1..=n_replicas:
    ///       - let H = hsdir_ring::onion_service_index(id, replica, rand,
    ///         period).
    ///       - Find the position of H within hsdir_ring.
    ///       - Take elements from hsdir_ring starting at that position,
    ///         adding them to Dirs until we have added `spread` new elements
    ///         that were not there before.
    #[cfg(feature = "hs-common")]
405
    fn select_hsdirs<'h, 'r: 'h>(
405
        &'r self,
405
        hsid: HsBlindId,
405
        ring: &'h HsDirRing,
405
        spread: usize,
405
    ) -> impl Iterator<Item = Relay<'r>> + 'h {
405
        let n_replicas = self.n_replicas();
405

            
405
        (1..=n_replicas) // 1-indexed !
405
            .flat_map({
405
                let mut selected_nodes = HashSet::new();
405

            
430
                move |replica: u8| {
36
                    let hsdir_idx = hsdir_ring::service_hsdir_index(&hsid, replica, ring.params());
36

            
36
                    let items = ring
170
                        .ring_items_at(hsdir_idx, spread, |(hsdir_idx, _)| {
170
                            // According to rend-spec 2.2.3:
170
                            //                                                  ... If any of those
170
                            // nodes have already been selected for a lower-numbered replica of the
170
                            // service, any nodes already chosen are disregarded (i.e. skipped over)
170
                            // when choosing a replica's hsdir_spread_store nodes.
170
                            selected_nodes.insert(*hsdir_idx)
170
                        })
36
                        .collect::<Vec<_>>();
36

            
36
                    items
430
                }
405
            })
530
            .filter_map(move |(_hsdir_idx, rs_idx)| {
140
                // This ought not to be None but let's not panic or bail if it is
140
                self.relay_by_rs_idx(*rs_idx)
530
            })
405
    }
    /// Replace the overridden parameters in this netdir with `new_replacement`.
    ///
    /// After this function is done, the netdir's parameters will be those in
    /// the consensus, overridden by settings from `new_replacement`.  Any
    /// settings in the old replacement parameters will be discarded.
    pub fn replace_overridden_parameters(&mut self, new_replacement: &netstatus::NetParams<i32>) {
        // TODO(nickm): This is largely duplicate code from PartialNetDir::new().
        let mut new_params = NetParameters::default();
        let _ = new_params.saturating_update(self.consensus.params().iter());
        for u in new_params.saturating_update(new_replacement.iter()) {
            warn!("Unrecognized option: override_net_params.{}", u);
        }
        self.params = new_params;
    }
    /// Return an iterator over all Relay objects, including invalid ones
    /// that we can't use.
852816
    pub fn all_relays(&self) -> impl Iterator<Item = UncheckedRelay<'_>> {
852816
        // TODO: I'd like if we could memoize this so we don't have to
852816
        // do so many hashtable lookups.
852816
        self.c_relays()
852816
            .iter_enumerated()
32531256
            .map(move |(rsidx, rs)| self.relay_from_rs_and_rsidx(rs, rsidx))
852816
    }
    /// Return an iterator over all [usable](NetDir#usable) Relays.
798067
    pub fn relays(&self) -> impl Iterator<Item = Relay<'_>> {
798067
        self.all_relays().filter_map(UncheckedRelay::into_relay)
798067
    }
    /// Look up a relay's [`Microdesc`] by its [`RouterStatusIdx`]
    #[cfg_attr(not(feature = "hs-common"), allow(dead_code))]
    pub(crate) fn md_by_rsidx(&self, rsidx: RouterStatusIdx) -> Option<&Microdesc> {
        self.mds.get(rsidx)?.as_deref()
    }
    /// Return a relay matching a given identity, if we have a
    /// _usable_ relay with that key.
    ///
    /// (Does not return [unusable](NetDir#usable) relays.)
    ///
    ///
    /// Note that a `None` answer is not always permanent: if a microdescriptor
    /// is subsequently added for a relay with this ID, the ID may become usable
    /// even if it was not usable before.
356693
    pub fn by_id<'a, T>(&self, id: T) -> Option<Relay<'_>>
356693
    where
356693
        T: Into<RelayIdRef<'a>>,
356693
    {
356693
        let id = id.into();
356693
        let answer = match id {
356550
            RelayIdRef::Ed25519(ed25519) => {
356550
                let rsidx = *self.rsidx_by_ed.get(ed25519)?;
352268
                let rs = self.c_relays().get(rsidx).expect("Corrupt index");
352268

            
352268
                self.relay_from_rs_and_rsidx(rs, rsidx).into_relay()?
            }
143
            RelayIdRef::Rsa(rsa) => self
143
                .by_rsa_id_unchecked(rsa)
143
                .and_then(UncheckedRelay::into_relay)?,
            other_type => self.relays().find(|r| r.has_identity(other_type))?,
        };
352282
        assert!(answer.has_identity(id));
352282
        Some(answer)
356693
    }
    /// Obtain a `Relay` given a `RouterStatusIdx`
    ///
    /// Differs from `relay_from_rs_and_rsi` as follows:
    ///  * That function expects the caller to already have an `MdConsensusRouterStatus`;
    ///    it checks with `debug_assert` that the relay in the netdir matches.
    ///  * That function panics if the `RouterStatusIdx` is invalid; this one returns `None`.
    ///  * That function returns an `UncheckedRelay`; this one a `Relay`.
    ///
    /// `None` could be returned here, even with a valid `rsi`,
    /// if `rsi` refers to an [unusable](NetDir#usable) relay.
    #[cfg_attr(not(feature = "hs-common"), allow(dead_code))]
3178
    pub(crate) fn relay_by_rs_idx(&self, rs_idx: RouterStatusIdx) -> Option<Relay<'_>> {
3178
        let rs = self.c_relays().get(rs_idx)?;
3178
        let md = self.mds.get(rs_idx)?.as_deref();
3178
        UncheckedRelay {
3178
            rs,
3178
            md,
3178
            #[cfg(feature = "geoip")]
3178
            cc: self.country_codes.get(rs_idx.0).copied().flatten(),
3178
        }
3178
        .into_relay()
3178
    }
    /// Return a relay with the same identities as those in `target`, if one
    /// exists.
    ///
    /// Does not return [unusable](NetDir#usable) relays.
    ///
    /// Note that a negative result from this method is not necessarily permanent:
    /// it may be the case that a relay exists,
    /// but we don't yet have enough information about it to know all of its IDs.
    /// To test whether a relay is *definitely* absent,
    /// use [`by_ids_detailed`](Self::by_ids_detailed)
    /// or [`ids_listed`](Self::ids_listed).
    ///
    /// # Limitations
    ///
    /// This will be very slow if `target` does not have an Ed25519 or RSA
    /// identity.
342586
    pub fn by_ids<T>(&self, target: &T) -> Option<Relay<'_>>
342586
    where
342586
        T: HasRelayIds + ?Sized,
342586
    {
342586
        let mut identities = target.identities();
        // Don't try if there are no identities.
342586
        let first_id = identities.next()?;
        // Since there is at most one relay with each given ID type,
        // we only need to check the first relay we find.
342586
        let candidate = self.by_id(first_id)?;
342374
        if identities.all(|wanted_id| candidate.has_identity(wanted_id)) {
342372
            Some(candidate)
        } else {
2
            None
        }
342586
    }
    /// Check whether there is a relay that has at least one identity from
    /// `target`, and which _could_ have every identity from `target`.
    /// If so, return such a relay.
    ///
    /// Return `Ok(None)` if we did not find a relay with any identity from `target`.
    ///
    /// Return `RelayLookupError::Impossible` if we found a relay with at least
    /// one identity from `target`, but that relay's other identities contradict
    /// what we learned from `target`.
    ///
    /// Does not return [unusable](NetDir#usable) relays.
    ///
    /// (This function is only useful if you need to distinguish the
    /// "impossible" case from the "no such relay known" case.)
    ///
    /// # Limitations
    ///
    /// This will be very slow if `target` does not have an Ed25519 or RSA
    /// identity.
    //
    // TODO HS: This function could use a better name.
    //
    // TODO: We could remove the feature restriction here once we think this API is
    // stable.
    #[cfg(feature = "hs-common")]
20
    pub fn by_ids_detailed<T>(
20
        &self,
20
        target: &T,
20
    ) -> std::result::Result<Option<Relay<'_>>, RelayLookupError>
20
    where
20
        T: HasRelayIds + ?Sized,
20
    {
20
        let candidate = target
20
            .identities()
20
            // Find all the relays that share any identity with this set of identities.
36
            .filter_map(|id| self.by_id(id))
20
            // We might find the same relay more than once under a different
20
            // identity, so we remove the duplicates.
20
            //
20
            // Since there is at most one relay per rsa identity per consensus,
20
            // this is a true uniqueness check under current construction rules.
20
            .unique_by(|r| r.rs.rsa_identity())
20
            // If we find two or more distinct relays, then have a contradiction.
20
            .at_most_one()
20
            .map_err(|_| RelayLookupError::Impossible)?;
        // If we have no candidate, return None early.
20
        let candidate = match candidate {
10
            Some(relay) => relay,
10
            None => return Ok(None),
        };
        // Now we know we have a single candidate.  Make sure that it does not have any
        // identity that does not match the target.
10
        if target
10
            .identities()
14
            .all(|wanted_id| match candidate.identity(wanted_id.id_type()) {
                None => true,
14
                Some(id) => id == wanted_id,
14
            })
        {
6
            Ok(Some(candidate))
        } else {
4
            Err(RelayLookupError::Impossible)
        }
20
    }
    /// Return a boolean if this consensus definitely has (or does not have) a
    /// relay matching the listed identities.
    ///
    /// `Some(true)` indicates that the relay exists.
    /// `Some(false)` indicates that the relay definitely does not exist.
    /// `None` indicates that we can't yet tell whether such a relay exists,
    ///  due to missing information.
23676
    fn id_pair_listed(&self, ed_id: &Ed25519Identity, rsa_id: &RsaIdentity) -> Option<bool> {
23676
        let r = self.by_rsa_id_unchecked(rsa_id);
23676
        match r {
23404
            Some(unchecked) => {
23404
                if !unchecked.rs.ed25519_id_is_usable() {
                    return Some(false);
23404
                }
23404
                // If md is present, then it's listed iff we have the right
23404
                // ed id.  Otherwise we don't know if it's listed.
23922
                unchecked.md.map(|md| md.ed25519_id() == ed_id)
            }
            None => {
                // Definitely not listed.
272
                Some(false)
            }
        }
23676
    }
    /// Check whether a relay exists (or may exist)
    /// with the same identities as those in `target`.
    ///
    /// `Some(true)` indicates that the relay exists.
    /// `Some(false)` indicates that the relay definitely does not exist.
    /// `None` indicates that we can't yet tell whether such a relay exists,
    ///  due to missing information.
7916
    pub fn ids_listed<T>(&self, target: &T) -> Option<bool>
7916
    where
7916
        T: HasRelayIds + ?Sized,
7916
    {
7916
        let rsa_id = target.rsa_identity();
7916
        let ed25519_id = target.ed_identity();
7916

