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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
// Copyright (C) 2020 Alibaba Cloud. All rights reserved. // SPDX-License-Identifier: Apache-2.0 //! Fuse passthrough file system, mirroring an existing FS hierarchy. //! //! This file system mirrors the existing file system hierarchy of the system, starting at the //! root file system. This is implemented by just "passing through" all requests to the //! corresponding underlying file system. //! //! The code is derived from the //! [CrosVM](https://chromium.googlesource.com/chromiumos/platform/crosvm/) project, //! with heavy modification/enhancements from Alibaba Cloud OS team. use std::any::Any; use std::collections::{btree_map, BTreeMap}; use std::ffi::{CStr, CString}; use std::fs::File; use std::io; use std::mem::MaybeUninit; use std::os::unix::io::{AsRawFd, FromRawFd, RawFd}; use std::str::FromStr; use std::sync::atomic::{AtomicBool, AtomicU64, Ordering}; use std::sync::{Arc, Mutex, MutexGuard, RwLock, RwLockWriteGuard}; use std::time::Duration; use vm_memory::ByteValued; use crate::abi::linux_abi as fuse; use crate::api::filesystem::Entry; use crate::api::{BackendFileSystem, VFS_MAX_INO}; mod sync_io; mod multikey; use multikey::MultikeyBTreeMap; const CURRENT_DIR_CSTR: &[u8] = b".\0"; const PARENT_DIR_CSTR: &[u8] = b"..\0"; const EMPTY_CSTR: &[u8] = b"\0"; const PROC_CSTR: &[u8] = b"/proc\0"; type Inode = u64; type Handle = u64; #[derive(Clone, Copy, Debug, PartialOrd, Ord, PartialEq, Eq)] struct InodeAltKey { ino: libc::ino64_t, dev: libc::dev_t, } impl InodeAltKey { fn from_stat(st: &libc::stat64) -> Self { InodeAltKey { ino: st.st_ino, dev: st.st_dev, } } } struct InodeData { inode: Inode, // Most of these aren't actually files but ¯\_(ツ)_/¯. file: File, refcount: AtomicU64, } impl InodeData { fn new(inode: Inode, file: File, refcount: u64) -> Self { InodeData { inode, file, refcount: AtomicU64::new(refcount), } } // When making use of the underlying RawFd, the caller must ensure that the Arc<InodeData> // object is within scope. Otherwise it may cause race window to access wrong target fd. // By introducing this method, we could explicitly audit all callers making use of the // underlying RawFd. fn get_raw_fd(&self) -> RawFd { self.file.as_raw_fd() } } /// Data structures to manage accessed inodes. struct InodeMap { inodes: RwLock<MultikeyBTreeMap<Inode, InodeAltKey, Arc<InodeData>>>, } impl InodeMap { fn new() -> Self { InodeMap { inodes: RwLock::new(MultikeyBTreeMap::new()), } } fn clear(&self) { self.inodes.write().unwrap().clear(); } fn get(&self, inode: Inode) -> io::Result<Arc<InodeData>> { self.inodes .read() .unwrap() .get(&inode) .map(Arc::clone) .ok_or_else(ebadf) } fn get_alt(&self, altkey: &InodeAltKey) -> Option<Arc<InodeData>> { self.inodes.read().unwrap().get_alt(altkey).map(Arc::clone) } fn get_map_mut( &self, ) -> RwLockWriteGuard<MultikeyBTreeMap<Inode, InodeAltKey, Arc<InodeData>>> { self.inodes.write().unwrap() } fn insert(&self, inode: Inode, altkey: InodeAltKey, data: InodeData) { self.inodes .write() .unwrap() .insert(inode, altkey, Arc::new(data)); } } struct HandleData { inode: Inode, file: File, lock: Mutex<()>, } impl HandleData { fn new(inode: Inode, file: File) -> Self { HandleData { inode, file, lock: Mutex::new(()), } } fn get_file_mut(&self) -> (MutexGuard<()>, &File) { (self.lock.lock().unwrap(), &self.file) } // When making use of the underlying RawFd, the caller must ensure that the Arc<HandleData> // object is within scope. Otherwise it may cause race window to access wrong target fd. // By introducing this method, we could explicitly audit all callers making use of the // underlying RawFd. fn get_handle_raw_fd(&self) -> RawFd { self.file.as_raw_fd() } } struct HandleMap { handles: RwLock<BTreeMap<Handle, Arc<HandleData>>>, } impl HandleMap { fn new() -> Self { HandleMap { handles: RwLock::new(BTreeMap::new()), } } fn clear(&self) { self.handles.write().unwrap().clear(); } fn insert(&self, handle: Handle, data: HandleData) { self.handles.write().unwrap().insert(handle, Arc::new(data)); } fn release(&self, handle: Handle, inode: Inode) -> io::Result<()> { let mut handles = self.handles.write().unwrap(); if let