xref: /linux/fs/xfs/scrub/dirtree.c (revision f0100363d8c374bd8e9ea7c9ba02744f0b802ca4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2023-2024 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
5  */
6 #include "xfs_platform.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans.h"
14 #include "xfs_inode.h"
15 #include "xfs_icache.h"
16 #include "xfs_dir2.h"
17 #include "xfs_dir2_priv.h"
18 #include "xfs_attr.h"
19 #include "xfs_parent.h"
20 #include "scrub/scrub.h"
21 #include "scrub/common.h"
22 #include "scrub/bitmap.h"
23 #include "scrub/ino_bitmap.h"
24 #include "scrub/xfile.h"
25 #include "scrub/xfarray.h"
26 #include "scrub/xfblob.h"
27 #include "scrub/listxattr.h"
28 #include "scrub/trace.h"
29 #include "scrub/repair.h"
30 #include "scrub/orphanage.h"
31 #include "scrub/dirtree.h"
32 
33 /*
34  * Directory Tree Structure Validation
35  * ===================================
36  *
37  * Validating the tree qualities of the directory tree structure can be
38  * difficult.  If the tree is frozen, running a depth (or breadth) first search
39  * and marking a bitmap suffices to determine if there is a cycle.  XORing the
40  * mark bitmap with the inode bitmap afterwards tells us if there are
41  * disconnected cycles.  If the tree is not frozen, directory updates can move
42  * subtrees across the scanner wavefront, which complicates the design greatly.
43  *
44  * Directory parent pointers change that by enabling an incremental approach to
45  * validation of the tree structure.  Instead of using one thread to scan the
46  * entire filesystem, we instead can have multiple threads walking individual
47  * subdirectories upwards to the root.  In a perfect world, the IOLOCK would
48  * suffice to stabilize two directories in a parent -> child relationship.
49  * Unfortunately, the VFS does not take the IOLOCK when moving a child
50  * subdirectory, so we instead synchronize on ILOCK and use dirent update hooks
51  * to detect a race.  If a race occurs in a path, we restart the scan.
52  *
53  * If the walk terminates without reaching the root, we know the path is
54  * disconnected and ought to be attached to the lost and found.  If on the walk
55  * we find the same subdir that we're scanning, we know this is a cycle and
56  * should delete an incoming edge.  If we find multiple paths to the root, we
57  * know to delete an incoming edge.
58  *
59  * There are two big hitches with this approach: first, all file link counts
60  * must be correct to prevent other writers from doing the wrong thing with the
61  * directory tree structure.  Second, because we're walking upwards in a tree
62  * of arbitrary depth, we cannot hold all the ILOCKs.  Instead, we will use a
63  * directory update hook to invalidate the scan results if one of the paths
64  * we've scanned has changed.
65  */
66 
67 /* Clean up the dirtree checking resources. */
68 STATIC void
xchk_dirtree_buf_cleanup(void * buf)69 xchk_dirtree_buf_cleanup(
70 	void			*buf)
71 {
72 	struct xchk_dirtree	*dl = buf;
73 	struct xchk_dirpath	*path, *n;
74 
75 	if (dl->scan_ino != NULLFSINO)
76 		xfs_dir_hook_del(dl->sc->mp, &dl->dhook);
77 
78 	xchk_dirtree_for_each_path_safe(dl, path, n) {
79 		list_del_init(&path->list);
80 		xino_bitmap_destroy(&path->seen_inodes);
81 		kfree(path);
82 	}
83 
84 	if (dl->path_names)
85 		xfblob_destroy(dl->path_names);
86 	dl->path_names = NULL;
87 	if (dl->path_steps)
88 		xfarray_destroy(dl->path_steps);
89 	dl->path_steps = NULL;
90 	mutex_destroy(&dl->lock);
91 }
92 
93 /* Set us up to look for directory loops. */
94 int
xchk_setup_dirtree(struct xfs_scrub * sc)95 xchk_setup_dirtree(
96 	struct xfs_scrub	*sc)
97 {
98 	struct xchk_dirtree	*dl;
99 	int			error;
100 
101 	xchk_fsgates_enable(sc, XCHK_FSGATES_DIRENTS);
102 
103 	if (xchk_could_repair(sc)) {
104 		error = xrep_setup_dirtree(sc);
105 		if (error)
106 			return error;
107 	}
108 
109 	dl = kvzalloc_obj(struct xchk_dirtree, XCHK_GFP_FLAGS);
110 	if (!dl)
111 		return -ENOMEM;
112 	dl->sc = sc;
113 	dl->xname.name = dl->namebuf;
114 	dl->hook_xname.name = dl->hook_namebuf;
115 	INIT_LIST_HEAD(&dl->path_list);
116 	dl->root_ino = NULLFSINO;
117 	dl->scan_ino = NULLFSINO;
118 	dl->parent_ino = NULLFSINO;
119 
120 	mutex_init(&dl->lock);
121 
122 	error = xfarray_create("dirtree path steps", 0,
123 			sizeof(struct xchk_dirpath_step), &dl->path_steps);
124 	if (error)
125 		goto out_dl;
126 
127 	error = xfblob_create("dirtree path names", &dl->path_names);
128 	if (error)
129 		goto out_steps;
130 
131 	error = xchk_setup_inode_contents(sc, 0);
132 	if (error)
133 		goto out_names;
134 
135 	sc->buf = dl;
136 	sc->buf_cleanup = xchk_dirtree_buf_cleanup;
137 	return 0;
138 
139 out_names:
140 	xfblob_destroy(dl->path_names);
141 out_steps:
142 	xfarray_destroy(dl->path_steps);
143 out_dl:
144 	mutex_destroy(&dl->lock);
145 	kvfree(dl);
146 	return error;
147 }
148 
149 /*
150  * Add the parent pointer described by @dl->pptr to the given path as a new
151  * step.  Returns -ELNRNG if the path is too deep.
