forked from opencontainers/runc
-
Notifications
You must be signed in to change notification settings - Fork 1
/
devices_emulator.go
386 lines (349 loc) · 12.1 KB
/
devices_emulator.go
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
// SPDX-License-Identifier: Apache-2.0
/*
* Copyright (C) 2020 Aleksa Sarai <cyphar@cyphar.com>
* Copyright (C) 2020 SUSE LLC
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package devices
import (
"bufio"
"fmt"
"io"
"sort"
"strconv"
"strings"
"github.com/opencontainers/runc/libcontainer/devices"
)
// deviceMeta is a Rule without the Allow or Permissions fields, and no
// wildcard-type support. It's effectively the "match" portion of a metadata
// rule, for the purposes of our emulation.
type deviceMeta struct {
node devices.Type
major int64
minor int64
}
// deviceRule is effectively the tuple (deviceMeta, Permissions).
type deviceRule struct {
meta deviceMeta
perms devices.Permissions
}
// deviceRules is a mapping of device metadata rules to the associated
// permissions in the ruleset.
type deviceRules map[deviceMeta]devices.Permissions
func (r deviceRules) orderedEntries() []deviceRule {
var rules []deviceRule
for meta, perms := range r {
rules = append(rules, deviceRule{meta: meta, perms: perms})
}
sort.Slice(rules, func(i, j int) bool {
// Sort by (major, minor, type).
a, b := rules[i].meta, rules[j].meta
return a.major < b.major ||
(a.major == b.major && a.minor < b.minor) ||
(a.major == b.major && a.minor == b.minor && a.node < b.node)
})
return rules
}
type Emulator struct {
defaultAllow bool
rules deviceRules
}
func (e *Emulator) IsBlacklist() bool {
return e.defaultAllow
}
func (e *Emulator) IsAllowAll() bool {
return e.IsBlacklist() && len(e.rules) == 0
}
func parseLine(line string) (*deviceRule, error) {
// Input: node major:minor perms.
fields := strings.FieldsFunc(line, func(r rune) bool {
return r == ' ' || r == ':'
})
if len(fields) != 4 {
return nil, fmt.Errorf("malformed devices.list rule %s", line)
}
var (
rule deviceRule
node = fields[0]
major = fields[1]
minor = fields[2]
perms = fields[3]
)
// Parse the node type.
switch node {
case "a":
// Super-special case -- "a" always means every device with every
// access mode. In fact, for devices.list this actually indicates that
// the cgroup is in black-list mode.
// TODO: Double-check that the entire file is "a *:* rwm".
return nil, nil
case "b":
rule.meta.node = devices.BlockDevice
case "c":
rule.meta.node = devices.CharDevice
default:
return nil, fmt.Errorf("unknown device type %q", node)
}
// Parse the major number.
if major == "*" {
rule.meta.major = devices.Wildcard
} else {
val, err := strconv.ParseUint(major, 10, 32)
if err != nil {
return nil, fmt.Errorf("invalid major number: %w", err)
}
rule.meta.major = int64(val)
}
// Parse the minor number.
if minor == "*" {
rule.meta.minor = devices.Wildcard
} else {
val, err := strconv.ParseUint(minor, 10, 32)
if err != nil {
return nil, fmt.Errorf("invalid minor number: %w", err)
}
rule.meta.minor = int64(val)
}
// Parse the access permissions.
rule.perms = devices.Permissions(perms)
if !rule.perms.IsValid() || rule.perms.IsEmpty() {
return nil, fmt.Errorf("parse access mode: contained unknown modes or is empty: %q", perms)
}
return &rule, nil
}
func (e *Emulator) addRule(rule deviceRule) error { //nolint:unparam
if e.rules == nil {
e.rules = make(map[deviceMeta]devices.Permissions)
}
// Merge with any pre-existing permissions.
oldPerms := e.rules[rule.meta]
newPerms := rule.perms.Union(oldPerms)
e.rules[rule.meta] = newPerms
return nil
}
func (e *Emulator) rmRule(rule deviceRule) error {
// Give an error if any of the permissions requested to be removed are
// present in a partially-matching wildcard rule, because such rules will
// be ignored by cgroupv1.
