proxmox-backup/src/tools.rs

612 lines
18 KiB
Rust
Raw Normal View History

2019-01-05 15:53:28 +00:00
//! Tools and utilities
//!
//! This is a collection of small and useful tools.
use std::any::Any;
use std::collections::HashMap;
2019-10-26 09:36:01 +00:00
use std::hash::BuildHasher;
use std::fs::File;
2020-09-17 08:27:04 +00:00
use std::io::{self, BufRead, ErrorKind, Read, Seek, SeekFrom};
use std::os::unix::io::RawFd;
use std::path::Path;
2018-12-19 09:02:24 +00:00
use anyhow::{bail, format_err, Error};
2019-03-03 08:52:06 +00:00
use serde_json::Value;
use openssl::hash::{hash, DigestBytes, MessageDigest};
use percent_encoding::AsciiSet;
use proxmox::tools::vec;
pub use proxmox::tools::fd::Fd;
pub mod acl;
pub mod async_io;
pub mod borrow;
pub mod cert;
pub mod daemon;
pub mod disks;
pub mod fs;
pub mod format;
pub mod lru_cache;
pub mod runtime;
pub mod ticket;
pub mod statistics;
pub mod systemd;
pub mod nom;
pub mod logrotate;
pub mod loopdev;
pub mod fuse_loop;
mod parallel_handler;
pub use parallel_handler::*;
mod wrapped_reader_stream;
pub use wrapped_reader_stream::*;
mod std_channel_writer;
pub use std_channel_writer::*;
pub mod xattr;
mod process_locker;
pub use process_locker::*;
mod file_logger;
pub use file_logger::*;
mod broadcast_future;
pub use broadcast_future::*;
/// The `BufferedRead` trait provides a single function
2019-01-05 16:28:20 +00:00
/// `buffered_read`. It returns a reference to an internal buffer. The
/// purpose of this traid is to avoid unnecessary data copies.
pub trait BufferedRead {
/// This functions tries to fill the internal buffers, then
/// returns a reference to the available data. It returns an empty
/// buffer if `offset` points to the end of the file.
2019-01-05 16:28:20 +00:00
fn buffered_read(&mut self, offset: u64) -> Result<&[u8], Error>;
}
2019-01-05 15:53:28 +00:00
/// Split a file into equal sized chunks. The last chunk may be
/// smaller. Note: We cannot implement an `Iterator`, because iterators
/// cannot return a borrowed buffer ref (we want zero-copy)
pub fn file_chunker<C, R>(mut file: R, chunk_size: usize, mut chunk_cb: C) -> Result<(), Error>
where
C: FnMut(usize, &[u8]) -> Result<bool, Error>,
R: Read,
2018-12-15 10:14:41 +00:00
{
const READ_BUFFER_SIZE: usize = 4 * 1024 * 1024; // 4M
2018-12-15 10:14:41 +00:00
if chunk_size > READ_BUFFER_SIZE {
bail!("chunk size too large!");
}
2018-12-15 10:14:41 +00:00
let mut buf = vec::undefined(READ_BUFFER_SIZE);
2018-12-15 10:14:41 +00:00
let mut pos = 0;
let mut file_pos = 0;
loop {
let mut eof = false;
let mut tmp = &mut buf[..];
// try to read large portions, at least chunk_size
2018-12-15 10:14:41 +00:00
while pos < chunk_size {
match file.read(tmp) {
Ok(0) => {
eof = true;
break;
}
2018-12-15 10:14:41 +00:00
Ok(n) => {
pos += n;
if pos > chunk_size {
break;
}
2018-12-15 10:14:41 +00:00
tmp = &mut tmp[n..];
}
Err(ref e) if e.kind() == ErrorKind::Interrupted => { /* try again */ }
2018-12-15 10:16:27 +00:00
Err(e) => bail!("read chunk failed - {}", e.to_string()),
2018-12-15 10:14:41 +00:00
}
}
let mut start = 0;
while start + chunk_size <= pos {
if !(chunk_cb)(file_pos, &buf[start..start + chunk_size])? {
break;
}
2018-12-15 10:14:41 +00:00
file_pos += chunk_size;
start += chunk_size;
}
if eof {
if start < pos {
(chunk_cb)(file_pos, &buf[start..pos])?;
//file_pos += pos - start;
}
break;
} else {
let rest = pos - start;
if rest > 0 {
let ptr = buf.as_mut_ptr();
unsafe {
std::ptr::copy_nonoverlapping(ptr.add(start), ptr, rest);
}
2018-12-15 10:14:41 +00:00
pos = rest;
} else {
pos = 0;
}
}
}
Ok(())
}
pub fn json_object_to_query(data: Value) -> Result<String, Error> {
let mut query = url::form_urlencoded::Serializer::new(String::new());
let object = data.as_object().ok_or_else(|| {
format_err!("json_object_to_query: got wrong data type (expected object).")