btree_map::Entry::Occupied(e) = handles.entry(handle) { if e.get().inode == inode { // We don't need to close the file here because that will happen automatically when // the last `Arc` is dropped. e.remove(); return Ok(()); } } Err(ebadf()) } fn get(&self, handle: Handle, inode: Inode) -> io::Result<Arc<HandleData>> { self.handles .read() .unwrap() .get(&handle) .filter(|hd| hd.inode == inode) .map(Arc::clone) .ok_or_else(ebadf) } } #[repr(C, packed)] #[derive(Clone, Copy, Debug, Default)] struct LinuxDirent64 { d_ino: libc::ino64_t, d_off: libc::off64_t, d_reclen: libc::c_ushort, d_ty: libc::c_uchar, } unsafe impl ByteValued for LinuxDirent64 {} /// The caching policy that the file system should report to the FUSE client. By default the FUSE /// protocol uses close-to-open consistency. This means that any cached contents of the file are /// invalidated the next time that file is opened. #[derive(Debug, Clone, PartialEq)] pub enum CachePolicy { /// The client should never cache file data and all I/O should be directly forwarded to the /// server. This policy must be selected when file contents may change without the knowledge of /// the FUSE client (i.e., the file system does not have exclusive access to the directory). Never, /// The client is free to choose when and how to cache file data. This is the default policy and /// uses close-to-open consistency as described in the enum documentation. Auto, /// The client should always cache file data. This means that the FUSE client will not /// invalidate any cached data that was returned by the file system the last time the file was /// opened. This policy should only be selected when the file system has exclusive access to the /// directory. Always, } impl FromStr for CachePolicy { type Err = &'static str; fn from_str(s: &str) -> Result<Self, Self::Err> { match s { "never" | "Never" | "NEVER" | "none" | "None" | "NONE" => Ok(CachePolicy::Never), "auto" | "Auto" | "AUTO" => Ok(CachePolicy::Auto), "always" | "Always" | "ALWAYS" => Ok(CachePolicy::Always), _ => Err("invalid cache policy"), } } } impl Default for CachePolicy { fn default() -> Self { CachePolicy::Auto } } /// Options that configure the behavior of the passthrough fuse file system. #[derive(Debug, Clone, PartialEq)] pub struct Config { /// How long the FUSE client should consider directory entries to be valid. If the contents of a /// directory can only be modified by the FUSE client (i.e., the file system has exclusive /// access), then this should be a large value. /// /// The default value for this option is 5 seconds. pub entry_timeout: Duration, /// How long the FUSE client should consider file and directory attributes to be valid. If the /// attributes of a file or directory can only be modified by the FUSE client (i.e., the file /// system has exclusive access), then this should be set to a large value. /// /// The default value for this option is 5 seconds. pub attr_timeout: Duration, /// The caching policy the file system should use. See the documentation of `CachePolicy` for /// more details. pub cache_policy: CachePolicy, /// Whether the file system should enabled writeback caching. This can improve performance as it /// allows the FUSE client to cache and coalesce multiple writes before sending them to the file /// system. However, enabling this option can increase the risk of data corruption if the file /// contents can change without the knowledge of the FUSE client (i.e., the server does **NOT** /// have exclusive access). Additionally, the file system should have read access to all files /// in the directory it is serving as the FUSE client may send read requests even for files /// opened with `O_WRONLY`. /// /// Therefore callers should only enable this option when they can guarantee that: 1) the file /// system has exclusive access to the directory and 2) the file system has read permissions for /// all files in that directory. /// /// The default value for this option is `false`. pub writeback: bool, /// The path of the root directory. /// /// The default is `/`. pub root_dir: String, /// Whether the file system should support Extended Attributes (xattr). Enabling this feature may /// have a significant impact on performance, especially