            
7916
        // TODO: If we later support more identity key types, this will
7916
        // become incorrect.  This assertion might help us recognize that case.
7916
        const_assert!(RelayIdType::COUNT == 2);
7916

            
7916
        match (rsa_id, ed25519_id) {
7916
            (Some(r), Some(e)) => self.id_pair_listed(e, r),
            (Some(r), None) => Some(self.rsa_id_is_listed(r)),
            (None, Some(e)) => {
                if self.rsidx_by_ed.contains_key(e) {
                    Some(true)
                } else {
                    None
                }
            }
            (None, None) => None,
        }
7916
    }
    /// Return a (possibly [unusable](NetDir#usable)) relay with a given RSA identity.
    ///
    /// This API can be used to find information about a relay that is listed in
    /// the current consensus, even if we don't yet have enough information
    /// (like a microdescriptor) about the relay to use it.
249794
    #[cfg_attr(feature = "experimental-api", visibility::make(pub))]
249794
    #[cfg_attr(docsrs, doc(cfg(feature = "experimental-api")))]
249794
    fn by_rsa_id_unchecked(&self, rsa_id: &RsaIdentity) -> Option<UncheckedRelay<'_>> {
249794
        let rsidx = *self.rsidx_by_rsa.get(rsa_id)?;
249377
        let rs = self.c_relays().get(rsidx).expect("Corrupt index");
249377
        assert_eq!(rs.rsa_identity(), rsa_id);
249377
        Some(self.relay_from_rs_and_rsidx(rs, rsidx))
249794
    }
    /// Return the relay with a given RSA identity, if we have one
    /// and it is [usable](NetDir#usable).
12
    fn by_rsa_id(&self, rsa_id: &RsaIdentity) -> Option<Relay<'_>> {
12
        self.by_rsa_id_unchecked(rsa_id)?.into_relay()
12
    }
    /// Return true if `rsa_id` is listed in this directory, even if it isn't
    /// currently usable.
    ///
    /// (An "[unusable](NetDir#usable)" relay in this context is one for which we don't have full
    /// directory information.)
6
    #[cfg_attr(feature = "experimental-api", visibility::make(pub))]
6
    #[cfg_attr(docsrs, doc(cfg(feature = "experimental-api")))]
6
    fn rsa_id_is_listed(&self, rsa_id: &RsaIdentity) -> bool {
6
        self.by_rsa_id_unchecked(rsa_id).is_some()
6
    }
    /// List the hsdirs in this NetDir, that should be in the HSDir rings
    ///
    /// The results are not returned in any particular order.
    #[cfg(feature = "hs-common")]
7768
    fn all_hsdirs(&self) -> impl Iterator<Item = (RouterStatusIdx, Relay<'_>)> {
293581
        self.c_relays().iter_enumerated().filter_map(|(rsidx, rs)| {
293395
            let relay = self.relay_from_rs_and_rsidx(rs, rsidx);
293395
            relay.is_hsdir_for_ring().then_some(())?;
71470
            let relay = relay.into_relay()?;
71462
            Some((rsidx, relay))
293581
        })
7768
    }
    /// Return the parameters from the consensus, clamped to the
    /// correct ranges, with defaults filled in.
    ///
    /// NOTE: that unsupported parameters aren't returned here; only those
    /// values configured in the `params` module are available.
392076
    pub fn params(&self) -> &NetParameters {
392076
        &self.params
392076
    }
    /// Return a [`ProtoStatus`](netstatus::ProtoStatus) that lists the
    /// network's current requirements and recommendations for the list of
    /// protocols that every relay must implement.
    //
    // TODO HS: I am not sure this is the right API; other alternatives would be:
    //    * To expose the _required_ relay protocol list instead (since that's all that
    //      onion service implementations need).
    //    * To expose the client protocol list as well (for symmetry).
    //    * To expose the MdConsensus instead (since that's more general, although
    //      it restricts the future evolution of this API).
    //
    // I think that this is a reasonably good compromise for now, but I'm going
    // to put it behind the `hs-common` feature to give us time to consider more.
    #[cfg(feature = "hs-common")]
135
    pub fn relay_protocol_status(&self) -> &netstatus::ProtoStatus {
135
        self.consensus.relay_protocol_status()
135
    }
    /// Return weighted the fraction of relays we can use.  We only
    /// consider relays that match the predicate `usable`.  We weight
    /// this bandwidth according to the provided `role`.
    ///
    /// If _no_ matching relays in the consensus have a nonzero
    /// weighted bandwidth value, we fall back to looking at the
    /// unweighted fraction of matching relays.
    ///
    /// If there are no matching relays in the consensus, we return 0.0.
22215
    fn frac_for_role<'a, F>(&'a self, role: WeightRole, usable: F) -> f64
22215
    where
22215
        F: Fn(&UncheckedRelay<'a>) -> bool,
22215
    {
22215
        let mut total_weight = 0_u64;
22215
        let mut have_weight = 0_u64;
22215
        let mut have_count = 0_usize;
22215
        let mut total_count = 0_usize;
832890
        for r in self.all_relays() {
832890
            if !usable(&r) {
275560
                continue;
557330
            }
557330
            let w = self.weights.weight_rs_for_role(r.rs, role);
557330
            total_weight += w;
557330
            total_count += 1;
557330
            if r.is_usable() {
554562
                have_weight += w;
554562
                have_count += 1;
554562
            }
        }
22215
        if total_weight > 0 {
            // The consensus lists some weighted bandwidth so return the
            // fraction of the weighted bandwidth for which we have
            // descriptors.
22215
            (have_weight as f64) / (total_weight as f64)
        } else if total_count > 0 {
            // The consensus lists no weighted bandwidth for these relays,
            // but at least it does list relays. Return the fraction of
            // relays for which it we have descriptors.
            (have_count as f64) / (total_count as f64)
        } else {
            // There are no relays of this kind in the consensus.  Return
            // 0.0, to avoid dividing by zero and giving NaN.
            0.0
        }
22215
    }
    /// Return the estimated fraction of possible paths that we have
    /// enough microdescriptors to build.
7598
    fn frac_usable_paths(&self) -> f64 {
7598
        // TODO #504, TODO SPEC: We may want to add a set of is_flagged_fast() and/or
7598
        // is_flagged_stable() checks here.  This will require spec clarification.
280392
        let f_g = self.frac_for_role(WeightRole::Guard, |u| {
280205
            u.low_level_details().is_suitable_as_guard()
280392
        });
280392
        let f_m = self.frac_for_role(WeightRole::Middle, |_| true);
82955
        let f_e = if self.all_relays().any(|u| u.rs.is_flagged_exit()) {
272414
            self.frac_for_role(WeightRole::Exit, |u| u.rs.is_flagged_exit())
        } else {
            // If there are no exits at all, we use f_m here.
585
            f_m
        };
7598
        f_g * f_m * f_e
7598
    }
    /// Return true if there is enough information in this NetDir to build
    /// multihop circuits.
7596
    fn have_enough_paths(&self) -> bool {
7596
        // TODO-A001: This should check for our guards as well, and
7596
        // make sure that if they're listed in the consensus, we have
7596
        // the descriptors for them.
7596

            
7596
        // If we can build a randomly chosen path with at least this
7596
        // probability, we know enough information to participate
7596
        // on the network.
7596

            
7596
        let min_frac_paths: f64 = self.params().min_circuit_path_threshold.as_fraction();
7596

            
7596
        // What fraction of paths can we build?
7596
        let available = self.frac_usable_paths();
7596

