File size: 15,263 Bytes
375880c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Shared adapter helpers: actions, rule questions, groups, connectors, exceptions, thresholds.



An adapter turns one Request into the exact body that type's dashboard form would POST.

It starts from the form's defaults and only changes what the request decided, so a policy

made by the copilot is indistinguishable from one made by hand.

"""
from __future__ import annotations

import difflib
from dataclasses import dataclass, field
from typing import Any

from ..interpret import Content, Request
from ..model import Draft, Question
from ..platform import catalog
from ..platform.context import PlatformContext
from ..rules import RuleDecision

VERB_WORD = {'BLOCK': 'block', 'WARN': 'warn', 'REDACT': 'redact', 'ALERT': 'alert', 'LOG': 'log/monitor',
             'DELETE': 'delete', 'REMEDIATE': 'recall', 'REVOKE': 'revoke', 'READONLY': 'make read-only',
             'ALLOW': 'allow', 'ENABLE': 'turn on', 'DISABLE': 'turn off'}


@dataclass
class Build:
    """What an adapter hands back."""
    draft: Draft
    consumed_exceptions: list[int] = field(default_factory=list)   # indexes into request.exceptions
    cancelled: bool = False


def answer_for(answers: dict, slot: str, index: int) -> Any:
    """An answer for this draft ('slot@index') or for all drafts ('slot')."""
    if not isinstance(answers, dict):
        return None
    v = answers.get(f'{slot}@{index}')
    return v if v is not None else answers.get(slot)


def choose_action(d: Draft, req: Request, index: int, answers: dict, *, mapping: dict[str, str],

                  allowed: list[tuple[str, str]], default: str, type_label: str,

                  impossible_default: str | None = None) -> str:
    """Pick the body action for the request's verb.



    mapping: verb -> action for verbs this type can honour.

    allowed: [(action, label)] the type accepts, for questions.

    A verb that is not in mapping is impossible on this type: the draft keeps a valid action

    and a blocking question asks which of the possible ones the admin wants.

    """
    valid = [a for a, _ in allowed]
    ans = answer_for(answers, 'action', index)
    if isinstance(ans, str) and ans in valid:
        return ans
    verb = req.verb
    options = [{'value': a, 'label': lbl} for a, lbl in allowed]
    if verb is None or verb in ('ENABLE', 'CREATE', 'DISABLE'):
        d.ask(Question('action', f'What should happen when this matches on {type_label}?', options, True, default))
        return default
    if req.verb_conf < 0.5:
        d.ask(Question('action', f'I could not tell whether you want to block or allow this. '
                                 f'What should happen on {type_label}?', options, True, default))
        return default
    if verb in mapping:
        return mapping[verb]
    fallback = impossible_default or default
    d.issue('ACTION_NOT_AVAILABLE', 'blocking',
            f'{type_label} policies cannot {VERB_WORD.get(verb, verb.lower())}. '
            f'They can: {", ".join(lbl for _, lbl in allowed)}.',
            requested=verb.lower(), type=d.policyType, available=valid)
    d.ask(Question('action', f'{type_label} cannot {VERB_WORD.get(verb, verb.lower())}. Which of these do you want?',
                   options, True, fallback))
    return fallback


def rule_questions(d: Draft, rd: RuleDecision, ctx: PlatformContext) -> None:
    """Questions about the detection-rule choice (never blocking unless the data was guessed)."""
    if rd.mode == 'superset' and rd.rule_ids:
        r = ctx.rule(rd.rule_ids[-1])
        extra = ', '.join(catalog.entity_label(e) for e in rd.extras)
        d.ask(Question('rule', f'Use the existing rule "{r.name if r else rd.rule_ids[-1]}" (it also detects {extra}), '
                               f'or create a new rule with only what you asked for?',
                       [{'value': rd.rule_ids[-1], 'label': f'Use "{r.name if r else rd.rule_ids[-1]}"'},
                        {'value': 'create', 'label': 'Create a new rule'}], False, rd.rule_ids[-1]))
    if rd.mode == 'create' and rd.alternative and rd.alternative.get('mode') == 'reuse':
        alt = rd.alternative
        d.ask(Question('rule', f'I will create a new detection rule. The built-in rule "{alt["ruleName"]}" '
                               f'already covers this and more. Use it instead?',
                       [{'value': 'create', 'label': 'Create the new rule'},
                        {'value': alt['ruleIds'], 'label': f'Use "{alt["ruleName"]}"'}], False, 'create'))
    if rd.assumed:
        names = [ctx.rule(r).name for r in rd.rule_ids if ctx.rule(r)]
        d.issue('DATA_ASSUMED', 'warning',
                f'"Sensitive data" can mean many things. I used these rules: {", ".join(names)}. '
                f'Change the list if you meant something narrower.', ruleIds=list(rd.rule_ids))
        d.ask(Question('data', 'Which data do you want this policy to look for?',
                       [{'value': r.id, 'label': r.name} for r in ctx.rules], False, list(rd.rule_ids)))