152  */
153 int
xchk_dirpath_append(struct xchk_dirtree * dl,struct xfs_inode * ip,struct xchk_dirpath * path,const struct xfs_name * name,const struct xfs_parent_rec * pptr)154 xchk_dirpath_append(
155 	struct xchk_dirtree		*dl,
156 	struct xfs_inode		*ip,
157 	struct xchk_dirpath		*path,
158 	const struct xfs_name		*name,
159 	const struct xfs_parent_rec	*pptr)
160 {
161 	struct xchk_dirpath_step	step = {
162 		.pptr_rec		= *pptr, /* struct copy */
163 		.name_len		= name->len,
164 	};
165 	int				error;
166 
167 	/*
168 	 * If this path is more than 2 billion steps long, this directory tree
169 	 * is too far gone to fix.
170 	 */
171 	if (path->nr_steps >= XFS_MAXLINK)
172 		return -ELNRNG;
173 
174 	error = xfblob_storename(dl->path_names, &step.name_cookie, name);
175 	if (error)
176 		return error;
177 
178 	error = xino_bitmap_set(&path->seen_inodes, I_INO(ip));
179 	if (error)
180 		return error;
181 
182 	error = xfarray_append(dl->path_steps, &step);
183 	if (error)
184 		return error;
185 
186 	path->nr_steps++;
187 	return 0;
188 }
189 
190 /*
191  * Create an xchk_path for each parent pointer of the directory that we're
192  * scanning.  For each path created, we will eventually try to walk towards the
193  * root with the goal of deleting all parents except for one that leads to the
194  * root.
195  *
196  * Returns -EFSCORRUPTED to signal that the inode being scanned has a corrupt
197  * parent pointer and hence there's no point in continuing; or -ENOSR if there
198  * are too many parent pointers for this directory.
199  */
200 STATIC int
xchk_dirtree_create_path(struct xfs_scrub * sc,struct xfs_inode * ip,unsigned int attr_flags,const unsigned char * name,unsigned int namelen,const void * value,unsigned int valuelen,void * priv)201 xchk_dirtree_create_path(
202 	struct xfs_scrub		*sc,
203 	struct xfs_inode		*ip,
204 	unsigned int			attr_flags,
205 	const unsigned char		*name,
206 	unsigned int			namelen,
207 	const void			*value,
208 	unsigned int			valuelen,
209 	void				*priv)
210 {
211 	struct xfs_name			xname = {
212 		.name			= name,
213 		.len			= namelen,
214 	};
215 	struct xchk_dirtree		*dl = priv;
216 	struct xchk_dirpath		*path;
217 	const struct xfs_parent_rec	*rec = value;
218 	int				error;
219 
220 	if (!(attr_flags & XFS_ATTR_PARENT))
221 		return 0;
222 
223 	error = xfs_parent_from_attr(sc->mp, attr_flags, name, namelen, value,
224 			valuelen, NULL, NULL);
225 	if (error)
226 		return error;
227 
228 	/*
229 	 * If there are more than 2 billion actual parent pointers for this
230 	 * subdirectory, this fs is too far gone to fix.
231 	 */
232 	if (dl->nr_paths >= XFS_MAXLINK)
233 		return -ENOSR;
234 
235 	trace_xchk_dirtree_create_path(sc, ip, dl->nr_paths, &xname, rec);
236 
237 	/*
238 	 * Create a new xchk_path structure to remember this parent pointer
239 	 * and record the first name step.
240 	 */
241 	path = kmalloc_obj(struct xchk_dirpath, XCHK_GFP_FLAGS);
242 	if (!path)
243 		return -ENOMEM;
244 
245 	INIT_LIST_HEAD(&path->list);
246 	xino_bitmap_init(&path->seen_inodes);
247 	path->nr_steps = 0;
248 	path->outcome = XCHK_DIRPATH_SCANNING;
249 
250 	error = xchk_dirpath_append(dl, sc->ip, path, &xname, rec);
251 	if (error)
252 		goto out_path;
253 
254 	path->first_step = xfarray_length(dl->path_steps) - 1;
255 	path->second_step = XFARRAY_NULLIDX;
256 	path->path_nr = dl->nr_paths;
257 
258 	list_add_tail(&path->list, &dl->path_list);
259 	dl->nr_paths++;
260 	return 0;
261 out_path:
262 	xino_bitmap_destroy(&path->seen_inodes);
263 	kfree(path);
264 	return error;
265 }
266 
267 /*
268  * Validate that the first step of this path still has a corresponding
269  * parent pointer in @sc->ip.  We probably dropped @sc->ip's ILOCK while
270  * walking towards the roots, which is why this is necessary.