on write parallelism. This is the result /// of FUSE attempting to remove the special file privileges after each write request. /// /// The default value for this options is `false`. pub xattr: bool, /// To be compatible with Vfs and PseudoFs, PassthroughFs needs to prepare /// root inode before accepting INIT request. /// /// The default value for this option is `true`. pub do_import: bool, /// Control whether no_open is allowed. /// /// The default value for this option is `false`. pub no_open: bool, /// Control whether no_opendir is allowed. /// /// The default value for this option is `false`. pub no_opendir: bool, } impl Default for Config { fn default() -> Self { Config { entry_timeout: Duration::from_secs(5), attr_timeout: Duration::from_secs(5), cache_policy: Default::default(), writeback: false, root_dir: String::from("/"), xattr: false, do_import: true, no_open: false, no_opendir: false, } } } /// A file system that simply "passes through" all requests it receives to the underlying file /// system. /// /// To keep the implementation simple it servers the contents of its root directory. Users /// that wish to serve only a specific directory should set up the environment so that that /// directory ends up as the root of the file system process. One way to accomplish this is via a /// combination of mount namespaces and the pivot_root system call. pub struct PassthroughFs { // File descriptors for various points in the file system tree. These fds are always opened with // the `O_PATH` option so they cannot be used for reading or writing any data. See the // documentation of the `O_PATH` flag in `open(2)` for more details on what one can and cannot // do with an fd opened with this flag. inode_map: InodeMap, next_inode: AtomicU64, // File descriptors for open files and directories. Unlike the fds in `inodes`, these _can_ be // used for reading and writing data. handle_map: HandleMap, next_handle: AtomicU64, // File descriptor pointing to the `/proc` directory. This is used to convert an fd from // `inodes` into one that can go into `handles`. This is accomplished by reading the // `self/fd/{}` symlink. We keep an open fd here in case the file system tree that we are meant // to be serving doesn't have access to `/proc`. proc: File, // Whether writeback caching is enabled for this directory. This will only be true when // `cfg.writeback` is true and `init` was called with `FsOptions::WRITEBACK_CACHE`. writeback: AtomicBool, // Whether no_open is enabled. no_open: AtomicBool, // Whether no_opendir is enabled. no_opendir: AtomicBool, cfg: Config, } impl PassthroughFs { /// Create a Passthrough file system instance. pub fn new(cfg: Config) -> io::Result<PassthroughFs> { // Safe because this is a constant value and a valid C string. let proc_cstr = unsafe { CStr::from_bytes_with_nul_unchecked(PROC_CSTR) }; let proc = Self::open_file( libc::AT_FDCWD, proc_cstr, libc::O_PATH | libc::O_NOFOLLOW | libc::O_CLOEXEC, 0, )?; Ok(PassthroughFs { inode_map: InodeMap::new(), next_inode: AtomicU64::new(fuse::ROOT_ID + 1), handle_map: HandleMap::new(), next_handle: AtomicU64::new(1), proc, writeback: AtomicBool::new(false), no_open: AtomicBool::new(false), no_opendir: AtomicBool::new(false), cfg, }) } /// Initialize the Passthrough file system. pub fn import(&self) -> io::Result<()> { let root = CString::new(self.cfg.root_dir.as_str()).expect("CString::new failed"); // We use `O_PATH` because we just want this for traversing the directory tree // and not for actually reading the contents. let f = Self::open_file( libc::AT_FDCWD, &root, libc::O_PATH | libc::O_NOFOLLOW | libc::O_CLOEXEC, 0, )?; let st = Self::stat(&f)?; // Safe because this doesn't modify any memory and there is no need to check the return // value because this system call always succeeds. We need to clear the umask here because // we want the client to be able to set all the bits in the mode. unsafe { libc::umask(0o000) }; // Not sure why the root inode gets a refcount of 2 but that's what libfuse does. self.inode_map.insert( fuse::ROOT_ID, InodeAltKey::from_stat(&st), InodeData::new(fuse::ROOT_ID, f, 2), ); Ok(()) } /// Get the list of file descriptors which should be reserved across live upgrade. pub fn