Files
rustfs/crates/utils/src/os/linux.rs
T
houseme 5d74637968 fix(utils): tolerate bavail greater than bfree on Linux (#3119)
* fix(utils): tolerate bavail greater than bfree on Linux

Treat f_bavail > f_bfree as a compatibility edge case instead of fatal corruption during startup disk info probing.

- keep corruption checks for other invalid counter relationships
- log a warning and clamp reserved blocks to 0 when bavail exceeds bfree
- add linux unit coverage for the compatibility and overflow paths

Refs #3025

* fix(utils): cap bavail to bfree for Linux statfs edge

* fix(utils): rate-limit Linux statfs compatibility warning

* test(utils): cover Linux statfs capacity math

Expand Linux statfs capacity math tests around normal, equal, zero, and fully-free block counter combinations.

Refs #3025

* fix(utils): avoid statfs warning guard reallocations

* test(utils): align statfs reserved block expectations

* fmt

* fix(utils): avoid warning guard path allocation

Check the warn-once set before allocating a PathBuf so repeated bavail/bfree compatibility warnings stay low-overhead.

Refs #3025
2026-05-29 11:20:47 +00:00

518 lines
17 KiB
Rust

// Copyright 2024 RustFS Team
//
// 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.
use super::{DiskInfo, IOStats, fs_type::get_fs_type};
use rustix::fs::statfs;
use std::collections::BTreeSet;
use std::fs;
use std::fs::File;
use std::io::{self, BufRead, Error, ErrorKind, Read};
use std::path::{Path, PathBuf};
use std::sync::{Mutex, OnceLock};
use tracing::warn;
static BAVAIL_GT_BFREE_WARNING_PATHS: OnceLock<Mutex<BTreeSet<PathBuf>>> = OnceLock::new();
/// Returns total and free bytes available in a directory, e.g. `/`.
pub fn get_info(p: impl AsRef<Path>) -> std::io::Result<DiskInfo> {
// Use statfs on Linux to get access to f_type (filesystem magic number)
let stat = statfs(p.as_ref())?;
// Linux statfs:
// f_bsize: Optimal transfer block size
// f_blocks: Total data blocks in file system
// f_frsize: Fragment size (since Linux 2.6) - unit for blocks
//
// If f_frsize is > 0, it is the unit for f_blocks, f_bfree, f_bavail.
// Otherwise f_bsize is used.
let bsize = if stat.f_frsize > 0 {
stat.f_frsize as u64
} else {
stat.f_bsize as u64
};
let bfree = stat.f_bfree as u64;
let bavail = stat.f_bavail as u64;
let blocks = stat.f_blocks as u64;
let (total, free, used) = calculate_space_usage(blocks, bfree, bavail, bsize, p.as_ref())?;
let st = rustix::fs::stat(p.as_ref())?;
Ok(DiskInfo {
total,
free,
used,
files: stat.f_files as u64,
ffree: stat.f_ffree as u64,
fstype: get_fs_type(stat.f_type as u64).to_string(),
major: rustix::fs::major(st.st_dev) as u64,
minor: rustix::fs::minor(st.st_dev) as u64,
..Default::default()
})
}
fn calculate_space_usage(blocks: u64, bfree: u64, bavail: u64, bsize: u64, path: &Path) -> std::io::Result<(u64, u64, u64)> {
let available = if bfree < bavail {
if should_warn_bavail_greater_than_bfree(path) {
warn!(
path = %path.display(),
f_bfree = bfree,
f_bavail = bavail,
"detected f_bavail greater than f_bfree, capping available blocks to f_bfree for compatibility"
);
}
bfree
} else {
bavail
};
let reserved = bfree - available;
let total = match blocks.checked_sub(reserved) {
Some(total) => total * bsize,
None => {
return Err(Error::other(format!(
"detected reserved space ({reserved}) > blocks space ({blocks}), fs corruption at ({}). please run 'fsck'",
path.display(),
)));
}
};
let free = available * bsize;
let used = match total.checked_sub(free) {
Some(used) => used,
None => {
return Err(Error::other(format!(
"detected free space ({free}) > total drive space ({total}), fs corruption at ({}). please run 'fsck'",
path.display(),
)));
}
};
Ok((total, free, used))
}
fn should_warn_bavail_greater_than_bfree(path: &Path) -> bool {
let warned_paths = BAVAIL_GT_BFREE_WARNING_PATHS.get_or_init(|| Mutex::new(BTreeSet::new()));