            
7596
        available >= min_frac_paths
7596
    }
    /// Choose a relay at random.
    ///
    /// Each relay is chosen with probability proportional to its weight
    /// in the role `role`, and is only selected if the predicate `usable`
    /// returns true for it.
    ///
    /// This function returns None if (and only if) there are no relays
    /// with nonzero weight where `usable` returned true.
    //
    // TODO this API, with the `usable` closure, invites mistakes where we fail to
    // check conditions that are implied by the role we have selected for the relay:
    // call sites must include a call to `Relay::is_polarity_inverter()` or whatever.
    // IMO the `WeightRole` ought to imply a condition (and it should therefore probably
    // be renamed.)  -Diziet
43396
    pub fn pick_relay<'a, R, P>(
43396
        &'a self,
43396
        rng: &mut R,
43396
        role: WeightRole,
43396
        usable: P,
43396
    ) -> Option<Relay<'a>>
43396
    where
43396
        R: rand::Rng,
43396
        P: FnMut(&Relay<'a>) -> bool,
43396
    {
43396
        let relays: Vec<_> = self.relays().filter(usable).collect();
43396
        // This algorithm uses rand::distr::WeightedIndex, and uses
43396
        // gives O(n) time and space  to build the index, plus O(log n)
43396
        // sampling time.
43396
        //
43396
        // We might be better off building a WeightedIndex in advance
43396
        // for each `role`, and then sampling it repeatedly until we
43396
        // get a relay that satisfies `usable`.  Or we might not --
43396
        // that depends heavily on the actual particulars of our
43396
        // inputs.  We probably shouldn't make any changes there
43396
        // unless profiling tells us that this function is in a hot
43396
        // path.
43396
        //
43396
        // The C Tor sampling implementation goes through some trouble
43396
        // here to try to make its path selection constant-time.  I
43396
        // believe that there is no actual remotely exploitable
43396
        // side-channel here however.  It could be worth analyzing in
43396
        // the future.
43396
        //
43396
        // This code will give the wrong result if the total of all weights
43396
        // can exceed u64::MAX.  We make sure that can't happen when we
43396
        // set up `self.weights`.
770126
        match relays[..].choose_weighted(rng, |r| self.weights.weight_rs_for_role(r.rs, role)) {
43346
            Ok(relay) => Some(relay.clone()),
            Err(WeightError::InsufficientNonZero) => {
                if relays.is_empty() {
                    None
                } else {
                    warn!(?self.weights, ?role,
                          "After filtering, all {} relays had zero weight. Choosing one at random. See bug #1907.",
                          relays.len());
                    relays.choose(rng).cloned()
                }
            }
50
            Err(e) => {
50
                warn_report!(e, "Unexpected error while sampling a relay");
50
                None
            }
        }
43396
    }
    /// Choose `n` relay at random.
    ///
    /// Each relay is chosen with probability proportional to its weight
    /// in the role `role`, and is only selected if the predicate `usable`
    /// returns true for it.
    ///
    /// Relays are chosen without replacement: no relay will be
    /// returned twice. Therefore, the resulting vector may be smaller
    /// than `n` if we happen to have fewer than `n` appropriate relays.
    ///
    /// This function returns an empty vector if (and only if) there
    /// are no relays with nonzero weight where `usable` returned
    /// true.
    #[allow(clippy::cognitive_complexity)] // all due to tracing crate.
26131
    pub fn pick_n_relays<'a, R, P>(
26131
        &'a self,
26131
        rng: &mut R,
26131
        n: usize,
26131
        role: WeightRole,
26131
        usable: P,
26131
    ) -> Vec<Relay<'a>>
26131
    where
26131
        R: rand::Rng,
26131
        P: FnMut(&Relay<'a>) -> bool,
26131
    {
26131
        let relays: Vec<_> = self.relays().filter(usable).collect();
        // NOTE: See discussion in pick_relay().
190016
        let mut relays = match relays[..].choose_multiple_weighted(rng, n, |r| {
190016
            self.weights.weight_rs_for_role(r.rs, role) as f64
190016
        }) {
            Err(WeightError::InsufficientNonZero) => {
                // Too few relays had nonzero weights: return all of those that are okay.
                // (This is behavior used to come up with rand 0.9; it no longer does.
                // We still detect it.)
                let remaining: Vec<_> = relays
                    .iter()
                    .filter(|r| self.weights.weight_rs_for_role(r.rs, role) > 0)
                    .cloned()
                    .collect();
                if remaining.is_empty() {
                    warn!(?self.weights, ?role,
                          "After filtering, all {} relays had zero weight! Picking some at random. See bug #1907.",
                          relays.len());
                    if relays.len() >= n {
                        relays.choose_multiple(rng, n).cloned().collect()
                    } else {
                        relays
                    }
                } else {
                    warn!(?self.weights, ?role,
                          "After filtering, only had {}/{} relays with nonzero weight. Returning them all. See bug #1907.",
                           remaining.len(), relays.len());
                    remaining
                }
            }
            Err(e) => {
                warn_report!(e, "Unexpected error while sampling a set of relays");
                Vec::new()
            }
26131
            Ok(iter) => {
26131
                let selection: Vec<_> = iter.map(Relay::clone).collect();
26131
                if selection.len() < n {
22078
                    warn!(?self.weights, ?role,
                          "After filtering, choose_multiple_weighted only returned {}/{} relays with nonzero weight. See bug #1907.",
                          selection.len(), relays.len());
4053
                }
26131
                selection
            }
        };
26131
        relays.shuffle(rng);
26131
        relays
26131
    }
    /// Compute the weight with which `relay` will be selected for a given
    /// `role`.
23402
    pub fn relay_weight<'a>(&'a self, relay: &Relay<'a>, role: WeightRole) -> RelayWeight {
23402
        RelayWeight(self.weights.weight_rs_for_role(relay.rs, role))
23402
    }
    /// Compute the total weight with which any relay matching `usable`
    /// will be selected for a given `role`.
    ///
    /// Note: because this function is used to assess the total
    /// properties of the consensus, the `usable` predicate takes a
    /// [`RouterStatus`] rather than a [`Relay`].
20859
    pub fn total_weight<P>(&self, role: WeightRole, usable: P) -> RelayWeight
20859
    where
20859
        P: Fn(&UncheckedRelay<'_>) -> bool,
20859
    {
20859
        self.all_relays()
446582
            .filter_map(|unchecked| {
446582
                if usable(&unchecked) {
199876
                    Some(RelayWeight(
199876
                        self.weights.weight_rs_for_role(unchecked.rs, role),
199876
                    ))
                } else {
246706
                    None
                }
446582
            })
20859
            .sum()
20859
    }
    /// Compute the weight with which a relay with ID `rsa_id` would be
    /// selected for a given `role`.
    ///
    /// Note that weight returned by this function assumes that the
    /// relay with that ID is actually [usable](NetDir#usable); if it isn't usable,
    /// then other weight-related functions will call its weight zero.
225769
    pub fn weight_by_rsa_id(&self, rsa_id: &RsaIdentity, role: WeightRole) -> Option<RelayWeight> {
225769
        self.by_rsa_id_unchecked(rsa_id)
230787
            .map(|unchecked| RelayWeight(self.weights.weight_rs_for_role(unchecked.rs, role)))
225769
    }
    /// Return all relays in this NetDir known to be in the same family as
    /// `relay`.
    ///
    /// This list of members will **not** necessarily include `relay` itself.
    ///
    /// # Limitations
    ///
    /// Two relays only belong to the same family if _each_ relay
    /// claims to share a family with the other.  But if we are
    /// missing a microdescriptor for one of the relays listed by this
    /// relay, we cannot know whether it acknowledges family
    /// membership with this relay or not.  Therefore, this function
    /// can omit family members for which there is not (as yet) any
    /// Relay object.
4
    pub fn known_family_members<'a>(
4
        &'a self,
4
        relay: &'a Relay<'a>,
4
    ) -> impl Iterator<Item = Relay<'a>> {
4
        let relay_rsa_id = relay.rsa_id();
10
        relay.md.family().members().filter_map(move |other_rsa_id| {
8
            self.by_rsa_id(other_rsa_id)
8
                .filter(|other_relay| other_relay.md.family().contains(relay_rsa_id))
10
        })
4
    }
    /// Return the current hidden service directory "time period".
    ///
    /// Specifically, this returns the time period that contains the beginning
    /// of the validity period of this `NetDir`'s consensus.  That time period
    /// is the one we use when acting as an hidden service client.
    #[cfg(feature = "hs-common")]
1350
    pub fn hs_time_period(&self) -> TimePeriod {
1350
        self.hsdir_rings.current.time_period()
1350
    }
    /// Return the [`HsDirParams`] of all the relevant hidden service directory "time periods"
    ///
    /// This includes the current time period (as from
    /// [`.hs_time_period`](NetDir::hs_time_period))
    /// plus additional time periods that we publish descriptors for when we are
    /// acting as a hidden service.
    #[cfg(feature = "hs-service")]
360
    pub fn hs_all_time_periods(&self) -> Vec<HsDirParams> {
360
        self.hsdir_rings
360
            .iter()
728
            .map(|r| r.params().clone())
360
            .collect()
360
    }
    /// Return the relays in this network directory that will be used as hidden service directories
    ///
    /// These are suitable to retrieve a given onion service's descriptor at a given time period.
    #[cfg(feature = "hs-common")]
2
    pub fn hs_dirs_download<'r, R>(
2
        &'r self,
2
        hsid: HsBlindId,
2
        period: TimePeriod,
2
        rng: &mut R,
2
    ) -> std::result::Result<Vec<Relay<'r>>, Bug>
2
    where
2
        R: rand::Rng,
2
    {
2
        // Algorithm:
2
        //
2
        // 1. Determine which HsDirRing to use, based on the time period.
2
        // 2. Find the shared random value that's associated with that HsDirRing.
2
        // 3. Choose spread = the parameter `hsdir_spread_fetch`
2
        // 4. Let n_replicas = the parameter `hsdir_n_replicas`.
2
        // 5. Initialize Dirs = []
2
        // 6. for idx in 1..=n_replicas:
2
        //       - let H = hsdir_ring::onion_service_index(id, replica, rand,
2
        //         period).
2
        //       - Find the position of H within hsdir_ring.
2
        //       - Take elements from hsdir_ring starting at that position,
2
        //         adding them to Dirs until we have added `spread` new elements
2
        //         that were not there before.
2
        // 7. Shuffle Dirs
2
        // 8. return Dirs.
2

            
2
        let spread = self.spread(HsDirOp::Download);
2

            
2
        // When downloading, only look at relays on current ring.
2
        let ring = &self.hsdir_rings.current;
2