keep_fds(&self) -> Vec<RawFd> { vec![self.proc.as_raw_fd()] } fn stat(f: &File) -> io::Result<libc::stat64> { // Safe because this is a constant value and a valid C string. let pathname = unsafe { CStr::from_bytes_with_nul_unchecked(EMPTY_CSTR) }; let mut st = MaybeUninit::<libc::stat64>::zeroed(); // Safe because the kernel will only write data in `st` and we check the return value. let res = unsafe { libc::fstatat64( f.as_raw_fd(), pathname.as_ptr(), st.as_mut_ptr(), libc::AT_EMPTY_PATH | libc::AT_SYMLINK_NOFOLLOW, ) }; if res >= 0 { // Safe because the kernel guarantees that the struct is now fully initialized. Ok(unsafe { st.assume_init() }) } else { Err(io::Error::last_os_error()) } } fn open_file(dfd: i32, pathname: &CStr, flags: i32, mode: u32) -> io::Result<File> { let fd = if flags & libc::O_CREAT == libc::O_CREAT { unsafe { libc::openat(dfd, pathname.as_ptr(), flags, mode) } } else { unsafe { libc::openat(dfd, pathname.as_ptr(), flags) } }; if fd < 0 { return Err(io::Error::last_os_error()); } // Safe because we just opened this fd. Ok(unsafe { File::from_raw_fd(fd) }) } fn do_lookup(&self, parent: Inode, name: &CStr) -> io::Result<Entry> { let p = self.inode_map.get(parent)?; let f = Self::open_file( p.get_raw_fd(), name, libc::O_PATH | libc::O_NOFOLLOW | libc::O_CLOEXEC, 0, )?; let st = Self::stat(&f)?; let altkey = InodeAltKey::from_stat(&st); let mut found = None; 'search: loop { match self.inode_map.get_alt(&altkey) { // No existing entry found None => break 'search, Some(data) => { let curr = data.refcount.load(Ordering::Acquire); // forgot_one() has just destroyed the entry, retry... if curr == 0 { continue 'search; } // Saturating add to avoid integer overflow, it's not realistic to saturate u64. let new = curr.saturating_add(1); // Synchronizes with the forgot_one() if data.refcount.compare_and_swap(curr, new, Ordering::AcqRel) == curr { found = Some(data.inode); break; } } } } let inode = if let Some(v) = found { v } else { // There is a possible race here where 2 threads end up adding the same file // into the inode list. However, since each of those will get a unique Inode // value and unique file descriptors this shouldn't be that much of a problem. let inode = self.next_inode.fetch_add(1, Ordering::Relaxed); if inode > VFS_MAX_INO { return Err(io::Error::new( io::ErrorKind::Other, format!("max inode number reached: {}", VFS_MAX_INO), )); } trace!("fuse: do_lookup new inode {} altkey {:?}", inode, altkey); self.inode_map .insert(inode, altkey, InodeData::new(inode, f, 1)); inode }; Ok(Entry { inode, generation: 0, attr: st, attr_timeout: self.cfg.attr_timeout, entry_timeout: self.cfg.entry_timeout, }) } fn forget_one( inodes: &mut MultikeyBTreeMap<Inode, InodeAltKey, Arc<InodeData>>, inode: Inode, count: u64, ) { // ROOT_ID should not be forgotten, or we're not able to access to files any more. if inode == fuse::ROOT_ID { return; } if let Some(data) = inodes.get(&inode) { // Acquiring the write lock on the inode map prevents new lookups from incrementing the // refcount but there is the possibility that a previous lookup already acquired a // reference to the inode data and is in the process of updating the refcount so we need // to loop here until we can decrement successfully. loop { let curr = data.refcount.load(Ordering::Acquire); // Saturating sub because it doesn't make sense for a refcount to go below zero and // we don't want misbehaving clients to cause integer overflow. let new = curr.saturating_sub(count); trace!( "fuse: forget inode {} refcount {}, count {}, new_count {}", inode, curr, count, new ); // Synchronizes with the acquire load in `do_lookup`. if data.refcount.compare_and_swap(curr, new, Ordering::AcqRel) == curr { if new == 0 { // We just removed the last refcount for this inode. inodes.remove(&inode); } break; } } } } fn do_release(&self, inode: Inode, handle: Handle) -> io::Result<()> { self.handle_map.release(handle, inode) } } impl BackendFileSystem for PassthroughFs { fn mount(&self) -> io::Result<(Entry, u64)> { let entry = self.do_lookup(fuse::ROOT_ID, &CString::new(".").unwrap())?; Ok((entry, VFS_MAX_INO)) } fn as_any(&self) -> &dyn Any { self } } fn ebadf() -> io::Error { io::Error::from_raw_os_error(libc::EBADF) }