def need_data(d: Draft, rd: RuleDecision, ctx: PlatformContext, why: str) -> None:
    d.issue('NO_DETECTION_RULES', 'blocking', why)
    d.ask(Question('data', 'What data should this policy look for?',
                   [{'value': r.id, 'label': r.name} for r in ctx.rules], True, []))


def resolve_group(d: Draft, req: Request, ctx: PlatformContext, index: int, answers: dict) -> str:
    """The groupId for an endpoint policy. An unknown group is asked about, never dropped silently."""
    ans = answer_for(answers, 'group', index)
    if isinstance(ans, str) and (ans == 'default' or ctx.group(ans)):
        return ans
    if req.group_unknown and req.group == 'default':
        names = [g.name for g in ctx.groups]
        close = difflib.get_close_matches(req.group_unknown, [n.lower() for n in names], n=3, cutoff=0.5)
        opts = [{'value': g.id, 'label': g.name} for g in ctx.groups if g.name.lower() in close]
        opts += [{'value': g.id, 'label': g.name} for g in ctx.groups if g.name.lower() not in close][:8]
        opts.append({'value': 'default', 'label': 'Everyone (org-wide)'})
        d.issue('UNKNOWN_GROUP', 'blocking', f'There is no group called "{req.group_unknown}".',
                name=req.group_unknown)
        d.ask(Question('group', f'There is no group called "{req.group_unknown}". Which group did you mean?',
                       opts, True, 'default'))
        return 'default'
    if req.group != 'default' and ctx.group(req.group):
        return req.group
    return 'default'


def channel_group_check(d: Draft, req: Request, ctx: PlatformContext, index: int, answers: dict) -> None:
    """Channel policies are org-wide in the dashboard: a group can't narrow them."""
    asked = req.group != 'default' or req.group_unknown
    if not asked or answer_for(answers, 'scope', index) == 'org':
        return
    name = ctx.group(req.group).name if ctx.group(req.group) else req.group_unknown
    d.issue('GROUP_NOT_SUPPORTED', 'blocking',
            f'{catalog.TYPE_LABEL[d.policyType]} policies apply to the whole organisation; they cannot be limited '
            f'to "{name}". Only endpoint policies are scoped to a group.', group=name)
    d.ask(Question('scope', f'This policy will apply to everyone, not only {name}. Continue?',
                   [{'value': 'org', 'label': 'Yes, apply to everyone'}, {'value': 'cancel', 'label': 'Cancel'}],
                   True, 'org'))


def pick_connectors(d: Draft, req: Request, ctx: PlatformContext, channel: str, index: int, answers: dict) -> list[str]:
    ans = answer_for(answers, 'connector', index)
    known = {c.id for c in ctx.connectors_for(channel)}
    if isinstance(ans, list):
        return [a for a in ans if a in known]
    if isinstance(ans, str) and ans in known:
        return [ans]
    ids = [cid for ch, cid in req.content.connectors if ch == channel and cid in known]
    if ctx.connectorsKnown and not ctx.connectors_for(channel):
        d.issue('NO_CONNECTOR', 'warning',
                f'No {catalog.TYPE_LABEL[channel]} connection is set up yet, so this policy will not see anything '
                f'until one is added in Settings.', channel=channel)
    return ids


def thresholds_unsupported(d: Draft, req: Request, index: int, answers: dict, handled: set[str] = frozenset()) -> None:
    rest = [(n, noun) for n, noun in req.content.thresholds if noun not in handled]
    if not rest or answer_for(answers, 'threshold', index) == 'ignore':
        return
    n, noun = rest[0]
    d.issue('THRESHOLD_NOT_SUPPORTED', 'blocking',
            f'{catalog.TYPE_LABEL[d.policyType]} policies cannot count ("more than {n} {noun}"). '
            f'The policy acts on the first match.', count=n, noun=noun)
    d.ask(Question('threshold', f'The policy will act on the first match, not after {n} {noun}. Continue?',
                   [{'value': 'ignore', 'label': 'Yes, act on every match'}, {'value': 'cancel', 'label': 'Cancel'}],
                   True, 'ignore'))


def schedule_unsupported(d: Draft, req: Request, index: int, answers: dict, after_hours_ok: bool = False) -> None:
    mods = req.content.mods
    if 'timed' in mods and answer_for(answers, 'schedule', index) != 'ignore':
        d.issue('START_END_NOT_SUPPORTED', 'blocking',
                'Policies apply from the moment they are switched on until someone switches them off; they cannot '
                'start or end on a date by themselves.')
        d.ask(Question('schedule', 'The policy will apply as soon as it is switched on, with no end date. Continue?',
                       [{'value': 'ignore', 'label': 'Yes'}, {'value': 'cancel', 'label': 'Cancel'}], True, 'ignore'))
        return
    timed = 'during_hours' in mods or (not after_hours_ok and (req.content.hours or 'after_hours' in mods))
    if not timed:
        return
    if answer_for(answers, 'schedule', index) == 'ignore':
        return
    d.issue('SCHEDULE_NOT_SUPPORTED', 'blocking',
            f'{catalog.TYPE_LABEL[d.policyType]} policies apply at all times; they cannot be limited to certain hours.')
    d.ask(Question('schedule', 'The policy will apply at all hours. Continue?',
                   [{'value': 'ignore', 'label': 'Yes, apply at all times'}, {'value': 'cancel', 'label': 'Cancel'}],
                   True, 'ignore'))


QUALIFIERS = {
    'executables': 'kind of file (executables)',
    'bulk': 'volume ("bulk")',
}


def qualifiers_unsupported(d: Draft, req: Request, index: int, answers: dict, handled: set[str] = frozenset()) -> None:
    """Words that narrow the request in a way this policy type cannot ("larger than 10 MB", "executables").