            
2
        if ring.params().time_period != period {
            return Err(internal!(
                "our current ring is not associated with the requested time period!"
            ));
2
        }
2

            
2
        let mut hs_dirs = self.select_hsdirs(hsid, ring, spread).collect_vec();
2

            
2
        // When downloading, the order of the returned relays is random.
2
        hs_dirs.shuffle(rng);
2

            
2
        Ok(hs_dirs)
2
    }
    /// Return the relays in this network directory that will be used as hidden service directories
    ///
    /// Returns the relays that are suitable for storing a given onion service's descriptors at the
    /// given time period.
    #[cfg(feature = "hs-service")]
360
    pub fn hs_dirs_upload(
360
        &self,
360
        hsid: HsBlindId,
360
        period: TimePeriod,
360
    ) -> std::result::Result<impl Iterator<Item = Relay<'_>>, Bug> {
360
        // Algorithm:
360
        //
360
        // 1. Choose spread = the parameter `hsdir_spread_store`
360
        // 2. Determine which HsDirRing to use, based on the time period.
360
        // 3. Find the shared random value that's associated with that HsDirRing.
360
        // 4. Let n_replicas = the parameter `hsdir_n_replicas`.
360
        // 5. Initialize Dirs = []
360
        // 6. for idx in 1..=n_replicas:
360
        //       - let H = hsdir_ring::onion_service_index(id, replica, rand,
360
        //         period).
360
        //       - Find the position of H within hsdir_ring.
360
        //       - Take elements from hsdir_ring starting at that position,
360
        //         adding them to Dirs until we have added `spread` new elements
360
        //         that were not there before.
360
        // 3. return Dirs.
360
        let spread = self.spread(HsDirOp::Upload);
360

            
360
        // For each HsBlindId, determine which HsDirRing to use.
360
        let rings = self
360
            .hsdir_rings
360
            .iter()
368
            .filter_map(move |ring| {
360
                // Make sure the ring matches the TP of the hsid it's matched with.
360
                (ring.params().time_period == period).then_some((ring, hsid, period))
368
            })
360
            .collect::<Vec<_>>();
360

            
360
        // The specified period should have an associated ring.
368
        if !rings.iter().any(|(_, _, tp)| *tp == period) {
            return Err(internal!(
                "the specified time period does not have an associated ring"
            ));
360
        };
360

            
360
        // Now that we've matched each `hsid` with the ring associated with its TP, we can start
360
        // selecting replicas from each ring.
368
        Ok(rings.into_iter().flat_map(move |(ring, hsid, period)| {
16
            assert_eq!(period, ring.params().time_period());
16
            self.select_hsdirs(hsid, ring, spread)
368
        }))
360
    }
    /// Return the relays in this network directory that will be used as hidden service directories
    ///
    /// Depending on `op`,
    /// these are suitable to either store, or retrieve, a
    /// given onion service's descriptor at a given time period.
    ///
    /// When `op` is `Download`, the order is random.
    /// When `op` is `Upload`, the order is not specified.
    ///
    /// Return an error if the time period is not one returned by
    /// `onion_service_time_period` or `onion_service_secondary_time_periods`.
    //
    // TODO: make HsDirOp pub(crate) once this is removed
    #[cfg(feature = "hs-common")]
    #[deprecated(note = "Use hs_dirs_upload or hs_dirs_download instead")]
4
    pub fn hs_dirs<'r, R>(&'r self, hsid: &HsBlindId, op: HsDirOp, rng: &mut R) -> Vec<Relay<'r>>
4
    where
4
        R: rand::Rng,
4
    {
4
        // Algorithm:
4
        //
4
        // 1. Determine which HsDirRing to use, based on the time period.
4
        // 2. Find the shared random value that's associated with that HsDirRing.
4
        // 3. Choose spread = the parameter `hsdir_spread_store` or
4
        //    `hsdir_spread_fetch` based on `op`.
4
        // 4. Let n_replicas = the parameter `hsdir_n_replicas`.
4
        // 5. Initialize Dirs = []
4
        // 6. for idx in 1..=n_replicas:
4
        //       - let H = hsdir_ring::onion_service_index(id, replica, rand,
4
        //         period).
4
        //       - Find the position of H within hsdir_ring.
4
        //       - Take elements from hsdir_ring starting at that position,
4
        //         adding them to Dirs until we have added `spread` new elements
4
        //         that were not there before.
4
        // 7. return Dirs.
4
        let n_replicas = self
4
            .params
4
            .hsdir_n_replicas
4
            .get()
4
            .try_into()
4
            .expect("BoundedInt did not enforce bounds");
4
        let spread = match op {
2
            HsDirOp::Download => self.params.hsdir_spread_fetch,
            #[cfg(feature = "hs-service")]
2
            HsDirOp::Upload => self.params.hsdir_spread_store,
        };
4
        let spread = spread
4
            .get()
4
            .try_into()
4
            .expect("BoundedInt did not enforce bounds!");
4

            
4
        // TODO: I may be wrong here but I suspect that this function may
4
        // need refactoring so that it does not look at _all_ of the HsDirRings,
4
        // but only at the ones that corresponds to time periods for which
4
        // HsBlindId is valid.  Or I could be mistaken, in which case we should
4
        // have a comment to explain why I am, since the logic is subtle.
4
        // (For clients, there is only one ring.) -nickm
4
        //
4
        // (Actually, there is no need to follow through with the above TODO,
4
        // since this function is deprecated, and not used anywhere but the
4
        // tests.)
4

            
4
        let mut hs_dirs = self
4
            .hsdir_rings
4
            .iter_for_op(op)
4
            .cartesian_product(1..=n_replicas) // 1-indexed !
4
            .flat_map({
4
                let mut selected_nodes = HashSet::new();
4

            
8
                move |(ring, replica): (&HsDirRing, u8)| {
8
                    let hsdir_idx = hsdir_ring::service_hsdir_index(hsid, replica, ring.params());
8

            
8
                    let items = ring
40
                        .ring_items_at(hsdir_idx, spread, |(hsdir_idx, _)| {
40
                            // According to rend-spec 2.2.3:
40
                            //                                                  ... If any of those
40
                            // nodes have already been selected for a lower-numbered replica of the
40
                            // service, any nodes already chosen are disregarded (i.e. skipped over)
40
                            // when choosing a replica's hsdir_spread_store nodes.
40
                            selected_nodes.insert(*hsdir_idx)
40
                        })
8
                        .collect::<Vec<_>>();
8

            
8
                    items
8
                }
4
            })
28
            .filter_map(|(_hsdir_idx, rs_idx)| {
28
                // This ought not to be None but let's not panic or bail if it is
28
                self.relay_by_rs_idx(*rs_idx)
28
            })
4
            .collect_vec();
4