})?;
for (key, value) in object {
match value {
Value::Bool(b) => {
query.append_pair(key, &b.to_string());
}
Value::Number(n) => {
query.append_pair(key, &n.to_string());
}
Value::String(s) => {
query.append_pair(key, &s);
}
Value::Array(arr) => {
for element in arr {
match element {
Value::Bool(b) => {
query.append_pair(key, &b.to_string());
}
Value::Number(n) => {
query.append_pair(key, &n.to_string());
}
Value::String(s) => {
query.append_pair(key, &s);
}
_ => bail!(
"json_object_to_query: unable to handle complex array data types."
),
}
}
}
_ => bail!("json_object_to_query: unable to handle complex data types."),
}
}
Ok(query.finish())
}
pub fn required_string_param<'a>(param: &'a Value, name: &str) -> Result<&'a str, Error> {
match param[name].as_str() {
Some(s) => Ok(s),
None => bail!("missing parameter '{}'", name),
}
}
2019-01-18 09:13:45 +00:00
pub fn required_string_property<'a>(param: &'a Value, name: &str) -> Result<&'a str, Error> {
match param[name].as_str() {
Some(s) => Ok(s),
None => bail!("missing property '{}'", name),
}
}
2019-01-18 09:13:45 +00:00
pub fn required_integer_param<'a>(param: &'a Value, name: &str) -> Result<i64, Error> {
match param[name].as_i64() {
2019-01-18 09:13:45 +00:00
Some(s) => Ok(s),
None => bail!("missing parameter '{}'", name),
}
}
pub fn required_integer_property<'a>(param: &'a Value, name: &str) -> Result<i64, Error> {
match param[name].as_i64() {
Some(s) => Ok(s),
None => bail!("missing property '{}'", name),
}
}
pub fn required_array_param<'a>(param: &'a Value, name: &str) -> Result<Vec<Value>, Error> {
match param[name].as_array() {
Some(s) => Ok(s.to_vec()),
None => bail!("missing parameter '{}'", name),
2019-01-18 09:13:45 +00:00
}
}
pub fn required_array_property<'a>(param: &'a Value, name: &str) -> Result<Vec<Value>, Error> {
match param[name].as_array() {
Some(s) => Ok(s.to_vec()),
None => bail!("missing property '{}'", name),
}
}
2019-10-26 09:36:01 +00:00
pub fn complete_file_name<S: BuildHasher>(arg: &str, _param: &HashMap<String, String, S>) -> Vec<String> {
let mut result = vec![];
use nix::fcntl::AtFlags;
use nix::fcntl::OFlag;
use nix::sys::stat::Mode;
2019-10-26 09:36:01 +00:00
let mut dirname = std::path::PathBuf::from(if arg.is_empty() { "./" } else { arg });
let is_dir = match nix::sys::stat::fstatat(libc::AT_FDCWD, &dirname, AtFlags::empty()) {
Ok(stat) => (stat.st_mode & libc::S_IFMT) == libc::S_IFDIR,
Err(_) => false,
};
if !is_dir {
if let Some(parent) = dirname.parent() {
dirname = parent.to_owned();
}
}
let mut dir =
match nix::dir::Dir::openat(libc::AT_FDCWD, &dirname, OFlag::O_DIRECTORY, Mode::empty()) {
Ok(d) => d,
Err(_) => return result,
};
for item in dir.iter() {
if let Ok(entry) = item {
if let Ok(name) = entry.file_name().to_str() {
if name == "." || name == ".." {
continue;
}
let mut newpath = dirname.clone();
newpath.push(name);
if let Ok(stat) =
nix::sys::stat::fstatat(libc::AT_FDCWD, &newpath, AtFlags::empty())
{
if (stat.st_mode & libc::S_IFMT) == libc::S_IFDIR {
newpath.push("");
if let Some(newpath) = newpath.to_str() {
result.push(newpath.to_owned());
}
continue;
}
}
if let Some(newpath) = newpath.to_str() {
result.push(newpath.to_owned());
}
}
}
}
result
}
2019-01-20 16:31:43 +00:00
/// Scan directory for matching file names.
