1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  */
6 #include "xfs.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_btree.h"
13 #include "xfs_bit.h"
14 #include "xfs_log_format.h"
15 #include "xfs_trans.h"
16 #include "xfs_inode.h"
17 #include "xfs_alloc.h"
18 #include "xfs_bmap.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_rmap.h"
21 #include "xfs_rmap_btree.h"
22 #include "scrub/scrub.h"
23 #include "scrub/common.h"
24 #include "scrub/btree.h"
25 #include "xfs_ag.h"
26 
27 /* Set us up with an inode's bmap. */
28 int
xchk_setup_inode_bmap(struct xfs_scrub * sc)29 xchk_setup_inode_bmap(
30 	struct xfs_scrub	*sc)
31 {
32 	int			error;
33 
34 	error = xchk_get_inode(sc);
35 	if (error)
36 		goto out;
37 
38 	sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
39 	xfs_ilock(sc->ip, sc->ilock_flags);
40 
41 	/*
42 	 * We don't want any ephemeral data fork updates sitting around
43 	 * while we inspect block mappings, so wait for directio to finish
44 	 * and flush dirty data if we have delalloc reservations.
45 	 */
46 	if (S_ISREG(VFS_I(sc->ip)->i_mode) &&
47 	    sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) {
48 		struct address_space	*mapping = VFS_I(sc->ip)->i_mapping;
49 
50 		inode_dio_wait(VFS_I(sc->ip));
51 
52 		/*
53 		 * Try to flush all incore state to disk before we examine the
54 		 * space mappings for the data fork.  Leave accumulated errors
55 		 * in the mapping for the writer threads to consume.
56 		 *
57 		 * On ENOSPC or EIO writeback errors, we continue into the
58 		 * extent mapping checks because write failures do not
59 		 * necessarily imply anything about the correctness of the file
60 		 * metadata.  The metadata and the file data could be on
61 		 * completely separate devices; a media failure might only
62 		 * affect a subset of the disk, etc.  We can handle delalloc
63 		 * extents in the scrubber, so leaving them in memory is fine.
64 		 */
65 		error = filemap_fdatawrite(mapping);
66 		if (!error)
67 			error = filemap_fdatawait_keep_errors(mapping);
68 		if (error && (error != -ENOSPC && error != -EIO))
69 			goto out;
70 	}
71 
72 	/* Got the inode, lock it and we're ready to go. */
73 	error = xchk_trans_alloc(sc, 0);
74 	if (error)
75 		goto out;
76 	sc->ilock_flags |= XFS_ILOCK_EXCL;
77 	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
78 
79 out:
80 	/* scrub teardown will unlock and release the inode */
81 	return error;
82 }
83 
84 /*
85  * Inode fork block mapping (BMBT) scrubber.
86  * More complex than the others because we have to scrub
87  * all the extents regardless of whether or not the fork
88  * is in btree format.
89  */
90 
91 struct xchk_bmap_info {
92 	struct xfs_scrub	*sc;
93 	struct xfs_iext_cursor	icur;
94 	xfs_fileoff_t		lastoff;
95 	bool			is_rt;
96 	bool			is_shared;
97 	bool			was_loaded;
98 	int			whichfork;
99 };
100 
101 /* Look for a corresponding rmap for this irec. */
102 static inline bool
xchk_bmap_get_rmap(struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec,xfs_agblock_t agbno,uint64_t owner,struct xfs_rmap_irec * rmap)103 xchk_bmap_get_rmap(
104 	struct xchk_bmap_info	*info,
105 	struct xfs_bmbt_irec	*irec,
106 	xfs_agblock_t		agbno,
107 	uint64_t		owner,
108 	struct xfs_rmap_irec	*rmap)
109 {
110 	xfs_fileoff_t		offset;
111 	unsigned int		rflags = 0;
112 	int			has_rmap;
113 	int			error;
114 
115 	if (info->whichfork == XFS_ATTR_FORK)
116 		rflags |= XFS_RMAP_ATTR_FORK;
117 	if (irec->br_state == XFS_EXT_UNWRITTEN)
118 		rflags |= XFS_RMAP_UNWRITTEN;
119 
120 	/*
121 	 * CoW staging extents are owned (on disk) by the refcountbt, so
122 	 * their rmaps do not have offsets.