            
4
        match op {
2
            HsDirOp::Download => {
2
                // When `op` is `Download`, the order is random.
2
                hs_dirs.shuffle(rng);
2
            }
            #[cfg(feature = "hs-service")]
2
            HsDirOp::Upload => {
2
                // When `op` is `Upload`, the order is not specified.
2
            }
        }
4
        hs_dirs
4
    }
}
impl MdReceiver for NetDir {
145
    fn missing_microdescs(&self) -> Box<dyn Iterator<Item = &MdDigest> + '_> {
145
        Box::new(self.rsidx_by_missing.keys())
145
    }
294738
    fn add_microdesc(&mut self, md: Microdesc) -> bool {
294738
        self.add_arc_microdesc(Arc::new(md))
294738
    }
270
    fn n_missing(&self) -> usize {
270
        self.rsidx_by_missing.len()
270
    }
}
impl<'a> UncheckedRelay<'a> {
    /// Return an [`UncheckedRelayDetails`](details::UncheckedRelayDetails) for this relay.
    ///
    /// Callers should generally avoid using this information directly if they can;
    /// it's better to use a higher-level function that exposes semantic information
    /// rather than these properties.
828395
    pub fn low_level_details(&self) -> details::UncheckedRelayDetails<'_> {
828395
        details::UncheckedRelayDetails(self)
828395
    }
    /// Return true if this relay is valid and [usable](NetDir#usable).
    ///
    /// This function should return `true` for every Relay we expose
    /// to the user.
32057161
    pub fn is_usable(&self) -> bool {
32057161
        // No need to check for 'valid' or 'running': they are implicit.
32057161
        self.md.is_some() && self.rs.ed25519_id_is_usable()
32057161
    }
    /// If this is [usable](NetDir#usable), return a corresponding Relay object.
31499831
    pub fn into_relay(self) -> Option<Relay<'a>> {
31499831
        if self.is_usable() {
            Some(Relay {
31499807
                rs: self.rs,
31499807
                md: self.md?,
                #[cfg(feature = "geoip")]
31499807
                cc: self.cc,
            })
        } else {
24
            None
        }
31499831
    }
    /// Return true if this relay is a hidden service directory
    ///
    /// Ie, if it is to be included in the hsdir ring.
    #[cfg(feature = "hs-common")]
293395
    pub(crate) fn is_hsdir_for_ring(&self) -> bool {
293395
        // TODO are there any other flags should we check?
293395
        // rend-spec-v3 2.2.3 says just
293395
        //   "each node listed in the current consensus with the HSDir flag"
293395
        // Do we need to check ed25519_id_is_usable ?
293395
        // See also https://gitlab.torproject.org/tpo/core/arti/-/issues/504
293395
        self.rs.is_flagged_hsdir()
293395
    }
}
impl<'a> Relay<'a> {
    /// Return a [`RelayDetails`](details::RelayDetails) for this relay.
    ///
    /// Callers should generally avoid using this information directly if they can;
    /// it's better to use a higher-level function that exposes semantic information
    /// rather than these properties.
64724131
    pub fn low_level_details(&self) -> details::RelayDetails<'_> {
64724131
        details::RelayDetails(self)
64724131
    }
    /// Return the Ed25519 ID for this relay.
95171722
    pub fn id(&self) -> &Ed25519Identity {
95171722
        self.md.ed25519_id()
95171722
    }
    /// Return the RsaIdentity for this relay.
61819014
    pub fn rsa_id(&self) -> &RsaIdentity {
61819014
        self.rs.rsa_identity()
61819014
    }
    /// Return a reference to this relay's "router status" entry in
    /// the consensus.
    ///
    /// The router status entry contains information about the relay
    /// that the authorities voted on directly.  For most use cases,
    /// you shouldn't need them.
    ///
    /// This function is only available if the crate was built with
    /// its `experimental-api` feature.
    #[cfg(feature = "experimental-api")]
4
    pub fn rs(&self) -> &netstatus::MdConsensusRouterStatus {
4
        self.rs
4
    }
    /// Return a reference to this relay's "microdescriptor" entry in
    /// the consensus.
    ///
    /// A "microdescriptor" is a synopsis of the information about a relay,
    /// used to determine its capabilities and route traffic through it.
    /// For most use cases, you shouldn't need it.
    ///
    /// This function is only available if the crate was built with
    /// its `experimental-api` feature.
    #[cfg(feature = "experimental-api")]
4
    pub fn md(&self) -> &Microdesc {
4
        self.md
4
    }
}
/// An error value returned from [`NetDir::by_ids_detailed`].
#[cfg(feature = "hs-common")]
#[derive(Clone, Debug, thiserror::Error)]
#[non_exhaustive]
pub enum RelayLookupError {
    /// We found a relay whose presence indicates that the provided set of
    /// identities is impossible to resolve.
    #[error("Provided set of identities is impossible according to consensus.")]
    Impossible,
}
impl<'a> HasAddrs for Relay<'a> {
86982882
    fn addrs(&self) -> &[std::net::SocketAddr] {
86982882
        self.rs.addrs()
86982882
    }
}
#[cfg(feature = "geoip")]
#[cfg_attr(docsrs, doc(cfg(feature = "geoip")))]
impl<'a> HasCountryCode for Relay<'a> {
540
    fn country_code(&self) -> Option<CountryCode> {
540
        self.cc
540
    }
}
impl<'a> tor_linkspec::HasRelayIdsLegacy for Relay<'a> {
95160460
    fn ed_identity(&self) -> &Ed25519Identity {
95160460
        self.id()
95160460
    }
29053318
    fn rsa_identity(&self) -> &RsaIdentity {
29053318
        self.rsa_id()
29053318
    }
}
impl<'a> HasRelayIds for UncheckedRelay<'a> {
    fn identity(&self, key_type: RelayIdType) -> Option<RelayIdRef<'_>> {
        match key_type {
            RelayIdType::Ed25519 if self.rs.ed25519_id_is_usable() => {
                self.md.map(|m| m.ed25519_id().into())
            }
            RelayIdType::Rsa => Some(self.rs.rsa_identity().into()),
            _ => None,
        }
    }
}
#[cfg(feature = "geoip")]
impl<'a> HasCountryCode for UncheckedRelay<'a> {
    fn country_code(&self) -> Option<CountryCode> {
        self.cc
    }
}
impl<'a> DirectChanMethodsHelper for Relay<'a> {}
impl<'a> ChanTarget for Relay<'a> {}
impl<'a> tor_linkspec::CircTarget for Relay<'a> {
1139445
    fn ntor_onion_key(&self) -> &ll::pk::curve25519::PublicKey {
1139445
        self.md.ntor_key()
1139445
    }
1139445
    fn protovers(&self) -> &tor_protover::Protocols {
1139445
        self.rs.protovers()
1139445
    }
}
#[cfg(test)]
mod test {
    // @@ begin test lint list maintained by maint/add_warning @@
    #![allow(clippy::bool_assert_comparison)]
    #![allow(clippy::clone_on_copy)]
    #![allow(clippy::dbg_macro)]
    #![allow(clippy::mixed_attributes_style)]
    #![allow(clippy::print_stderr)]
    #![allow(clippy::print_stdout)]
    #![allow(clippy::single_char_pattern)]
    #![allow(clippy::unwrap_used)]
    #![allow(clippy::unchecked_duration_subtraction)]
    #![allow(clippy::useless_vec)]
    #![allow(clippy::needless_pass_by_value)]
    //! <!-- @@ end test lint list maintained by maint/add_warning @@ -->
    #![allow(clippy::cognitive_complexity)]
    use super::*;
    use crate::testnet::*;
    use float_eq::assert_float_eq;
    use std::collections::HashSet;
    use std::time::Duration;
    use tor_basic_utils::test_rng::{self, testing_rng};
    use tor_linkspec::{RelayIdType, RelayIds};
    #[cfg(feature = "hs-common")]
    fn dummy_hs_blind_id() -> HsBlindId {
        let hsid = [2, 1, 1, 1].iter().cycle().take(32).cloned().collect_vec();
        let hsid = Ed25519Identity::new(hsid[..].try_into().unwrap());
        HsBlindId::from(hsid)
    }
    // Basic functionality for a partial netdir: Add microdescriptors,
    // then you have a netdir.
    #[test]
    fn partial_netdir() {
        let (consensus, microdescs) = construct_network().unwrap();
        let dir = PartialNetDir::new(consensus, None);
        // Check the lifetime
        let lifetime = dir.lifetime();
        assert_eq!(
            lifetime
                .valid_until()
                .duration_since(lifetime.valid_after())
                .unwrap(),
            Duration::new(86400, 0)
        );
        // No microdescriptors, so we don't have enough paths, and can't
        // advance.
        assert!(!dir.have_enough_paths());
        let mut dir = match dir.unwrap_if_sufficient() {
            Ok(_) => panic!(),
            Err(d) => d,
        };
        let missing: HashSet<_> = dir.missing_microdescs().collect();
        assert_eq!(missing.len(), 40);
        assert_eq!(missing.len(), dir.netdir.c_relays().len());
        for md in &microdescs {
            assert!(missing.contains(md.digest()));
        }
        // Now add all the mds and try again.
        for md in microdescs {
            let wanted = dir.add_microdesc(md);
            assert!(wanted);
        }
        let missing: HashSet<_> = dir.missing_microdescs().collect();
        assert!(missing.is_empty());
        assert!(dir.have_enough_paths());
        let _complete = match dir.unwrap_if_sufficient() {
            Ok(d) => d,
            Err(_) => panic!(),
        };
    }
    #[test]
    fn override_params() {
        let (consensus, _microdescs) = construct_network().unwrap();
        let override_p = "bwweightscale=2 doesnotexist=77 circwindow=500"
            .parse()
            .unwrap();
        let dir = PartialNetDir::new(consensus.clone(), Some(&override_p));
        let params = &dir.netdir.params;
        assert_eq!(params.bw_weight_scale.get(), 2);
        assert_eq!(params.circuit_window.get(), 500_i32);
        // try again without the override.
        let dir = PartialNetDir::new(consensus, None);
        let params = &dir.netdir.params;
        assert_eq!(params.bw_weight_scale.get(), 1_i32);
        assert_eq!(params.circuit_window.get(), 1000_i32);
    }
    #[test]
    fn fill_from_previous() {
