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
#![allow(mismatched_lifetime_syntaxes)] // temporary workaround for arti#2060
45
//! <!-- @@ end lint list maintained by maint/add_warning @@ -->
46

            
47
pub mod details;
48
mod err;
49
#[cfg(feature = "hs-common")]
50
mod hsdir_params;
51
#[cfg(feature = "hs-common")]
52
mod hsdir_ring;
53
pub mod params;
54
mod weight;
55

            
56
#[cfg(any(test, feature = "testing"))]
57
pub mod testnet;
58
#[cfg(feature = "testing")]
59
pub mod testprovider;
60

            
61
use async_trait::async_trait;
62
#[cfg(feature = "hs-service")]
63
use itertools::chain;
64
use static_assertions::const_assert;
65
use tor_error::warn_report;
66
use tor_linkspec::{
67
    ChanTarget, DirectChanMethodsHelper, HasAddrs, HasRelayIds, RelayIdRef, RelayIdType,
68
};
69
use tor_llcrypto as ll;
70
use tor_llcrypto::pk::{ed25519::Ed25519Identity, rsa::RsaIdentity};
71
use tor_netdoc::doc::microdesc::{MdDigest, Microdesc};
72
use tor_netdoc::doc::netstatus::{self, MdConsensus, MdConsensusRouterStatus, RouterStatus};
73
#[cfg(feature = "hs-common")]
74
use {hsdir_ring::HsDirRing, std::iter};
75

            
76
use derive_more::{From, Into};
77
use futures::{StreamExt, stream::BoxStream};
78
use num_enum::{IntoPrimitive, TryFromPrimitive};
79
use rand::seq::{IndexedRandom as _, SliceRandom as _, WeightError};
80
use serde::Deserialize;
81
use std::collections::HashMap;
82
use std::net::IpAddr;
83
use std::ops::Deref;
84
use std::sync::Arc;
85
use std::time::SystemTime;
86
use strum::{EnumCount, EnumIter};
87
use tracing::warn;
88
use typed_index_collections::{TiSlice, TiVec};
89

            
90
#[cfg(feature = "hs-common")]
91
use {
92
    itertools::Itertools,
93
    std::collections::HashSet,
94
    tor_error::{Bug, internal},
95
    tor_hscrypto::{pk::HsBlindId, time::TimePeriod},
96
};
97

            
98
pub use err::Error;
99
pub use weight::WeightRole;
100
/// A Result using the Error type from the tor-netdir crate
101
pub type Result<T> = std::result::Result<T, Error>;
102

            
103
#[cfg(feature = "hs-common")]
104
pub use err::OnionDirLookupError;
105

            
106
use params::NetParameters;
107
#[cfg(feature = "geoip")]
108
use tor_geoip::{CountryCode, GeoipDb, HasCountryCode};
109

            
110
#[cfg(feature = "hs-common")]
111
#[cfg_attr(docsrs, doc(cfg(feature = "hs-common")))]
112
pub use hsdir_params::HsDirParams;
113

            
114
/// Index into the consensus relays
115
///
116
/// This is an index into the list of relays returned by
117
/// [`.c_relays()`](ConsensusRelays::c_relays)
118
/// (on the corresponding consensus or netdir).
119
///
120
/// This is just a `usize` inside, but using a newtype prevents getting a relay index
121
/// confused with other kinds of slice indices or counts.
122
///
123
/// If you are in a part of the code which needs to work with multiple consensuses,
124
/// the typechecking cannot tell if you try to index into the wrong consensus.
125
#[derive(Debug, From, Into, Copy, Clone, Ord, PartialOrd, Eq, PartialEq, Hash)]
126
pub(crate) struct RouterStatusIdx(usize);
127

            
128
/// Extension trait to provide index-type-safe `.c_relays()` method
129
//
130
// TODO: Really it would be better to have MdConsensns::relays() return TiSlice,
131
// but that would be an API break there.
132
pub(crate) trait ConsensusRelays {
133
    /// Obtain the list of relays in the consensus
134
    //
135
    fn c_relays(&self) -> &TiSlice<RouterStatusIdx, MdConsensusRouterStatus>;
136
}
137
impl ConsensusRelays for MdConsensus {
138
35306667
    fn c_relays(&self) -> &TiSlice<RouterStatusIdx, MdConsensusRouterStatus> {
139
35306667
        TiSlice::from_ref(MdConsensus::relays(self))
140
35306667
    }
141
}
142
impl ConsensusRelays for NetDir {
143
35230536
    fn c_relays(&self) -> &TiSlice<RouterStatusIdx, MdConsensusRouterStatus> {
144
35230536
        self.consensus.c_relays()
145
35230536
    }
146
}
147

            
148
/// Configuration for determining when two relays have addresses "too close" in
149
/// the network.
150
///
151
/// Used by `Relay::low_level_details().in_same_subnet()`.
152
#[derive(Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
153
#[serde(deny_unknown_fields)]
154
pub struct SubnetConfig {
155
    /// Consider IPv4 nodes in the same /x to be the same family.
156
    ///
157
    /// If this value is 0, all nodes with IPv4 addresses will be in the
158
    /// same family.  If this value is above 32, then no nodes will be
159
    /// placed im the same family based on their IPv4 addresses.
160
    subnets_family_v4: u8,
161
    /// Consider IPv6 nodes in the same /x to be the same family.
162
    ///
163
    /// If this value is 0, all nodes with IPv6 addresses will be in the
164
    /// same family.  If this value is above 128, then no nodes will be
165
    /// placed im the same family based on their IPv6 addresses.
166
    subnets_family_v6: u8,
167
}
168

            
169
impl Default for SubnetConfig {
170
364099
    fn default() -> Self {
171
364099
        Self::new(16, 32)
172
364099
    }
173
}
174

            
175
impl SubnetConfig {
176
    /// Construct a new SubnetConfig from a pair of bit prefix lengths.
177
    ///
178
    /// The values are clamped to the appropriate ranges if they are
179
    /// out-of-bounds.
180
1111774
    pub fn new(subnets_family_v4: u8, subnets_family_v6: u8) -> Self {
181
1111774
        Self {
182
1111774
            subnets_family_v4,
183
1111774
            subnets_family_v6,
184
1111774
        }
185
1111774
    }
186

            
187
    /// Construct a new SubnetConfig such that addresses are not in the same
188
    /// family with anything--not even with themselves.
189
2246312
    pub fn no_addresses_match() -> SubnetConfig {
190
2246312
        SubnetConfig {
191
2246312
            subnets_family_v4: 33,
192
2246312
            subnets_family_v6: 129,
193
2246312
        }
194
2246312
    }
195

            
196
    /// Return true if the two addresses in the same subnet, according to this
197
    /// configuration.
198
31295003
    pub fn addrs_in_same_subnet(&self, a: &IpAddr, b: &IpAddr) -> bool {
199
31295003
        match (a, b) {
200
31294971
            (IpAddr::V4(a), IpAddr::V4(b)) => {
201
31294971
                let bits = self.subnets_family_v4;
202
31294971
                if bits > 32 {
203
6668
                    return false;
204
31288303
                }
205
31288303
                let a = u32::from_be_bytes(a.octets());
206
31288303
                let b = u32::from_be_bytes(b.octets());
207
31288303
                (a >> (32 - bits)) == (b >> (32 - bits))
208
            }
209
20
            (IpAddr::V6(a), IpAddr::V6(b)) => {
210
20
                let bits = self.subnets_family_v6;
211
20
                if bits > 128 {
212
4
                    return false;
213
16
                }
214
16
                let a = u128::from_be_bytes(a.octets());
215
16
                let b = u128::from_be_bytes(b.octets());
216
16
                (a >> (128 - bits)) == (b >> (128 - bits))
217
            }
218
12
            _ => false,
219
        }
220
31295003
    }
221