let mut warned_paths = match warned_paths.lock() {
Ok(guard) => guard,
Err(poisoned) => poisoned.into_inner(),
};
if warned_paths.contains(path) {
false
} else {
warned_paths.insert(path.to_path_buf())
}
}
pub fn same_disk(disk1: &str, disk2: &str) -> std::io::Result<bool> {
let stat1 = rustix::fs::stat(disk1)?;
let stat2 = rustix::fs::stat(disk2)?;
Ok(stat1.st_dev == stat2.st_dev)
}
/// Resolve the leaf physical device identities backing a local filesystem path.
///
/// Linux block stacks such as partitions, `dm-*`, or software RAID can all
/// expose a filesystem through an intermediate device node. This helper walks
/// sysfs until it reaches the leaf backing devices so the caller can compare
/// physical failure domains instead of only filesystem device numbers.
pub fn get_physical_device_ids(disk: &str) -> std::io::Result<Vec<String>> {
let stat = rustix::fs::stat(disk)?;
let major = rustix::fs::major(stat.st_dev) as u64;
let minor = rustix::fs::minor(stat.st_dev) as u64;
let devices = resolve_block_device_ids(major, minor)?;
Ok(devices.into_iter().collect())
}
fn resolve_block_device_ids(major: u64, minor: u64) -> std::io::Result<BTreeSet<String>> {
let sysfs_path = PathBuf::from(format!("/sys/dev/block/{major}:{minor}"));
let resolved = match fs::canonicalize(&sysfs_path) {
Ok(path) => path,
Err(err) if err.kind() == ErrorKind::NotFound => {
return Ok(BTreeSet::from([format!("{major}:{minor}")]));
}
Err(err) => return Err(err),
};
let devices = collect_block_device_ids(&resolved)?;
if devices.is_empty() {
Ok(BTreeSet::from([format!("{major}:{minor}")]))
} else {
Ok(devices)
}
}
fn collect_block_device_ids(device_path: &Path) -> std::io::Result<BTreeSet<String>> {
let mut ids = BTreeSet::new();
let slaves_dir = device_path.join("slaves");
match fs::read_dir(&slaves_dir) {
Ok(entries) => {
let mut found_slave = false;
for entry in entries {
let entry = entry?;
found_slave = true;
let resolved = fs::canonicalize(entry.path())?;
ids.extend(collect_block_device_ids(&resolved)?);
}
if found_slave {
return Ok(ids);
}
}
Err(err) if err.kind() == ErrorKind::NotFound => {}
Err(err) => return Err(err),
}
ids.insert(normalize_block_device_name(device_path));
Ok(ids)
}
fn normalize_block_device_name(device_path: &Path) -> String {
if device_path.join("partition").exists()
&& let Some(parent_name) = device_path.parent().and_then(|parent| parent.file_name())
{
return parent_name.to_string_lossy().into_owned();
}
device_path
.file_name()
.map(|name| name.to_string_lossy().into_owned())
.unwrap_or_else(|| device_path.display().to_string())
}
/// Check whether any configured export path contains nested mount points.
///
/// This mirrors the intent of MinIO's cross-device mount guardrail: once an
/// export path is selected, RustFS should not silently traverse into child
/// mount points hosted by other devices.
pub fn check_cross_device_mounts(paths: &[String]) -> std::io::Result<()> {
check_cross_device_mounts_with_reader(paths, File::open("/proc/mounts")?)
}
/// Parse `/proc/mounts`-style content and validate each export path against it.
fn check_cross_device_mounts_with_reader(paths: &[String], mut reader: impl Read) -> std::io::Result<()> {
let mut content = String::new();
reader.read_to_string(&mut content)?;
let mount_paths = parse_mount_paths(&content);
for path in paths {
ensure_no_sub_mounts(path, &mount_paths)?;
}
Ok(())
}
/// Extract mount paths from `/proc/mounts` content, decoding escaped spaces.
fn parse_mount_paths(content: &str) -> Vec<String> {
content
.lines()
.filter_map(|line| {
let fields = line.split_whitespace().collect::<Vec<_>>();
if fields.len() != 6 {
return None;
}
Some(fields[1].replace("\\040", " "))
})
.collect()
}
/// Validate that `path` does not contain nested child mount points.
fn ensure_no_sub_mounts(path: &str, mount_paths: &[String]) -> std::io::Result<()> {
if !Path::new(path).is_absolute() {