//
// This is a diversion from cgroupv1, but is necessary to avoid leading
// users into a false sense of security. cgroupv1 will silently(!) ignore
// requests to remove partial exceptions, but we really shouldn't do that.
//
// It may seem like we could just "split" wildcard rules which hit this
// issue, but unfortunately there are 2^32 possible major and minor
// numbers, which would exhaust kernel memory quickly if we did this. Not
// to mention it'd be really slow (the kernel side is implemented as a
// linked-list of exceptions).
for _, partialMeta := range []deviceMeta{
{node: rule.meta.node, major: devices.Wildcard, minor: rule.meta.minor},
{node: rule.meta.node, major: rule.meta.major, minor: devices.Wildcard},
{node: rule.meta.node, major: devices.Wildcard, minor: devices.Wildcard},
} {
// This wildcard rule is equivalent to the requested rule, so skip it.
if rule.meta == partialMeta {
continue
}
// Only give an error if the set of permissions overlap.
partialPerms := e.rules[partialMeta]
if !partialPerms.Intersection(rule.perms).IsEmpty() {
return fmt.Errorf("requested rule [%v %v] not supported by devices cgroupv1 (cannot punch hole in existing wildcard rule [%v %v])", rule.meta, rule.perms, partialMeta, partialPerms)
}
}
// Subtract all of the permissions listed from the full match rule. If the
// rule didn't exist, all of this is a no-op.
newPerms := e.rules[rule.meta].Difference(rule.perms)
if newPerms.IsEmpty() {
delete(e.rules, rule.meta)
} else {
e.rules[rule.meta] = newPerms
}
// TODO: The actual cgroup code doesn't care if an exception didn't exist
// during removal, so not erroring out here is /accurate/ but quite
// worrying. Maybe we should do additional validation, but again we
// have to worry about backwards-compatibility.
return nil
}
func (e *Emulator) allow(rule *deviceRule) error {
// This cgroup is configured as a black-list. Reset the entire emulator,
// and put is into black-list mode.
if rule == nil || rule.meta.node == devices.WildcardDevice {
*e = Emulator{
defaultAllow: true,
rules: nil,
}
return nil
}
var err error
if e.defaultAllow {
err = wrapErr(e.rmRule(*rule), "unable to remove 'deny' exception")
} else {
err = wrapErr(e.addRule(*rule), "unable to add 'allow' exception")
}
return err
}
func (e *Emulator) deny(rule *deviceRule) error {
// This cgroup is configured as a white-list. Reset the entire emulator,
// and put is into white-list mode.
if rule == nil || rule.meta.node == devices.WildcardDevice {
*e = Emulator{
defaultAllow: false,
rules: nil,
}
return nil
}
var err error
if e.defaultAllow {
err = wrapErr(e.addRule(*rule), "unable to add 'deny' exception")
} else {
err = wrapErr(e.rmRule(*rule), "unable to remove 'allow' exception")
}
return err
}
func (e *Emulator) Apply(rule devices.Rule) error {
if !rule.Type.CanCgroup() {
return fmt.Errorf("cannot add rule [%#v] with non-cgroup type %q", rule, rule.Type)
}
innerRule := &deviceRule{
meta: deviceMeta{
node: rule.Type,
major: rule.Major,
minor: rule.Minor,
},
perms: rule.Permissions,
}
if innerRule.meta.node == devices.WildcardDevice {
innerRule = nil
}
if rule.Allow {
return e.allow(innerRule)
}
return e.deny(innerRule)
}
// EmulatorFromList takes a reader to a "devices.list"-like source, and returns
// a new Emulator that represents the state of the devices cgroup. Note that
// black-list devices cgroups cannot be fully reconstructed, due to limitations
// in the devices cgroup API. Instead, such cgroups are always treated as
// "allow all" cgroups.