///
/// Scan through all directory entries and call `callback()` function
/// if the entry name matches the regular expression. This function
/// used unix `openat()`, so you can pass absolute or relative file
/// names. This function simply skips non-UTF8 encoded names.
pub fn scandir<P, F>(
dirfd: RawFd,
path: &P,
2019-01-20 16:31:43 +00:00
regex: &regex::Regex,
mut callback: F,
2019-01-20 16:31:43 +00:00
) -> Result<(), Error>
where
F: FnMut(RawFd, &str, nix::dir::Type) -> Result<(), Error>,
P: ?Sized + nix::NixPath,
2019-01-20 16:31:43 +00:00
{
for entry in self::fs::scan_subdir(dirfd, path, regex)? {
2019-01-20 16:31:43 +00:00
let entry = entry?;
let file_type = match entry.file_type() {
Some(file_type) => file_type,
None => bail!("unable to detect file type"),
};
callback(
entry.parent_fd(),
unsafe { entry.file_name_utf8_unchecked() },
file_type,
)?;
2019-01-20 16:31:43 +00:00
}
Ok(())
}
/// Shortcut for md5 sums.
pub fn md5sum(data: &[u8]) -> Result<DigestBytes, Error> {
hash(MessageDigest::md5(), data).map_err(Error::from)
}
pub fn get_hardware_address() -> Result<String, Error> {
static FILENAME: &str = "/etc/ssh/ssh_host_rsa_key.pub";
let contents = proxmox::tools::fs::file_get_contents(FILENAME)?;
let digest = md5sum(&contents)?;
Ok(proxmox::tools::bin_to_hex(&digest))
}
pub fn assert_if_modified(digest1: &str, digest2: &str) -> Result<(), Error> {
if digest1 != digest2 {
bail!("detected modified configuration - file changed by other user? Try again.");
}
Ok(())
}
2019-01-31 11:22:00 +00:00
/// Extract a specific cookie from cookie header.
2019-01-31 11:22:00 +00:00
/// We assume cookie_name is already url encoded.
pub fn extract_cookie(cookie: &str, cookie_name: &str) -> Option<String> {
2019-01-31 11:22:00 +00:00
for pair in cookie.split(';') {
let (name, value) = match pair.find('=') {
Some(i) => (pair[..i].trim(), pair[(i + 1)..].trim()),
None => return None, // Cookie format error
};
if name == cookie_name {
use percent_encoding::percent_decode;
2019-01-31 11:22:00 +00:00
if let Ok(value) = percent_decode(value.as_bytes()).decode_utf8() {
return Some(value.into());
} else {
return None; // Cookie format error
}
}
}
None
}
2019-03-02 15:12:34 +00:00
pub fn join(data: &Vec<String>, sep: char) -> String {
let mut list = String::new();
for item in data {
2019-10-26 09:36:01 +00:00
if !list.is_empty() {
list.push(sep);
}
2019-03-02 15:12:34 +00:00
list.push_str(item);
}
list
}
2020-01-15 11:27:05 +00:00
/// Detect modified configuration files
///
/// This function fails with a reasonable error message if checksums do not match.
