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 	xfs_fileoff_t		lastoff;
94 	bool			is_rt;
95 	bool			is_shared;
96 	bool			was_loaded;
97 	int			whichfork;
98 };
99 
100 /* Look for a corresponding rmap for this irec. */
101 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)102 xchk_bmap_get_rmap(
103 	struct xchk_bmap_info	*info,
104 	struct xfs_bmbt_irec	*irec,
105 	xfs_agblock_t		agbno,
106 	uint64_t		owner,
107 	struct xfs_rmap_irec	*rmap)
108 {
109 	xfs_fileoff_t		offset;
110 	unsigned int		rflags = 0;
111 	int			has_rmap;
112 	int			error;
113 
114 	if (info->whichfork == XFS_ATTR_FORK)
115 		rflags |= XFS_RMAP_ATTR_FORK;
116 	if (irec->br_state == XFS_EXT_UNWRITTEN)
117 		rflags |= XFS_RMAP_UNWRITTEN;
118 
119 	/*
120 	 * CoW staging extents are owned (on disk) by the refcountbt, so
121 	 * their rmaps do not have offsets.
122 	 */
123 	if (info->whichfork == XFS_COW_FORK)
124 		offset = 0;
125 	else
126 		offset = irec->br_startoff;
127 
128 	/*
129 	 * If the caller thinks this could be a shared bmbt extent (IOWs,
130 	 * any data fork extent of a reflink inode) then we have to use the
131 	 * range rmap lookup to make sure we get the correct owner/offset.
132 	 */
133 	if (info->is_shared) {
134 		error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno,
135 				owner, offset, rflags, rmap, &has_rmap);
136 	} else {
137 		error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno,
138 				owner, offset, rflags, rmap, &has_rmap);
139 	}
140 	if (!xchk_should_check_xref(info->sc, &error, &info->sc->sa.rmap_cur))
141 		return false;
142 
143 	if (!has_rmap)
144 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
145 			irec->br_startoff);
146 	return has_rmap;
147 }
148 
149 /* Make sure that we have rmapbt records for this extent. */
150 STATIC void
xchk_bmap_xref_rmap(struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec,xfs_agblock_t agbno)151 xchk_bmap_xref_rmap(
152 	struct xchk_bmap_info	*info,
153 	struct xfs_bmbt_irec	*irec,
154 	xfs_agblock_t		agbno)
155 {
156 	struct xfs_rmap_irec	rmap;
157 	unsigned long long	rmap_end;
158 	uint64_t		owner;
159 
160 	if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm))
161 		return;
162 
163 	if (info->whichfork == XFS_COW_FORK)
164 		owner = XFS_RMAP_OWN_COW;
165 	else
166 		owner = info->sc->ip->i_ino;
167 
168 	/* Find the rmap record for this irec. */
169 	if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap))
170 		return;
171 
172 	/* Check the rmap. */
173 	rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
174 	if (rmap.rm_startblock > agbno ||
175 	    agbno + irec->br_blockcount > rmap_end)
176 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
177 				irec->br_startoff);
178 
179 	/*
180 	 * Check the logical offsets if applicable.  CoW staging extents
181 	 * don't track logical offsets since the mappings only exist in
182 	 * memory.
183 	 */
184 	if (info->whichfork != XFS_COW_FORK) {
185 		rmap_end = (unsigned long long)rmap.rm_offset +
186 				rmap.rm_blockcount;
187 		if (rmap.rm_offset > irec->br_startoff ||
188 		    irec->br_startoff + irec->br_blockcount > rmap_end)
189 			xchk_fblock_xref_set_corrupt(info->sc,
190 					info->whichfork, irec->br_startoff);
191 	}
192 
193 	if (rmap.rm_owner != owner)
194 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
195 				irec->br_startoff);
196 
197 	/*
198 	 * Check for discrepancies between the unwritten flag in the irec and
199 	 * the rmap.  Note that the (in-memory) CoW fork distinguishes between
200 	 * unwritten and written extents, but we don't track that in the rmap
201 	 * records because the blocks are owned (on-disk) by the refcountbt,
202 	 * which doesn't track unwritten state.
