File: backend_inotify.go

package info (click to toggle)
golang-fsnotify 1.9.0-1
  • links: PTS, VCS
  • area: main
  • in suites: experimental, forky, sid
  • size: 716 kB
  • sloc: ansic: 98; makefile: 4
file content (583 lines) | stat: -rw-r--r-- 15,066 bytes parent folder | download
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
//go:build linux && !appengine

package fsnotify

import (
	"errors"
	"fmt"
	"io"
	"io/fs"
	"os"
	"path/filepath"
	"strings"
	"sync"
	"time"
	"unsafe"

	"github.com/fsnotify/fsnotify/internal"
	"golang.org/x/sys/unix"
)

type inotify struct {
	*shared
	Events chan Event
	Errors chan error

	// Store fd here as os.File.Read() will no longer return on close after
	// calling Fd(). See: https://github.com/golang/go/issues/26439
	fd          int
	inotifyFile *os.File
	watches     *watches
	doneResp    chan struct{} // Channel to respond to Close

	// Store rename cookies in an array, with the index wrapping to 0. Almost
	// all of the time what we get is a MOVED_FROM to set the cookie and the
	// next event inotify sends will be MOVED_TO to read it. However, this is
	// not guaranteed – as described in inotify(7) – and we may get other events
	// between the two MOVED_* events (including other MOVED_* ones).
	//
	// A second issue is that moving a file outside the watched directory will
	// trigger a MOVED_FROM to set the cookie, but we never see the MOVED_TO to
	// read and delete it. So just storing it in a map would slowly leak memory.
	//
	// Doing it like this gives us a simple fast LRU-cache that won't allocate.
	// Ten items should be more than enough for our purpose, and a loop over
	// such a short array is faster than a map access anyway (not that it hugely
	// matters since we're talking about hundreds of ns at the most, but still).
	cookies     [10]koekje
	cookieIndex uint8
	cookiesMu   sync.Mutex
}

type (
	watches struct {
		wd   map[uint32]*watch // wd → watch
		path map[string]uint32 // pathname → wd
	}
	watch struct {
		wd      uint32 // Watch descriptor (as returned by the inotify_add_watch() syscall)
		flags   uint32 // inotify flags of this watch (see inotify(7) for the list of valid flags)
		path    string // Watch path.
		recurse bool   // Recursion with ./...?
	}
	koekje struct {
		cookie uint32
		path   string
	}
)

func newWatches() *watches {
	return &watches{
		wd:   make(map[uint32]*watch),
		path: make(map[string]uint32),
	}
}

func (w *watches) byPath(path string) *watch { return w.wd[w.path[path]] }
func (w *watches) byWd(wd uint32) *watch     { return w.wd[wd] }
func (w *watches) len() int                  { return len(w.wd) }
func (w *watches) add(ww *watch)             { w.wd[ww.wd] = ww; w.path[ww.path] = ww.wd }
func (w *watches) remove(watch *watch)       { delete(w.path, watch.path); delete(w.wd, watch.wd) }

func (w *watches) removePath(path string) ([]uint32, error) {
	path, recurse := recursivePath(path)
	wd, ok := w.path[path]
	if !ok {
		return nil, fmt.Errorf("%w: %s", ErrNonExistentWatch, path)
	}

	watch := w.wd[wd]
	if recurse && !watch.recurse {
		return nil, fmt.Errorf("can't use /... with non-recursive watch %q", path)
	}

	delete(w.path, path)
	delete(w.wd, wd)
	if !watch.recurse {
		return []uint32{wd}, nil
	}

	wds := make([]uint32, 0, 8)
	wds = append(wds, wd)
	for p, rwd := range w.path {
		if strings.HasPrefix(p, path) {
			delete(w.path, p)
			delete(w.wd, rwd)
			wds = append(wds, rwd)
		}
	}
	return wds, nil
}

func (w *watches) updatePath(path string, f func(*watch) (*watch, error)) error {
	var existing *watch
	wd, ok := w.path[path]
	if ok {
		existing = w.wd[wd]
	}

	upd, err := f(existing)
	if err != nil {
		return err
	}
	if upd != nil {
		w.wd[upd.wd] = upd
		w.path[upd.path] = upd.wd

		if upd.wd != wd {
			delete(w.wd, wd)
		}
	}

	return nil
}

var defaultBufferSize = 0

func newBackend(ev chan Event, errs chan error) (backend, error) {
	// Need to set nonblocking mode for SetDeadline to work, otherwise blocking
	// I/O operations won't terminate on close.
	fd, errno := unix.InotifyInit1(unix.IN_CLOEXEC | unix.IN_NONBLOCK)
	if fd == -1 {
		return nil, errno
	}

	w := &inotify{
		shared:      newShared(ev, errs),
		Events:      ev,
		Errors:      errs,
		fd:          fd,
		inotifyFile: os.NewFile(uintptr(fd), ""),
		watches:     newWatches(),
		doneResp:    make(chan struct{}),
	}

	go w.readEvents()
	return w, nil
}

func (w *inotify) Close() error {
	if w.shared.close() {
		return nil
	}

	// Causes any blocking reads to return with an error, provided the file
	// still supports deadline operations.
	err := w.inotifyFile.Close()
	if err != nil {
		return err
	}

