aboutsummaryrefslogtreecommitdiff
path: root/doc/dev/notes/client-auth.md
blob: eed283d98658bebc4707a592fc117e14a249d625 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
# HS client auth

> This document uses deprecated terminology: "client authorization" is
> now known as "restricted discovery".

This based on our previous discussions from #1028, #1027, #696.

It presents a simplified version of what is proposed in #1028, and an
implementation plan (in the form of action items and tickets).

# Proposal A (rejected)

## Generating keys

### Using the `arti hsc` subcommand

Client authorization keys can be manually generated using the `arti hsc
generate-key stealth` command. In addition to generating a client auth
keypair in the keystore, this command also exports the public part of key,
in the format specified using `--pub-format`.

To generate a client authorization key for service xyz.onion in keystore foo:

```
 arti hsc --config arti.toml generate-key stealth --keystore foo  \
     --nickname alice                                             \
     --hsid xyz.onion                                             \
     --pub-format arti
     # or --pub-format ctor (the format is documented in the tor manpage
     # under CLIENT AUTHORIZATION)
```

Some other possible names for `generate-key stealth`:
  * `generate-key hs-auth-stealth`
  * `generate-key hs-auth-desc`
  * `generate-hs-auth-key stealth`
  * ..

Initially, `arti hsc generate-key` will only support `stealth` keys, for use
with services running in **stealth mode** (as defined in #1028). If we implement
other types of client authorization in the future, we'll likely need to also
extend `arti hsc generate-key`.

This command displays an error if the client already has a keypair of the
specified kind. The public part of a preexisting keypair can be extracted using
`arti hsc export-pubkey`.

(Another possibility would be to let the service generate the client auth keys,
but then we'd need to come up with a secure way for it to communicate the
private part of the key to the client; we decided this is a nonstarter)

#### Public key output format

##### `--pub-format ctor`

With `--pub-format ctor`, `arti hsc generate-key stealth` will generate a
`<client_nickname>.auth` file, with the content in the
`<auth-type>:<key-type>:<base32-encoded-public-key>` format, as per the `CLIENT
AUTHORIZATION` section from `tor(1)`.

##### `--pub-format arti`

Regardless of how we choose to implement client auth configuration on the
service side, with `--pub-format arti`, `arti hsc generate-key stealth` will
generate a `kp_hsc_desc_enc.x25519_public` file containing an OpenSSH key (using
our custom `[email protected]` algorithm name):

```
[email protected] AAAAGngyNTUxOUBzcGVjLnRvcnByb2plY3Qub3JnAAAAIGmMjbhv/HldaPDU3zGl4YspW84XMqiEoNon1Tre14Eh
```

This file can then be shared with HS operators through a secure channel (there
are many ways to peel this orange, but they're outside the scope of this
document).

**Suggested action items**:
  * [ ] Decide if `arti hsc generate-key stealth` is a good name for the
    command, and come up with a better one if it isn't
  * [ ] Implement the `arti hsc generate-key` subcommand (#1281)
  * [ ] Implement the `arti hsc export-pubkey` subcommand

### Auto-generating client auth keys

We could also provide an HS client config option for auto-generating the client
authorization keys for specific hidden services:

```toml
   [hs_client]
   # Generate client authorization keys for these services, if needed.
   foo_bar_onion_svc_auth = [
       "foo.onion",
       "bar.onion"
   ]
```

However, I think the UX for this would be bad:
  * it's not obvious at all that the client might also have authorization keys for
    services not listed under `foo_bar_onion_svc_auth`
  * we will need to provide a CLI for extracting the public key in a format that
    can be used by an Arti or C Tor hidden service (`arti hsc
    extract-pub-auth-key-foo-bar`). So auto-generating the
    client auth keys doesn't even save us from having to invoke `arti hsc`

**Suggested action items**:
  * [x] Do not implement this

## Configuring client authorization (client side)

Clients wanting to connect to services that require client authorization don't
need to be explicitly configured to do so: the presence of
`client/<client_id>/<hsid>/ks_hsc_desc_enc.x25519_private` in the client
keystore is enough for the client to be able to connect to `<hsid>` (assuming `<hsid>`
is configured to allow connections from this client).

We might need to revisit this decision if we implement additional types of
client authentication (other types of client auth will potentially need to be
enabled selectively, only for the services that expect it). If we do, we need to
make sure the config changes are backwards-compatible (i.e. clients default to
using their `<hsid>/ks_hsc_desc_enc.x25519_private`, if any, when connecting to
`<hsid>`).