2020-01-15 11:27:05 +00:00
pub fn detect_modified_configuration_file(digest1: &[u8;32], digest2: &[u8;32]) -> Result<(), Error> {
if digest1 != digest2 {
bail!("detected modified configuration - file changed by other user? Try again.");
}
Ok(())
}
/// normalize uri path
///
/// Do not allow ".", "..", or hidden files ".XXXX"
/// Also remove empty path components
pub fn normalize_uri_path(path: &str) -> Result<(String, Vec<&str>), Error> {
let items = path.split('/');
let mut path = String::new();
let mut components = vec![];
for name in items {
if name.is_empty() {
continue;
}
2019-10-26 09:36:01 +00:00
if name.starts_with('.') {
bail!("Path contains illegal components.");
}
path.push('/');
path.push_str(name);
components.push(name);
}
Ok((path, components))
}
/// Helper to check result from std::process::Command output
///
/// The exit_code_check() function should return true if the exit code
/// is considered successful.
pub fn command_output(
output: std::process::Output,
exit_code_check: Option<fn(i32) -> bool>,
) -> Result<Vec<u8>, Error> {
if !output.status.success() {
match output.status.code() {
Some(code) => {
let is_ok = match exit_code_check {
Some(check_fn) => check_fn(code),
None => code == 0,
};
if !is_ok {
let msg = String::from_utf8(output.stderr)
.map(|m| if m.is_empty() { String::from("no error message") } else { m })
.unwrap_or_else(|_| String::from("non utf8 error message (suppressed)"));
bail!("status code: {} - {}", code, msg);
}
}
None => bail!("terminated by signal"),
}
}
Ok(output.stdout)
}
/// Helper to check result from std::process::Command output, returns String.
///
/// The exit_code_check() function should return true if the exit code
/// is considered successful.
pub fn command_output_as_string(
output: std::process::Output,
exit_code_check: Option<fn(i32) -> bool>,
) -> Result<String, Error> {
let output = command_output(output, exit_code_check)?;
let output = String::from_utf8(output)?;
Ok(output)
}
pub fn run_command(
mut command: std::process::Command,
exit_code_check: Option<fn(i32) -> bool>,
) -> Result<String, Error> {
let output = command.output()
.map_err(|err| format_err!("failed to execute {:?} - {}", command, err))?;
let output = crate::tools::command_output_as_string(output, exit_code_check)
.map_err(|err| format_err!("command {:?} failed - {}", command, err))?;
Ok(output)
}
pub fn fd_change_cloexec(fd: RawFd, on: bool) -> Result<(), Error> {
use nix::fcntl::{fcntl, FdFlag, F_GETFD, F_SETFD};
let mut flags = FdFlag::from_bits(fcntl(fd, F_GETFD)?)
.ok_or_else(|| format_err!("unhandled file flags"))?; // nix crate is stupid this way...
flags.set(FdFlag::FD_CLOEXEC, on);
fcntl(fd, F_SETFD(flags))?;
Ok(())
}
static mut SHUTDOWN_REQUESTED: bool = false;
pub fn request_shutdown() {
unsafe {
SHUTDOWN_REQUESTED = true;
}
crate::server::server_shutdown();
}
#[inline(always)]
pub fn shutdown_requested() -> bool {
unsafe { SHUTDOWN_REQUESTED }
}
2019-04-01 10:13:02 +00:00
pub fn fail_on_shutdown() -> Result<(), Error> {
if shutdown_requested() {
bail!("Server shutdown requested - aborting task");
}
Ok(())
}
/// safe wrapper for `nix::unistd::pipe2` defaulting to `O_CLOEXEC` and guarding the file
/// descriptors.