func EmulatorFromList(list io.Reader) (*Emulator, error) {
// Normally cgroups are in black-list mode by default, but the way we
// figure out the current mode is whether or not devices.list has an
// allow-all rule. So we default to a white-list, and the existence of an
// "a *:* rwm" entry will tell us otherwise.
e := &Emulator{
defaultAllow: false,
}
// Parse the "devices.list".
s := bufio.NewScanner(list)
for s.Scan() {
line := s.Text()
deviceRule, err := parseLine(line)
if err != nil {
return nil, fmt.Errorf("error parsing line %q: %w", line, err)
}
// "devices.list" is an allow list. Note that this means that in
// black-list mode, we have no idea what rules are in play. As a
// result, we need to be very careful in Transition().
if err := e.allow(deviceRule); err != nil {
return nil, fmt.Errorf("error adding devices.list rule: %w", err)
}
}
if err := s.Err(); err != nil {
return nil, fmt.Errorf("error reading devices.list lines: %w", err)
}
return e, nil
}
// Transition calculates what is the minimally-disruptive set of rules need to
// be applied to a devices cgroup in order to transition to the given target.
// This means that any already-existing rules will not be applied, and
// disruptive rules (like denying all device access) will only be applied if
// necessary.
//
// This function is the sole reason for all of Emulator -- to allow us
// to figure out how to update a containers' cgroups without causing spurious
// device errors (if possible).
func (source *Emulator) Transition(target *Emulator) ([]*devices.Rule, error) {
var transitionRules []*devices.Rule
oldRules := source.rules
// If the default policy doesn't match, we need to include a "disruptive"
// rule (either allow-all or deny-all) in order to switch the cgroup to the
// correct default policy.
//
// However, due to a limitation in "devices.list" we cannot be sure what
// deny rules are in place in a black-list cgroup. Thus if the source is a
// black-list we also have to include a disruptive rule.
if source.IsBlacklist() || source.defaultAllow != target.defaultAllow {
transitionRules = append(transitionRules, &devices.Rule{
Type: 'a',
Major: -1,
Minor: -1,
Permissions: devices.Permissions("rwm"),
Allow: target.defaultAllow,
})
// The old rules are only relevant if we aren't starting out with a
// disruptive rule.
oldRules = nil
}
// NOTE: We traverse through the rules in a sorted order so we always write
// the same set of rules (this is to aid testing).
// First, we create inverse rules for any old rules not in the new set.
// This includes partial-inverse rules for specific permissions. This is a
// no-op if we added a disruptive rule, since oldRules will be empty.
for _, rule := range oldRules.orderedEntries() {
meta, oldPerms := rule.meta, rule.perms
newPerms := target.rules[meta]
droppedPerms := oldPerms.Difference(newPerms)
if !droppedPerms.IsEmpty() {
transitionRules = append(transitionRules, &devices.Rule{
Type: meta.node,
Major: meta.major,
Minor: meta.minor,
Permissions: droppedPerms,
Allow: target.defaultAllow,
})
}
}
// Add any additional rules which weren't in the old set. We happen to
// filter out rules which are present in both sets, though this isn't
// strictly necessary.
for _, rule := range target.rules.orderedEntries() {
meta, newPerms := rule.meta, rule.perms
oldPerms := oldRules[meta]
gainedPerms := newPerms.Difference(oldPerms)
if !gainedPerms.IsEmpty() {
transitionRules = append(transitionRules, &devices.Rule{
Type: meta.node,
Major: meta.major,
Minor: meta.minor,
Permissions: gainedPerms,
Allow: !target.defaultAllow,
})
}
}
return transitionRules, nil
}
// Rules returns the minimum set of rules necessary to convert a *deny-all*
// cgroup to the emulated filter state (note that this is not the same as a
// default cgroupv1 cgroup -- which is allow-all). This is effectively just a
// wrapper around Transition() with the source emulator being an empty cgroup.
func (e *Emulator) Rules() ([]*devices.Rule, error) {
defaultCgroup := &Emulator{defaultAllow: false}
return defaultCgroup.Transition(e)
}
func wrapErr(err error, text string) error {
if err == nil {
return nil
}
return fmt.Errorf(text+": %w", err)
}