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
|
// Copyright 2020 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package poll
import (
"internal/syscall/unix"
"sync/atomic"
"syscall"
)
var copyFileRangeSupported int32 = 1 // accessed atomically
const maxCopyFileRangeRound = 1 << 30
// CopyFileRange copies at most remain bytes of data from src to dst, using
// the copy_file_range system call. dst and src must refer to regular files.
func CopyFileRange(dst, src *FD, remain int64) (written int64, handled bool, err error) {
if atomic.LoadInt32(©FileRangeSupported) == 0 {
return 0, false, nil
}
for remain > 0 {
max := remain
if max > maxCopyFileRangeRound {
max = maxCopyFileRangeRound
}
n, err := copyFileRange(dst, src, int(max))
switch err {
case syscall.ENOSYS:
// copy_file_range(2) was introduced in Linux 4.5.
// Go supports Linux >= 2.6.33, so the system call
// may not be present.
//
// If we see ENOSYS, we have certainly not transfered
// any data, so we can tell the caller that we
// couldn't handle the transfer and let them fall
// back to more generic code.
//
// Seeing ENOSYS also means that we will not try to
// use copy_file_range(2) again.
atomic.StoreInt32(©FileRangeSupported, 0)
return 0, false, nil
case syscall.EXDEV, syscall.EINVAL, syscall.EIO, syscall.EOPNOTSUPP, syscall.EPERM:
// Prior to Linux 5.3, it was not possible to
// copy_file_range across file systems. Similarly to
// the ENOSYS case above, if we see EXDEV, we have
// not transfered any data, and we can let the caller
// fall back to generic code.
//
// As for EINVAL, that is what we see if, for example,
// dst or src refer to a pipe rather than a regular
// file. This is another case where no data has been
// transfered, so we consider it unhandled.
//
// If src and dst are on CIFS, we can see EIO.
// See issue #42334.
//
// If the file is on NFS, we can see EOPNOTSUPP.
// See issue #40731.
//
// If the process is running inside a Docker container,
// we might see EPERM instead of ENOSYS. See issue
// #40893. Since EPERM might also be a legitimate error,
// don't mark copy_file_range(2) as unsupported.
return 0, false, nil
case nil:
if n == 0 {
// src is at EOF, which means we are done.
return written, true, nil
}
remain -= n
written += n
default:
return written, true, err
}
}
return written, true, nil
}
// copyFileRange performs one round of copy_file_range(2).
func copyFileRange(dst, src *FD, max int) (written int64, err error) {
// The signature of copy_file_range(2) is:
//
// ssize_t copy_file_range(int fd_in, loff_t *off_in,
// int fd_out, loff_t *off_out,
// size_t len, unsigned int flags);
//
// Note that in the call to unix.CopyFileRange below, we use nil
// values for off_in and off_out. For the system call, this means
// "use and update the file offsets". That is why we must acquire
// locks for both file descriptors (and why this whole machinery is
// in the internal/poll package to begin with).
if err := dst.writeLock(); err != nil {
return 0, err
}
defer dst.writeUnlock()
if err := src.readLock(); err != nil {
return 0, err
}
defer src.readUnlock()
var n int
for {
n, err = unix.CopyFileRange(src.Sysfd, nil, dst.Sysfd, nil, max, 0)
if err != syscall.EINTR {
break
}
}
return int64(n), err
}
|