pub fn pipe() -> Result<(Fd, Fd), Error> {
let (pin, pout) = nix::unistd::pipe2(nix::fcntl::OFlag::O_CLOEXEC)?;
Ok((Fd(pin), Fd(pout)))
}
2019-05-08 09:05:38 +00:00
/// safe wrapper for `nix::sys::socket::socketpair` defaulting to `O_CLOEXEC` and guarding the file
/// descriptors.
pub fn socketpair() -> Result<(Fd, Fd), Error> {
use nix::sys::socket;
let (pa, pb) = socket::socketpair(
socket::AddressFamily::Unix,
socket::SockType::Stream,
None,
socket::SockFlag::SOCK_CLOEXEC,
)?;
Ok((Fd(pa), Fd(pb)))
}
2019-05-08 09:05:38 +00:00
/// An easy way to convert types to Any
///
/// Mostly useful to downcast trait objects (see RpcEnvironment).
pub trait AsAny {
fn as_any(&self) -> &dyn Any;
2019-05-08 09:05:38 +00:00
}
impl<T: Any> AsAny for T {
fn as_any(&self) -> &dyn Any {
self
}
2019-05-08 09:05:38 +00:00
}
/// This used to be: `SIMPLE_ENCODE_SET` plus space, `"`, `#`, `<`, `>`, backtick, `?`, `{`, `}`
pub const DEFAULT_ENCODE_SET: &AsciiSet = &percent_encoding::CONTROLS // 0..1f and 7e
// The SIMPLE_ENCODE_SET adds space and anything >= 0x7e (7e itself is already included above)
.add(0x20)
.add(0x7f)
// the DEFAULT_ENCODE_SET added:
.add(b' ')
.add(b'"')
.add(b'#')
.add(b'<')
.add(b'>')
.add(b'`')
.add(b'?')
.add(b'{')
.add(b'}');
/// Get an iterator over lines of a file, skipping empty lines and comments (lines starting with a
/// `#`).
pub fn file_get_non_comment_lines<P: AsRef<Path>>(
path: P,
) -> Result<impl Iterator<Item = io::Result<String>>, Error> {
let path = path.as_ref();
Ok(io::BufReader::new(
File::open(path).map_err(|err| format_err!("error opening {:?}: {}", path, err))?,
)
.lines()
.filter_map(|line| match line {
Ok(line) => {
let line = line.trim();
if line.is_empty() || line.starts_with('#') {
None
} else {
Some(Ok(line.to_string()))
}
}
Err(err) => Some(Err(err)),
}))
}
pub fn setup_safe_path_env() {
std::env::set_var("PATH", "/sbin:/bin:/usr/sbin:/usr/bin");
// Make %ENV safer - as suggested by https://perldoc.perl.org/perlsec.html
for name in &["IFS", "CDPATH", "ENV", "BASH_ENV"] {
std::env::remove_var(name);
}
}
pub fn strip_ascii_whitespace(line: &[u8]) -> &[u8] {
let line = match line.iter().position(|&b| !b.is_ascii_whitespace()) {
Some(n) => &line[n..],
None => return &[],
};
match line.iter().rev().position(|&b| !b.is_ascii_whitespace()) {
Some(n) => &line[..(line.len() - n)],
None => &[],
}
}
2020-09-17 08:27:04 +00:00
/// Seeks to start of file and computes the SHA256 hash
pub fn compute_file_csum(file: &mut File) -> Result<([u8; 32], u64), Error> {
file.seek(SeekFrom::Start(0))?;
let mut hasher = openssl::sha::Sha256::new();
let mut buffer = proxmox::tools::vec::undefined(256*1024);
let mut size: u64 = 0;
loop {
let count = match file.read(&mut buffer) {
Ok(count) => count,
Err(ref err) if err.kind() == std::io::ErrorKind::Interrupted => {
continue;
}
Err(err) => return Err(err.into()),
};
if count == 0 {
break;
}
size += count as u64;
hasher.update(&buffer[..count]);
}
let csum = hasher.finish();
Ok((csum, size))
}