    Ignoring them would make the policy broader than asked, so the admin confirms first."""
    c = req.content
    found = []
    if c.sizes and 'sizes' not in handled:
        found.append(f'size ({", ".join(c.sizes)})')
    for mod in ('executables', 'bulk'):
        if mod in c.mods and mod not in handled:
            found.append(QUALIFIERS[mod])
    for s in c.situations:
        found.append(f'situation ("{s}")')
    if not found or answer_for(answers, 'qualifier', index) == 'ignore':
        return
    d.issue('QUALIFIER_NOT_SUPPORTED', 'blocking',
            f'This policy cannot be limited by {" or ".join(found)}; it would apply to everything it covers.',
            qualifiers=found)
    d.ask(Question('qualifier', f'The policy cannot be limited by {" or ".join(found)}. Apply it to everything?',
                   [{'value': 'ignore', 'label': 'Yes, apply to everything'}, {'value': 'cancel', 'label': 'Cancel'}],
                   True, 'ignore'))


def size_bytes(s: str) -> int | None:
    """'5mb' -> 5242880."""
    import re
    m = re.match(r'^([\d.]+)(kb|mb|gb|tb|kilobytes|megabytes|gigabytes)$', s)
    if not m:
        return None
    mult = {'kb': 1024, 'kilobytes': 1024, 'mb': 1024 ** 2, 'megabytes': 1024 ** 2, 'gb': 1024 ** 3,
            'gigabytes': 1024 ** 3, 'tb': 1024 ** 4}[m.group(2)]
    try:
        return max(1, int(float(m.group(1)) * mult))
    except ValueError:
        return None


def domains_question(d: Draft, req: Request, index: int, answers: dict, doms: list[str],

                     options: list[tuple[str, str]]) -> str | None:
    """What a domain named in a channel request is for. Returns the chosen role, or None (asked)."""
    if 'our_domain' in req.content.mods:
        return 'owned'
    ans = answer_for(answers, 'domains', index)
    if ans in [v for v, _ in options] or ans == 'ignore':
        return ans
    d.issue('DOMAINS_UNCLEAR', 'blocking', f'I am not sure what {", ".join(doms)} should do in this policy.',
            domains=doms)
    d.ask(Question('domains', f'What is {", ".join(doms)}?',
                   [{'value': v, 'label': lbl} for v, lbl in options] + [{'value': 'ignore', 'label': 'Ignore it'}],
                   True, 'ignore'))
    return None


def exceptions_unsupported(d: Draft, req: Request, consumed: list[int], index: int, answers: dict,

                           ctx: PlatformContext) -> None:
    left = [e for k, e in enumerate(req.exceptions) if k not in consumed and not e.says_nothing()]
    if not left or answer_for(answers, 'exception', index) == 'ignore':
        return
    desc = describe_content(left[0], ctx)
    groups = [ctx.group(g).name for e in left for g in e.groups if ctx.group(g)]
    hint = ''
    if groups and d.policyType == 'endpoint':
        hint = (f' For AI prompt rules, CLOAK\'s way to do this is a separate policy for {", ".join(groups)}: a '
                f'group\'s own policies replace the org-wide prompt rules for that group. Device and browser '
                f'controls (USB, printing, block lists) set org-wide cannot be lifted for one group.')
    d.issue('EXCEPTION_NOT_SUPPORTED', 'blocking',
            f'This kind of policy has no way to exempt {desc}.{hint}', exception=desc)
    d.ask(Question('exception', f'The policy cannot exempt {desc}; it would apply to them too. Continue?',
                   [{'value': 'ignore', 'label': 'Yes, without the exception'}, {'value': 'cancel', 'label': 'Cancel'}],
                   True, 'ignore'))


def describe_content(c: Content, ctx: PlatformContext) -> str:
    parts: list[str] = []
    parts += [ctx.group(g).name for g in c.groups if ctx.group(g)]
    parts += c.unknown_groups
    parts += [catalog.entity_label(e) for e in c.data.entity_set()]
    parts += c.domains + c.emails + c.serials + c.sites + c.apps + c.ai_sites + c.mods + c.words
    return ', '.join(dict.fromkeys(parts)) or 'that'


def domains_of(c: Content) -> list[str]:
    hosts = set(catalog.ai_hosts())
    return [x for x in c.domains if x not in hosts]


def base_body(policy_type: str) -> dict:
    body = catalog.default_body(policy_type)
    # Copilot proposals are always saved inactive; the confirm screen has an explicit switch.
    body['active'] = False
    body['enabled'] = False
    return body