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
|
// Copyright 2018 The gVisor Authors.
//
// 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 mm
import (
"fmt"
"sync/atomic"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/memmap"
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
"gvisor.dev/gvisor/pkg/sentry/platform"
"gvisor.dev/gvisor/pkg/usermem"
)
// NewMemoryManager returns a new MemoryManager with no mappings and 1 user.
func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider, sleepForActivation bool) *MemoryManager {
return &MemoryManager{
p: p,
mfp: mfp,
haveASIO: p.SupportsAddressSpaceIO(),
privateRefs: &privateRefs{},
users: 1,
auxv: arch.Auxv{},
dumpability: UserDumpable,
aioManager: aioManager{contexts: make(map[uint64]*AIOContext)},
sleepForActivation: sleepForActivation,
}
}
// SetMmapLayout initializes mm's layout from the given arch.Context.
//
// Preconditions: mm contains no mappings and is not used concurrently.
func (mm *MemoryManager) SetMmapLayout(ac arch.Context, r *limits.LimitSet) (arch.MmapLayout, error) {
layout, err := ac.NewMmapLayout(mm.p.MinUserAddress(), mm.p.MaxUserAddress(), r)
if err != nil {
return arch.MmapLayout{}, err
}
mm.layout = layout
return layout, nil
}
// Fork creates a copy of mm with 1 user, as for Linux syscalls fork() or
// clone() (without CLONE_VM).
func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) {
mm.metadataMu.Lock()
defer mm.metadataMu.Unlock()
mm.mappingMu.RLock()
defer mm.mappingMu.RUnlock()
mm2 := &MemoryManager{
p: mm.p,
mfp: mm.mfp,
haveASIO: mm.haveASIO,
layout: mm.layout,
privateRefs: mm.privateRefs,
users: 1,
brk: mm.brk,
usageAS: mm.usageAS,
dataAS: mm.dataAS,
// "The child does not inherit its parent's memory locks (mlock(2),
// mlockall(2))." - fork(2). So lockedAS is 0 and defMLockMode is
// MLockNone, both of which are zero values. vma.mlockMode is reset
// when copied below.
captureInvalidations: true,
argv: mm.argv,
envv: mm.envv,
auxv: append(arch.Auxv(nil), mm.auxv...),
// IncRef'd below, once we know that there isn't an error.
executable: mm.executable,
dumpability: mm.dumpability,
aioManager: aioManager{contexts: make(map[uint64]*AIOContext)},
sleepForActivation: mm.sleepForActivation,
}
// Copy vmas.
dontforks := false
dstvgap := mm2.vmas.FirstGap()
for srcvseg := mm.vmas.FirstSegment(); srcvseg.Ok(); srcvseg = srcvseg.NextSegment() {
vma := srcvseg.Value() // makes a copy of the vma
vmaAR := srcvseg.Range()
if vma.dontfork {
length := uint64(vmaAR.Length())
mm2.usageAS -= length
if vma.isPrivateDataLocked() {
mm2.dataAS -= length
}
dontforks = true
continue
}
// Inform the Mappable, if any, of the new mapping.
if vma.mappable != nil {
if err := vma.mappable.AddMapping(ctx, mm2, vmaAR, vma.off, vma.canWriteMappableLocked()); err != nil {
mm2.removeVMAsLocked(ctx, mm2.applicationAddrRange())
return nil, err
}
}
if vma.id != nil {
vma.id.IncRef()
}
vma.mlockMode = memmap.MLockNone
dstvgap = mm2.vmas.Insert(dstvgap, vmaAR, vma).NextGap()
// We don't need to update mm2.usageAS since we copied it from mm
// above.
}
// Copy pmas. We have to lock mm.activeMu for writing to make existing
// private pmas copy-on-write. We also have to lock mm2.activeMu since
// after copying vmas above, memmap.Mappables may call mm2.Invalidate. We
// only copy private pmas, since in the common case where fork(2) is
// immediately followed by execve(2), copying non-private pmas that can be
// regenerated by calling memmap.Mappable.Translate is a waste of time.