123 	 */
124 	if (info->whichfork == XFS_COW_FORK)
125 		offset = 0;
126 	else
127 		offset = irec->br_startoff;
128 
129 	/*
130 	 * If the caller thinks this could be a shared bmbt extent (IOWs,
131 	 * any data fork extent of a reflink inode) then we have to use the
132 	 * range rmap lookup to make sure we get the correct owner/offset.
133 	 */
134 	if (info->is_shared) {
135 		error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno,
136 				owner, offset, rflags, rmap, &has_rmap);
137 	} else {
138 		error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno,
139 				owner, offset, rflags, rmap, &has_rmap);
140 	}
141 	if (!xchk_should_check_xref(info->sc, &error, &info->sc->sa.rmap_cur))
142 		return false;
143 
144 	if (!has_rmap)
145 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
146 			irec->br_startoff);
147 	return has_rmap;
148 }
149 
150 static inline bool
xchk_bmap_has_prev(struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)151 xchk_bmap_has_prev(
152 	struct xchk_bmap_info	*info,
153 	struct xfs_bmbt_irec	*irec)
154 {
155 	struct xfs_bmbt_irec	got;
156 	struct xfs_ifork	*ifp;
157 
158 	ifp = xfs_ifork_ptr(info->sc->ip, info->whichfork);
159 
160 	if (!xfs_iext_peek_prev_extent(ifp, &info->icur, &got))
161 		return false;
162 	if (got.br_startoff + got.br_blockcount != irec->br_startoff)
163 		return false;
164 	if (got.br_startblock + got.br_blockcount != irec->br_startblock)
165 		return false;
166 	if (got.br_state != irec->br_state)
167 		return false;
168 	return true;
169 }
170 
171 static inline bool
xchk_bmap_has_next(struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)172 xchk_bmap_has_next(
173 	struct xchk_bmap_info	*info,
174 	struct xfs_bmbt_irec	*irec)
175 {
176 	struct xfs_bmbt_irec	got;
177 	struct xfs_ifork	*ifp;
178 
179 	ifp = xfs_ifork_ptr(info->sc->ip, info->whichfork);
180 
181 	if (!xfs_iext_peek_next_extent(ifp, &info->icur, &got))
182 		return false;
183 	if (irec->br_startoff + irec->br_blockcount != got.br_startoff)
184 		return false;
185 	if (irec->br_startblock + irec->br_blockcount != got.br_startblock)
186 		return false;
187 	if (got.br_state != irec->br_state)
188 		return false;
189 	return true;
190 }
191 
192 /* Make sure that we have rmapbt records for this extent. */
193 STATIC void
xchk_bmap_xref_rmap(struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec,xfs_agblock_t agbno)194 xchk_bmap_xref_rmap(
195 	struct xchk_bmap_info	*info,
196 	struct xfs_bmbt_irec	*irec,
197 	xfs_agblock_t		agbno)
198 {
199 	struct xfs_rmap_irec	rmap;
200 	unsigned long long	rmap_end;
201 	uint64_t		owner;
202 
203 	if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm))
204 		return;
205 
206 	if (info->whichfork == XFS_COW_FORK)
207 		owner = XFS_RMAP_OWN_COW;
208 	else
209 		owner = info->sc->ip->i_ino;
210 
211 	/* Find the rmap record for this irec. */
212 	if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap))
213 		return;
214 
215 	/* Check the rmap. */
216 	rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
217 	if (rmap.rm_startblock > agbno ||
218 	    agbno + irec->br_blockcount > rmap_end)
219 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
220 				irec->br_startoff);
221 
222 	/*
223 	 * Check the logical offsets if applicable.  CoW staging extents
224 	 * don't track logical offsets since the mappings only exist in
225 	 * memory.
226 	 */
227 	if (info->whichfork != XFS_COW_FORK) {
228 		rmap_end = (unsigned long long)rmap.rm_offset +
229 				rmap.rm_blockcount;
230 		if (rmap.rm_offset > irec->br_startoff ||
231 		    irec->br_startoff + irec->br_blockcount > rmap_end)
232 			xchk_fblock_xref_set_corrupt(info->sc,
233 					info->whichfork, irec->br_startoff);
234 	}
235 
236 	if (rmap.rm_owner != owner)
237 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
238 				irec->br_startoff);
239 
240 	/*
241 	 * Check for discrepancies between the unwritten flag in the irec and
242 	 * the rmap.  Note that the (in-memory) CoW fork distinguishes between
243 	 * unwritten and written extents, but we don't track that in the rmap
244 	 * records because the blocks are owned (on-disk) by the refcountbt,
245 	 * which doesn't track unwritten state.