            
222
    /// Return true if any of the addresses in `a` shares a subnet with any of
223
    /// the addresses in `b`, according to this configuration.
224
31294953
    pub fn any_addrs_in_same_subnet<T, U>(&self, a: &T, b: &U) -> bool
225
31294953
    where
226
31294953
        T: tor_linkspec::HasAddrs,
227
31294953
        U: tor_linkspec::HasAddrs,
228
31294953
    {
229
31294959
        a.addrs().iter().any(|aa| {
230
31294959
            b.addrs()
231
31294959
                .iter()
232
31294967
                .any(|bb| self.addrs_in_same_subnet(&aa.ip(), &bb.ip()))
233
31294959
        })
234
31294953
    }
235

            
236
    /// Return a new subnet configuration that is the union of `self` and
237
    /// `other`.
238
    ///
239
    /// That is, return a subnet configuration that puts all addresses in the
240
    /// same subnet if and only if at least one of `self` and `other` would put
241
    /// them in the same subnet.
242
1120506
    pub fn union(&self, other: &Self) -> Self {
243
        use std::cmp::min;
244
1120506
        Self {
245
1120506
            subnets_family_v4: min(self.subnets_family_v4, other.subnets_family_v4),
246
1120506
            subnets_family_v6: min(self.subnets_family_v6, other.subnets_family_v6),
247
1120506
        }
248
1120506
    }
249
}
250

            
251
/// Configuration for which listed family information to use when deciding
252
/// whether relays belong to the same family.
253
///
254
/// Derived from network parameters.
255
#[derive(Clone, Copy, Debug)]
256
pub struct FamilyRules {
257
    /// If true, we use family information from lists of family members.
258
    use_family_lists: bool,
259
    /// If true, we use family information from lists of family IDs and from family certs.
260
    use_family_ids: bool,
261
}
262

            
263
impl<'a> From<&'a NetParameters> for FamilyRules {
264
373770
    fn from(params: &'a NetParameters) -> Self {
265
373770
        FamilyRules {
266
373770
            use_family_lists: bool::from(params.use_family_lists),
267
373770
            use_family_ids: bool::from(params.use_family_ids),
268
373770
        }
269
373770
    }
270
}
271

            
272
impl FamilyRules {
273
    /// Return a `FamilyRules` that will use all recognized kinds of family information.
274
587
    pub fn all_family_info() -> Self {
275
587
        Self {
276
587
            use_family_lists: true,
277
587
            use_family_ids: true,
278
587
        }
279
587
    }
280

            
281
    /// Return a `FamilyRules` that will ignore all family information declared by relays.
282
2246400
    pub fn ignore_declared_families() -> Self {
283
2246400
        Self {
284
2246400
            use_family_lists: false,
285
2246400
            use_family_ids: false,
286
2246400
        }
287
2246400
    }
288

            
289
    /// Configure this `FamilyRules` to use (or not use) family information from
290
    /// lists of family members.
291
45
    pub fn use_family_lists(&mut self, val: bool) -> &mut Self {
292
45
        self.use_family_lists = val;
293
45
        self
294
45
    }
295

            
296
    /// Configure this `FamilyRules` to use (or not use) family information from
297
    /// family IDs and family certs.
298
45
    pub fn use_family_ids(&mut self, val: bool) -> &mut Self {
299
45
        self.use_family_ids = val;
300
45
        self
301
45
    }
302

            
303
    /// Return a `FamilyRules` that will look at every source of information
304
    /// requested by `self` or by `other`.
305
1120500
    pub fn union(&self, other: &Self) -> Self {
306
1120500
        Self {
307
1120500
            use_family_lists: self.use_family_lists || other.use_family_lists,
308
1120500
            use_family_ids: self.use_family_ids || other.use_family_ids,
309
        }
310
1120500
    }
311
}
312

            
313
/// An opaque type representing the weight with which a relay or set of
314
/// relays will be selected for a given role.
315
///
316
/// Most users should ignore this type, and just use pick_relay instead.
317
#[derive(
318
    Copy,
319
    Clone,
320
    Debug,
321
    derive_more::Add,
322
    derive_more::Sum,
323
    derive_more::AddAssign,
324
    Eq,
325
    PartialEq,
326
    Ord,
327
    PartialOrd,
328
)]
329
pub struct RelayWeight(u64);
330

            
331
impl RelayWeight {
332
    /// Try to divide this weight by `rhs`.
333
    ///
334
    /// Return a ratio on success, or None on division-by-zero.
335
2299
    pub fn checked_div(&self, rhs: RelayWeight) -> Option<f64> {
336
2299
        if rhs.0 == 0 {
337
2
            None
338
        } else {
339
2297
            Some((self.0 as f64) / (rhs.0 as f64))
340
        }
341
2299
    }
342

            
343
    /// Compute a ratio `frac` of this weight.
344
    ///
345
    /// Return None if frac is less than zero, since negative weights
346
    /// are impossible.
347
24349
    pub fn ratio(&self, frac: f64) -> Option<RelayWeight> {
348
24349
        let product = (self.0 as f64) * frac;
349
24349
        if product >= 0.0 && product.is_finite() {
350
24347
            Some(RelayWeight(product as u64))
351
        } else {
352
2
            None
353
        }
354
24349
    }
355
}
356

            
357
impl From<u64> for RelayWeight {
358
4590
    fn from(val: u64) -> Self {
359
4590
        RelayWeight(val)
360
4590
    }
361
}
362

            
363
/// An operation for which we might be requesting a hidden service directory.
364
#[derive(Copy, Clone, Debug, PartialEq)]
365
// TODO: make this pub(crate) once NetDir::hs_dirs is removed
366
#[non_exhaustive]
367
pub enum HsDirOp {
368
    /// Uploading an onion service descriptor.
369
    #[cfg(feature = "hs-service")]
370
    Upload,
371
    /// Downloading an onion service descriptor.
372
    Download,
373
}
374

