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
|
// 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 pgalloc
import (
"bytes"
"context"
"fmt"
"io"
"runtime"
"sync/atomic"
"golang.org/x/sys/unix"
"gvisor.dev/gvisor/pkg/hostarch"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sentry/usage"
"gvisor.dev/gvisor/pkg/state"
"gvisor.dev/gvisor/pkg/state/wire"
)
// SaveTo writes f's state to the given stream.
func (f *MemoryFile) SaveTo(ctx context.Context, w wire.Writer) error {
// Wait for reclaim.
f.mu.Lock()
defer f.mu.Unlock()
for f.reclaimable {
f.reclaimCond.Signal()
f.mu.Unlock()
runtime.Gosched()
f.mu.Lock()
}
// Ensure that there are no pending evictions.
if len(f.evictable) != 0 {
panic(fmt.Sprintf("evictions still pending for %d users; call StartEvictions and WaitForEvictions before SaveTo", len(f.evictable)))
}
// Ensure that all pages that contain data have knownCommitted set, since
// we only store knownCommitted pages below.
zeroPage := make([]byte, hostarch.PageSize)
err := f.updateUsageLocked(0, func(bs []byte, committed []byte) error {
for pgoff := 0; pgoff < len(bs); pgoff += hostarch.PageSize {
i := pgoff / hostarch.PageSize
pg := bs[pgoff : pgoff+hostarch.PageSize]
if !bytes.Equal(pg, zeroPage) {
committed[i] = 1
continue
}
committed[i] = 0
// Reading the page caused it to be committed; decommit it to
// reduce memory usage.
//
// "MADV_REMOVE [...] Free up a given range of pages and its
// associated backing store. This is equivalent to punching a hole
// in the corresponding byte range of the backing store (see
// fallocate(2))." - madvise(2)
if err := unix.Madvise(pg, unix.MADV_REMOVE); err != nil {
// This doesn't impact the correctness of saved memory, it
// just means that we're incrementally more likely to OOM.
// Complain, but don't abort saving.
log.Warningf("Decommitting page %p while saving failed: %v", pg, err)
}
}
return nil
})
if err != nil {
return err
}
// Save metadata.
if _, err := state.Save(ctx, w, &f.fileSize); err != nil {
return err
}
if _, err := state.Save(ctx, w, &f.usage); err != nil {
return err
}
// Dump out committed pages.
for seg := f.usage.FirstSegment(); seg.Ok(); seg = seg.NextSegment() {
if !seg.Value().knownCommitted {
continue
}
// Write a header to distinguish from objects.
if err := state.WriteHeader(w, uint64(seg.Range().Length()), false); err != nil {
return err
}
// Write out data.
var ioErr error
err := f.forEachMappingSlice(seg.Range(), func(s []byte) {
if ioErr != nil {
return
}
_, ioErr = w.Write(s)
})
if ioErr != nil {
return ioErr
}
if err != nil {
return err
}
}
return nil
}
// LoadFrom loads MemoryFile state from the given stream.
func (f *MemoryFile) LoadFrom(ctx context.Context, r wire.Reader) error {
// Load metadata.
if _, err := state.Load(ctx, r, &f.fileSize); err != nil {
return err
}
if err := f.file.Truncate(f.fileSize); err != nil {
return err
}
newMappings := make([]uintptr, f.fileSize>>chunkShift)
f.mappings.Store(newMappings)
if _, err := state.Load(ctx, r, &f.usage); err != nil {
return err
}
// Try to map committed chunks concurrently: For any given chunk, either
// this loop or the following one will mmap the chunk first and cache it in
// f.mappings for the other, but this loop is likely to run ahead of the
// other since it doesn't do any work between mmaps. The rest of this
// function doesn't mutate f.usage, so it's safe to iterate concurrently.
mapperDone := make(chan struct{})
mapperCanceled := int32(0)
go func() { // S/R-SAFE: see comment
defer func() { close(mapperDone) }()
for seg := f.usage.FirstSegment(); seg.Ok(); seg = seg.NextSegment() {
if atomic.LoadInt32(&mapperCanceled) != 0 {
return
}
if seg.Value().knownCommitted {
f.forEachMappingSlice(seg.Range(), func(s []byte) {})
}
}
}()
defer func() {
atomic.StoreInt32(&mapperCanceled, 1)
<-mapperDone
}()
// Load committed pages.
for seg := f.usage.FirstSegment(); seg.Ok(); seg = seg.NextSegment() {
if !seg.Value().knownCommitted {
continue
}
// Verify header.
length, object, err := state.ReadHeader(r)
if err != nil {
return err
}
if object {
// Not expected.
return fmt.Errorf("unexpected object")
}
if expected := uint64(seg.Range().Length()); length != expected {
// Size mismatch.
return fmt.Errorf("mismatched segment: expected %d, got %d", expected, length)
}
// Read data.
var ioErr error
err = f.forEachMappingSlice(seg.Range(), func(s []byte) {
if ioErr != nil {
return
}
_, ioErr = io.ReadFull(r, s)
})
if ioErr != nil {
return ioErr
}
if err != nil {
return err
}
// Update accounting for restored pages. We need to do this here since
// these segments are marked as "known committed", and will be skipped
// over on accounting scans.
usage.MemoryAccounting.Inc(seg.End()-seg.Start(), seg.Value().kind)
}
return nil
}
// MemoryFileProvider provides the MemoryFile method.
//
// This type exists to work around a save/restore defect. The only object in a
// saved object graph that S/R allows to be replaced at time of restore is the
// starting point of the restore, kernel.Kernel. However, the MemoryFile
// changes between save and restore as well, so objects that need persistent
// access to the MemoryFile must instead store a pointer to the Kernel and call
// Kernel.MemoryFile() as required. In most cases, depending on the kernel
// package directly would create a package dependency loop, so the stored
// pointer must instead be a MemoryProvider interface object. Correspondingly,
// kernel.Kernel is the only implementation of this interface.
type MemoryFileProvider interface {
// MemoryFile returns the Kernel MemoryFile.
MemoryFile() *MemoryFile
}
|