246 	 */
247 	if (owner != XFS_RMAP_OWN_COW &&
248 	    !!(irec->br_state == XFS_EXT_UNWRITTEN) !=
249 	    !!(rmap.rm_flags & XFS_RMAP_UNWRITTEN))
250 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
251 				irec->br_startoff);
252 
253 	if (!!(info->whichfork == XFS_ATTR_FORK) !=
254 	    !!(rmap.rm_flags & XFS_RMAP_ATTR_FORK))
255 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
256 				irec->br_startoff);
257 	if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK)
258 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
259 				irec->br_startoff);
260 
261 	/*
262 	 * If the rmap starts before this bmbt record, make sure there's a bmbt
263 	 * record for the previous offset that is contiguous with this mapping.
264 	 * Skip this for CoW fork extents because the refcount btree (and not
265 	 * the inode) is the ondisk owner for those extents.
266 	 */
267 	if (info->whichfork != XFS_COW_FORK && rmap.rm_startblock < agbno &&
268 	    !xchk_bmap_has_prev(info, irec)) {
269 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
270 				irec->br_startoff);
271 		return;
272 	}
273 
274 	/*
275 	 * If the rmap ends after this bmbt record, make sure there's a bmbt
276 	 * record for the next offset that is contiguous with this mapping.
277 	 * Skip this for CoW fork extents because the refcount btree (and not
278 	 * the inode) is the ondisk owner for those extents.
279 	 */
280 	rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
281 	if (info->whichfork != XFS_COW_FORK &&
282 	    rmap_end > agbno + irec->br_blockcount &&
283 	    !xchk_bmap_has_next(info, irec)) {
284 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
285 				irec->br_startoff);
286 		return;
287 	}
288 }
289 
290 /* Cross-reference a single rtdev extent record. */
291 STATIC void
xchk_bmap_rt_iextent_xref(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)292 xchk_bmap_rt_iextent_xref(
293 	struct xfs_inode	*ip,
294 	struct xchk_bmap_info	*info,
295 	struct xfs_bmbt_irec	*irec)
296 {
297 	xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
298 			irec->br_blockcount);
299 }
300 
301 /* Cross-reference a single datadev extent record. */
302 STATIC void
xchk_bmap_iextent_xref(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)303 xchk_bmap_iextent_xref(
304 	struct xfs_inode	*ip,
305 	struct xchk_bmap_info	*info,
306 	struct xfs_bmbt_irec	*irec)
307 {
308 	struct xfs_mount	*mp = info->sc->mp;
309 	xfs_agnumber_t		agno;
310 	xfs_agblock_t		agbno;
311 	xfs_extlen_t		len;
312 	int			error;
313 
314 	agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
315 	agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
316 	len = irec->br_blockcount;
317 
318 	error = xchk_ag_init_existing(info->sc, agno, &info->sc->sa);
319 	if (!xchk_fblock_process_error(info->sc, info->whichfork,
320 			irec->br_startoff, &error))
321 		goto out_free;
322 
323 	xchk_xref_is_used_space(info->sc, agbno, len);
324 	xchk_xref_is_not_inode_chunk(info->sc, agbno, len);
325 	xchk_bmap_xref_rmap(info, irec, agbno);
326 	switch (info->whichfork) {
327 	case XFS_DATA_FORK:
328 		if (xfs_is_reflink_inode(info->sc->ip))
329 			break;
330 		fallthrough;
331 	case XFS_ATTR_FORK:
332 		xchk_xref_is_not_shared(info->sc, agbno,
333 				irec->br_blockcount);
334 		break;
335 	case XFS_COW_FORK:
336 		xchk_xref_is_cow_staging(info->sc, agbno,
337 				irec->br_blockcount);
338 		xchk_xref_is_not_shared(info->sc, agbno,
339 				irec->br_blockcount);
340 		break;
341 	}
342 
343 out_free:
344 	xchk_ag_free(info->sc, &info->sc->sa);
345 }
346 
347 /*
348  * Directories and attr forks should never have blocks that can't be addressed
349  * by a xfs_dablk_t.