            
375
/// A view of the Tor directory, suitable for use in building circuits.
376
///
377
/// Abstractly, a [`NetDir`] is a set of usable public [`Relay`]s, each of which
378
/// has its own properties, identity, and correct weighted probability for use
379
/// under different circumstances.
380
///
381
/// A [`NetDir`] is constructed by making a [`PartialNetDir`] from a consensus
382
/// document, and then adding enough microdescriptors to that `PartialNetDir` so
383
/// that it can be used to build paths. (Thus, if you have a NetDir, it is
384
/// definitely adequate to build paths.)
385
///
386
/// # "Usable" relays
387
///
388
/// Many methods on NetDir are defined in terms of <a name="usable">"Usable"</a> relays.  Unless
389
/// otherwise stated, a relay is "usable" if it is listed in the consensus,
390
/// if we have full directory information for that relay (including a
391
/// microdescriptor), and if that relay does not have any flags indicating that
392
/// we should never use it. (Currently, `NoEdConsensus` is the only such flag.)
393
///
394
/// # Limitations
395
///
396
/// The current NetDir implementation assumes fairly strongly that every relay
397
/// has an Ed25519 identity and an RSA identity, that the consensus is indexed
398
/// by RSA identities, and that the Ed25519 identities are stored in
399
/// microdescriptors.
400
///
401
/// If these assumptions someday change, then we'll have to revise the
402
/// implementation.
403
#[derive(Debug, Clone)]
404
pub struct NetDir {
405
    /// A microdescriptor consensus that lists the members of the network,
406
    /// and maps each one to a 'microdescriptor' that has more information
407
    /// about it
408
    consensus: Arc<MdConsensus>,
409
    /// A map from keys to integer values, distributed in the consensus,
410
    /// and clamped to certain defaults.
411
    params: NetParameters,
412
    /// Map from routerstatus index, to that routerstatus's microdescriptor (if we have one.)
413
    mds: TiVec<RouterStatusIdx, Option<Arc<Microdesc>>>,
414
    /// Map from SHA256 of _missing_ microdescriptors to the index of their
415
    /// corresponding routerstatus.
416
    rsidx_by_missing: HashMap<MdDigest, RouterStatusIdx>,
417
    /// Map from ed25519 identity to index of the routerstatus.
418
    ///
419
    /// Note that we don't know the ed25519 identity of a relay until
420
    /// we get the microdescriptor for it, so this won't be filled in
421
    /// until we get the microdescriptors.
422
    ///
423
    /// # Implementation note
424
    ///
425
    /// For this field, and for `rsidx_by_rsa`,
426
    /// it might be cool to have references instead.
427
    /// But that would make this into a self-referential structure,
428
    /// which isn't possible in safe rust.
429
    rsidx_by_ed: HashMap<Ed25519Identity, RouterStatusIdx>,
430
    /// Map from RSA identity to index of the routerstatus.
431
    ///
432
    /// This is constructed at the same time as the NetDir object, so it
433
    /// can be immutable.
434
    rsidx_by_rsa: Arc<HashMap<RsaIdentity, RouterStatusIdx>>,
435

            
436
    /// Hash ring(s) describing the onion service directory.
437
    ///
438
    /// This is empty in a PartialNetDir, and is filled in before the NetDir is
439
    /// built.
440
    //
441
    // TODO hs: It is ugly to have this exist in a partially constructed state
442
    // in a PartialNetDir.
443
    // Ideally, a PartialNetDir would contain only an HsDirs<HsDirParams>,
444
    // or perhaps nothing at all, here.
445
    #[cfg(feature = "hs-common")]
446
    hsdir_rings: Arc<HsDirs<HsDirRing>>,
447

            
448
    /// Weight values to apply to a given relay when deciding how frequently
449
    /// to choose it for a given role.
450
    weights: weight::WeightSet,
451

            
452
    #[cfg(feature = "geoip")]
453
    /// Country codes for each router in our consensus.
454
    ///
455
    /// This is indexed by the `RouterStatusIdx` (i.e. a router idx of zero has
456
    /// the country code at position zero in this array).
457
    country_codes: Vec<Option<CountryCode>>,
458
}
459

            
460
/// Collection of hidden service directories (or parameters for them)
461
///
462
/// In [`NetDir`] this is used to store the actual hash rings.
463
/// (But, in a NetDir in a [`PartialNetDir`], it contains [`HsDirRing`]s
464
/// where only the `params` are populated, and the `ring` is empty.)
465
///
466
/// This same generic type is used as the return type from
467
/// [`HsDirParams::compute`](HsDirParams::compute),
468
/// where it contains the *parameters* for the primary and secondary rings.
469
#[derive(Debug, Clone)]
470
#[cfg(feature = "hs-common")]
471
pub(crate) struct HsDirs<D> {
472
    /// The current ring
473
    ///
474
    /// It corresponds to the time period containing the `valid-after` time in
475
    /// the consensus. Its SRV is whatever SRV was most current at the time when
476
    /// that time period began.
477
    ///
478
    /// This is the hash ring that we should use whenever we are fetching an
479
    /// onion service descriptor.
480
    current: D,
481

            
482
    /// Secondary rings (based on the parameters for the previous and next time periods)
483
    ///
484
    /// Onion services upload to positions on these ring as well, based on how
485
    /// far into the current time period this directory is, so that
486
    /// not-synchronized clients can still find their descriptor.
487
    ///
488
    /// Note that with the current (2023) network parameters, with
489
    /// `hsdir_interval = SRV lifetime = 24 hours` at most one of these
490
    /// secondary rings will be active at a time.  We have two here in order
491
    /// to conform with a more flexible regime in proposal 342.
492
    //
493
    // TODO: hs clients never need this; so I've made it not-present for them.
494
    // But does that risk too much with respect to side channels?
495
    //
496
    // TODO: Perhaps we should refactor this so that it is clear that these
497
    // are immutable?  On the other hand, the documentation for this type
498
    // declares that it is immutable, so we are likely okay.
499
    //
500
    // TODO: this `Vec` is only ever 0,1,2 elements.
501
    // Maybe it should be an ArrayVec or something.
502
    #[cfg(feature = "hs-service")]
503
    secondary: Vec<D>,
504
}
505

            
506
#[cfg(feature = "hs-common")]
507
impl<D> HsDirs<D> {
508
    /// Convert an `HsDirs<D>` to `HsDirs<D2>` by mapping each contained `D`
509
15096
    pub(crate) fn map<D2>(self, mut f: impl FnMut(D) -> D2) -> HsDirs<D2> {
510
15096
        HsDirs {
511
15096
            current: f(self.current),
512
15096
            #[cfg(feature = "hs-service")]
513
15096
            secondary: self.secondary.into_iter().map(f).collect(),
514
15096
        }
515
15096
    }
516

            
517
    /// Iterate over some of the contained hsdirs, according to `secondary`
518
    ///
519
    /// The current ring is always included.
520
    /// Secondary rings are included iff `secondary` and the `hs-service` feature is enabled.
521
184
    fn iter_filter_secondary(&self, secondary: bool) -> impl Iterator<Item = &D> {
522
184
        let i = iter::once(&self.current);
523
184

            
524
184
        // With "hs-service" disabled, there are no secondary rings,
525
184
        // so we don't care.
526
184
        let _ = secondary;
527
184

            
528
184
        #[cfg(feature = "hs-service")]
529
364
        let i = chain!(i, self.secondary.iter().filter(move |_| secondary));
530
184

            
531
184
        i
532
184
    }
533

            
534
    /// Iterate over all the contained hsdirs
535
180
    pub(crate) fn iter(&self) -> impl Iterator<Item = &D> {
536
180
        self.iter_filter_secondary(true)
537
180
    }
538

            
539
    /// Iterate over the hsdirs relevant for `op`
540
4
    pub(crate) fn iter_for_op(&self, op: HsDirOp) -> impl Iterator<Item = &D> {
541
4
        self.iter_filter_secondary(match op {
542
            #[cfg(feature = "hs-service")]
543
2
            HsDirOp::Upload => true,
544
2
            HsDirOp::Download => false,
545
        })
546
4
    }
547
}
548