271  *
272  * This function has a side effect of loading the first parent pointer of this
273  * path into the parent pointer scratch pad.  This prepares us to walk up the
274  * directory tree towards the root.  Returns -ESTALE if the scan data is now
275  * out of date.
276  */
277 STATIC int
xchk_dirpath_revalidate(struct xchk_dirtree * dl,struct xchk_dirpath * path)278 xchk_dirpath_revalidate(
279 	struct xchk_dirtree		*dl,
280 	struct xchk_dirpath		*path)
281 {
282 	struct xfs_scrub		*sc = dl->sc;
283 	int				error;
284 
285 	/*
286 	 * Look up the parent pointer that corresponds to the start of this
287 	 * path.  If the parent pointer has disappeared on us, dump all the
288 	 * scan results and try again.
289 	 */
290 	error = xfs_parent_lookup(sc->tp, sc->ip, &dl->xname, &dl->pptr_rec,
291 			&dl->pptr_args);
292 	if (error == -ENOATTR) {
293 		trace_xchk_dirpath_disappeared(dl->sc, sc->ip, path->path_nr,
294 				path->first_step, &dl->xname, &dl->pptr_rec);
295 		dl->stale = true;
296 		return -ESTALE;
297 	}
298 
299 	return error;
300 }
301 
302 /*
303  * Walk the parent pointers of a directory at the end of a path and record
304  * the parent that we find in @dl->xname/pptr_rec.
305  */
306 STATIC int
xchk_dirpath_find_next_step(struct xfs_scrub * sc,struct xfs_inode * ip,unsigned int attr_flags,const unsigned char * name,unsigned int namelen,const void * value,unsigned int valuelen,void * priv)307 xchk_dirpath_find_next_step(
308 	struct xfs_scrub		*sc,
309 	struct xfs_inode		*ip,
310 	unsigned int			attr_flags,
311 	const unsigned char		*name,
312 	unsigned int			namelen,
313 	const void			*value,
314 	unsigned int			valuelen,
315 	void				*priv)
316 {
317 	struct xchk_dirtree		*dl = priv;
318 	const struct xfs_parent_rec	*rec = value;
319 	int				error;
320 
321 	if (!(attr_flags & XFS_ATTR_PARENT))
322 		return 0;
323 
324 	error = xfs_parent_from_attr(sc->mp, attr_flags, name, namelen, value,
325 			valuelen, NULL, NULL);
326 	if (error)
327 		return error;
328 
329 	/*
330 	 * If we've already set @dl->pptr_rec, then this directory has multiple
331 	 * parents.  Signal this back to the caller via -EMLINK.
332 	 */
333 	if (dl->parents_found > 0)
334 		return -EMLINK;
335 
336 	dl->parents_found++;
337 	memcpy(dl->namebuf, name, namelen);
338 	dl->xname.len = namelen;
339 	dl->pptr_rec = *rec; /* struct copy */
340 	return 0;
341 }
342 
343 /* Set and log the outcome of a path walk. */
344 static inline void
xchk_dirpath_set_outcome(struct xchk_dirtree * dl,struct xchk_dirpath * path,enum xchk_dirpath_outcome outcome)345 xchk_dirpath_set_outcome(
346 	struct xchk_dirtree		*dl,
347 	struct xchk_dirpath		*path,
348 	enum xchk_dirpath_outcome	outcome)
349 {
350 	trace_xchk_dirpath_set_outcome(dl->sc, path->path_nr, path->nr_steps,
351 			outcome);
352 
353 	path->outcome = outcome;
354 }
355 
356 /*
357  * Scan the directory at the end of this path for its parent directory link.
358  * If we find one, extend the path.  Returns -ESTALE if the scan data out of
359  * date.  Returns -EFSCORRUPTED if the parent pointer is bad; or -ELNRNG if
360  * the path got too deep.