350  */
351 STATIC void
xchk_bmap_dirattr_extent(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)352 xchk_bmap_dirattr_extent(
353 	struct xfs_inode	*ip,
354 	struct xchk_bmap_info	*info,
355 	struct xfs_bmbt_irec	*irec)
356 {
357 	struct xfs_mount	*mp = ip->i_mount;
358 	xfs_fileoff_t		off;
359 
360 	if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK)
361 		return;
362 
363 	if (!xfs_verify_dablk(mp, irec->br_startoff))
364 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
365 				irec->br_startoff);
366 
367 	off = irec->br_startoff + irec->br_blockcount - 1;
368 	if (!xfs_verify_dablk(mp, off))
369 		xchk_fblock_set_corrupt(info->sc, info->whichfork, off);
370 }
371 
372 /* Scrub a single extent record. */
373 STATIC void
xchk_bmap_iextent(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)374 xchk_bmap_iextent(
375 	struct xfs_inode	*ip,
376 	struct xchk_bmap_info	*info,
377 	struct xfs_bmbt_irec	*irec)
378 {
379 	struct xfs_mount	*mp = info->sc->mp;
380 
381 	/*
382 	 * Check for out-of-order extents.  This record could have come
383 	 * from the incore list, for which there is no ordering check.
384 	 */
385 	if (irec->br_startoff < info->lastoff)
386 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
387 				irec->br_startoff);
388 
389 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
390 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
391 				irec->br_startoff);
392 
393 	xchk_bmap_dirattr_extent(ip, info, irec);
394 
395 	/* There should never be a "hole" extent in either extent list. */
396 	if (irec->br_startblock == HOLESTARTBLOCK)
397 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
398 				irec->br_startoff);
399 
400 	/* Make sure the extent points to a valid place. */
401 	if (irec->br_blockcount > XFS_MAX_BMBT_EXTLEN)
402 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
403 				irec->br_startoff);
404 	if (info->is_rt &&
405 	    !xfs_verify_rtext(mp, irec->br_startblock, irec->br_blockcount))
406 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
407 				irec->br_startoff);
408 	if (!info->is_rt &&
409 	    !xfs_verify_fsbext(mp, irec->br_startblock, irec->br_blockcount))
410 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
411 				irec->br_startoff);
412 
413 	/* We don't allow unwritten extents on attr forks. */
414 	if (irec->br_state == XFS_EXT_UNWRITTEN &&
415 	    info->whichfork == XFS_ATTR_FORK)
416 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
417 				irec->br_startoff);
418 
419 	if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
420 		return;
421 
422 	if (info->is_rt)
423 		xchk_bmap_rt_iextent_xref(ip, info, irec);
424 	else
425 		xchk_bmap_iextent_xref(ip, info, irec);
426 }
427 
428 /* Scrub a bmbt record. */
429 STATIC int
xchk_bmapbt_rec(struct xchk_btree * bs,const union xfs_btree_rec * rec)430 xchk_bmapbt_rec(
431 	struct xchk_btree	*bs,
432 	const union xfs_btree_rec *rec)
433 {
434 	struct xfs_bmbt_irec	irec;
435 	struct xfs_bmbt_irec	iext_irec;
436 	struct xfs_iext_cursor	icur;
437 	struct xchk_bmap_info	*info = bs->private;
438 	struct xfs_inode	*ip = bs->cur->bc_ino.ip;
439 	struct xfs_buf		*bp = NULL;
440 	struct xfs_btree_block	*block;
441 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, info->whichfork);
442 	uint64_t		owner;
443 	int			i;
444 
445 	/*
446 	 * Check the owners of the btree blocks up to the level below
447 	 * the root since the verifiers don't do that.
448 	 */
449 	if (xfs_has_crc(bs->cur->bc_mp) &&
450 	    bs->cur->bc_levels[0].ptr == 1) {
451 		for (i = 0; i < bs->cur->bc_nlevels - 1; i++) {
452 			block = xfs_btree_get_block(bs->cur, i, &bp);
453 			owner = be64_to_cpu(block->bb_u.l.bb_owner);
454 			if (owner != ip->i_ino)
455 				xchk_fblock_set_corrupt(bs->sc,
456 						info->whichfork, 0);
457 		}
458 	}
459 
460 	/*
461 	 * Check that the incore extent tree contains an extent that matches
462 	 * this one exactly.  We validate those cached bmaps later, so we don't
463 	 * need to check them here.  If the incore extent tree was just loaded
464 	 * from disk by the scrubber, we assume that its contents match what's
465 	 * on disk (we still hold the ILOCK) and skip the equivalence check.