            
549
/// An event that a [`NetDirProvider`] can broadcast to indicate that a change in
550
/// the status of its directory.
551
#[derive(
552
3238
    Debug, Clone, Copy, PartialEq, Eq, EnumIter, EnumCount, IntoPrimitive, TryFromPrimitive,
553
)]
554
#[non_exhaustive]
555
#[repr(u16)]
556
pub enum DirEvent {
557
    /// A new consensus has been received, and has enough information to be
558
    /// used.
559
    ///
560
    /// This event is also broadcast when a new set of consensus parameters is
561
    /// available, even if that set of parameters comes from a configuration
562
    /// change rather than from the latest consensus.
563
    NewConsensus,
564

            
565
    /// New descriptors have been received for the current consensus.
566
    ///
567
    /// (This event is _not_ broadcast when receiving new descriptors for a
568
    /// consensus which is not yet ready to replace the current consensus.)
569
    NewDescriptors,
570

            
571
    /// We have received updated recommendations and requirements
572
    /// for which subprotocols we should have to use the network.
573
    NewProtocolRecommendation,
574
}
575

            
576
/// The network directory provider is shutting down without giving us the
577
/// netdir we asked for.
578
#[derive(Clone, Copy, Debug, thiserror::Error)]
579
#[error("Network directory provider is shutting down")]
580
#[non_exhaustive]
581
pub struct NetdirProviderShutdown;
582

            
583
impl tor_error::HasKind for NetdirProviderShutdown {
584
    fn kind(&self) -> tor_error::ErrorKind {
585
        tor_error::ErrorKind::ArtiShuttingDown
586
    }
587
}
588

            
589
/// How "timely" must a network directory be?
590
///
591
/// This enum is used as an argument when requesting a [`NetDir`] object from
592
/// [`NetDirProvider`] and other APIs, to specify how recent the information
593
/// must be in order to be useful.
594
#[derive(Copy, Clone, Eq, PartialEq, Debug)]
595
#[allow(clippy::exhaustive_enums)]
596
pub enum Timeliness {
597
    /// The network directory must be strictly timely.
598
    ///
599
    /// That is, it must be based on a consensus that valid right now, with no
600
    /// tolerance for skew or consensus problems.
601
    ///
602
    /// Avoid using this option if you could use [`Timeliness::Timely`] instead.
603
    Strict,
604
    /// The network directory must be roughly timely.
605
    ///
606
    /// This is, it must be be based on a consensus that is not _too_ far in the
607
    /// future, and not _too_ far in the past.
608
    ///
609
    /// (The tolerances for "too far" will depend on configuration.)
610
    ///
611
    /// This is almost always the option that you want to use.
612
    Timely,
613
    /// Any network directory is permissible, regardless of how untimely.
614
    ///
615
    /// Avoid using this option if you could use [`Timeliness::Timely`] instead.
616
    Unchecked,
617
}
618

            
619
/// An object that can provide [`NetDir`]s, as well as inform consumers when
620
/// they might have changed.
621
///
622
/// It is the responsibility of the implementor of `NetDirProvider`
623
/// to try to obtain an up-to-date `NetDir`,
624
/// and continuously to maintain and update it.
625
///
626
/// In usual configurations, Arti uses `tor_dirmgr::DirMgr`
627
/// as its `NetDirProvider`.
628
#[async_trait]
629
pub trait NetDirProvider: UpcastArcNetDirProvider + Send + Sync {
630
    /// Return a network directory that's live according to the provided
631
    /// `timeliness`.
632
    fn netdir(&self, timeliness: Timeliness) -> Result<Arc<NetDir>>;
633

            
634
    /// Return a reasonable netdir for general usage.
635
    ///
636
    /// This is an alias for
637
    /// [`NetDirProvider::netdir`]`(`[`Timeliness::Timely`]`)`.
638
680
    fn timely_netdir(&self) -> Result<Arc<NetDir>> {
639
680
        self.netdir(Timeliness::Timely)
640
680
    }
641

            
642
    /// Return a new asynchronous stream that will receive notification
643
    /// whenever the consensus has changed.
644
    ///
645
    /// Multiple events may be batched up into a single item: each time
646
    /// this stream yields an event, all you can assume is that the event has
647
    /// occurred at least once.
648
    fn events(&self) -> BoxStream<'static, DirEvent>;
649

            
650
    /// Return the latest network parameters.
651
    ///
652
    /// If we have no directory, return a reasonable set of defaults.
653
    fn params(&self) -> Arc<dyn AsRef<NetParameters>>;
654

            
655
    /// Get a NetDir from `provider`, waiting until one exists.
656
    async fn wait_for_netdir(
657
        &self,
658
        timeliness: Timeliness,
659
24
    ) -> std::result::Result<Arc<NetDir>, NetdirProviderShutdown> {
660
24
        if let Ok(nd) = self.netdir(timeliness) {
661
24
            return Ok(nd);
662
        }
663

            
664
        let mut stream = self.events();
665
        loop {
666
            // We need to retry `self.netdir()` before waiting for any stream events, to
667
            // avoid deadlock.
668
            //
669
            // We ignore all errors here: they can all potentially be fixed by
670
            // getting a fresh consensus, and they will all get warned about
671
            // by the NetDirProvider itself.
672
            if let Ok(nd) = self.netdir(timeliness) {
673
                return Ok(nd);
674
            }
675
            match stream.next().await {
676
                Some(_) => {}
677
                None => {
678
                    return Err(NetdirProviderShutdown);
679
                }
680
            }
681
        }
682
48
    }
683

            
684
    /// Wait until `provider` lists `target`.
685
    ///
686
    /// NOTE: This might potentially wait indefinitely, if `target` is never actually
687
    /// becomes listed in the directory.  It will exit if the `NetDirProvider` shuts down.
688
    async fn wait_for_netdir_to_list(
689
        &self,
690
        target: &tor_linkspec::RelayIds,
691
        timeliness: Timeliness,
692
    ) -> std::result::Result<(), NetdirProviderShutdown> {
693
        let mut events = self.events();
694
        loop {
695
            // See if the desired relay is in the netdir.
696
            //
697
            // We do this before waiting for any events, to avoid race conditions.
698
            {
699
                let netdir = self.wait_for_netdir(timeliness).await?;
700
                if netdir.ids_listed(target) == Some(true) {
701
                    return Ok(());
702
                }
703
                // If we reach this point, then ids_listed returned `Some(false)`,
704
                // meaning "This relay is definitely not in the current directory";
705
                // or it returned `None`, meaning "waiting for more information
706
                // about this network directory.
707
                // In both cases, it's reasonable to just wait for another netdir
708
                // event and try again.
709
            }
710
            // We didn't find the relay; wait for the provider to have a new netdir
711
            // or more netdir information.
712
            if events.next().await.is_none() {
713
                // The event stream is closed; the provider has shut down.
714
                return Err(NetdirProviderShutdown);
715
            }
716
        }
717
    }
718

            
719
    /// Return the latest set of recommended and required protocols, if there is one.
720
    ///
721
    /// This may be more recent (or more available) than this provider's associated NetDir.
722
    fn protocol_statuses(&self) -> Option<(SystemTime, Arc<netstatus::ProtoStatuses>)>;
723
}
724

            
725
impl<T> NetDirProvider for Arc<T>
726
where
727
    T: NetDirProvider,
728
{
729
50
    fn netdir(&self, timeliness: Timeliness) -> Result<Arc<NetDir>> {
730
50
        self.deref().netdir(timeliness)
731
50
    }
732