361  */
362 STATIC int
xchk_dirpath_step_up(struct xchk_dirtree * dl,struct xchk_dirpath * path,bool is_metadir)363 xchk_dirpath_step_up(
364 	struct xchk_dirtree	*dl,
365 	struct xchk_dirpath	*path,
366 	bool			is_metadir)
367 {
368 	struct xfs_scrub	*sc = dl->sc;
369 	struct xfs_inode	*dp;
370 	xfs_ino_t		parent_ino = be64_to_cpu(dl->pptr_rec.p_ino);
371 	unsigned int		lock_mode;
372 	int			error = 0;
373 
374 	if (xchk_should_terminate(sc, &error))
375 		return error;
376 
377 	/* Grab and lock the parent directory. */
378 	error = xchk_iget(sc, parent_ino, &dp);
379 	switch (error) {
380 	case -EINVAL:
381 	case -ENOENT:
382 		mutex_lock(&dl->lock);
383 
384 		if (dl->stale) {
385 			/* live update detected a change in this path */
386 			error = -ESTALE;
387 		} else {
388 			/* inode doesn't exist, path invalid */
389 			error = -EFSCORRUPTED;
390 
391 			trace_xchk_dirpath_badino(dl->sc, path->path_nr,
392 					path->nr_steps, &dl->xname,
393 					&dl->pptr_rec);
394 		}
395 
396 		mutex_unlock(&dl->lock);
397 		return error;
398 	case 0:
399 		/* keep going */
400 		break;
401 	default:
402 		return error;
403 	}
404 
405 	lock_mode = xfs_ilock_attr_map_shared(dp);
406 	mutex_lock(&dl->lock);
407 
408 	if (dl->stale) {
409 		error = -ESTALE;
410 		goto out_scanlock;
411 	}
412 
413 	/* The handle encoded in the parent pointer must match. */
414 	if (VFS_I(dp)->i_generation != be32_to_cpu(dl->pptr_rec.p_gen)) {
415 		trace_xchk_dirpath_badgen(dl->sc, dp, path->path_nr,
416 				path->nr_steps, &dl->xname, &dl->pptr_rec);
417 		error = -EFSCORRUPTED;
418 		goto out_scanlock;
419 	}
420 
421 	/* We've reached the root directory; the path is ok. */
422 	if (parent_ino == dl->root_ino) {
423 		xchk_dirpath_set_outcome(dl, path, XCHK_DIRPATH_OK);
424 		error = 0;
425 		goto out_scanlock;
426 	}
427 
428 	/*
429 	 * The inode being scanned is its own distant ancestor!  Get rid of
430 	 * this path.
431 	 */
432 	if (parent_ino == I_INO(sc->ip)) {
433 		xchk_dirpath_set_outcome(dl, path, XCHK_DIRPATH_DELETE);
434 		error = 0;
435 		goto out_scanlock;
436 	}
437 
438 	/*
439 	 * We've seen this inode before during the path walk.  There's a loop
440 	 * above us in the directory tree.  This probably means that we cannot
441 	 * continue, but let's keep walking paths to get a full picture.
442 	 */
443 	if (xino_bitmap_test(&path->seen_inodes, parent_ino)) {
444 		xchk_dirpath_set_outcome(dl, path, XCHK_DIRPATH_LOOP);
445 		error = 0;
446 		goto out_scanlock;
447 	}
448 
449 	/* Parent pointer must point up to a directory. */
450 	if (!S_ISDIR(VFS_I(dp)->i_mode)) {
451 		trace_xchk_dirpath_nondir_parent(dl->sc, dp, path->path_nr,
452 				path->nr_steps, &dl->xname, &dl->pptr_rec);
453 		error = -EFSCORRUPTED;
454 		goto out_scanlock;
455 	}
456 
457 	/* Parent cannot be an unlinked directory. */
458 	if (VFS_I(dp)->i_nlink == 0) {
459 		trace_xchk_dirpath_unlinked_parent(dl->sc, dp, path->path_nr,
460 				path->nr_steps, &dl->xname, &dl->pptr_rec);
461 		error = -EFSCORRUPTED;
462 		goto out_scanlock;
463 	}
464 
465 	/* Parent must be in the same directory tree. */
466 	if (is_metadir != xfs_is_metadir_inode(dp)) {
467 		trace_xchk_dirpath_crosses_tree(dl->sc, dp, path->path_nr,
468 				path->nr_steps, &dl->xname, &dl->pptr_rec);
469 		error = -EFSCORRUPTED;
470 		goto out_scanlock;
471 	}
472 
473 	/*
474 	 * If the extended attributes look as though they has been zapped by
475 	 * the inode record repair code, we cannot scan for parent pointers.
476 	 */
477 	if (xchk_pptr_looks_zapped(dp)) {
478 		error = -EBUSY;
479 		xchk_set_incomplete(sc);
480 		goto out_scanlock;
481 	}
482 
483 	/*
484 	 * Walk the parent pointers of @dp to find the parent of this directory
485 	 * to find the next step in our walk.  If we find that @dp has exactly
486 	 * one parent, the parent pointer information will be stored in
487 	 * @dl->pptr_rec.  This prepares us for the next step of the walk.
488 	 */
489 	mutex_unlock(&dl->lock);
490 	dl->parents_found = 0;
491 	error = xchk_xattr_walk(sc, dp, xchk_dirpath_find_next_step, NULL, dl);
492 	mutex_lock(&dl->lock);
493 	if (error == -EFSCORRUPTED || error == -EMLINK ||
494 	    (!error && dl->parents_found == 0)) {
495 		/*
496 		 * Further up the directory tree from @sc->ip, we found a
497 		 * corrupt parent pointer, multiple parent pointers while
498 		 * finding this directory's parent, or zero parents despite
499 		 * having a nonzero link count.  Keep looking for other paths.