466 	 */
467 	if (!info->was_loaded)
468 		return 0;
469 
470 	xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
471 	if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur,
472 				&iext_irec) ||
473 	    irec.br_startoff != iext_irec.br_startoff ||
474 	    irec.br_startblock != iext_irec.br_startblock ||
475 	    irec.br_blockcount != iext_irec.br_blockcount ||
476 	    irec.br_state != iext_irec.br_state)
477 		xchk_fblock_set_corrupt(bs->sc, info->whichfork,
478 				irec.br_startoff);
479 	return 0;
480 }
481 
482 /* Scan the btree records. */
483 STATIC int
xchk_bmap_btree(struct xfs_scrub * sc,int whichfork,struct xchk_bmap_info * info)484 xchk_bmap_btree(
485 	struct xfs_scrub	*sc,
486 	int			whichfork,
487 	struct xchk_bmap_info	*info)
488 {
489 	struct xfs_owner_info	oinfo;
490 	struct xfs_ifork	*ifp = xfs_ifork_ptr(sc->ip, whichfork);
491 	struct xfs_mount	*mp = sc->mp;
492 	struct xfs_inode	*ip = sc->ip;
493 	struct xfs_btree_cur	*cur;
494 	int			error;
495 
496 	/* Load the incore bmap cache if it's not loaded. */
497 	info->was_loaded = !xfs_need_iread_extents(ifp);
498 
499 	error = xfs_iread_extents(sc->tp, ip, whichfork);
500 	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
501 		goto out;
502 
503 	/* Check the btree structure. */
504 	cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
505 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
506 	error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
507 	xfs_btree_del_cursor(cur, error);
508 out:
509 	return error;
510 }
511 
512 struct xchk_bmap_check_rmap_info {
513 	struct xfs_scrub	*sc;
514 	int			whichfork;
515 	struct xfs_iext_cursor	icur;
516 };
517 
518 /* Can we find bmaps that fit this rmap? */
519 STATIC int
xchk_bmap_check_rmap(struct xfs_btree_cur * cur,const struct xfs_rmap_irec * rec,void * priv)520 xchk_bmap_check_rmap(
521 	struct xfs_btree_cur		*cur,
522 	const struct xfs_rmap_irec	*rec,
523 	void				*priv)
524 {
525 	struct xfs_bmbt_irec		irec;
526 	struct xfs_rmap_irec		check_rec;
527 	struct xchk_bmap_check_rmap_info	*sbcri = priv;
528 	struct xfs_ifork		*ifp;
529 	struct xfs_scrub		*sc = sbcri->sc;
530 	bool				have_map;
531 
532 	/* Is this even the right fork? */
533 	if (rec->rm_owner != sc->ip->i_ino)
534 		return 0;
535 	if ((sbcri->whichfork == XFS_ATTR_FORK) ^
536 	    !!(rec->rm_flags & XFS_RMAP_ATTR_FORK))
537 		return 0;
538 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
539 		return 0;
540 
541 	/* Now look up the bmbt record. */
542 	ifp = xfs_ifork_ptr(sc->ip, sbcri->whichfork);
543 	if (!ifp) {
544 		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
545 				rec->rm_offset);
546 		goto out;
547 	}
548 	have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset,
549 			&sbcri->icur, &irec);
550 	if (!have_map)
551 		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
552 				rec->rm_offset);
553 	/*
554 	 * bmap extent record lengths are constrained to 2^21 blocks in length
555 	 * because of space constraints in the on-disk metadata structure.
556 	 * However, rmap extent record lengths are constrained only by AG
557 	 * length, so we have to loop through the bmbt to make sure that the
558 	 * entire rmap is covered by bmbt records.