**Suggested action items**: none

## Configuring client authorization (service side)

The authorized clients will be configured using the `authorized_clients` service
option. As mentioned in #1028, we might want to support dynamic HS client
providers at some point, but for now we're only going to allow statically
configured clients.

We have several options for the static authorized clients configuration.

### Option 1: Place authorized client keys in the state dir

We could put the authorized client keys in a directory
within the state dir (`<state_dir>/authorized_clients/<client_nick>`).

We will provide a `arti hss auth-clients` CLI (described under `Extra CLI
subcommands for managing authorized clients` below) for managing client
authorization.

We will use the same naming convention as we do for the keys in an
`ArtiNativeKeystore`, so the paths of client authorization keys
will be of the form
```
<state_dir>/authorized_clients/<client_nick>/kp_hsc_desc_enc.x25519_public
```

but we will additionally support keys in the format used by C Tor. So
`authorized_clients` can also contain entries of the form
```
<state_dir>/authorized_clients/<client_nick>/<client_nick>.auth
```

If both `kp_hsc_desc_enc.x25519_public` and `<client_nick>.auth` are present,
the service will use `kp_hsc_desc_enc.x25519_public` and log a warning.

In addition to provisioning the `authorized_clients` directory, HS operators
wanting to enable client authorization must explicitly set `enabled = true` in
the toml config:

```toml
[onion_service."allium-cepa"]
authorized_clients.enabled = true
```


If `authorized_clients` is empty, no clients are authorized to access the
service. Alternatively, we could declare an empty directory means no
authorization is required (this is what C Tor does), but that would be redundant
with the `enabled` option.

Pros:
  * this simplifies the distribution and management of client keys: service
    operators can grant/revoke client authorization by simply moving the client
    keys to/from the `authorized_clients` directory
  * the keys are stored in a familiar format (the same one we use in the
    keystore)

Cons:
  * the presence of an empty `authorized_clients` directory can be interpreted
    in multiple ways ("nobody is authorized" or "everyone is authorized").
    However, this is probably disambiguated by the existence
    `authorized_clients.enabled = true` (`authorized_clients.enabled = false`
    means "everyone is authorized")
  * services need to watch the `authorized_clients` directory for changes (and
    update their view of which clients are authorized accordingly)


### Option 2: Encode the authorized clients as a JSON blob

#1028 suggests encoding the authorized clients in a semi-opaque format, and
embedding

```toml
[onion_service."allium-cepa"]
authorized_clients.enabled = true

authorized_clients.static = {
  "alice": "{...}"
}
```

in the config (or reading each client's config from a separate
`<state_dir>/authorized_clients/config.json`)

However, I'm not sure I see the benefit of using JSON here.

Pros:
  * the authorized clients can be reloaded along with the rest of the config in
    `watch_for_config_changes`
  * it might be more user-friendly (readable) than the alternative. OTOH, I'm
    not sure we want it to be readable (we don't want to encourage users to
    manually modify it)

Cons:
  * it complicates the distribution and management of client keys: service
    operators have to fiddle with the config to authorize new clients (they need
    to paste the contents of `kp_hsc_desc_enc.x25519_public` in the config).
    This can be alleviated by providing an `arti hss auth-clients` subcommand
    for managing authorized clients (see `Extra CLI subcommands for managing
    authorized clients` below)

**Suggested action items**:
  * [ ] Implement Option 1 for static authorized client configuration
  * [ ] Make sure the service reloads its authorized clients if there are
    changes to the `authorized_clients` directory

## Extra CLI subcommands for managing authorized clients

We might want to provide an `arti hss auth-clients` command for managing a
service's authorized clients (that is, assuming we implement `Option 1` from
above).

```
NAME
       arti-hss-auth-clients - Manage the authorized clients of this hidden service

SYNOPSIS
       arti hss auth-clients [SUBCOMMAND]

DESCRIPTION
      A command for managing the authorized clients of an Arti hidden service.

      TODO: document how these commands are supposed to work after we reach a
      conclusion in #1028

SUBCOMMANDS
       help                  Print this message or the help of the given subcommand(s)
       list                  List the authorized clients
       import                Import the public keys of a client
       disable               Un-authorize a previously authorized client
       remove                Purge the client authorization keys of a client, unauthorizing them
       enable                Authorize a new client
```

For example, `arti hss auth-clients import --nickname client-foo
~/downloads/kp_hsc_desc_enc.x25519_public` would create an authorized client
called `client-foo`.