        let (consensus, microdescs) = construct_network().unwrap();
        let mut dir = PartialNetDir::new(consensus.clone(), None);
        for md in microdescs.iter().skip(2) {
            let wanted = dir.add_microdesc(md.clone());
            assert!(wanted);
        }
        let dir1 = dir.unwrap_if_sufficient().unwrap();
        assert_eq!(dir1.missing_microdescs().count(), 2);
        let mut dir = PartialNetDir::new(consensus, None);
        assert_eq!(dir.missing_microdescs().count(), 40);
        dir.fill_from_previous_netdir(Arc::new(dir1));
        assert_eq!(dir.missing_microdescs().count(), 2);
    }
    #[test]
    fn path_count() {
        let low_threshold = "min_paths_for_circs_pct=64".parse().unwrap();
        let high_threshold = "min_paths_for_circs_pct=65".parse().unwrap();
        let (consensus, microdescs) = construct_network().unwrap();
        let mut dir = PartialNetDir::new(consensus.clone(), Some(&low_threshold));
        for (pos, md) in microdescs.iter().enumerate() {
            if pos % 7 == 2 {
                continue; // skip a few relays.
            }
            dir.add_microdesc(md.clone());
        }
        let dir = dir.unwrap_if_sufficient().unwrap();
        // We  have 40 relays that we know about from the consensus.
        assert_eq!(dir.all_relays().count(), 40);
        // But only 34 are usable.
        assert_eq!(dir.relays().count(), 34);
        // For guards: mds 20..=39 correspond to Guard relays.
        // Their bandwidth is 2*(1000+2000+...10000) = 110_000.
        // We skipped 23, 30, and 37.  They have bandwidth
        // 4000 + 1000 + 8000 = 13_000.  So our fractional bandwidth
        // should be (110-13)/110.
        let f = dir.frac_for_role(WeightRole::Guard, |u| u.rs.is_flagged_guard());
        assert!(((97.0 / 110.0) - f).abs() < 0.000001);
        // For exits: mds 10..=19 and 30..=39 correspond to Exit relays.
        // We skipped 16, 30,  and 37. Per above our fractional bandwidth is
        // (110-16)/110.
        let f = dir.frac_for_role(WeightRole::Exit, |u| u.rs.is_flagged_exit());
        assert!(((94.0 / 110.0) - f).abs() < 0.000001);
        // For middles: all relays are middles. We skipped 2, 9, 16,
        // 23, 30, and 37. Per above our fractional bandwidth is
        // (220-33)/220
        let f = dir.frac_for_role(WeightRole::Middle, |_| true);
        assert!(((187.0 / 220.0) - f).abs() < 0.000001);
        // Multiplying those together, we get the fraction of paths we can
        // build at ~0.64052066, which is above the threshold we set above for
        // MinPathsForCircsPct.
        let f = dir.frac_usable_paths();
        assert!((f - 0.64052066).abs() < 0.000001);
        // But if we try again with a slightly higher threshold...
        let mut dir = PartialNetDir::new(consensus, Some(&high_threshold));
        for (pos, md) in microdescs.into_iter().enumerate() {
            if pos % 7 == 2 {
                continue; // skip a few relays.
            }
            dir.add_microdesc(md);
        }
        assert!(dir.unwrap_if_sufficient().is_err());
    }
    /// Return a 3-tuple for use by `test_pick_*()` of an Rng, a number of
    /// iterations, and a tolerance.
    ///
    /// If the Rng is deterministic (the default), we can use a faster setup,
    /// with a higher tolerance and fewer iterations.  But if you've explicitly
    /// opted into randomization (or are replaying a seed from an earlier
    /// randomized test), we give you more iterations and a tighter tolerance.
    fn testing_rng_with_tolerances() -> (impl rand::Rng, usize, f64) {
        // Use a deterministic RNG if none is specified, since this is slow otherwise.
        let config = test_rng::Config::from_env().unwrap_or(test_rng::Config::Deterministic);
        let (iters, tolerance) = match config {
            test_rng::Config::Deterministic => (5000, 0.02),
            _ => (50000, 0.01),
        };
        (config.into_rng(), iters, tolerance)
    }
    #[test]
    fn test_pick() {
        let (consensus, microdescs) = construct_network().unwrap();
        let mut dir = PartialNetDir::new(consensus, None);
        for md in microdescs.into_iter() {
            let wanted = dir.add_microdesc(md.clone());
            assert!(wanted);
        }
        let dir = dir.unwrap_if_sufficient().unwrap();
        let (mut rng, total, tolerance) = testing_rng_with_tolerances();
        let mut picked = [0_isize; 40];
        for _ in 0..total {
            let r = dir.pick_relay(&mut rng, WeightRole::Middle, |r| {
                r.low_level_details().supports_exit_port_ipv4(80)
            });
            let r = r.unwrap();
            let id_byte = r.identity(RelayIdType::Rsa).unwrap().as_bytes()[0];
            picked[id_byte as usize] += 1;
        }
        // non-exits should never get picked.
        picked[0..10].iter().for_each(|x| assert_eq!(*x, 0));
        picked[20..30].iter().for_each(|x| assert_eq!(*x, 0));
        let picked_f: Vec<_> = picked.iter().map(|x| *x as f64 / total as f64).collect();
        // We didn't we any non-default weights, so the other relays get
        // weighted proportional to their bandwidth.
        assert_float_eq!(picked_f[19], (10.0 / 110.0), abs <= tolerance);
        assert_float_eq!(picked_f[38], (9.0 / 110.0), abs <= tolerance);
        assert_float_eq!(picked_f[39], (10.0 / 110.0), abs <= tolerance);
    }
    #[test]
    fn test_pick_multiple() {
        // This is mostly a copy of test_pick, except that it uses
        // pick_n_relays to pick several relays at once.
        let dir = construct_netdir().unwrap_if_sufficient().unwrap();
        let (mut rng, total, tolerance) = testing_rng_with_tolerances();
        let mut picked = [0_isize; 40];
        for _ in 0..total / 4 {
            let relays = dir.pick_n_relays(&mut rng, 4, WeightRole::Middle, |r| {
                r.low_level_details().supports_exit_port_ipv4(80)
            });
            assert_eq!(relays.len(), 4);
            for r in relays {
                let id_byte = r.identity(RelayIdType::Rsa).unwrap().as_bytes()[0];
                picked[id_byte as usize] += 1;
            }
        }
        // non-exits should never get picked.
        picked[0..10].iter().for_each(|x| assert_eq!(*x, 0));
        picked[20..30].iter().for_each(|x| assert_eq!(*x, 0));
        let picked_f: Vec<_> = picked.iter().map(|x| *x as f64 / total as f64).collect();
        // We didn't we any non-default weights, so the other relays get
        // weighted proportional to their bandwidth.
        assert_float_eq!(picked_f[19], (10.0 / 110.0), abs <= tolerance);
        assert_float_eq!(picked_f[36], (7.0 / 110.0), abs <= tolerance);
        assert_float_eq!(picked_f[39], (10.0 / 110.0), abs <= tolerance);
    }
    #[test]
    fn subnets() {
        let cfg = SubnetConfig::default();
        fn same_net(cfg: &SubnetConfig, a: &str, b: &str) -> bool {
            cfg.addrs_in_same_subnet(&a.parse().unwrap(), &b.parse().unwrap())
        }
        assert!(same_net(&cfg, "127.15.3.3", "127.15.9.9"));
        assert!(!same_net(&cfg, "127.15.3.3", "127.16.9.9"));
        assert!(!same_net(&cfg, "127.15.3.3", "127::"));
        assert!(same_net(&cfg, "ffff:ffff:90:33::", "ffff:ffff:91:34::"));
        assert!(!same_net(&cfg, "ffff:ffff:90:33::", "ffff:fffe:91:34::"));
        let cfg = SubnetConfig {
            subnets_family_v4: 32,
            subnets_family_v6: 128,
        };
        assert!(!same_net(&cfg, "127.15.3.3", "127.15.9.9"));
        assert!(!same_net(&cfg, "ffff:ffff:90:33::", "ffff:ffff:91:34::"));
        assert!(same_net(&cfg, "127.0.0.1", "127.0.0.1"));
        assert!(!same_net(&cfg, "127.0.0.1", "127.0.0.2"));
        assert!(same_net(&cfg, "ffff:ffff:90:33::", "ffff:ffff:90:33::"));
        let cfg = SubnetConfig {
            subnets_family_v4: 33,
            subnets_family_v6: 129,
        };
        assert!(!same_net(&cfg, "127.0.0.1", "127.0.0.1"));
        assert!(!same_net(&cfg, "::", "::"));
    }
    #[test]
    fn subnet_union() {
        let cfg1 = SubnetConfig {
            subnets_family_v4: 16,
            subnets_family_v6: 64,
        };
        let cfg2 = SubnetConfig {
            subnets_family_v4: 24,
            subnets_family_v6: 32,
        };
        let a1 = "1.2.3.4".parse().unwrap();
        let a2 = "1.2.10.10".parse().unwrap();
        let a3 = "ffff:ffff::7".parse().unwrap();
        let a4 = "ffff:ffff:1234::8".parse().unwrap();
        assert_eq!(cfg1.addrs_in_same_subnet(&a1, &a2), true);
        assert_eq!(cfg2.addrs_in_same_subnet(&a1, &a2), false);
        assert_eq!(cfg1.addrs_in_same_subnet(&a3, &a4), false);
        assert_eq!(cfg2.addrs_in_same_subnet(&a3, &a4), true);
        let cfg_u = cfg1.union(&cfg2);
        assert_eq!(
            cfg_u,
            SubnetConfig {
                subnets_family_v4: 16,
                subnets_family_v6: 32,
            }
        );
        assert_eq!(cfg_u.addrs_in_same_subnet(&a1, &a2), true);
        assert_eq!(cfg_u.addrs_in_same_subnet(&a3, &a4), true);
        assert_eq!(cfg1.union(&cfg1), cfg1);
        assert_eq!(cfg1.union(&SubnetConfig::no_addresses_match()), cfg1);
    }
    #[test]
    fn relay_funcs() {
        let (consensus, microdescs) = construct_custom_network(
            |pos, nb, _| {
                if pos == 15 {
                    nb.rs.add_or_port("[f0f0::30]:9001".parse().unwrap());
                } else if pos == 20 {
                    nb.rs.add_or_port("[f0f0::3131]:9001".parse().unwrap());
                }
            },
            None,
        )
        .unwrap();
        let subnet_config = SubnetConfig::default();
        let all_family_info = FamilyRules::all_family_info();
        let mut dir = PartialNetDir::new(consensus, None);
        for md in microdescs.into_iter() {