return Err(Error::new(
ErrorKind::InvalidInput,
format!("Invalid argument, path ({path}) is expected to be absolute"),
));
}
if path == "/" {
return Err(Error::new(
ErrorKind::InvalidInput,
"Invalid argument, path (/) cannot be the filesystem root for export validation",
));
}
let base = normalize_mount_path(path);
let mut cross_mounts = Vec::new();
for mount_path in mount_paths {
let mount_base = normalize_mount_path(mount_path);
if mount_base.starts_with(&base) && mount_base != base {
cross_mounts.push(mount_path.clone());
}
}
if cross_mounts.is_empty() {
return Ok(());
}
cross_mounts.sort();
cross_mounts.dedup();
Err(Error::other(format!(
"Nested mount points detected under path ({path}) at the following locations: {}. Export path should not have any sub-mounts, refusing to start.",
cross_mounts.join(", ")
)))
}
/// Normalize mount paths so prefix checks treat `/a/b` and `/a/b/` identically.
fn normalize_mount_path(path: &str) -> String {
let trimmed = path.trim_end_matches('/');
if trimmed.is_empty() {
"/".to_string()
} else {
format!("{trimmed}/")
}
}
pub fn get_drive_stats(major: u32, minor: u32) -> std::io::Result<IOStats> {
read_drive_stats(&format!("/sys/dev/block/{major}:{minor}/stat"))
}
fn read_drive_stats(stats_file: &str) -> std::io::Result<IOStats> {
let stats = read_stat(stats_file)?;
if stats.len() < 11 {
return Err(Error::new(
ErrorKind::InvalidData,
format!("found invalid format while reading {stats_file}"),
));
}
let mut io_stats = IOStats {
read_ios: stats[0],
read_merges: stats[1],
read_sectors: stats[2],
read_ticks: stats[3],
write_ios: stats[4],
write_merges: stats[5],
write_sectors: stats[6],
write_ticks: stats[7],
current_ios: stats[8],
total_ticks: stats[9],
req_ticks: stats[10],
..Default::default()
};
if stats.len() > 14 {
io_stats.discard_ios = stats[11];
io_stats.discard_merges = stats[12];
io_stats.discard_sectors = stats[13];
io_stats.discard_ticks = stats[14];
}
Ok(io_stats)
}
fn read_stat(file_name: &str) -> std::io::Result<Vec<u64>> {
// Open file
let path = Path::new(file_name);
let file = File::open(path)?;
// Create a BufReader
let reader = io::BufReader::new(file);
// Read first line
let mut stats = Vec::new();
if let Some(line) = reader.lines().next() {
let line = line?;
// Split line and parse as u64
// https://rust-lang.github.io/rust-clippy/master/index.html#trim_split_whitespace
for token in line.split_whitespace() {
let ui64: u64 = token
.parse()
.map_err(|e| Error::new(ErrorKind::InvalidData, format!("failed to parse '{token}' as u64: {e}")))?;
stats.push(ui64);
}
}
Ok(stats)
}
#[cfg(test)]
mod tests {
use super::*;
use tempfile::tempdir;
#[test]
fn normalize_partition_device_to_parent_disk() {
let dir = tempdir().unwrap();
let block = dir.path().join("block");
let disk = block.join("nvme0n1");
let partition = disk.join("nvme0n1p1");
let slaves = partition.join("slaves");
fs::create_dir_all(&slaves).unwrap();
fs::write(partition.join("partition"), "1").unwrap();
let ids = collect_block_device_ids(&partition).unwrap();
assert_eq!(ids.into_iter().collect::<Vec<_>>(), vec!["nvme0n1".to_string()]);
}
#[test]
fn flatten_device_mapper_slaves_to_leaf_devices() {
let dir = tempdir().unwrap();
let block = dir.path().join("block");
let dm = block.join("dm-0");
let dm_slaves = dm.join("slaves");
let nvme0 = block.join("nvme0n1");
let nvme1 = block.join("nvme1n1");
fs::create_dir_all(&dm_slaves).unwrap();
fs::create_dir_all(&nvme0).unwrap();
fs::create_dir_all(&nvme1).unwrap();
#[cfg(unix)]
{
std::os::unix::fs::symlink(&nvme0, dm_slaves.join("nvme0n1")).unwrap();
std::os::unix::fs::symlink(&nvme1, dm_slaves.join("nvme1n1")).unwrap();
}
let ids = collect_block_device_ids(&dm).unwrap();
assert_eq!(ids.into_iter().collect::<Vec<_>>(), vec!["nvme0n1".to_string(), "nvme1n1".to_string()]);
}
#[test]
fn detect_cross_device_sub_mounts() {
let mounts = "\
/dev/root / ext4 rw 0 0