Since `import`, `disable`, `enable` are essentially just wrappers around `cp`
and `mv`, this subcommand may not be particularly useful. OTOH, `enable` and
`disable` could be useful for managing temporarily unauthorized clients: arti
would maintain a separate `revoked_clients` (`disabled_clients`?) directory, and
`disable` and `enable` would move keys to and from it.

In addition to the commands listed under `SUBCOMMANDS` above, we might also want
to provide subcommands for:
  * retrieving the absolute path of `<state_dir>/authorized_clients`
  * retrieving the absolute path of
    `<state_dir>/authorized_clients/<client_nick>`

**Suggested action items**:
  * [ ] Make this a low-priority item (and implement it if time permits)


# Proposal B

## Generating keys

### Using the `arti hsc` subcommand

Client authorization keys can be manually generated using the `arti hsc
prepare-restricted-mode-key` command. In addition to generating a client auth
keypair in the keystore, this command also exports the public part of the key
(in C Tor format).

```
arti hsc prepare-restricted-mode-key
   --hsid ...                 # no default
   [ --config arti.toml ]     # default is default arti.toml
   [ --output FOO.auth ]      # default is <hs-nickname>.auth, use `-` for stdout
   [ --overwrite ]            # overwrites any existing output file; default is to refuse
   [ --generate=no|yes|if-needed ]     # if-needed is the default; otherwise, can error
```

**Suggested action items**:
  * [ ] Implement the `prepare-restricted-mode-key` command (#1281)

#### Public key output format

For now, we will only support C Tor format for restricted mode client public keys.

**Suggested action items**:
  * [ ] Support encoding x25519 public keys in C Tor format

## HS nickname -> HsId mapping (rejected)

Clients will have nicknames for the services they have authorization keys for.
This will allow clients to refer to services by nickname rather than
by HsId (`torproject` vs
`2gzyxa5ihm7nsggfxnu52rck2vv4rvmdlkiu3zzui5du4xyclen53wid`).

The client will need to maintain a mapping from HS nickname to HsId. The reverse
mapping is also going to be needed (at least conceptually), because when
connecting to `<hsid>.onion`, the client needs to "look up" the corresponding
`<hs-nickname>` in order to be able to compute the key specifier of that
particular authorization key (if it exists).

During HsId rotation, clients will need to be able to connect to both the old
and the new HsId. This is needed to support e.g. load balancing setups where
multiple hosts run the "same" hidden service (i.e. they all use the same HsId),
and race to publish/republish the descriptor.

This mapping will need to be:
  * as persistent as the keystore
  * compatible with the non-disk keystore types we plan to implement in
    the future

I don't think this mapping belongs in the state dir. Putting it there would
create more opportunities for synchronization bugs where the keystore is out of
sync with the state dir.

Instead, I propose we encode the mapping in the `ArtiPath`s of the client keys.
(Alternatively, it could be encoded in the comment of the OpenSSH key.)

`ArtiPaths` of the form
```
client/<hsid>/ks_hsc_desc_enc.x25519_private
```
will become
```
client/<hs-nickname>+<hsid>/ks_hsc_desc_enc.x25519_private
```

We will need to restrict the `<hs-nickname>` charset (for instance, it cannot
include the `+` symbol), as well as its length (to avoid running into
platform-specific file path length limits).

Pros:
  * the keys associated with a given `<hs-nickname>` can be listed/removed using
    the `client/<hs-nickname>+*/ks_hsc_desc_enc.x25519_private`
    `KeyPathPattern`. In fact, the entire mapping can be derived by listing
    all the key specifiers matching `client/*+*/ks_hsc_desc_enc.x25519_private`
  * the mapping cannot go out of sync with the keystore
  * when asked to connect to `<hsid.onion>`, the client doesn't need to know the
    nickname of the service: it just needs to find the key matching
    `client/*+<hsid>/ks_hsc_desc_enc.x25519_private`, and bail if there is more
    than 1 such key (in the future we might decide to allow many-to-many
    nickname -> hsid mappings, but for now they are forbidden)

Cons:
  * in practice, the length of the nickname is going to be limited to about 147
    characters. I think this is fine.
  * in the case of C Tor keystores, the mapping can't be extracted from the
    `CTorPath`s of the client keys alone (the HsId needs to be read from
    `<hs-nickname>.auth`). This asymmetry might mean we need to split
    `KeyMgr::list_matching`  into `KeyMgr::list_matching_arti` and
    `KeyMgr::list_matching_ctor` (because `ArtiPath`s are going to be handled
    very differently from `CTorPath`s). (I think this is actually a pervasive
    issue that we haven't tackled yet: a number of other callsites/APIs will
    likely need to change when we add support for C Tor keystores).