            let wanted = dir.add_microdesc(md.clone());
            assert!(wanted);
        }
        let dir = dir.unwrap_if_sufficient().unwrap();
        // Pick out a few relays by ID.
        let k0 = Ed25519Identity::from([0; 32]);
        let k1 = Ed25519Identity::from([1; 32]);
        let k2 = Ed25519Identity::from([2; 32]);
        let k3 = Ed25519Identity::from([3; 32]);
        let k10 = Ed25519Identity::from([10; 32]);
        let k15 = Ed25519Identity::from([15; 32]);
        let k20 = Ed25519Identity::from([20; 32]);
        let r0 = dir.by_id(&k0).unwrap();
        let r1 = dir.by_id(&k1).unwrap();
        let r2 = dir.by_id(&k2).unwrap();
        let r3 = dir.by_id(&k3).unwrap();
        let r10 = dir.by_id(&k10).unwrap();
        let r15 = dir.by_id(&k15).unwrap();
        let r20 = dir.by_id(&k20).unwrap();
        assert_eq!(r0.id(), &[0; 32].into());
        assert_eq!(r0.rsa_id(), &[0; 20].into());
        assert_eq!(r1.id(), &[1; 32].into());
        assert_eq!(r1.rsa_id(), &[1; 20].into());
        assert!(r0.same_relay_ids(&r0));
        assert!(r1.same_relay_ids(&r1));
        assert!(!r1.same_relay_ids(&r0));
        assert!(r0.low_level_details().is_dir_cache());
        assert!(!r1.low_level_details().is_dir_cache());
        assert!(r2.low_level_details().is_dir_cache());
        assert!(!r3.low_level_details().is_dir_cache());
        assert!(!r0.low_level_details().supports_exit_port_ipv4(80));
        assert!(!r1.low_level_details().supports_exit_port_ipv4(80));
        assert!(!r2.low_level_details().supports_exit_port_ipv4(80));
        assert!(!r3.low_level_details().supports_exit_port_ipv4(80));
        assert!(!r0.low_level_details().policies_allow_some_port());
        assert!(!r1.low_level_details().policies_allow_some_port());
        assert!(!r2.low_level_details().policies_allow_some_port());
        assert!(!r3.low_level_details().policies_allow_some_port());
        assert!(r10.low_level_details().policies_allow_some_port());
        assert!(r0.low_level_details().in_same_family(&r0, all_family_info));
        assert!(r0.low_level_details().in_same_family(&r1, all_family_info));
        assert!(r1.low_level_details().in_same_family(&r0, all_family_info));
        assert!(r1.low_level_details().in_same_family(&r1, all_family_info));
        assert!(!r0.low_level_details().in_same_family(&r2, all_family_info));
        assert!(!r2.low_level_details().in_same_family(&r0, all_family_info));
        assert!(r2.low_level_details().in_same_family(&r2, all_family_info));
        assert!(r2.low_level_details().in_same_family(&r3, all_family_info));
        assert!(r0.low_level_details().in_same_subnet(&r10, &subnet_config));
        assert!(r10.low_level_details().in_same_subnet(&r10, &subnet_config));
        assert!(r0.low_level_details().in_same_subnet(&r0, &subnet_config));
        assert!(r1.low_level_details().in_same_subnet(&r1, &subnet_config));
        assert!(!r1.low_level_details().in_same_subnet(&r2, &subnet_config));
        assert!(!r2.low_level_details().in_same_subnet(&r3, &subnet_config));
        // Make sure IPv6 families work.
        let subnet_config = SubnetConfig {
            subnets_family_v4: 128,
            subnets_family_v6: 96,
        };
        assert!(r15.low_level_details().in_same_subnet(&r20, &subnet_config));
        assert!(!r15.low_level_details().in_same_subnet(&r1, &subnet_config));
        // Make sure that subnet configs can be disabled.
        let subnet_config = SubnetConfig {
            subnets_family_v4: 255,
            subnets_family_v6: 255,
        };
        assert!(!r15.low_level_details().in_same_subnet(&r20, &subnet_config));
    }
    #[test]
    fn test_badexit() {
        // make a netdir where relays 10-19 are badexit, and everybody
        // exits to 443 on IPv6.
        use tor_netdoc::doc::netstatus::RelayFlags;
        let netdir = construct_custom_netdir(|pos, nb, _| {
            if (10..20).contains(&pos) {
                nb.rs.add_flags(RelayFlags::BAD_EXIT);
            }
            nb.md.parse_ipv6_policy("accept 443").unwrap();
        })
        .unwrap()
        .unwrap_if_sufficient()
        .unwrap();
        let e12 = netdir.by_id(&Ed25519Identity::from([12; 32])).unwrap();
        let e32 = netdir.by_id(&Ed25519Identity::from([32; 32])).unwrap();
        assert!(!e12.low_level_details().supports_exit_port_ipv4(80));
        assert!(e32.low_level_details().supports_exit_port_ipv4(80));
        assert!(!e12.low_level_details().supports_exit_port_ipv6(443));
        assert!(e32.low_level_details().supports_exit_port_ipv6(443));
        assert!(!e32.low_level_details().supports_exit_port_ipv6(555));
        assert!(!e12.low_level_details().policies_allow_some_port());
        assert!(e32.low_level_details().policies_allow_some_port());
        assert!(!e12.low_level_details().ipv4_policy().allows_some_port());
        assert!(!e12.low_level_details().ipv6_policy().allows_some_port());
        assert!(e32.low_level_details().ipv4_policy().allows_some_port());
        assert!(e32.low_level_details().ipv6_policy().allows_some_port());
        assert!(e12
            .low_level_details()
            .ipv4_declared_policy()
            .allows_some_port());
        assert!(e12
            .low_level_details()
            .ipv6_declared_policy()
            .allows_some_port());
    }
    #[cfg(feature = "experimental-api")]
    #[test]
    fn test_accessors() {
        let netdir = construct_netdir().unwrap_if_sufficient().unwrap();
        let r4 = netdir.by_id(&Ed25519Identity::from([4; 32])).unwrap();
        let r16 = netdir.by_id(&Ed25519Identity::from([16; 32])).unwrap();
        assert!(!r4.md().ipv4_policy().allows_some_port());
        assert!(r16.md().ipv4_policy().allows_some_port());
        assert!(!r4.rs().is_flagged_exit());
        assert!(r16.rs().is_flagged_exit());
    }
    #[test]
    fn test_by_id() {
        // Make a netdir that omits the microdescriptor for 0xDDDDDD...
        let netdir = construct_custom_netdir(|pos, nb, _| {
            nb.omit_md = pos == 13;
        })
        .unwrap();
        let netdir = netdir.unwrap_if_sufficient().unwrap();
        let r = netdir.by_id(&Ed25519Identity::from([0; 32])).unwrap();
        assert_eq!(r.id().as_bytes(), &[0; 32]);
        assert!(netdir.by_id(&Ed25519Identity::from([13; 32])).is_none());
        let r = netdir.by_rsa_id(&[12; 20].into()).unwrap();
        assert_eq!(r.rsa_id().as_bytes(), &[12; 20]);
        assert!(netdir.rsa_id_is_listed(&[12; 20].into()));
        assert!(netdir.by_rsa_id(&[13; 20].into()).is_none());
        assert!(netdir.by_rsa_id_unchecked(&[99; 20].into()).is_none());
        assert!(!netdir.rsa_id_is_listed(&[99; 20].into()));
        let r = netdir.by_rsa_id_unchecked(&[13; 20].into()).unwrap();
        assert_eq!(r.rs.rsa_identity().as_bytes(), &[13; 20]);
        assert!(netdir.rsa_id_is_listed(&[13; 20].into()));
        let pair_13_13 = RelayIds::builder()
            .ed_identity([13; 32].into())
            .rsa_identity([13; 20].into())
            .build()
            .unwrap();
        let pair_14_14 = RelayIds::builder()
            .ed_identity([14; 32].into())
            .rsa_identity([14; 20].into())
            .build()
            .unwrap();
        let pair_14_99 = RelayIds::builder()
            .ed_identity([14; 32].into())
            .rsa_identity([99; 20].into())
            .build()
            .unwrap();
        let r = netdir.by_ids(&pair_13_13);
        assert!(r.is_none());
        let r = netdir.by_ids(&pair_14_14).unwrap();
        assert_eq!(r.identity(RelayIdType::Rsa).unwrap().as_bytes(), &[14; 20]);
        assert_eq!(
            r.identity(RelayIdType::Ed25519).unwrap().as_bytes(),
            &[14; 32]
        );
        let r = netdir.by_ids(&pair_14_99);
        assert!(r.is_none());
        assert_eq!(
            netdir.id_pair_listed(&[13; 32].into(), &[13; 20].into()),
            None
        );
        assert_eq!(
            netdir.id_pair_listed(&[15; 32].into(), &[15; 20].into()),
            Some(true)
        );
        assert_eq!(
            netdir.id_pair_listed(&[15; 32].into(), &[99; 20].into()),
            Some(false)
        );
    }
    #[test]
    #[cfg(feature = "hs-common")]
    fn test_by_ids_detailed() {
        // Make a netdir that omits the microdescriptor for 0xDDDDDD...
        let netdir = construct_custom_netdir(|pos, nb, _| {
            nb.omit_md = pos == 13;
        })
        .unwrap();
        let netdir = netdir.unwrap_if_sufficient().unwrap();
        let id13_13 = RelayIds::builder()
            .ed_identity([13; 32].into())
            .rsa_identity([13; 20].into())
            .build()
            .unwrap();
        let id15_15 = RelayIds::builder()
            .ed_identity([15; 32].into())
            .rsa_identity([15; 20].into())
            .build()
            .unwrap();
        let id15_99 = RelayIds::builder()
            .ed_identity([15; 32].into())
            .rsa_identity([99; 20].into())
            .build()
            .unwrap();
        let id99_15 = RelayIds::builder()
            .ed_identity([99; 32].into())
            .rsa_identity([15; 20].into())
            .build()
            .unwrap();
        let id99_99 = RelayIds::builder()
            .ed_identity([99; 32].into())
            .rsa_identity([99; 20].into())
            .build()
            .unwrap();
        let id15_xx = RelayIds::builder()
            .ed_identity([15; 32].into())
            .build()
            .unwrap();
        let idxx_15 = RelayIds::builder()
            .rsa_identity([15; 20].into())
            .build()
            .unwrap();
        assert!(matches!(netdir.by_ids_detailed(&id13_13), Ok(None)));
        assert!(matches!(netdir.by_ids_detailed(&id15_15), Ok(Some(_))));