203 	 */
204 	if (owner != XFS_RMAP_OWN_COW &&
205 	    !!(irec->br_state == XFS_EXT_UNWRITTEN) !=
206 	    !!(rmap.rm_flags & XFS_RMAP_UNWRITTEN))
207 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
208 				irec->br_startoff);
209 
210 	if (!!(info->whichfork == XFS_ATTR_FORK) !=
211 	    !!(rmap.rm_flags & XFS_RMAP_ATTR_FORK))
212 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
213 				irec->br_startoff);
214 	if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK)
215 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
216 				irec->br_startoff);
217 }
218 
219 /* Cross-reference a single rtdev extent record. */
220 STATIC void
xchk_bmap_rt_iextent_xref(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)221 xchk_bmap_rt_iextent_xref(
222 	struct xfs_inode	*ip,
223 	struct xchk_bmap_info	*info,
224 	struct xfs_bmbt_irec	*irec)
225 {
226 	xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
227 			irec->br_blockcount);
228 }
229 
230 /* Cross-reference a single datadev extent record. */
231 STATIC void
xchk_bmap_iextent_xref(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)232 xchk_bmap_iextent_xref(
233 	struct xfs_inode	*ip,
234 	struct xchk_bmap_info	*info,
235 	struct xfs_bmbt_irec	*irec)
236 {
237 	struct xfs_mount	*mp = info->sc->mp;
238 	xfs_agnumber_t		agno;
239 	xfs_agblock_t		agbno;
240 	xfs_extlen_t		len;
241 	int			error;
242 
243 	agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
244 	agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
245 	len = irec->br_blockcount;
246 
247 	error = xchk_ag_init_existing(info->sc, agno, &info->sc->sa);
248 	if (!xchk_fblock_process_error(info->sc, info->whichfork,
249 			irec->br_startoff, &error))
250 		goto out_free;
251 
252 	xchk_xref_is_used_space(info->sc, agbno, len);
253 	xchk_xref_is_not_inode_chunk(info->sc, agbno, len);
254 	xchk_bmap_xref_rmap(info, irec, agbno);
255 	switch (info->whichfork) {
256 	case XFS_DATA_FORK:
257 		if (xfs_is_reflink_inode(info->sc->ip))
258 			break;
259 		fallthrough;
260 	case XFS_ATTR_FORK:
261 		xchk_xref_is_not_shared(info->sc, agbno,
262 				irec->br_blockcount);
263 		break;
264 	case XFS_COW_FORK:
265 		xchk_xref_is_cow_staging(info->sc, agbno,
266 				irec->br_blockcount);
267 		break;
268 	}
269 
270 out_free:
271 	xchk_ag_free(info->sc, &info->sc->sa);
272 }
273 
274 /*
275  * Directories and attr forks should never have blocks that can't be addressed
276  * by a xfs_dablk_t.
277  */
278 STATIC void
xchk_bmap_dirattr_extent(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)279 xchk_bmap_dirattr_extent(
280 	struct xfs_inode	*ip,
281 	struct xchk_bmap_info	*info,
282 	struct xfs_bmbt_irec	*irec)
283 {
284 	struct xfs_mount	*mp = ip->i_mount;
285 	xfs_fileoff_t		off;
286 
287 	if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK)
288 		return;
289 
290 	if (!xfs_verify_dablk(mp, irec->br_startoff))
291 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
292 				irec->br_startoff);
293 
294 	off = irec->br_startoff + irec->br_blockcount - 1;
295 	if (!xfs_verify_dablk(mp, off))
296 		xchk_fblock_set_corrupt(info->sc, info->whichfork, off);
297 }
298 
299 /* Scrub a single extent record. */
300 STATIC int
xchk_bmap_iextent(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)301 xchk_bmap_iextent(
302 	struct xfs_inode	*ip,
303 	struct xchk_bmap_info	*info,
304 	struct xfs_bmbt_irec	*irec)
305 {
306 	struct xfs_mount	*mp = info->sc->mp;
307 	int			error = 0;
308 
309 	/*
310 	 * Check for out-of-order extents.  This record could have come
311 	 * from the incore list, for which there is no ordering check.