            
733
    fn timely_netdir(&self) -> Result<Arc<NetDir>> {
734
        self.deref().timely_netdir()
735
    }
736

            
737
72
    fn events(&self) -> BoxStream<'static, DirEvent> {
738
72
        self.deref().events()
739
72
    }
740

            
741
12
    fn params(&self) -> Arc<dyn AsRef<NetParameters>> {
742
12
        self.deref().params()
743
12
    }
744

            
745
8
    fn protocol_statuses(&self) -> Option<(SystemTime, Arc<netstatus::ProtoStatuses>)> {
746
8
        self.deref().protocol_statuses()
747
8
    }
748
}
749

            
750
/// Helper trait: allows any `Arc<X>` to be upcast to a `Arc<dyn
751
/// NetDirProvider>` if X is an implementation or supertrait of NetDirProvider.
752
///
753
/// This trait exists to work around a limitation in rust: when trait upcasting
754
/// coercion is stable, this will be unnecessary.
755
///
756
/// The Rust tracking issue is <https://github.com/rust-lang/rust/issues/65991>.
757
pub trait UpcastArcNetDirProvider {
758
    /// Return a view of this object as an `Arc<dyn NetDirProvider>`
759
    fn upcast_arc<'a>(self: Arc<Self>) -> Arc<dyn NetDirProvider + 'a>
760
    where
761
        Self: 'a;
762
}
763

            
764
impl<T> UpcastArcNetDirProvider for T
765
where
766
    T: NetDirProvider + Sized,
767
{
768
48
    fn upcast_arc<'a>(self: Arc<Self>) -> Arc<dyn NetDirProvider + 'a>
769
48
    where
770
48
        Self: 'a,
771
48
    {
772
48
        self
773
48
    }
774
}
775

            
776
impl AsRef<NetParameters> for NetDir {
777
855
    fn as_ref(&self) -> &NetParameters {
778
855
        self.params()
779
855
    }
780
}
781

            
782
/// A partially build NetDir -- it can't be unwrapped until it has
783
/// enough information to build safe paths.
784
#[derive(Debug, Clone)]
785
pub struct PartialNetDir {
786
    /// The netdir that's under construction.
787
    netdir: NetDir,
788

            
789
    /// The previous netdir, if we had one
790
    ///
791
    /// Used as a cache, so we can reuse information
792
    #[cfg(feature = "hs-common")]
793
    prev_netdir: Option<Arc<NetDir>>,
794
}
795

            
796
/// A view of a relay on the Tor network, suitable for building circuits.
797
// TODO: This should probably be a more specific struct, with a trait
798
// that implements it.
799
#[derive(Clone)]
800
pub struct Relay<'a> {
801
    /// A router descriptor for this relay.
802
    rs: &'a netstatus::MdConsensusRouterStatus,
803
    /// A microdescriptor for this relay.
804
    md: &'a Microdesc,
805
    /// The country code this relay is in, if we know one.
806
    #[cfg(feature = "geoip")]
807
    cc: Option<CountryCode>,
808
}
809

            
810
/// A relay that we haven't checked for validity or usability in
811
/// routing.
812
#[derive(Debug)]
813
pub struct UncheckedRelay<'a> {
814
    /// A router descriptor for this relay.
815
    rs: &'a netstatus::MdConsensusRouterStatus,
816
    /// A microdescriptor for this relay, if there is one.
817
    md: Option<&'a Microdesc>,
818
    /// The country code this relay is in, if we know one.
819
    #[cfg(feature = "geoip")]
820
    cc: Option<CountryCode>,
821
}
822

            
823
/// A partial or full network directory that we can download
824
/// microdescriptors for.
825
pub trait MdReceiver {
826
    /// Return an iterator over the digests for all of the microdescriptors
827
    /// that this netdir is missing.
828
    fn missing_microdescs(&self) -> Box<dyn Iterator<Item = &MdDigest> + '_>;
829
    /// Add a microdescriptor to this netdir, if it was wanted.
830
    ///
831
    /// Return true if it was indeed wanted.
832
    fn add_microdesc(&mut self, md: Microdesc) -> bool;
833
    /// Return the number of missing microdescriptors.
834
    fn n_missing(&self) -> usize;
835
}
836

            
837
impl PartialNetDir {
838
    /// Create a new PartialNetDir with a given consensus, and no
839
    /// microdescriptors loaded.
840
    ///
841
    /// If `replacement_params` is provided, override network parameters from
842
    /// the consensus with those from `replacement_params`.
843
7461
    pub fn new(
844
7461
        consensus: MdConsensus,
845
7461
        replacement_params: Option<&netstatus::NetParams<i32>>,
846
7461
    ) -> Self {
847
7461
        Self::new_inner(
848
7461
            consensus,
849
7461
            replacement_params,
850
7461
            #[cfg(feature = "geoip")]
851
7461
            None,
852
7461
        )
853
7461
    }
854

            
855
    /// Create a new PartialNetDir with GeoIP support.
856
    ///
857
    /// This does the same thing as `new()`, except the provided GeoIP database is used to add
858
    /// country codes to relays.
859
    #[cfg(feature = "geoip")]
860
    #[cfg_attr(docsrs, doc(cfg(feature = "geoip")))]
861
137
    pub fn new_with_geoip(
862
137
        consensus: MdConsensus,
863
137
        replacement_params: Option<&netstatus::NetParams<i32>>,
864
137
        geoip_db: &GeoipDb,
865
137
    ) -> Self {
866
137
        Self::new_inner(consensus, replacement_params, Some(geoip_db))
867
137
    }
868

            
869
    /// Implementation of the `new()` functions.
870
7598
    fn new_inner(
871
7598
        consensus: MdConsensus,
872
7598
        replacement_params: Option<&netstatus::NetParams<i32>>,
873
7598
        #[cfg(feature = "geoip")] geoip_db: Option<&GeoipDb>,
874
7598
    ) -> Self {
875
7598
        let mut params = NetParameters::default();
876
7598

            
877
7598
        // (We ignore unrecognized options here, since they come from
878
7598
        // the consensus, and we don't expect to recognize everything
879
7598
        // there.)
880
7598
        let _ = params.saturating_update(consensus.params().iter());
881

            
882
        // Now see if the user has any parameters to override.
883
        // (We have to do this now, or else changes won't be reflected in our
884
        // weights.)
885
7598
        if let Some(replacement) = replacement_params {
886
7586
            for u in params.saturating_update(replacement.iter()) {
887
2
                warn!("Unrecognized option: override_net_params.{}", u);
888
            }
889
12
        }
890

            
891
        // Compute the weights we'll want to use for these relays.
892
7598
        let weights = weight::WeightSet::from_consensus(&consensus, &params);
893
7598

            
894
7598
        let n_relays = consensus.c_relays().len();
895
7598

            
896
7598
        let rsidx_by_missing = consensus
897
7598
            .c_relays()
898
7598
            .iter_enumerated()
899
283632
            .map(|(rsidx, rs)| (*rs.md_digest(), rsidx))
900
7598
            .collect();
901
7598