        assert!(matches!(
            netdir.by_ids_detailed(&id15_99),
            Err(RelayLookupError::Impossible)
        ));
        assert!(matches!(
            netdir.by_ids_detailed(&id99_15),
            Err(RelayLookupError::Impossible)
        ));
        assert!(matches!(netdir.by_ids_detailed(&id99_99), Ok(None)));
        assert!(matches!(netdir.by_ids_detailed(&id15_xx), Ok(Some(_))));
        assert!(matches!(netdir.by_ids_detailed(&idxx_15), Ok(Some(_))));
    }
    #[test]
    fn weight_type() {
        let r0 = RelayWeight(0);
        let r100 = RelayWeight(100);
        let r200 = RelayWeight(200);
        let r300 = RelayWeight(300);
        assert_eq!(r100 + r200, r300);
        assert_eq!(r100.checked_div(r200), Some(0.5));
        assert!(r100.checked_div(r0).is_none());
        assert_eq!(r200.ratio(0.5), Some(r100));
        assert!(r200.ratio(-1.0).is_none());
    }
    #[test]
    fn weight_accessors() {
        // Make a netdir that omits the microdescriptor for 0xDDDDDD...
        let netdir = construct_netdir().unwrap_if_sufficient().unwrap();
        let g_total = netdir.total_weight(WeightRole::Guard, |r| r.rs.is_flagged_guard());
        // This is just the total guard weight, since all our Wxy = 1.
        assert_eq!(g_total, RelayWeight(110_000));
        let g_total = netdir.total_weight(WeightRole::Guard, |_| false);
        assert_eq!(g_total, RelayWeight(0));
        let relay = netdir.by_id(&Ed25519Identity::from([35; 32])).unwrap();
        assert!(relay.rs.is_flagged_guard());
        let w = netdir.relay_weight(&relay, WeightRole::Guard);
        assert_eq!(w, RelayWeight(6_000));
        let w = netdir
            .weight_by_rsa_id(&[33; 20].into(), WeightRole::Guard)
            .unwrap();
        assert_eq!(w, RelayWeight(4_000));
        assert!(netdir
            .weight_by_rsa_id(&[99; 20].into(), WeightRole::Guard)
            .is_none());
    }
    #[test]
    fn family_list() {
        let netdir = construct_custom_netdir(|pos, n, _| {
            if pos == 0x0a {
                n.md.family(
                    "$0B0B0B0B0B0B0B0B0B0B0B0B0B0B0B0B0B0B0B0B \
                     $0C0C0C0C0C0C0C0C0C0C0C0C0C0C0C0C0C0C0C0C \
                     $0D0D0D0D0D0D0D0D0D0D0D0D0D0D0D0D0D0D0D0D"
                        .parse()
                        .unwrap(),
                );
            } else if pos == 0x0c {
                n.md.family("$0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A".parse().unwrap());
            }
        })
        .unwrap()
        .unwrap_if_sufficient()
        .unwrap();
        // In the testing netdir, adjacent members are in the same family by default...
        let r0 = netdir.by_id(&Ed25519Identity::from([0; 32])).unwrap();
        let family: Vec<_> = netdir.known_family_members(&r0).collect();
        assert_eq!(family.len(), 1);
        assert_eq!(family[0].id(), &Ed25519Identity::from([1; 32]));
        // But we've made this relay claim membership with several others.
        let r10 = netdir.by_id(&Ed25519Identity::from([10; 32])).unwrap();
        let family: HashSet<_> = netdir.known_family_members(&r10).map(|r| *r.id()).collect();
        assert_eq!(family.len(), 2);
        assert!(family.contains(&Ed25519Identity::from([11; 32])));
        assert!(family.contains(&Ed25519Identity::from([12; 32])));
        // Note that 13 doesn't get put in, even though it's listed, since it doesn't claim
        //  membership with 10.
    }
    #[test]
    #[cfg(feature = "geoip")]
    fn relay_has_country_code() {
        let src_v6 = r#"
        fe80:dead:beef::,fe80:dead:ffff::,US
        fe80:feed:eeee::1,fe80:feed:eeee::2,AT
        fe80:feed:eeee::2,fe80:feed:ffff::,DE
        "#;
        let db = GeoipDb::new_from_legacy_format("", src_v6).unwrap();
        let netdir = construct_custom_netdir_with_geoip(
            |pos, n, _| {
                if pos == 0x01 {
                    n.rs.add_or_port("[fe80:dead:beef::1]:42".parse().unwrap());
                }
                if pos == 0x02 {
                    n.rs.add_or_port("[fe80:feed:eeee::1]:42".parse().unwrap());
                    n.rs.add_or_port("[fe80:feed:eeee::2]:42".parse().unwrap());
                }
                if pos == 0x03 {
                    n.rs.add_or_port("[fe80:dead:beef::1]:42".parse().unwrap());
                    n.rs.add_or_port("[fe80:dead:beef::2]:42".parse().unwrap());
                }
            },
            &db,
        )
        .unwrap()
        .unwrap_if_sufficient()
        .unwrap();
        // No GeoIP data available -> None
        let r0 = netdir.by_id(&Ed25519Identity::from([0; 32])).unwrap();
        assert_eq!(r0.cc, None);
        // Exactly one match -> Some
        let r1 = netdir.by_id(&Ed25519Identity::from([1; 32])).unwrap();
        assert_eq!(r1.cc.as_ref().map(|x| x.as_ref()), Some("US"));
        // Conflicting matches -> None
        let r2 = netdir.by_id(&Ed25519Identity::from([2; 32])).unwrap();
        assert_eq!(r2.cc, None);
        // Multiple agreeing matches -> Some
        let r3 = netdir.by_id(&Ed25519Identity::from([3; 32])).unwrap();
        assert_eq!(r3.cc.as_ref().map(|x| x.as_ref()), Some("US"));
    }
    #[test]
    #[cfg(feature = "hs-common")]
    #[allow(deprecated)]
    fn hs_dirs_selection() {
        use tor_basic_utils::test_rng::testing_rng;
        const HSDIR_SPREAD_STORE: i32 = 6;
        const HSDIR_SPREAD_FETCH: i32 = 2;
        const PARAMS: [(&str, i32); 2] = [
            ("hsdir_spread_store", HSDIR_SPREAD_STORE),
            ("hsdir_spread_fetch", HSDIR_SPREAD_FETCH),
        ];
        let netdir: Arc<NetDir> =
            crate::testnet::construct_custom_netdir_with_params(|_, _, _| {}, PARAMS, None)
                .unwrap()
                .unwrap_if_sufficient()
                .unwrap()
                .into();
        let hsid = dummy_hs_blind_id();
        const OP_RELAY_COUNT: &[(HsDirOp, usize)] = &[
            // We can't upload to (hsdir_n_replicas * hsdir_spread_store) = 12, relays because there
            // are only 10 relays with the HsDir flag in the consensus.
            #[cfg(feature = "hs-service")]
            (HsDirOp::Upload, 10),
            (HsDirOp::Download, 4),
        ];
        for (op, relay_count) in OP_RELAY_COUNT {
            let relays = netdir.hs_dirs(&hsid, *op, &mut testing_rng());
            assert_eq!(relays.len(), *relay_count);
            // There should be no duplicates (the filtering function passed to
            // HsDirRing::ring_items_at() ensures the relays that are already in use for
            // lower-numbered replicas aren't considered a second time for a higher-numbered
            // replica).
            let unique = relays
                .iter()
                .map(|relay| relay.ed_identity())
                .collect::<HashSet<_>>();
            assert_eq!(unique.len(), relays.len());
        }
        // TODO: come up with a test that checks that HsDirRing::ring_items_at() skips over the
        // expected relays.
        //
        // For example, let's say we have the following hsdir ring:
        //
        //         A  -  B
        //        /       \
        //       F         C
        //        \       /
        //         E  -  D
        //
        // Let's also assume that:
        //
        //   * hsdir_spread_store = 3
        //   * the ordering of the relays on the ring is [A, B, C, D, E, F]
        //
        // If we use relays [A, B, C] for replica 1, and hs_index(2) = E, then replica 2 _must_ get
        // relays [E, F, D]. We should have a test that checks this.
    }
    #[test]
    fn zero_weights() {
        // Here we check the behavior of IndexedRandom::{choose_weighted, choose_multiple_weighted}
        // in the presence of items whose weight is 0.
        //
        // We think that the behavior is:
        //   - An item with weight 0 is never returned.
        //   - If all items have weight 0, choose_weighted returns an error.
        //   - If all items have weight 0, choose_multiple_weighted returns an empty list.
        //   - If we request n items from choose_multiple_weighted,
        //     but only m<n items have nonzero weight, we return all m of those items.
        //   - if the request for n items can't be completely satisfied with n items of weight >= 0,
        //     we get InsufficientNonZero.
        let items = vec![1, 2, 3];
        let mut rng = testing_rng();
        let a = items.choose_weighted(&mut rng, |_| 0);
        assert!(matches!(a, Err(WeightError::InsufficientNonZero)));
        let x = items.choose_multiple_weighted(&mut rng, 2, |_| 0);
        let xs: Vec<_> = x.unwrap().collect();
        assert!(xs.is_empty());
        let only_one = |n: &i32| if *n == 1 { 1 } else { 0 };
        let x = items.choose_multiple_weighted(&mut rng, 2, only_one);
        let xs: Vec<_> = x.unwrap().collect();
        assert_eq!(&xs[..], &[&1]);
        for _ in 0..100 {
            let a = items.choose_weighted(&mut rng, only_one);
            assert_eq!(a.unwrap(), &1);
            let x = items
                .choose_multiple_weighted(&mut rng, 1, only_one)
                .unwrap()
                .collect::<Vec<_>>();
            assert_eq!(x, vec![&1]);
        }
    }
    #[test]
    fn insufficient_but_nonzero() {
        // Here we check IndexedRandom::choose_multiple_weighted when there no zero values,
        // but there are insufficient values.
        // (If this behavior changes, we need to change our usage.)
        let items = vec![1, 2, 3];
        let mut rng = testing_rng();
        let mut a = items
            .choose_multiple_weighted(&mut rng, 10, |_| 1)
            .unwrap()
            .copied()
            .collect::<Vec<_>>();
        a.sort();
        assert_eq!(a, items);
    }
}