312 	 */
313 	if (irec->br_startoff < info->lastoff)
314 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
315 				irec->br_startoff);
316 
317 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
318 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
319 				irec->br_startoff);
320 
321 	xchk_bmap_dirattr_extent(ip, info, irec);
322 
323 	/* There should never be a "hole" extent in either extent list. */
324 	if (irec->br_startblock == HOLESTARTBLOCK)
325 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
326 				irec->br_startoff);
327 
328 	/*
329 	 * Check for delalloc extents.  We never iterate the ones in the
330 	 * in-core extent scan, and we should never see these in the bmbt.
331 	 */
332 	if (isnullstartblock(irec->br_startblock))
333 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
334 				irec->br_startoff);
335 
336 	/* Make sure the extent points to a valid place. */
337 	if (irec->br_blockcount > XFS_MAX_BMBT_EXTLEN)
338 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
339 				irec->br_startoff);
340 	if (info->is_rt &&
341 	    !xfs_verify_rtext(mp, irec->br_startblock, irec->br_blockcount))
342 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
343 				irec->br_startoff);
344 	if (!info->is_rt &&
345 	    !xfs_verify_fsbext(mp, irec->br_startblock, irec->br_blockcount))
346 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
347 				irec->br_startoff);
348 
349 	/* We don't allow unwritten extents on attr forks. */
350 	if (irec->br_state == XFS_EXT_UNWRITTEN &&
351 	    info->whichfork == XFS_ATTR_FORK)
352 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
353 				irec->br_startoff);
354 
355 	if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
356 		return 0;
357 
358 	if (info->is_rt)
359 		xchk_bmap_rt_iextent_xref(ip, info, irec);
360 	else
361 		xchk_bmap_iextent_xref(ip, info, irec);
362 
363 	info->lastoff = irec->br_startoff + irec->br_blockcount;
364 	return error;
365 }
366 
367 /* Scrub a bmbt record. */
368 STATIC int
xchk_bmapbt_rec(struct xchk_btree * bs,const union xfs_btree_rec * rec)369 xchk_bmapbt_rec(
370 	struct xchk_btree	*bs,
371 	const union xfs_btree_rec *rec)
372 {
373 	struct xfs_bmbt_irec	irec;
374 	struct xfs_bmbt_irec	iext_irec;
375 	struct xfs_iext_cursor	icur;
376 	struct xchk_bmap_info	*info = bs->private;
377 	struct xfs_inode	*ip = bs->cur->bc_ino.ip;
378 	struct xfs_buf		*bp = NULL;
379 	struct xfs_btree_block	*block;
380 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, info->whichfork);
381 	uint64_t		owner;
382 	int			i;
383 
384 	/*
385 	 * Check the owners of the btree blocks up to the level below
386 	 * the root since the verifiers don't do that.
387 	 */
388 	if (xfs_has_crc(bs->cur->bc_mp) &&
389 	    bs->cur->bc_levels[0].ptr == 1) {
390 		for (i = 0; i < bs->cur->bc_nlevels - 1; i++) {
391 			block = xfs_btree_get_block(bs->cur, i, &bp);
392 			owner = be64_to_cpu(block->bb_u.l.bb_owner);
393 			if (owner != ip->i_ino)
394 				xchk_fblock_set_corrupt(bs->sc,
395 						info->whichfork, 0);
396 		}
397 	}
398 
399 	/*
400 	 * Check that the incore extent tree contains an extent that matches
401 	 * this one exactly.  We validate those cached bmaps later, so we don't
402 	 * need to check them here.  If the incore extent tree was just loaded
403 	 * from disk by the scrubber, we assume that its contents match what's
404 	 * on disk (we still hold the ILOCK) and skip the equivalence check.