            
902
7598
        let rsidx_by_rsa = consensus
903
7598
            .c_relays()
904
7598
            .iter_enumerated()
905
283632
            .map(|(rsidx, rs)| (*rs.rsa_identity(), rsidx))
906
7598
            .collect();
907

            
908
        #[cfg(feature = "geoip")]
909
7598
        let country_codes = if let Some(db) = geoip_db {
910
137
            consensus
911
137
                .c_relays()
912
137
                .iter()
913
714
                .map(|rs| {
914
990
                    db.lookup_country_code_multi(rs.addrs().iter().map(|x| x.ip()))
915
710
                        .cloned()
916
714
                })
917
137
                .collect()
918
        } else {
919
7461
            Default::default()
920
        };
921

            
922
        #[cfg(feature = "hs-common")]
923
7598
        let hsdir_rings = Arc::new({
924
7598
            let params = HsDirParams::compute(&consensus, &params).expect("Invalid consensus!");
925
7598
            // TODO: It's a bit ugly to use expect above, but this function does
926
7598
            // not return a Result. On the other hand, the error conditions under which
927
7598
            // HsDirParams::compute can return Err are _very_ narrow and hard to
928
7598
            // hit; see documentation in that function.  As such, we probably
929
7598
            // don't need to have this return a Result.
930
7598

            
931
7598
            params.map(HsDirRing::empty_from_params)
932
7598
        });
933
7598

            
934
7598
        let netdir = NetDir {
935
7598
            consensus: Arc::new(consensus),
936
7598
            params,
937
7598
            mds: vec![None; n_relays].into(),
938
7598
            rsidx_by_missing,
939
7598
            rsidx_by_rsa: Arc::new(rsidx_by_rsa),
940
7598
            rsidx_by_ed: HashMap::with_capacity(n_relays),
941
7598
            #[cfg(feature = "hs-common")]
942
7598
            hsdir_rings,
943
7598
            weights,
944
7598
            #[cfg(feature = "geoip")]
945
7598
            country_codes,
946
7598
        };
947
7598

            
948
7598
        PartialNetDir {
949
7598
            netdir,
950
7598
            #[cfg(feature = "hs-common")]
951
7598
            prev_netdir: None,
952
7598
        }
953
7598
    }
954

            
955
    /// Return the declared lifetime of this PartialNetDir.
956
2
    pub fn lifetime(&self) -> &netstatus::Lifetime {
957
2
        self.netdir.lifetime()
958
2
    }
959

            
960
    /// Record a previous netdir, which can be used for reusing cached information
961
    //
962
    // Fills in as many missing microdescriptors as possible in this
963
    // netdir, using the microdescriptors from the previous netdir.
964
    //
965
    // With HS enabled, stores the netdir for reuse of relay hash ring index values.
966
    #[allow(clippy::needless_pass_by_value)] // prev might, or might not, be stored
967
2
    pub fn fill_from_previous_netdir(&mut self, prev: Arc<NetDir>) {
968
76
        for md in prev.mds.iter().flatten() {
969
76
            self.netdir.add_arc_microdesc(md.clone());
970
76
        }
971

            
972
        #[cfg(feature = "hs-common")]
973
2
        {
974
2
            self.prev_netdir = Some(prev);
975
2
        }
976
2
    }
977

            
978
    /// Compute the hash ring(s) for this NetDir
979
    #[cfg(feature = "hs-common")]
980
7498
    fn compute_rings(&mut self) {
981
7498
        let params = HsDirParams::compute(&self.netdir.consensus, &self.netdir.params)
982
7498
            .expect("Invalid consensus");
983
7498
        // TODO: see TODO by similar expect in new()
984
7498

            
985
7498
        self.netdir.hsdir_rings =
986
8038
            Arc::new(params.map(|params| {
987
7858
                HsDirRing::compute(params, &self.netdir, self.prev_netdir.as_deref())
988
8038
            }));
989
7498
    }
990

            
991
    /// Return true if this are enough information in this directory
992
    /// to build multihop paths.
993
4
    pub fn have_enough_paths(&self) -> bool {
994
4
        self.netdir.have_enough_paths()
995
4
    }
996
    /// If this directory has enough information to build multihop
997
    /// circuits, return it.
998
7682
    pub fn unwrap_if_sufficient(
999
7682
        #[allow(unused_mut)] mut self,
7682
    ) -> std::result::Result<NetDir, PartialNetDir> {
7682
        if self.netdir.have_enough_paths() {
            #[cfg(feature = "hs-common")]
7498
            self.compute_rings();
7498
            Ok(self.netdir)
        } else {
184
            Err(self)
        }
7682
    }
}
impl MdReceiver for PartialNetDir {
98
    fn missing_microdescs(&self) -> Box<dyn Iterator<Item = &MdDigest> + '_> {
98
        self.netdir.missing_microdescs()
98
    }
298338
    fn add_microdesc(&mut self, md: Microdesc) -> bool {
298338
        self.netdir.add_microdesc(md)
298338
    }
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.
298414
    fn add_arc_microdesc(&mut self, md: Arc<Microdesc>) -> bool {
298414
        if let Some(rsidx) = self.rsidx_by_missing.remove(md.digest()) {
282709
            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.
282709
            self.rsidx_by_ed.insert(*md.ed25519_id(), rsidx);
282709

            
282709
            // Happy path: we did indeed want this one.
282709
            self.mds[rsidx] = Some(md);
282709

            
282709
            // Save some space in the missing-descriptor list.
282709
            if self.rsidx_by_missing.len() < self.rsidx_by_missing.capacity() / 4 {
14239
                self.rsidx_by_missing.shrink_to_fit();
268470
            }
282709
            return true;
15705
        }
15705

            
15705
        // Either we already had it, or we never wanted it at all.
15705
        false
298414
    }
    /// Construct a (possibly invalid) Relay object from a routerstatus and its
    /// index within the consensus.
33449888
    fn relay_from_rs_and_rsidx<'a>(
33449888
        &'a self,
33449888
        rs: &'a netstatus::MdConsensusRouterStatus,
33449888
        rsidx: RouterStatusIdx,
33449888
    ) -> UncheckedRelay<'a> {
33449888
        debug_assert_eq!(self.c_relays()[rsidx].rsa_identity(), rs.rsa_identity());
33449888
        let md = self.mds[rsidx].as_deref();
33449888
        if let Some(md) = md {
33446201
            debug_assert_eq!(rs.md_digest(), md.digest());
3687
        }
33449888
        UncheckedRelay {
33449888
            rs,
33449888
            md,
33449888
            #[cfg(feature = "geoip")]
33449888
            cc: self.country_codes.get(rsidx.0).copied().flatten(),
33449888
        }
33449888
    }
    /// Return the value of the hsdir_n_replicas param.
    #[cfg(feature = "hs-common")]
45
    fn n_replicas(&self) -> u8 {
45
        self.params
45
            .hsdir_n_replicas
45
            .get()
45
            .try_into()
45
            .expect("BoundedInt did not enforce bounds")
45
    }
    /// Return the spread parameter for the specified `op`.
    #[cfg(feature = "hs-common")]
45
    fn spread(&self, op: HsDirOp) -> usize {
45
        let spread = match op {
45
            HsDirOp::Download => self.params.hsdir_spread_fetch,
            #[cfg(feature = "hs-service")]
            HsDirOp::Upload => self.params.hsdir_spread_store,
        };
45
        spread
45
            .get()
45
            .try_into()
45
            .expect("BoundedInt did not enforce bounds!")
45
    }
    /// 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")]
45
    fn select_hsdirs<'h, 'r: 'h>(
45
        &'r self,
45
        hsid: HsBlindId,
45
        ring: &'h HsDirRing,
45
        spread: usize,
45
    ) -> impl Iterator<Item = Relay<'r>> + 'h {
45
        let n_replicas = self.n_replicas();
45