// (Linux does the same; compare kernel/fork.c:dup_mmap() =>
// mm/memory.c:copy_page_range().)
mm2.activeMu.Lock()
defer mm2.activeMu.Unlock()
mm.activeMu.Lock()
defer mm.activeMu.Unlock()
if dontforks {
defer mm.pmas.MergeRange(mm.applicationAddrRange())
}
srcvseg := mm.vmas.FirstSegment()
dstpgap := mm2.pmas.FirstGap()
var unmapAR usermem.AddrRange
for srcpseg := mm.pmas.FirstSegment(); srcpseg.Ok(); srcpseg = srcpseg.NextSegment() {
pma := srcpseg.ValuePtr()
if !pma.private {
continue
}
if dontforks {
// Find the 'vma' that contains the starting address
// associated with the 'pma' (there must be one).
srcvseg = srcvseg.seekNextLowerBound(srcpseg.Start())
if checkInvariants {
if !srcvseg.Ok() {
panic(fmt.Sprintf("no vma covers pma range %v", srcpseg.Range()))
}
if srcpseg.Start() < srcvseg.Start() {
panic(fmt.Sprintf("vma %v ran ahead of pma %v", srcvseg.Range(), srcpseg.Range()))
}
}
srcpseg = mm.pmas.Isolate(srcpseg, srcvseg.Range())
if srcvseg.ValuePtr().dontfork {
continue
}
pma = srcpseg.ValuePtr()
}
if !pma.needCOW {
pma.needCOW = true
if pma.effectivePerms.Write {
// We don't want to unmap the whole address space, even though
// doing so would reduce calls to unmapASLocked(), because mm
// will most likely continue to be used after the fork, so
// unmapping pmas unnecessarily will result in extra page
// faults. But we do want to merge consecutive AddrRanges
// across pma boundaries.
if unmapAR.End == srcpseg.Start() {
unmapAR.End = srcpseg.End()
} else {
if unmapAR.Length() != 0 {
mm.unmapASLocked(unmapAR)
}
unmapAR = srcpseg.Range()
}
pma.effectivePerms.Write = false
}
pma.maxPerms.Write = false
}
fr := srcpseg.fileRange()
mm2.incPrivateRef(fr)
srcpseg.ValuePtr().file.IncRef(fr)
addrRange := srcpseg.Range()
mm2.addRSSLocked(addrRange)
dstpgap = mm2.pmas.Insert(dstpgap, addrRange, *pma).NextGap()
}
if unmapAR.Length() != 0 {
mm.unmapASLocked(unmapAR)
}
// Between when we call memmap.Mappable.AddMapping while copying vmas and
// when we lock mm2.activeMu to copy pmas, calls to mm2.Invalidate() are
// ineffective because the pmas they invalidate haven't yet been copied,
// possibly allowing mm2 to get invalidated translations:
//
// Invalidating Mappable mm.Fork
// --------------------- -------
//
// mm2.Invalidate()
// mm.activeMu.Lock()
// mm.Invalidate() /* blocks */
// mm2.activeMu.Lock()
// (mm copies invalidated pma to mm2)
//
// This would technically be both safe (since we only copy private pmas,
// which will still hold a reference on their memory) and consistent with
// Linux, but we avoid it anyway by setting mm2.captureInvalidations during
// construction, causing calls to mm2.Invalidate() to be captured in
// mm2.capturedInvalidations, to be replayed after pmas are copied - i.e.
// here.
mm2.captureInvalidations = false
for _, invArgs := range mm2.capturedInvalidations {
mm2.invalidateLocked(invArgs.ar, invArgs.opts.InvalidatePrivate, true)
}
mm2.capturedInvalidations = nil
if mm2.executable != nil {
mm2.executable.IncRef()
}
return mm2, nil
}
// IncUsers increments mm's user count and returns true. If the user count is
// already 0, IncUsers does nothing and returns false.
func (mm *MemoryManager) IncUsers() bool {
for {
users := atomic.LoadInt32(&mm.users)
if users == 0 {
return false
}
if atomic.CompareAndSwapInt32(&mm.users, users, users+1) {
return true
}
}
}
// DecUsers decrements mm's user count. If the user count reaches 0, all
// mappings in mm are unmapped.
func (mm *MemoryManager) DecUsers(ctx context.Context) {
if users := atomic.AddInt32(&mm.users, -1); users > 0 {
return
} else if users < 0 {
panic(fmt.Sprintf("Invalid MemoryManager.users: %d", users))
}
mm.aioManager.destroy()
mm.metadataMu.Lock()
exe := mm.executable
mm.executable = nil
mm.metadataMu.Unlock()
if exe != nil {
exe.DecRef()
}
mm.activeMu.Lock()
// Sanity check.
if atomic.LoadInt32(&mm.active) != 0 {
panic("active address space lost?")
}
// Make sure the AddressSpace is returned.
if mm.as != nil {
mm.as.Release()
mm.as = nil
}
mm.activeMu.Unlock()
mm.mappingMu.Lock()
defer mm.mappingMu.Unlock()
// If mm is being dropped before mm.SetMmapLayout was called,
// mm.applicationAddrRange() will be empty.
if ar := mm.applicationAddrRange(); ar.Length() != 0 {
mm.unmapLocked(ctx, ar)
}
}
|