### Handling HsId changes

If a service `<hsid1>` running in "restricted mode" rotates its identity keys
(`<hsid1>` -> `<hsid2>`), on the client side, `client/<hs-nickname>+<hsid1>`
needs to be copied to `client/<hs-nickname>+<hsid2>` for the duration of the
transition period. After the transition period, `client/<hs-nickname>+<hsid1>`
can be removed. We can provide an `arti hsc` subcommand for handling HsId
changes, but it will need to be run manually.

This is inconvenient but unavoidable: the nickname -> HsId mapping needs to be
manually updated regardless of whether it's encoded in the `ArtiPath` or stored
separately.

**Suggested action items**:
  * [x] For now, do not implement any of this. When we have a concrete use case
    for it, we should come up with an alternative way to map nicknames to
    HsIds and/or multiple HsIds to the same service identity. For context, see
    https://gitlab.torproject.org/tpo/core/arti/-/merge_requests/1987#note_2999114

## Configuring client authorization (service side)

The authorized clients are going be part of the service configuration.

```toml
[onion_service."allium-cepa".restricted_mode]
# TODO: The naming and values of this field are provisional
enabled = auto | on | off

[onion_service."allium-cepa".restricted_mode.authorized_clients.static]
alice = "descriptor:x25519:PU63REQUH4PP464E2Y7AVQ35HBB5DXDH5XEUVUNP3KCPNOXZGIBA"
bob   = "descriptor:x25519:B5ZQGTPERMMUDA6VC63LHJUF5IHPOKJMUK26LY2XKSF7VG52AESQ"

# Alternatively, you can specify a directory of authorized clients.
# Each authorized client is represented by an .auth file, as specified
# under CLIENT AUTHORIZATION in tor(1).
#
# [onion_service."allium-cepa".restricted_mode.authorized_clients.keydirectory]
# path = "/etc/allium/authorized_clients"
```

`restricted_mode.enabled = off` disables "restricted mode", even if the
list of authorized clients is non-empty.

As per #1028, in the future we might extend this with support for pluggable
client auth key databases:
```toml
[onion_service."allium-cepa".restricted_mode]

[onion_service."allium-cepa".restricted_mode.authorized_clients.static]
alice = "descriptor:x25519:PU63REQUH4PP464E2Y7AVQ35HBB5DXDH5XEUVUNP3KCPNOXZGIBA"
bob   = "descriptor:x25519:B5ZQGTPERMMUDA6VC63LHJUF5IHPOKJMUK26LY2XKSF7VG52AESQ"

[onion_service."allium-cepa".restricted_mode.provider]
driver = "postgresql"
database = "..."
query = "SELECT nick, pubkey AS kp_desc_enc FROM clients JOIN client_keys WHERE clients.enabled"
```

If we later introduce new client auth protocols, we will also add
corresponding service configuration modes:
```toml
[onion_service."allium-cepa".restricted_mode]
enabled = on

[onion_service."allium-cepa".foobar_mode]
enabled = on

[onion_service."allium-cepa".foobar_mode.provider]
...
```

Each mode can be toggled on or off independently of the others. Some modes may
be incompatible. The service will error if the enabled authorization modes are
mutually incompatible.

If we want to add an authorization mechanism that uses the "restricted mode"
x25519 public keys, we can simply nest its configuration within the
`restricted_mode` section:

```toml
[onion_service."allium-cepa".restricted_mode]
enabled = true

[onion_service."allium-cepa".extra_foobar_checks]
enabled = auto
...
```

**Suggested action items**:
  * [  ] Choose a name for the `enabled` option, and decide what values it
    should take (`BoolOrAuto` may not be the right type for it)
  * [  ] Implement the service configuration for configuring "restricted" mode
    with static `authorized_clients`