559 	 */
560 	check_rec = *rec;
561 	while (have_map) {
562 		if (irec.br_startoff != check_rec.rm_offset)
563 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
564 					check_rec.rm_offset);
565 		if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp,
566 				cur->bc_ag.pag->pag_agno,
567 				check_rec.rm_startblock))
568 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
569 					check_rec.rm_offset);
570 		if (irec.br_blockcount > check_rec.rm_blockcount)
571 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
572 					check_rec.rm_offset);
573 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
574 			break;
575 		check_rec.rm_startblock += irec.br_blockcount;
576 		check_rec.rm_offset += irec.br_blockcount;
577 		check_rec.rm_blockcount -= irec.br_blockcount;
578 		if (check_rec.rm_blockcount == 0)
579 			break;
580 		have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec);
581 		if (!have_map)
582 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
583 					check_rec.rm_offset);
584 	}
585 
586 out:
587 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
588 		return -ECANCELED;
589 	return 0;
590 }
591 
592 /* Make sure each rmap has a corresponding bmbt entry. */
593 STATIC int
xchk_bmap_check_ag_rmaps(struct xfs_scrub * sc,int whichfork,struct xfs_perag * pag)594 xchk_bmap_check_ag_rmaps(
595 	struct xfs_scrub		*sc,
596 	int				whichfork,
597 	struct xfs_perag		*pag)
598 {
599 	struct xchk_bmap_check_rmap_info	sbcri;
600 	struct xfs_btree_cur		*cur;
601 	struct xfs_buf			*agf;
602 	int				error;
603 
604 	error = xfs_alloc_read_agf(pag, sc->tp, 0, &agf);
605 	if (error)
606 		return error;
607 
608 	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, pag);
609 
610 	sbcri.sc = sc;
611 	sbcri.whichfork = whichfork;
612 	error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri);
613 	if (error == -ECANCELED)
614 		error = 0;
615 
616 	xfs_btree_del_cursor(cur, error);
617 	xfs_trans_brelse(sc->tp, agf);
618 	return error;
619 }
620 
621 /* Make sure each rmap has a corresponding bmbt entry. */
622 STATIC int
xchk_bmap_check_rmaps(struct xfs_scrub * sc,int whichfork)623 xchk_bmap_check_rmaps(
624 	struct xfs_scrub	*sc,
625 	int			whichfork)
626 {
627 	struct xfs_ifork	*ifp = xfs_ifork_ptr(sc->ip, whichfork);
628 	struct xfs_perag	*pag;
629 	xfs_agnumber_t		agno;
630 	bool			zero_size;
631 	int			error;
632 
633 	if (!xfs_has_rmapbt(sc->mp) ||
634 	    whichfork == XFS_COW_FORK ||
635 	    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
636 		return 0;
637 
638 	/* Don't support realtime rmap checks yet. */
639 	if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK)
640 		return 0;
641 
642 	ASSERT(xfs_ifork_ptr(sc->ip, whichfork) != NULL);
643 
644 	/*
645 	 * Only do this for complex maps that are in btree format, or for
646 	 * situations where we would seem to have a size but zero extents.
647 	 * The inode repair code can zap broken iforks, which means we have
648 	 * to flag this bmap as corrupt if there are rmaps that need to be
649 	 * reattached.
650 	 */
651 
652 	if (whichfork == XFS_DATA_FORK)
653 		zero_size = i_size_read(VFS_I(sc->ip)) == 0;
654 	else
655 		zero_size = false;
656 
657 	if (ifp->if_format != XFS_DINODE_FMT_BTREE &&
658 	    (zero_size || ifp->if_nextents > 0))
659 		return 0;
660 
661 	for_each_perag(sc->mp, agno, pag) {
662 		error = xchk_bmap_check_ag_rmaps(sc, whichfork, pag);
663 		if (error ||
664 		    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
665 			xfs_perag_rele(pag);
666 			return error;
667 		}
668 	}
669 
670 	return 0;
671 }
672 
673 /* Scrub a delalloc reservation from the incore extent map tree. */
674 STATIC void
xchk_bmap_iextent_delalloc(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)675 xchk_bmap_iextent_delalloc(
676 	struct xfs_inode	*ip,
677 	struct xchk_bmap_info	*info,
678 	struct xfs_bmbt_irec	*irec)
679 {
680 	struct xfs_mount	*mp = info->sc->mp;
681 
682 	/*
683 	 * Check for out-of-order extents.  This record could have come
684 	 * from the incore list, for which there is no ordering check.
685 	 */
686 	if (irec->br_startoff < info->lastoff)
687 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
688 				irec->br_startoff);
689 
690 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
691 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
692 				irec->br_startoff);
693 
694 	/* Make sure the extent points to a valid place. */
695 	if (irec->br_blockcount > XFS_MAX_BMBT_EXTLEN)
696 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
697 				irec->br_startoff);
698 }
699 
700 /*
701  * Scrub an inode fork's block mappings.
702  *
703  * First we scan every record in every btree block, if applicable.
704  * Then we unconditionally scan the incore extent cache.
