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
//! Random-access destination file (PRD §9).
//!
//! Workers write straight into the final file at the right offset. There are no
//! per-chunk temp files and no merge pass, so a 500 GB download needs 500 GB of
//! disk, not 1 TB, and finishing costs nothing.
//!
//! Positioned writes (`pwrite`) are used rather than seek+write so that
//! concurrent writers share one file descriptor without a lock and without
//! racing on the file cursor.
use std::fs::{File, OpenOptions};
use std::io;
use std::path::{Path, PathBuf};
/// Identifies the *file*, not the path. Used on resume to detect that the file
/// we recorded progress against has been replaced by a different one.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct FileIdentity {
pub dev: u64,
pub ino: u64,
}
pub struct DestFile {
file: File,
path: PathBuf,
}
impl DestFile {
/// Open (or create) the destination for random-access writing without
/// truncating: an existing partial file is exactly what we want to resume
/// into.
pub fn open(path: &Path) -> io::Result<Self> {
if let Some(parent) = path.parent() {
if !parent.as_os_str().is_empty() {
std::fs::create_dir_all(parent)?;
}
}
let file = OpenOptions::new()
.read(true)
.write(true)
.create(true)
.truncate(false)
.open(path)?;
Ok(Self {
file,
path: path.to_path_buf(),
})
}
/// Size the file up front. This turns "out of disk" into an error now
/// rather than at 90%, and gives the filesystem a chance to lay the file
/// out contiguously.
///
/// `set_len` creates a sparse file on every filesystem we target; it
/// reserves the *size*, not necessarily the *blocks*. We accept that: the
/// alternative (writing zeroes over the whole range) would double the I/O
/// for a 500 GB download.
pub fn preallocate(&self, size: u64) -> io::Result<()> {
if self.file.metadata()?.len() != size {
self.file.set_len(size)?;
}
Ok(())
}
/// Write every byte of `buf` at `offset`. Short writes are retried, so a
/// successful return means the whole buffer reached the kernel.
pub fn write_at(&self, buf: &[u8], offset: u64) -> io::Result<()> {
#[cfg(unix)]
{
use std::os::unix::fs::FileExt;
self.file.write_all_at(buf, offset)
}
#[cfg(windows)]
{
use std::os::windows::fs::FileExt;
let mut written = 0;
while written < buf.len() {
let n = self
.file
.seek_write(&buf[written..], offset + written as u64)?;
if n == 0 {
return Err(io::Error::new(
io::ErrorKind::WriteZero,
"failed to write whole buffer",
));
}
written += n;
}
Ok(())
}
}
pub fn read_at(&self, buf: &mut [u8], offset: u64) -> io::Result<usize> {
#[cfg(unix)]
{
use std::os::unix::fs::FileExt;
self.file.read_at(buf, offset)
}
#[cfg(windows)]
{
use std::os::windows::fs::FileExt;
self.file.seek_read(buf, offset)
}
}
/// The durability barrier. See `docs/CRASH_CONSISTENCY.md`: this must
/// return before any of the bytes it covers may be recorded as complete.
///
/// `sync_data` rather than `sync_all` — we need the data and the block map
/// durable, not the mtime.
pub fn sync_data(&self) -> io::Result<()> {
self.file.sync_data()
}
/// Named `size` rather than `len` because a file has no meaningful
/// `is_empty` counterpart and the pair would only mislead.
pub fn size(&self) -> io::Result<u64> {
Ok(self.file.metadata()?.len())
}
pub fn truncate(&self, size: u64) -> io::Result<()> {
self.file.set_len(size)
}
pub fn identity(&self) -> io::Result<FileIdentity> {
identity_of(&self.file)
}
pub fn path(&self) -> &Path {
&self.path
}
/// Reopen a fresh handle for hashing, so verification reads do not disturb
/// the write handle.
pub fn open_for_read(&self) -> io::Result<File> {
File::open(&self.path)
}
}
fn identity_of(file: &File) -> io::Result<FileIdentity> {
#[cfg(unix)]
{
use std::os::unix::fs::MetadataExt;
let m = file.metadata()?;
Ok(FileIdentity {
dev: m.dev(),
ino: m.ino(),
})
}
#[cfg(windows)]
{
// Windows has no stable inode in `Metadata`; fall back to a
// size+created pair, which is weaker but still catches replacement.
use std::os::windows::fs::MetadataExt;
let m = file.metadata()?;
Ok(FileIdentity {
dev: m.volume_serial_number().unwrap_or(0) as u64,
ino: m.file_index().unwrap_or(0),
})
}
}
#[cfg(test)]
mod tests {
use super::*;
fn tmpdir(name: &str) -> PathBuf {
let dir =
std::env::temp_dir().join(format!("rget-file-test-{name}-{}", std::process::id()));
std::fs::create_dir_all(&dir).unwrap();
dir
}
#[test]
fn writes_at_offsets_out_of_order() {
let dir = tmpdir("offsets");
let path = dir.join("out.bin");
let f = DestFile::open(&path).unwrap();
f.preallocate(10).unwrap();
f.write_at(b"world", 5).unwrap();
f.write_at(b"hello", 0).unwrap();
f.sync_data().unwrap();
assert_eq!(std::fs::read(&path).unwrap(), b"helloworld");
std::fs::remove_dir_all(&dir).ok();
}
#[test]
fn opening_does_not_truncate() {
let dir = tmpdir("notrunc");
let path = dir.join("keep.bin");
std::fs::write(&path, b"existing").unwrap();
let f = DestFile::open(&path).unwrap();
assert_eq!(f.size().unwrap(), 8);
drop(f);
assert_eq!(std::fs::read(&path).unwrap(), b"existing");
std::fs::remove_dir_all(&dir).ok();
}
#[test]
fn preallocate_sets_exact_size() {
let dir = tmpdir("prealloc");
let path = dir.join("big.bin");
let f = DestFile::open(&path).unwrap();
f.preallocate(1024 * 1024).unwrap();
assert_eq!(f.size().unwrap(), 1024 * 1024);
// Idempotent.
f.preallocate(1024 * 1024).unwrap();
assert_eq!(f.size().unwrap(), 1024 * 1024);
std::fs::remove_dir_all(&dir).ok();
}
#[test]
fn identity_distinguishes_files() {
let dir = tmpdir("identity");
let a = DestFile::open(&dir.join("a.bin")).unwrap();
let b = DestFile::open(&dir.join("b.bin")).unwrap();
// Two different files never share an identity...
assert_ne!(a.identity().unwrap(), b.identity().unwrap());
// ...and reopening the same file yields the same one.
let a_again = DestFile::open(&dir.join("a.bin")).unwrap();
assert_eq!(a.identity().unwrap(), a_again.identity().unwrap());
// Note: delete-then-recreate can legitimately reuse an inode on Linux,
// which is why resume also checks the recorded size, not identity alone.
std::fs::remove_dir_all(&dir).ok();
}
#[test]
fn creates_missing_parents() {
let dir = tmpdir("parents");
let path = dir.join("a/b/c/deep.bin");
let f = DestFile::open(&path).unwrap();
f.write_at(b"x", 0).unwrap();
assert!(path.exists());
std::fs::remove_dir_all(&dir).ok();
}
}