500 		 */
501 		xchk_dirpath_set_outcome(dl, path, XCHK_DIRPATH_CORRUPT);
502 		error = 0;
503 		goto out_scanlock;
504 	}
505 	if (error)
506 		goto out_scanlock;
507 
508 	if (dl->stale) {
509 		error = -ESTALE;
510 		goto out_scanlock;
511 	}
512 
513 	trace_xchk_dirpath_found_next_step(sc, dp, path->path_nr,
514 			path->nr_steps, &dl->xname, &dl->pptr_rec);
515 
516 	/* Append to the path steps */
517 	error = xchk_dirpath_append(dl, dp, path, &dl->xname, &dl->pptr_rec);
518 	if (error)
519 		goto out_scanlock;
520 
521 	if (path->second_step == XFARRAY_NULLIDX)
522 		path->second_step = xfarray_length(dl->path_steps) - 1;
523 
524 out_scanlock:
525 	mutex_unlock(&dl->lock);
526 	xfs_iunlock(dp, lock_mode);
527 	xchk_irele(sc, dp);
528 	return error;
529 }
530 
531 /*
532  * Walk the directory tree upwards towards what is hopefully the root
533  * directory, recording path steps as we go.  The current path components are
534  * stored in dl->pptr_rec and dl->xname.
535  *
536  * Returns -ESTALE if the scan data are out of date.  Returns -EFSCORRUPTED
537  * only if the direct parent pointer of @sc->ip associated with this path is
538  * corrupt.
539  */
540 STATIC int
xchk_dirpath_walk_upwards(struct xchk_dirtree * dl,struct xchk_dirpath * path)541 xchk_dirpath_walk_upwards(
542 	struct xchk_dirtree	*dl,
543 	struct xchk_dirpath	*path)
544 {
545 	struct xfs_scrub	*sc = dl->sc;
546 	bool			is_metadir;
547 	int			error;
548 
549 	ASSERT(sc->ilock_flags & XFS_ILOCK_EXCL);
550 
551 	/* Reload the start of this path and make sure it's still there. */
552 	error = xchk_dirpath_revalidate(dl, path);
553 	if (error)
554 		return error;
555 
556 	trace_xchk_dirpath_walk_upwards(sc, sc->ip, path->path_nr, &dl->xname,
557 			&dl->pptr_rec);
558 
559 	/*
560 	 * The inode being scanned is its own direct ancestor!
561 	 * Get rid of this path.
562 	 */
563 	if (be64_to_cpu(dl->pptr_rec.p_ino) == I_INO(sc->ip)) {
564 		xchk_dirpath_set_outcome(dl, path, XCHK_DIRPATH_DELETE);
565 		return 0;
566 	}
567 
568 	/*
569 	 * Drop ILOCK_EXCL on the inode being scanned.  We still hold
570 	 * IOLOCK_EXCL on it, so it cannot move around or be renamed.
571 	 *
572 	 * Beyond this point we're walking up the directory tree, which means
573 	 * that we can acquire and drop the ILOCK on an alias of sc->ip.  The
574 	 * ILOCK state is no longer tracked in the scrub context.  Hence we
575 	 * must drop @sc->ip's ILOCK during the walk.
576 	 */
577 	is_metadir = xfs_is_metadir_inode(sc->ip);
578 	mutex_unlock(&dl->lock);
579 	xchk_iunlock(sc, XFS_ILOCK_EXCL);
580 
581 	/*
582 	 * Take the first step in the walk towards the root by checking the
583 	 * start of this path, which is a direct parent pointer of @sc->ip.
584 	 * If we see any kind of error here (including corruptions), the parent
585 	 * pointer of @sc->ip is corrupt.  Stop the whole scan.
586 	 */
587 	error = xchk_dirpath_step_up(dl, path, is_metadir);
588 	if (error) {
589 		xchk_ilock(sc, XFS_ILOCK_EXCL);
590 		mutex_lock(&dl->lock);
591 		return error;
592 	}
593 
594 	/*
595 	 * Take steps upward from the second step in this path towards the
596 	 * root.  If we hit corruption errors here, there's a problem
597 	 * *somewhere* in the path, but we don't need to stop scanning.
598 	 */
599 	while (!error && path->outcome == XCHK_DIRPATH_SCANNING)
600 		error = xchk_dirpath_step_up(dl, path, is_metadir);
601 
602 	/* Retake the locks we had, mark paths, etc. */
603 	xchk_ilock(sc, XFS_ILOCK_EXCL);
604 	mutex_lock(&dl->lock);
605 	if (error == -EFSCORRUPTED) {
606 		xchk_dirpath_set_outcome(dl, path, XCHK_DIRPATH_CORRUPT);
607 		error = 0;
608 	}
609 	if (!error && dl->stale)
610 		return -ESTALE;
611 	return error;
612 }
613 
614 /*
615  * Decide if this path step has been touched by this live update.  Returns
616  * 1 for yes, 0 for no, or a negative errno.