705  */
706 STATIC int
xchk_bmap(struct xfs_scrub * sc,int whichfork)707 xchk_bmap(
708 	struct xfs_scrub	*sc,
709 	int			whichfork)
710 {
711 	struct xfs_bmbt_irec	irec;
712 	struct xchk_bmap_info	info = { NULL };
713 	struct xfs_mount	*mp = sc->mp;
714 	struct xfs_inode	*ip = sc->ip;
715 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
716 	xfs_fileoff_t		endoff;
717 	int			error = 0;
718 
719 	/* Non-existent forks can be ignored. */
720 	if (!ifp)
721 		goto out;
722 
723 	info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
724 	info.whichfork = whichfork;
725 	info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
726 	info.sc = sc;
727 
728 	switch (whichfork) {
729 	case XFS_COW_FORK:
730 		/* No CoW forks on non-reflink inodes/filesystems. */
731 		if (!xfs_is_reflink_inode(ip)) {
732 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
733 			goto out;
734 		}
735 		break;
736 	case XFS_ATTR_FORK:
737 		if (!xfs_has_attr(mp) && !xfs_has_attr2(mp))
738 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
739 		break;
740 	default:
741 		ASSERT(whichfork == XFS_DATA_FORK);
742 		break;
743 	}
744 
745 	/* Check the fork values */
746 	switch (ifp->if_format) {
747 	case XFS_DINODE_FMT_UUID:
748 	case XFS_DINODE_FMT_DEV:
749 	case XFS_DINODE_FMT_LOCAL:
750 		/* No mappings to check. */
751 		if (whichfork == XFS_COW_FORK)
752 			xchk_fblock_set_corrupt(sc, whichfork, 0);
753 		goto out;
754 	case XFS_DINODE_FMT_EXTENTS:
755 		break;
756 	case XFS_DINODE_FMT_BTREE:
757 		if (whichfork == XFS_COW_FORK) {
758 			xchk_fblock_set_corrupt(sc, whichfork, 0);
759 			goto out;
760 		}
761 
762 		error = xchk_bmap_btree(sc, whichfork, &info);
763 		if (error)
764 			goto out;
765 		break;
766 	default:
767 		xchk_fblock_set_corrupt(sc, whichfork, 0);
768 		goto out;
769 	}
770 
771 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
772 		goto out;
773 
774 	/* Find the offset of the last extent in the mapping. */
775 	error = xfs_bmap_last_offset(ip, &endoff, whichfork);
776 	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
777 		goto out;
778 
779 	/* Scrub extent records. */
780 	info.lastoff = 0;
781 	ifp = xfs_ifork_ptr(ip, whichfork);
782 	for_each_xfs_iext(ifp, &info.icur, &irec) {
783 		if (xchk_should_terminate(sc, &error) ||
784 		    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
785 			goto out;
786 
787 		if (irec.br_startoff >= endoff) {
788 			xchk_fblock_set_corrupt(sc, whichfork,
789 					irec.br_startoff);
790 			goto out;
791 		}
792 
793 		if (isnullstartblock(irec.br_startblock))
794 			xchk_bmap_iextent_delalloc(ip, &info, &irec);
795 		else
796 			xchk_bmap_iextent(ip, &info, &irec);
797 		info.lastoff = irec.br_startoff + irec.br_blockcount;
798 	}
799 
800 	error = xchk_bmap_check_rmaps(sc, whichfork);
801 	if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error))
802 		goto out;
803 out:
804 	return error;
805 }
806 
807 /* Scrub an inode's data fork. */
808 int
xchk_bmap_data(struct xfs_scrub * sc)809 xchk_bmap_data(
810 	struct xfs_scrub	*sc)
811 {
812 	return xchk_bmap(sc, XFS_DATA_FORK);
813 }
814 
815 /* Scrub an inode's attr fork. */
816 int
xchk_bmap_attr(struct xfs_scrub * sc)817 xchk_bmap_attr(
818 	struct xfs_scrub	*sc)
819 {
820 	return xchk_bmap(sc, XFS_ATTR_FORK);
821 }
822 
823 /* Scrub an inode's CoW fork. */
824 int
xchk_bmap_cow(struct xfs_scrub * sc)825 xchk_bmap_cow(
826 	struct xfs_scrub	*sc)
827 {
828 	if (!xfs_is_reflink_inode(sc->ip))
829 		return -ENOENT;
830 
831 	return xchk_bmap(sc, XFS_COW_FORK);
832 }
833