	<-w.doneResp // Wait for readEvents() to finish.
	return nil
}

func (w *inotify) Add(name string) error { return w.AddWith(name) }

func (w *inotify) AddWith(path string, opts ...addOpt) error {
	if w.isClosed() {
		return ErrClosed
	}
	if debug {
		fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s  AddWith(%q)\n",
			time.Now().Format("15:04:05.000000000"), path)
	}

	with := getOptions(opts...)
	if !w.xSupports(with.op) {
		return fmt.Errorf("%w: %s", xErrUnsupported, with.op)
	}

	add := func(path string, with withOpts, recurse bool) error {
		var flags uint32
		if with.noFollow {
			flags |= unix.IN_DONT_FOLLOW
		}
		if with.op.Has(Create) {
			flags |= unix.IN_CREATE
		}
		if with.op.Has(Write) {
			flags |= unix.IN_MODIFY
		}
		if with.op.Has(Remove) {
			flags |= unix.IN_DELETE | unix.IN_DELETE_SELF
		}
		if with.op.Has(Rename) {
			flags |= unix.IN_MOVED_TO | unix.IN_MOVED_FROM | unix.IN_MOVE_SELF
		}
		if with.op.Has(Chmod) {
			flags |= unix.IN_ATTRIB
		}
		if with.op.Has(xUnportableOpen) {
			flags |= unix.IN_OPEN
		}
		if with.op.Has(xUnportableRead) {
			flags |= unix.IN_ACCESS
		}
		if with.op.Has(xUnportableCloseWrite) {
			flags |= unix.IN_CLOSE_WRITE
		}
		if with.op.Has(xUnportableCloseRead) {
			flags |= unix.IN_CLOSE_NOWRITE
		}
		return w.register(path, flags, recurse)
	}

	w.mu.Lock()
	defer w.mu.Unlock()
	path, recurse := recursivePath(path)
	if recurse {
		return filepath.WalkDir(path, func(root string, d fs.DirEntry, err error) error {
			if err != nil {
				return err
			}
			if !d.IsDir() {
				if root == path {
					return fmt.Errorf("fsnotify: not a directory: %q", path)
				}
				return nil
			}

			// Send a Create event when adding new directory from a recursive
			// watch; this is for "mkdir -p one/two/three". Usually all those
			// directories will be created before we can set up watchers on the
			// subdirectories, so only "one" would be sent as a Create event and
			// not "one/two" and "one/two/three" (inotifywait -r has the same
			// problem).
			if with.sendCreate && root != path {
				w.sendEvent(Event{Name: root, Op: Create})
			}

			return add(root, with, true)
		})
	}

	return add(path, with, false)
}

func (w *inotify) register(path string, flags uint32, recurse bool) error {
	return w.watches.updatePath(path, func(existing *watch) (*watch, error) {
		if existing != nil {
			flags |= existing.flags | unix.IN_MASK_ADD
		}

		wd, err := unix.InotifyAddWatch(w.fd, path, flags)
		if wd == -1 {
			return nil, err
		}

		if e, ok := w.watches.wd[uint32(wd)]; ok {
			return e, nil
		}

		if existing == nil {
			return &watch{
				wd:      uint32(wd),
				path:    path,
				flags:   flags,
				recurse: recurse,
			}, nil
		}

		existing.wd = uint32(wd)
		existing.flags = flags
		return existing, nil
	})
}

func (w *inotify) Remove(name string) error {
	if w.isClosed() {
		return nil
	}
	if debug {
		fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s  Remove(%q)\n",
			time.Now().Format("15:04:05.000000000"), name)
	}

	w.mu.Lock()
	defer w.mu.Unlock()
	return w.remove(filepath.Clean(name))
}

func (w *inotify) remove(name string) error {
	wds, err := w.watches.removePath(name)
	if err != nil {
		return err
	}