617  */
618 STATIC int
xchk_dirpath_step_is_stale(struct xchk_dirtree * dl,struct xchk_dirpath * path,unsigned int step_nr,xfarray_idx_t step_idx,struct xfs_dir_update_params * p,xfs_ino_t * cursor)619 xchk_dirpath_step_is_stale(
620 	struct xchk_dirtree		*dl,
621 	struct xchk_dirpath		*path,
622 	unsigned int			step_nr,
623 	xfarray_idx_t			step_idx,
624 	struct xfs_dir_update_params	*p,
625 	xfs_ino_t			*cursor)
626 {
627 	struct xchk_dirpath_step	step;
628 	xfs_ino_t			child_ino = *cursor;
629 	int				error;
630 
631 	error = xfarray_load(dl->path_steps, step_idx, &step);
632 	if (error)
633 		return error;
634 	*cursor = be64_to_cpu(step.pptr_rec.p_ino);
635 
636 	/*
637 	 * If the parent and child being updated are not the ones mentioned in
638 	 * this path step, the scan data is still ok.
639 	 */
640 	if (I_INO(p->ip) != child_ino || I_INO(p->dp) != *cursor)
641 		return 0;
642 
643 	/*
644 	 * If the dirent name lengths or byte sequences are different, the scan
645 	 * data is still ok.
646 	 */
647 	if (p->name->len != step.name_len)
648 		return 0;
649 
650 	error = xfblob_loadname(dl->path_names, step.name_cookie,
651 			&dl->hook_xname, step.name_len);
652 	if (error)
653 		return error;
654 
655 	if (memcmp(dl->hook_xname.name, p->name->name, p->name->len) != 0)
656 		return 0;
657 
658 	/*
659 	 * If the update comes from the repair code itself, walk the state
660 	 * machine forward.
661 	 */
662 	if (I_INO(p->ip) == dl->scan_ino &&
663 	    path->outcome == XREP_DIRPATH_ADOPTING) {
664 		xchk_dirpath_set_outcome(dl, path, XREP_DIRPATH_ADOPTED);
665 		return 0;
666 	}
667 
668 	if (I_INO(p->ip) == dl->scan_ino &&
669 	    path->outcome == XREP_DIRPATH_DELETING) {
670 		xchk_dirpath_set_outcome(dl, path, XREP_DIRPATH_DELETED);
671 		return 0;
672 	}
673 
674 	/* Exact match, scan data is out of date. */
675 	trace_xchk_dirpath_changed(dl->sc, path->path_nr, step_nr, p->dp,
676 			p->ip, p->name);
677 	return 1;
678 }
679 
680 /*
681  * Decide if this path has been touched by this live update.  Returns 1 for
682  * yes, 0 for no, or a negative errno.
683  */
684 STATIC int
xchk_dirpath_is_stale(struct xchk_dirtree * dl,struct xchk_dirpath * path,struct xfs_dir_update_params * p)685 xchk_dirpath_is_stale(
686 	struct xchk_dirtree		*dl,
687 	struct xchk_dirpath		*path,
688 	struct xfs_dir_update_params	*p)
689 {
690 	xfs_ino_t			cursor = dl->scan_ino;
691 	xfarray_idx_t			idx = path->first_step;
692 	unsigned int			i;
693 	int				ret;
694 
695 	/*
696 	 * The child being updated has not been seen by this path at all; this
697 	 * path cannot be stale.
698 	 */
699 	if (!xino_bitmap_test(&path->seen_inodes, I_INO(p->ip)))
700 		return 0;
701 
702 	ret = xchk_dirpath_step_is_stale(dl, path, 0, idx, p, &cursor);
703 	if (ret != 0)
704 		return ret;
705 
706 	for (i = 1, idx = path->second_step; i < path->nr_steps; i++, idx++) {
707 		ret = xchk_dirpath_step_is_stale(dl, path, i, idx, p, &cursor);
708 		if (ret != 0)
709 			return ret;
710 	}
711 
712 	return 0;
713 }
714 
715 /*
716  * Decide if a directory update from the regular filesystem touches any of the
717  * paths we've scanned, and invalidate the scan data if true.
718  */
719 STATIC int
xchk_dirtree_live_update(struct notifier_block * nb,unsigned long action,void * data)720 xchk_dirtree_live_update(
721 	struct notifier_block		*nb,
722 	unsigned long			action,
723 	void				*data)
724 {
725 	struct xfs_dir_update_params	*p = data;
726 	struct xchk_dirtree		*dl;
727 	struct xchk_dirpath		*path;
728 	int				ret;
729 
730 	dl = container_of(nb, struct xchk_dirtree, dhook.dirent_hook.nb);
731 
732 	trace_xchk_dirtree_live_update(dl->sc, p->dp, action, p->ip, p->delta,
733 			p->name);
734 
735 	mutex_lock(&dl->lock);
736 
737 	if (dl->stale || dl->aborted)
738 		goto out_unlock;
739 
740 	xchk_dirtree_for_each_path(dl, path) {
741 		ret = xchk_dirpath_is_stale(dl, path, p);
742 		if (ret < 0) {
743 			dl->aborted = true;
744 			break;
745 		}
746 		if (ret == 1) {
747 			dl->stale = true;
748 			break;
749 		}
750 	}
751 
752 out_unlock:
753 	mutex_unlock(&dl->lock);
754 	return NOTIFY_DONE;
755 }
756 
757 /* Delete all the collected path information. */
758 STATIC void
xchk_dirtree_reset(void * buf)759 xchk_dirtree_reset(
760 	void			*buf)
761 {
762 	struct xchk_dirtree	*dl = buf;
763 	struct xchk_dirpath	*path, *n;
764 
765 	ASSERT(dl->sc->ilock_flags & XFS_ILOCK_EXCL);
766 
767 	xchk_dirtree_for_each_path_safe(dl, path, n) {
768 		list_del_init(&path->list);
769 		xino_bitmap_destroy(&path->seen_inodes);
770 		kfree(path);
771 	}
772 	dl->nr_paths = 0;
773 
774 	xfarray_truncate(dl->path_steps);
775 	xfblob_truncate(dl->path_names);
776 
777 	dl->stale = false;
778 }
779 
780 /*
781  * Load the name/pptr from the first step in this path into @dl->pptr_rec and
782  * @dl->xname.