405 	 */
406 	if (!info->was_loaded)
407 		return 0;
408 
409 	xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
410 	if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur,
411 				&iext_irec) ||
412 	    irec.br_startoff != iext_irec.br_startoff ||
413 	    irec.br_startblock != iext_irec.br_startblock ||
414 	    irec.br_blockcount != iext_irec.br_blockcount ||
415 	    irec.br_state != iext_irec.br_state)
416 		xchk_fblock_set_corrupt(bs->sc, info->whichfork,
417 				irec.br_startoff);
418 	return 0;
419 }
420 
421 /* Scan the btree records. */
422 STATIC int
xchk_bmap_btree(struct xfs_scrub * sc,int whichfork,struct xchk_bmap_info * info)423 xchk_bmap_btree(
424 	struct xfs_scrub	*sc,
425 	int			whichfork,
426 	struct xchk_bmap_info	*info)
427 {
428 	struct xfs_owner_info	oinfo;
429 	struct xfs_ifork	*ifp = xfs_ifork_ptr(sc->ip, whichfork);
430 	struct xfs_mount	*mp = sc->mp;
431 	struct xfs_inode	*ip = sc->ip;
432 	struct xfs_btree_cur	*cur;
433 	int			error;
434 
435 	/* Load the incore bmap cache if it's not loaded. */
436 	info->was_loaded = !xfs_need_iread_extents(ifp);
437 
438 	error = xfs_iread_extents(sc->tp, ip, whichfork);
439 	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
440 		goto out;
441 
442 	/* Check the btree structure. */
443 	cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
444 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
445 	error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
446 	xfs_btree_del_cursor(cur, error);
447 out:
448 	return error;
449 }
450 
451 struct xchk_bmap_check_rmap_info {
452 	struct xfs_scrub	*sc;
453 	int			whichfork;
454 	struct xfs_iext_cursor	icur;
455 };
456 
457 /* Can we find bmaps that fit this rmap? */
458 STATIC int
xchk_bmap_check_rmap(struct xfs_btree_cur * cur,const struct xfs_rmap_irec * rec,void * priv)459 xchk_bmap_check_rmap(
460 	struct xfs_btree_cur		*cur,
461 	const struct xfs_rmap_irec	*rec,
462 	void				*priv)
463 {
464 	struct xfs_bmbt_irec		irec;
465 	struct xfs_rmap_irec		check_rec;
466 	struct xchk_bmap_check_rmap_info	*sbcri = priv;
467 	struct xfs_ifork		*ifp;
468 	struct xfs_scrub		*sc = sbcri->sc;
469 	bool				have_map;
470 
471 	/* Is this even the right fork? */
472 	if (rec->rm_owner != sc->ip->i_ino)
473 		return 0;
474 	if ((sbcri->whichfork == XFS_ATTR_FORK) ^
475 	    !!(rec->rm_flags & XFS_RMAP_ATTR_FORK))
476 		return 0;
477 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
478 		return 0;
479 
480 	/* Now look up the bmbt record. */
481 	ifp = xfs_ifork_ptr(sc->ip, sbcri->whichfork);
482 	if (!ifp) {
483 		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
484 				rec->rm_offset);
485 		goto out;
486 	}
487 	have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset,
488 			&sbcri->icur, &irec);
489 	if (!have_map)
490 		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
491 				rec->rm_offset);
492 	/*
493 	 * bmap extent record lengths are constrained to 2^21 blocks in length
494 	 * because of space constraints in the on-disk metadata structure.
495 	 * However, rmap extent record lengths are constrained only by AG
496 	 * length, so we have to loop through the bmbt to make sure that the
497 	 * entire rmap is covered by bmbt records.