            
45
        (1..=n_replicas) // 1-indexed !
45
            .flat_map({
45
                let mut selected_nodes = HashSet::new();
45

            
48
                move |replica: u8| {
4
                    let hsdir_idx = hsdir_ring::service_hsdir_index(&hsid, replica, ring.params());
4

            
18
                    ring.ring_items_at(hsdir_idx, spread, |(hsdir_idx, _)| {
18
                        // According to rend-spec 2.2.3:
18
                        //                                                  ... If any of those
18
                        // nodes have already been selected for a lower-numbered replica of the
18
                        // service, any nodes already chosen are disregarded (i.e. skipped over)
18
                        // when choosing a replica's hsdir_spread_store nodes.
18
                        selected_nodes.insert(*hsdir_idx)
18
                    })
4
                    .collect::<Vec<_>>()
48
                }
45
            })
56
            .filter_map(move |(_hsdir_idx, rs_idx)| {
12
                // This ought not to be None but let's not panic or bail if it is
12
                self.relay_by_rs_idx(*rs_idx)
56
            })
45
    }
    /// 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.
853131
    pub fn all_relays(&self) -> impl Iterator<Item = UncheckedRelay<'_>> {
853131
        // TODO: I'd like if we could memoize this so we don't have to
853131
        // do so many hashtable lookups.
853131
        self.c_relays()
853131
            .iter_enumerated()
32541253
            .map(move |(rsidx, rs)| self.relay_from_rs_and_rsidx(rs, rsidx))
853131
    }
    /// Return an iterator over all [usable](NetDir#usable) Relays.
798022
    pub fn relays(&self) -> impl Iterator<Item = Relay<'_>> {
798022
        self.all_relays().filter_map(UncheckedRelay::into_relay)
798022
    }
    /// 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.
353741
    pub fn by_id<'a, T>(&self, id: T) -> Option<Relay<'_>>
353741
    where
353741
        T: Into<RelayIdRef<'a>>,
353741
    {
353741
        let id = id.into();
353741
        let answer = match id {
353598
            RelayIdRef::Ed25519(ed25519) => {
353598
                let rsidx = *self.rsidx_by_ed.get(ed25519)?;
349316
                let rs = self.c_relays().get(rsidx).expect("Corrupt index");
349316

            
349316
                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))?,
        };
349330
        assert!(answer.has_identity(id));
349330
        Some(answer)
353741
    }
    /// 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))]
298
    pub(crate) fn relay_by_rs_idx(&self, rs_idx: RouterStatusIdx) -> Option<Relay<'_>> {
298
        let rs = self.c_relays().get(rs_idx)?;
298
        let md = self.mds.get(rs_idx)?.as_deref();
298
        UncheckedRelay {
298
            rs,
298
            md,
298
            #[cfg(feature = "geoip")]
298
            cc: self.country_codes.get(rs_idx.0).copied().flatten(),
298
        }
298
        .into_relay()
298
    }
    /// 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.
342442
    pub fn by_ids<T>(&self, target: &T) -> Option<Relay<'_>>
342442
    where
342442
        T: HasRelayIds + ?Sized,
342442
    {
342442
        let mut identities = target.identities();
        // Don't try if there are no identities.
342442
        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.
342442
        let candidate = self.by_id(first_id)?;
342230
        if identities.all(|wanted_id| candidate.has_identity(wanted_id)) {
342228
            Some(candidate)
        } else {
2
            None
        }
342442
    }
    /// 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.
23451
    fn id_pair_listed(&self, ed_id: &Ed25519Identity, rsa_id: &RsaIdentity) -> Option<bool> {
23451
        let r = self.by_rsa_id_unchecked(rsa_id);
23451
        match r {
23134
            Some(unchecked) => {
23134
                if !unchecked.rs.ed25519_id_is_usable() {
                    return Some(false);
23134
                }
23134
                // If md is present, then it's listed iff we have the right
23134
                // ed id.  Otherwise we don't know if it's listed.
23646
                unchecked.md.map(|md| md.ed25519_id() == ed_id)
            }
            None => {
                // Definitely not listed.
317
                Some(false)
            }
        }
23451
    }
    /// 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.
7906
    pub fn ids_listed<T>(&self, target: &T) -> Option<bool>
7906
    where
7906
        T: HasRelayIds + ?Sized,
7906
    {
7906
        let rsa_id = target.rsa_identity();
7906
        let ed25519_id = target.ed_identity();
7906

            
7906
        // TODO: If we later support more identity key types, this will
7906
        // become incorrect.  This assertion might help us recognize that case.
7906
        const_assert!(RelayIdType::COUNT == 2);
7906

            
7906
        match (rsa_id, ed25519_id) {
7906
            (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,
        }
7906
    }
    /// 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.
249569
    #[cfg_attr(feature = "experimental-api", visibility::make(pub))]
249569
    #[cfg_attr(docsrs, doc(cfg(feature = "experimental-api")))]
249569
    fn by_rsa_id_unchecked(&self, rsa_id: &RsaIdentity) -> Option<UncheckedRelay<'_>> {
249569
        let rsidx = *self.rsidx_by_rsa.get(rsa_id)?;
249107
        let rs = self.c_relays().get(rsidx).expect("Corrupt index");
249107
        assert_eq!(rs.rsa_identity(), rsa_id);
249107
        Some(self.relay_from_rs_and_rsidx(rs, rsidx))
249569
    }
    /// 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")]
7858
    fn all_hsdirs(&self) -> impl Iterator<Item = (RouterStatusIdx, Relay<'_>)> {
297183
        self.c_relays().iter_enumerated().filter_map(|(rsidx, rs)| {
296995
            let relay = self.relay_from_rs_and_rsidx(rs, rsidx);
296995
            relay.is_hsdir_for_ring().then_some(())?;
72370
            let relay = relay.into_relay()?;
72362
            Some((rsidx, relay))
297183
        })
7858
    }
    /// 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.
392931
    pub fn params(&self) -> &NetParameters {
392931
        &self.params
392931
    }
    /// 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 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: See notes on relay_protocol_status above.
    #[cfg(feature = "hs-common")]
    pub fn client_protocol_status(&self) -> &netstatus::ProtoStatus {
        self.consensus.client_protocol_status()
    }
    /// 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.
22485
    fn frac_for_role<'a, F>(&'a self, role: WeightRole, usable: F) -> f64
22485
    where
22485
        F: Fn(&UncheckedRelay<'a>) -> bool,
22485
    {
22485
        let mut total_weight = 0_u64;
22485
        let mut have_weight = 0_u64;
22485
        let mut have_count = 0_usize;
22485
        let mut total_count = 0_usize;
843690
        for r in self.all_relays() {
843690
            if !usable(&r) {
279250
                continue;
564440
            }
564440
            let w = self.weights.weight_rs_for_role(r.rs, role);
564440
            total_weight += w;
564440
            total_count += 1;
564440
            if r.is_usable() {
561672
                have_weight += w;
561672
                have_count += 1;
561672
            }
        }
22485
        if total_weight > 0 {
            // The consensus lists some weighted bandwidth so return the
            // fraction of the weighted bandwidth for which we have
            // descriptors.
22485
            (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
        }
22485
    }
    /// Return the estimated fraction of possible paths that we have
    /// enough microdescriptors to build.
7688
    fn frac_usable_paths(&self) -> f64 {
7688
        // TODO #504, TODO SPEC: We may want to add a set of is_flagged_fast() and/or
7688
        // is_flagged_stable() checks here.  This will require spec clarification.
283994
        let f_g = self.frac_for_role(WeightRole::Guard, |u| {
283805
            u.low_level_details().is_suitable_as_guard()
283994
        });
283994
        let f_m = self.frac_for_role(WeightRole::Middle, |_| true);
83947
        let f_e = if self.all_relays().any(|u| u.rs.is_flagged_exit()) {
276016
            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
        };
7688
        f_g * f_m * f_e
7688
    }
    /// Return true if there is enough information in this NetDir to build
    /// multihop circuits.
7686
    fn have_enough_paths(&self) -> bool {
7686
        // TODO-A001: This should check for our guards as well, and
7686
        // make sure that if they're listed in the consensus, we have
7686
        // the descriptors for them.
7686