783  */
784 STATIC int
xchk_dirtree_load_path(struct xchk_dirtree * dl,struct xchk_dirpath * path)785 xchk_dirtree_load_path(
786 	struct xchk_dirtree		*dl,
787 	struct xchk_dirpath		*path)
788 {
789 	struct xchk_dirpath_step	step;
790 	int				error;
791 
792 	error = xfarray_load(dl->path_steps, path->first_step, &step);
793 	if (error)
794 		return error;
795 
796 	error = xfblob_loadname(dl->path_names, step.name_cookie, &dl->xname,
797 			step.name_len);
798 	if (error)
799 		return error;
800 
801 	dl->pptr_rec = step.pptr_rec; /* struct copy */
802 	return 0;
803 }
804 
805 /*
806  * For each parent pointer of this subdir, trace a path upwards towards the
807  * root directory and record what we find.  Returns 0 for success;
808  * -EFSCORRUPTED if walking the parent pointers of @sc->ip failed, -ELNRNG if a
809  * path was too deep; -ENOSR if there were too many parent pointers; or
810  * a negative errno.
811  */
812 int
xchk_dirtree_find_paths_to_root(struct xchk_dirtree * dl)813 xchk_dirtree_find_paths_to_root(
814 	struct xchk_dirtree	*dl)
815 {
816 	struct xfs_scrub	*sc = dl->sc;
817 	struct xchk_dirpath	*path;
818 	int			error = 0;
819 
820 	do {
821 		if (xchk_should_terminate(sc, &error))
822 			return error;
823 
824 		xchk_dirtree_reset(dl);
825 
826 		/*
827 		 * If the extended attributes look as though they has been
828 		 * zapped by the inode record repair code, we cannot scan for
829 		 * parent pointers.
830 		 */
831 		if (xchk_pptr_looks_zapped(sc->ip)) {
832 			xchk_set_incomplete(sc);
833 			return -EBUSY;
834 		}
835 
836 		/*
837 		 * Create path walk contexts for each parent of the directory
838 		 * that is being scanned.  Directories are supposed to have
839 		 * only one parent, but this is how we detect multiple parents.
840 		 */
841 		error = xchk_xattr_walk(sc, sc->ip, xchk_dirtree_create_path,
842 				NULL, dl);
843 		if (error)
844 			return error;
845 
846 		xchk_dirtree_for_each_path(dl, path) {
847 			/* Load path components into dl->pptr/xname */
848 			error = xchk_dirtree_load_path(dl, path);
849 			if (error)
850 				return error;
851 
852 			/*
853 			 * Try to walk up each path to the root.  This enables
854 			 * us to find directory loops in ancestors, and the
855 			 * like.
856 			 */
857 			error = xchk_dirpath_walk_upwards(dl, path);
858 			if (error == -EFSCORRUPTED) {
859 				/*
860 				 * A parent pointer of @sc->ip is bad, don't
861 				 * bother continuing.
862 				 */
863 				break;
864 			}
865 			if (error == -ESTALE) {
866 				/* This had better be an invalidation. */
867 				ASSERT(dl->stale);
868 				break;
869 			}
870 			if (error)
871 				return error;
872 			if (dl->aborted)
873 				return 0;
874 		}
875 	} while (dl->stale);
876 
877 	return error;
878 }
879 
880 /*
881  * Figure out what to do with the paths we tried to find.  Do not call this
882  * if the scan results are stale.