	for _, wd := range wds {
		_, err := unix.InotifyRmWatch(w.fd, wd)
		if err != nil {
			// TODO: Perhaps it's not helpful to return an error here in every
			// case; the only two possible errors are:
			//
			// EBADF, which happens when w.fd is not a valid file descriptor of
			// any kind.
			//
			// EINVAL, which is when fd is not an inotify descriptor or wd is
			// not a valid watch descriptor. Watch descriptors are invalidated
			// when they are removed explicitly or implicitly; explicitly by
			// inotify_rm_watch, implicitly when the file they are watching is
			// deleted.
			return err
		}
	}
	return nil
}

func (w *inotify) WatchList() []string {
	if w.isClosed() {
		return nil
	}

	w.mu.Lock()
	defer w.mu.Unlock()
	entries := make([]string, 0, w.watches.len())
	for pathname := range w.watches.path {
		entries = append(entries, pathname)
	}
	return entries
}

// readEvents reads from the inotify file descriptor, converts the
// received events into Event objects and sends them via the Events channel
func (w *inotify) readEvents() {
	defer func() {
		close(w.doneResp)
		close(w.Errors)
		close(w.Events)
	}()

	var buf [unix.SizeofInotifyEvent * 4096]byte // Buffer for a maximum of 4096 raw events
	for {
		if w.isClosed() {
			return
		}

		n, err := w.inotifyFile.Read(buf[:])
		if err != nil {
			if errors.Is(err, os.ErrClosed) {
				return
			}
			if !w.sendError(err) {
				return
			}
			continue
		}

		if n < unix.SizeofInotifyEvent {
			err := errors.New("notify: short read in readEvents()") // Read was too short.
			if n == 0 {
				err = io.EOF // If EOF is received. This should really never happen.
			}
			if !w.sendError(err) {
				return
			}
			continue
		}

		// We don't know how many events we just read into the buffer While the
		// offset points to at least one whole event.
		var offset uint32
		for offset <= uint32(n-unix.SizeofInotifyEvent) {
			// Point to the event in the buffer.
			inEvent := (*unix.InotifyEvent)(unsafe.Pointer(&buf[offset]))

			if inEvent.Mask&unix.IN_Q_OVERFLOW != 0 {
				if !w.sendError(ErrEventOverflow) {
					return
				}
			}

			ev, ok := w.handleEvent(inEvent, &buf, offset)
			if !ok {
				return
			}
			if !w.sendEvent(ev) {
				return
			}

			// Move to the next event in the buffer
			offset += unix.SizeofInotifyEvent + inEvent.Len
		}
	}
}

func (w *inotify) handleEvent(inEvent *unix.InotifyEvent, buf *[65536]byte, offset uint32) (Event, bool) {
	w.mu.Lock()
	defer w.mu.Unlock()

	/// If the event happened to the watched directory or the watched file, the
	/// kernel doesn't append the filename to the event, but we would like to
	/// always fill the the "Name" field with a valid filename. We retrieve the
	/// path of the watch from the "paths" map.
	///
	/// Can be nil if Remove() was called in another goroutine for this path
	/// inbetween reading the events from the kernel and reading the internal
	/// state. Not much we can do about it, so just skip. See #616.
	watch := w.watches.byWd(uint32(inEvent.Wd))
	if watch == nil {
		return Event{}, true
	}

	var (
		name    = watch.path
		nameLen = uint32(inEvent.Len)
	)
	if nameLen > 0 {
		/// Point "bytes" at the first byte of the filename
		bb := *buf
		bytes := (*[unix.PathMax]byte)(unsafe.Pointer(&bb[offset+unix.SizeofInotifyEvent]))[:nameLen:nameLen]
		/// The filename is padded with NULL bytes. TrimRight() gets rid of those.
		name += "/" + strings.TrimRight(string(bytes[0:nameLen]), "\x00")
	}

	if debug {
		internal.Debug(name, inEvent.Mask, inEvent.Cookie)
	}

	if inEvent.Mask&unix.IN_IGNORED != 0 || inEvent.Mask&unix.IN_UNMOUNT != 0 {
		w.watches.remove(watch)
		return Event{}, true
	}

	// inotify will automatically remove the watch on deletes; just need
	// to clean our state here.
	if inEvent.Mask&unix.IN_DELETE_SELF == unix.IN_DELETE_SELF {
		w.watches.remove(watch)
	}

	// We can't really update the state when a watched path is moved; only
	// IN_MOVE_SELF is sent and not IN_MOVED_{FROM,TO}. So remove the watch.
	if inEvent.Mask&unix.IN_MOVE_SELF == unix.IN_MOVE_SELF {
		if watch.recurse { // Do nothing
			return Event{}, true
		}

		err := w.remove(watch.path)
		if err != nil && !errors.Is(err, ErrNonExistentWatch) {
			if !w.sendError(err) {
				return Event{}, false
			}
		}
	}