/dev/sdb1 /data ext4 rw 0 0
/dev/sdc1 /data/disk1/sub ext4 rw 0 0
";
let err = check_cross_device_mounts_with_reader(&["/data/disk1".to_string()], mounts.as_bytes()).unwrap_err();
assert!(err.to_string().contains("Nested mount points detected under path"));
assert!(err.to_string().contains("/data/disk1/sub"));
}
#[test]
fn allow_mount_path_without_sub_mounts() {
let mounts = "\
/dev/root / ext4 rw 0 0
/dev/sdb1 /data/disk1 ext4 rw 0 0
";
check_cross_device_mounts_with_reader(&["/data/disk1".to_string()], mounts.as_bytes()).unwrap();
}
#[test]
fn parse_mount_paths_decodes_escaped_spaces() {
let mounts = "/dev/sdb1 /data/my\\040disk ext4 rw 0 0\n";
let paths = parse_mount_paths(mounts);
assert_eq!(paths, vec!["/data/my disk".to_string()]);
}
#[test]
fn reject_relative_path_for_cross_device_validation() {
let err = ensure_no_sub_mounts("relative/path", &[]).unwrap_err();
assert_eq!(err.kind(), ErrorKind::InvalidInput);
assert!(err.to_string().contains("expected to be absolute"));
}
#[test]
fn reject_root_path_for_cross_device_validation() {
let err = ensure_no_sub_mounts("/", &[]).unwrap_err();
assert_eq!(err.kind(), ErrorKind::InvalidInput);
assert!(err.to_string().contains("cannot be the filesystem root"));
}
#[test]
fn fallback_to_major_minor_when_sysfs_link_missing() {
let major = u64::MAX;
let minor = u64::MAX;
let ids = resolve_block_device_ids(major, minor).unwrap();
assert_eq!(ids.into_iter().collect::<Vec<_>>(), vec![format!("{major}:{minor}")]);
}
#[test]
fn calculate_space_usage_normal_bfree_greater_than_bavail() {
// Typical ext4/xfs scenario: bavail < bfree due to reserved blocks
let blocks = 1_000_u64;
let bfree = 900_u64;
let bavail = 850_u64;
let bsize = 4_096_u64;
let (total, free, used) = calculate_space_usage(blocks, bfree, bavail, bsize, Path::new("/data")).unwrap();
let reserved = bfree - bavail;
assert_eq!(total, (blocks - reserved) * bsize);
assert_eq!(free, bavail * bsize);
assert_eq!(used, total - free);
}
#[test]
fn calculate_space_usage_bfree_equals_bavail() {
// No reserved blocks: bfree == bavail (e.g. FAT/exFAT or root user)
let blocks = 500_u64;
let bfree = 400_u64;
let bavail = 400_u64;
let bsize = 4_096_u64;
let (total, free, used) = calculate_space_usage(blocks, bfree, bavail, bsize, Path::new("/data")).unwrap();
assert_eq!(total, blocks * bsize);
assert_eq!(free, bfree * bsize);
assert_eq!(used, (blocks - bfree) * bsize);
}
#[test]
fn calculate_space_usage_all_zero_blocks() {
let (total, free, used) = calculate_space_usage(0, 0, 0, 4_096, Path::new("/data")).unwrap();
assert_eq!(total, 0);
assert_eq!(free, 0);
assert_eq!(used, 0);
}
#[test]
fn calculate_space_usage_all_blocks_free() {
let blocks = 1_000_u64;
let bfree = 1_000_u64;
let bavail = 1_000_u64;
let bsize = 4_096_u64;
let (total, free, used) = calculate_space_usage(blocks, bfree, bavail, bsize, Path::new("/data")).unwrap();
assert_eq!(total, blocks * bsize);
assert_eq!(free, blocks * bsize);
assert_eq!(used, 0);
}
#[test]
fn calculate_space_usage_allows_bavail_greater_than_bfree() {
let blocks = 1_000_u64;
let bfree = 900_u64;
let bavail = 920_u64;
let bsize = 4_096_u64;
let (total, free, used) = calculate_space_usage(blocks, bfree, bavail, bsize, Path::new("/data")).unwrap();
assert_eq!(total, blocks * bsize);
assert_eq!(free, bfree * bsize);
assert_eq!(used, total - free);
}
#[test]
fn calculate_space_usage_rejects_free_greater_than_total() {
let err = calculate_space_usage(100, 120, 120, 4_096, Path::new("/data")).unwrap_err();
assert!(err.to_string().contains("detected free space"));
}
#[test]
fn bavail_greater_than_bfree_warning_is_once_per_path() {
let path = Path::new("/data/rustfs-bavail-warning-once");
assert!(should_warn_bavail_greater_than_bfree(path));
assert!(!should_warn_bavail_greater_than_bfree(path));
assert!(should_warn_bavail_greater_than_bfree(Path::new("/data/rustfs-bavail-warning-other")));
}
}