883  */
884 void
xchk_dirtree_evaluate(struct xchk_dirtree * dl,struct xchk_dirtree_outcomes * oc)885 xchk_dirtree_evaluate(
886 	struct xchk_dirtree		*dl,
887 	struct xchk_dirtree_outcomes	*oc)
888 {
889 	struct xchk_dirpath		*path;
890 
891 	ASSERT(!dl->stale);
892 
893 	/* Scan the paths we have to decide what to do. */
894 	memset(oc, 0, sizeof(struct xchk_dirtree_outcomes));
895 	xchk_dirtree_for_each_path(dl, path) {
896 		trace_xchk_dirpath_evaluate_path(dl->sc, path->path_nr,
897 				path->nr_steps, path->outcome);
898 
899 		switch (path->outcome) {
900 		case XCHK_DIRPATH_SCANNING:
901 			/* shouldn't get here */
902 			ASSERT(0);
903 			break;
904 		case XCHK_DIRPATH_DELETE:
905 			/* This one is already going away. */
906 			oc->bad++;
907 			break;
908 		case XCHK_DIRPATH_CORRUPT:
909 		case XCHK_DIRPATH_LOOP:
910 			/* Couldn't find the end of this path. */
911 			oc->suspect++;
912 			break;
913 		case XCHK_DIRPATH_STALE:
914 			/* shouldn't get here either */
915 			ASSERT(0);
916 			break;
917 		case XCHK_DIRPATH_OK:
918 			/* This path got all the way to the root. */
919 			oc->good++;
920 			break;
921 		case XREP_DIRPATH_DELETING:
922 		case XREP_DIRPATH_DELETED:
923 		case XREP_DIRPATH_ADOPTING:
924 		case XREP_DIRPATH_ADOPTED:
925 			/* These should not be in progress! */
926 			ASSERT(0);
927 			break;
928 		}
929 	}
930 
931 	trace_xchk_dirtree_evaluate(dl, oc);
932 }
933 
934 /* Look for directory loops. */
935 int
xchk_dirtree(struct xfs_scrub * sc)936 xchk_dirtree(
937 	struct xfs_scrub		*sc)
938 {
939 	struct xchk_dirtree_outcomes	oc;
940 	struct xchk_dirtree		*dl = sc->buf;
941 	int				error;
942 
943 	/*
944 	 * Nondirectories do not point downwards to other files, so they cannot
945 	 * cause a cycle in the directory tree.
946 	 */
947 	if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
948 		return -ENOENT;
949 
950 	ASSERT(xfs_has_parent(sc->mp));
951 
952 	/*
953 	 * Find the root of the directory tree.  Remember which directory to
954 	 * scan, because the hook doesn't detach until after sc->ip gets
955 	 * released during teardown.
956 	 */
957 	dl->root_ino = xchk_inode_rootdir_inum(sc->ip);
958 	dl->scan_ino = I_INO(sc->ip);
959 
960 	trace_xchk_dirtree_start(sc->ip, sc->sm, 0);
961 
962 	/*
963 	 * Hook into the directory entry code so that we can capture updates to
964 	 * paths that we have already scanned.  The scanner thread takes each
965 	 * directory's ILOCK, which means that any in-progress directory update
966 	 * will finish before we can scan the directory.
967 	 */
968 	ASSERT(sc->flags & XCHK_FSGATES_DIRENTS);
969 	xfs_dir_hook_setup(&dl->dhook, xchk_dirtree_live_update);
970 	error = xfs_dir_hook_add(sc->mp, &dl->dhook);
971 	if (error)
972 		goto out;
973 
974 	mutex_lock(&dl->lock);
975 
976 	/* Trace each parent pointer's path to the root. */
977 	error = xchk_dirtree_find_paths_to_root(dl);
978 	if (error == -EFSCORRUPTED || error == -ELNRNG || error == -ENOSR) {
979 		/*
980 		 * Don't bother walking the paths if the xattr structure or the
981 		 * parent pointers are corrupt; this scan cannot be completed
982 		 * without full information.
983 		 */
984 		xchk_ip_xref_set_corrupt(sc, sc->ip);
985 		error = 0;
986 		goto out_scanlock;
987 	}
988 	if (error == -EBUSY) {
989 		/*
990 		 * We couldn't scan some directory's parent pointers because
991 		 * the attr fork looked like it had been zapped.  The
992 		 * scan was marked incomplete, so no further error code
993 		 * is necessary.
994 		 */
995 		error = 0;
996 		goto out_scanlock;
997 	}
998 	if (error)
999 		goto out_scanlock;
1000 	if (dl->aborted) {
1001 		xchk_set_incomplete(sc);
1002 		goto out_scanlock;
1003 	}
1004 
1005 	/* Assess what we found in our path evaluation. */
1006 	xchk_dirtree_evaluate(dl, &oc);
1007 	if (xchk_dirtree_parentless(dl)) {
1008 		if (oc.good || oc.bad || oc.suspect)
1009 			xchk_ip_set_corrupt(sc, sc->ip);
1010 	} else {
1011 		if (oc.bad || oc.good + oc.suspect != 1)
1012 			xchk_ip_set_corrupt(sc, sc->ip);
1013 		if (oc.suspect)
1014 			xchk_ip_xref_set_corrupt(sc, sc->ip);
1015 	}
1016 
1017 out_scanlock:
1018 	mutex_unlock(&dl->lock);
1019 out:
1020 	trace_xchk_dirtree_done(sc->ip, sc->sm, error);
1021 	return error;
1022 }
1023 
1024 /* Does the directory targeted by this scrub have no parents? */
1025 bool
xchk_dirtree_parentless(const struct xchk_dirtree * dl)1026 xchk_dirtree_parentless(const struct xchk_dirtree *dl)
1027 {
1028 	struct xfs_scrub	*sc = dl->sc;
1029 
1030 	if (xchk_inode_is_dirtree_root(sc->ip))
1031 		return true;
1032 	if (VFS_I(sc->ip)->i_nlink == 0)
1033 		return true;
1034 	return false;
1035 }
1036