	/// Skip if we're watching both this path and the parent; the parent will
	/// already send a delete so no need to do it twice.
	if inEvent.Mask&unix.IN_DELETE_SELF != 0 {
		_, ok := w.watches.path[filepath.Dir(watch.path)]
		if ok {
			return Event{}, true
		}
	}

	ev := w.newEvent(name, inEvent.Mask, inEvent.Cookie)
	// Need to update watch path for recurse.
	if watch.recurse {
		isDir := inEvent.Mask&unix.IN_ISDIR == unix.IN_ISDIR
		/// New directory created: set up watch on it.
		if isDir && ev.Has(Create) {
			err := w.register(ev.Name, watch.flags, true)
			if !w.sendError(err) {
				return Event{}, false
			}

			// This was a directory rename, so we need to update all the
			// children.
			//
			// TODO: this is of course pretty slow; we should use a better data
			// structure for storing all of this, e.g. store children in the
			// watch. I have some code for this in my kqueue refactor we can use
			// in the future. For now I'm okay with this as it's not publicly
			// available. Correctness first, performance second.
			if ev.renamedFrom != "" {
				for k, ww := range w.watches.wd {
					if k == watch.wd || ww.path == ev.Name {
						continue
					}
					if strings.HasPrefix(ww.path, ev.renamedFrom) {
						ww.path = strings.Replace(ww.path, ev.renamedFrom, ev.Name, 1)
						w.watches.wd[k] = ww
					}
				}
			}
		}
	}

	return ev, true
}

func (w *inotify) isRecursive(path string) bool {
	ww := w.watches.byPath(path)
	if ww == nil { // path could be a file, so also check the Dir.
		ww = w.watches.byPath(filepath.Dir(path))
	}
	return ww != nil && ww.recurse
}

func (w *inotify) newEvent(name string, mask, cookie uint32) Event {
	e := Event{Name: name}
	if mask&unix.IN_CREATE == unix.IN_CREATE || mask&unix.IN_MOVED_TO == unix.IN_MOVED_TO {
		e.Op |= Create
	}
	if mask&unix.IN_DELETE_SELF == unix.IN_DELETE_SELF || mask&unix.IN_DELETE == unix.IN_DELETE {
		e.Op |= Remove
	}
	if mask&unix.IN_MODIFY == unix.IN_MODIFY {
		e.Op |= Write
	}
	if mask&unix.IN_OPEN == unix.IN_OPEN {
		e.Op |= xUnportableOpen
	}
	if mask&unix.IN_ACCESS == unix.IN_ACCESS {
		e.Op |= xUnportableRead
	}
	if mask&unix.IN_CLOSE_WRITE == unix.IN_CLOSE_WRITE {
		e.Op |= xUnportableCloseWrite
	}
	if mask&unix.IN_CLOSE_NOWRITE == unix.IN_CLOSE_NOWRITE {
		e.Op |= xUnportableCloseRead
	}
	if mask&unix.IN_MOVE_SELF == unix.IN_MOVE_SELF || mask&unix.IN_MOVED_FROM == unix.IN_MOVED_FROM {
		e.Op |= Rename
	}
	if mask&unix.IN_ATTRIB == unix.IN_ATTRIB {
		e.Op |= Chmod
	}

	if cookie != 0 {
		if mask&unix.IN_MOVED_FROM == unix.IN_MOVED_FROM {
			w.cookiesMu.Lock()
			w.cookies[w.cookieIndex] = koekje{cookie: cookie, path: e.Name}
			w.cookieIndex++
			if w.cookieIndex > 9 {
				w.cookieIndex = 0
			}
			w.cookiesMu.Unlock()
		} else if mask&unix.IN_MOVED_TO == unix.IN_MOVED_TO {
			w.cookiesMu.Lock()
			var prev string
			for _, c := range w.cookies {
				if c.cookie == cookie {
					prev = c.path
					break
				}
			}
			w.cookiesMu.Unlock()
			e.renamedFrom = prev
		}
	}
	return e
}

func (w *inotify) xSupports(op Op) bool {
	return true // Supports everything.
}

func (w *inotify) state() {
	w.mu.Lock()
	defer w.mu.Unlock()
	for wd, ww := range w.watches.wd {
		fmt.Fprintf(os.Stderr, "%4d: recurse=%t %q\n", wd, ww.recurse, ww.path)
	}
}