            
7686
        // If we can build a randomly chosen path with at least this
7686
        // probability, we know enough information to participate
7686
        // on the network.
7686

            
7686
        let min_frac_paths: f64 = self.params().min_circuit_path_threshold.as_fraction();
7686

            
7686
        // What fraction of paths can we build?
7686
        let available = self.frac_usable_paths();
7686

            
7686
        available >= min_frac_paths
7686
    }
    /// 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`.
770414
        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 && selection.len() < relays.len() {
                    warn!(?self.weights, ?role,
                          "choose_multiple_weighted returned only {returned}, despite requesting {n}, \
                          and having {filtered_len} available after filtering. See bug #1907.",
                          returned=selection.len(), filtered_len=relays.len());
26131
                }
26131
                selection
            }
        };
26131
        relays.shuffle(rng);
26131
        relays
26131
    }
    /// Compute the weight with which `relay` will be selected for a given
    /// `role`.
22952
    pub fn relay_weight<'a>(&'a self, relay: &Relay<'a>, role: WeightRole) -> RelayWeight {
22952
        RelayWeight(self.weights.weight_rs_for_role(relay.rs, role))
22952
    }
    /// 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")]
675
    pub fn hs_time_period(&self) -> TimePeriod {
675
        self.hsdir_rings.current.time_period()
675
    }
    /// 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")]
180
    pub fn hs_all_time_periods(&self) -> Vec<HsDirParams> {
180
        self.hsdir_rings
180
            .iter()
544
            .map(|r| r.params().clone())
180
            .collect()
180
    }
    /// 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")]
    pub fn hs_dirs_upload(
        &self,
        hsid: HsBlindId,
        period: TimePeriod,
    ) -> std::result::Result<impl Iterator<Item = Relay<'_>>, Bug> {
        // Algorithm:
        //
        // 1. Choose spread = the parameter `hsdir_spread_store`
        // 2. Determine which HsDirRing to use, based on the time period.
        // 3. Find the shared random value that's associated with that HsDirRing.
        // 4. Let n_replicas = the parameter `hsdir_n_replicas`.
        // 5. Initialize Dirs = []
        // 6. 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.
        // 3. return Dirs.
        let spread = self.spread(HsDirOp::Upload);
        // For each HsBlindId, determine which HsDirRing to use.
        let rings = self
            .hsdir_rings
            .iter()
            .filter_map(move |ring| {
                // Make sure the ring matches the TP of the hsid it's matched with.
                (ring.params().time_period == period).then_some((ring, hsid, period))
            })
            .collect::<Vec<_>>();
        // The specified period should have an associated ring.
        if !rings.iter().any(|(_, _, tp)| *tp == period) {
            return Err(internal!(
                "the specified time period does not have an associated ring"
            ));
        };
        // Now that we've matched each `hsid` with the ring associated with its TP, we can start
        // selecting replicas from each ring.
        Ok(rings.into_iter().flat_map(move |(ring, hsid, period)| {
            assert_eq!(period, ring.params().time_period());
            self.select_hsdirs(hsid, ring, spread)
        }))
    }
    /// 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

            
40
                    ring.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
                }
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
    }
298338
    fn add_microdesc(&mut self, md: Microdesc) -> bool {
298338
        self.add_arc_microdesc(Arc::new(md))
298338
    }
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.
831995
    pub fn low_level_details(&self) -> details::UncheckedRelayDetails<'_> {
831995
        details::UncheckedRelayDetails(self)
831995
    }
    /// Return true if this relay is valid and [usable](NetDir#usable).
    ///
    /// This function should return `true` for every Relay we expose
    /// to the user.
32057251
    pub fn is_usable(&self) -> bool {
32057251
        // No need to check for 'valid' or 'running': they are implicit.
32057251
        self.md.is_some() && self.rs.ed25519_id_is_usable()
32057251
    }
    /// If this is [usable](NetDir#usable), return a corresponding Relay object.
31492811
    pub fn into_relay(self) -> Option<Relay<'a>> {
31492811
        if self.is_usable() {
            Some(Relay {
31492787
                rs: self.rs,
31492787
                md: self.md?,
                #[cfg(feature = "geoip")]
31492787
                cc: self.cc,
            })
        } else {
24
            None
        }
31492811
    }
    /// 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")]
296995
    pub(crate) fn is_hsdir_for_ring(&self) -> bool {
296995
        // TODO are there any other flags should we check?
296995
        // rend-spec-v3 2.2.3 says just
296995
        //   "each node listed in the current consensus with the HSDir flag"
296995
        // Do we need to check ed25519_id_is_usable ?
296995
        // See also https://gitlab.torproject.org/tpo/core/arti/-/issues/504
296995
        self.rs.is_flagged_hsdir()
296995
    }
}
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.
64732591
    pub fn low_level_details(&self) -> details::RelayDetails<'_> {
64732591
        details::RelayDetails(self)
64732591
    }
    /// Return the Ed25519 ID for this relay.
95178472
    pub fn id(&self) -> &Ed25519Identity {
95178472
        self.md.ed25519_id()
95178472
    }
    /// Return the RsaIdentity for this relay.
61806009
    pub fn rsa_id(&self) -> &RsaIdentity {
61806009
        self.rs.rsa_identity()
61806009
    }
    /// 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> {
87033912
    fn addrs(&self) -> &[std::net::SocketAddr] {
87033912
        self.rs.addrs()
87033912
    }
}
#[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> {
95171170
    fn ed_identity(&self) -> &Ed25519Identity {
95171170
        self.id()
95171170
    }
29067403
    fn rsa_identity(&self) -> &RsaIdentity {
29067403
        self.rsa_id()
29067403
    }
}
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> {
1135485
    fn ntor_onion_key(&self) -> &ll::pk::curve25519::PublicKey {
1135485
        self.md.ntor_key()
1135485
    }
1135485
    fn protovers(&self) -> &tor_protover::Protocols {
1135485
        self.rs.protovers()
1135485
    }
}
#[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);
    }
}