498 	 */
499 	check_rec = *rec;
500 	while (have_map) {
501 		if (irec.br_startoff != check_rec.rm_offset)
502 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
503 					check_rec.rm_offset);
504 		if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp,
505 				cur->bc_ag.pag->pag_agno,
506 				check_rec.rm_startblock))
507 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
508 					check_rec.rm_offset);
509 		if (irec.br_blockcount > check_rec.rm_blockcount)
510 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
511 					check_rec.rm_offset);
512 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
513 			break;
514 		check_rec.rm_startblock += irec.br_blockcount;
515 		check_rec.rm_offset += irec.br_blockcount;
516 		check_rec.rm_blockcount -= irec.br_blockcount;
517 		if (check_rec.rm_blockcount == 0)
518 			break;
519 		have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec);
520 		if (!have_map)
521 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
522 					check_rec.rm_offset);
523 	}
524 
525 out:
526 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
527 		return -ECANCELED;
528 	return 0;
529 }
530 
531 /* Make sure each rmap has a corresponding bmbt entry. */
532 STATIC int
xchk_bmap_check_ag_rmaps(struct xfs_scrub * sc,int whichfork,struct xfs_perag * pag)533 xchk_bmap_check_ag_rmaps(
534 	struct xfs_scrub		*sc,
535 	int				whichfork,
536 	struct xfs_perag		*pag)
537 {
538 	struct xchk_bmap_check_rmap_info	sbcri;
539 	struct xfs_btree_cur		*cur;
540 	struct xfs_buf			*agf;
541 	int				error;
542 
543 	error = xfs_alloc_read_agf(pag, sc->tp, 0, &agf);
544 	if (error)
545 		return error;
546 
547 	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, pag);
548 
549 	sbcri.sc = sc;
550 	sbcri.whichfork = whichfork;
551 	error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri);
552 	if (error == -ECANCELED)
553 		error = 0;
554 
555 	xfs_btree_del_cursor(cur, error);
556 	xfs_trans_brelse(sc->tp, agf);
557 	return error;
558 }
559 
560 /* Make sure each rmap has a corresponding bmbt entry. */
561 STATIC int
xchk_bmap_check_rmaps(struct xfs_scrub * sc,int whichfork)562 xchk_bmap_check_rmaps(
563 	struct xfs_scrub	*sc,
564 	int			whichfork)
565 {
566 	struct xfs_ifork	*ifp = xfs_ifork_ptr(sc->ip, whichfork);
567 	struct xfs_perag	*pag;
568 	xfs_agnumber_t		agno;
569 	bool			zero_size;
570 	int			error;
571 
572 	if (!xfs_has_rmapbt(sc->mp) ||
573 	    whichfork == XFS_COW_FORK ||
574 	    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
575 		return 0;
576 
577 	/* Don't support realtime rmap checks yet. */
578 	if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK)
579 		return 0;
580 
581 	ASSERT(xfs_ifork_ptr(sc->ip, whichfork) != NULL);
582 
583 	/*
584 	 * Only do this for complex maps that are in btree format, or for
585 	 * situations where we would seem to have a size but zero extents.
586 	 * The inode repair code can zap broken iforks, which means we have
587 	 * to flag this bmap as corrupt if there are rmaps that need to be
588 	 * reattached.
589 	 */
590 
591 	if (whichfork == XFS_DATA_FORK)
592 		zero_size = i_size_read(VFS_I(sc->ip)) == 0;
593 	else
594 		zero_size = false;
595 
596 	if (ifp->if_format != XFS_DINODE_FMT_BTREE &&
597 	    (zero_size || ifp->if_nextents > 0))
598 		return 0;
599 
600 	for_each_perag(sc->mp, agno, pag) {
601 		error = xchk_bmap_check_ag_rmaps(sc, whichfork, pag);
602 		if (error)
603 			break;
604 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
605 			break;
606 	}
607 	if (pag)
608 		xfs_perag_put(pag);
609 	return error;
610 }
611 
612 /*
613  * Scrub an inode fork's block mappings.
614  *
615  * First we scan every record in every btree block, if applicable.
616  * Then we unconditionally scan the incore extent cache.
617  */
618 STATIC int
xchk_bmap(struct xfs_scrub * sc,int whichfork)619 xchk_bmap(
620 	struct xfs_scrub	*sc,
621 	int			whichfork)
622 {
623 	struct xfs_bmbt_irec	irec;
624 	struct xchk_bmap_info	info = { NULL };
625 	struct xfs_mount	*mp = sc->mp;
626 	struct xfs_inode	*ip = sc->ip;
627 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
628 	xfs_fileoff_t		endoff;
629 	struct xfs_iext_cursor	icur;
630 	int			error = 0;
631 
632 	/* Non-existent forks can be ignored. */
633 	if (!ifp)
634 		goto out;
635 
636 	info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
637 	info.whichfork = whichfork;
638 	info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
639 	info.sc = sc;
640 
641 	switch (whichfork) {
642 	case XFS_COW_FORK:
643 		/* No CoW forks on non-reflink inodes/filesystems. */
644 		if (!xfs_is_reflink_inode(ip)) {
645 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
646 			goto out;
647 		}
648 		break;
649 	case XFS_ATTR_FORK:
650 		if (!xfs_has_attr(mp) && !xfs_has_attr2(mp))
651 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
652 		break;
653 	default:
654 		ASSERT(whichfork == XFS_DATA_FORK);
655 		break;
656 	}
657 
658 	/* Check the fork values */
659 	switch (ifp->if_format) {
660 	case XFS_DINODE_FMT_UUID:
661 	case XFS_DINODE_FMT_DEV:
662 	case XFS_DINODE_FMT_LOCAL:
663 		/* No mappings to check. */
664 		goto out;
665 	case XFS_DINODE_FMT_EXTENTS:
666 		break;
667 	case XFS_DINODE_FMT_BTREE:
668 		if (whichfork == XFS_COW_FORK) {
669 			xchk_fblock_set_corrupt(sc, whichfork, 0);
670 			goto out;
671 		}
672 
673 		error = xchk_bmap_btree(sc, whichfork, &info);
674 		if (error)
675 			goto out;
676 		break;
677 	default:
678 		xchk_fblock_set_corrupt(sc, whichfork, 0);
679 		goto out;
680 	}
681 
682 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
683 		goto out;
684 
685 	/* Find the offset of the last extent in the mapping. */
686 	error = xfs_bmap_last_offset(ip, &endoff, whichfork);
687 	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
688 		goto out;
689 
690 	/* Scrub extent records. */
691 	info.lastoff = 0;
692 	ifp = xfs_ifork_ptr(ip, whichfork);
693 	for_each_xfs_iext(ifp, &icur, &irec) {
694 		if (xchk_should_terminate(sc, &error) ||
695 		    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
696 			goto out;
697 		if (isnullstartblock(irec.br_startblock))
698 			continue;
699 		if (irec.br_startoff >= endoff) {
700 			xchk_fblock_set_corrupt(sc, whichfork,
701 					irec.br_startoff);
702 			goto out;
703 		}
704 		error = xchk_bmap_iextent(ip, &info, &irec);
705 		if (error)
706 			goto out;
707 	}
708 
709 	error = xchk_bmap_check_rmaps(sc, whichfork);
710 	if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error))
711 		goto out;
712 out:
713 	return error;
714 }
715 
716 /* Scrub an inode's data fork. */
717 int
xchk_bmap_data(struct xfs_scrub * sc)718 xchk_bmap_data(
719 	struct xfs_scrub	*sc)
720 {
721 	return xchk_bmap(sc, XFS_DATA_FORK);
722 }
723 
724 /* Scrub an inode's attr fork. */
725 int
xchk_bmap_attr(struct xfs_scrub * sc)726 xchk_bmap_attr(
727 	struct xfs_scrub	*sc)
728 {
729 	return xchk_bmap(sc, XFS_ATTR_FORK);
730 }
731 
732 /* Scrub an inode's CoW fork. */
733 int
xchk_bmap_cow(struct xfs_scrub * sc)734 xchk_bmap_cow(
735 	struct xfs_scrub	*sc)
736 {
737 	if (!xfs_is_reflink_inode(sc->ip))
738 		return -ENOENT;
739 
740 	return xchk_bmap(sc, XFS_COW_FORK);
741 }
742