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 
26 /* Set us up with an inode's bmap. */
27 int
xchk_setup_inode_bmap(struct xfs_scrub * sc,struct xfs_inode * ip)28 xchk_setup_inode_bmap(
29 	struct xfs_scrub	*sc,
30 	struct xfs_inode	*ip)
31 {
32 	int			error;
33 
34 	error = xchk_get_inode(sc, ip);
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 		inode_dio_wait(VFS_I(sc->ip));
49 		error = filemap_write_and_wait(VFS_I(sc->ip)->i_mapping);
50 		if (error)
51 			goto out;
52 	}
53 
54 	/* Got the inode, lock it and we're ready to go. */
55 	error = xchk_trans_alloc(sc, 0);
56 	if (error)
57 		goto out;
58 	sc->ilock_flags |= XFS_ILOCK_EXCL;
59 	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
60 
61 out:
62 	/* scrub teardown will unlock and release the inode */
63 	return error;
64 }
65 
66 /*
67  * Inode fork block mapping (BMBT) scrubber.
68  * More complex than the others because we have to scrub
69  * all the extents regardless of whether or not the fork
70  * is in btree format.
71  */
72 
73 struct xchk_bmap_info {
74 	struct xfs_scrub	*sc;
75 	xfs_fileoff_t		lastoff;
76 	bool			is_rt;
77 	bool			is_shared;
78 	bool			was_loaded;
79 	int			whichfork;
80 };
81 
82 /* Look for a corresponding rmap for this irec. */
83 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)84 xchk_bmap_get_rmap(
85 	struct xchk_bmap_info	*info,
86 	struct xfs_bmbt_irec	*irec,
87 	xfs_agblock_t		agbno,
88 	uint64_t		owner,
89 	struct xfs_rmap_irec	*rmap)
90 {
91 	xfs_fileoff_t		offset;
92 	unsigned int		rflags = 0;
93 	int			has_rmap;
94 	int			error;
95 
96 	if (info->whichfork == XFS_ATTR_FORK)
97 		rflags |= XFS_RMAP_ATTR_FORK;
98 
99 	/*
100 	 * CoW staging extents are owned (on disk) by the refcountbt, so
101 	 * their rmaps do not have offsets.
102 	 */
103 	if (info->whichfork == XFS_COW_FORK)
104 		offset = 0;
105 	else
106 		offset = irec->br_startoff;
107 
108 	/*
109 	 * If the caller thinks this could be a shared bmbt extent (IOWs,
110 	 * any data fork extent of a reflink inode) then we have to use the
111 	 * range rmap lookup to make sure we get the correct owner/offset.
112 	 */
113 	if (info->is_shared) {
114 		error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno,
115 				owner, offset, rflags, rmap, &has_rmap);
116 		if (!xchk_should_check_xref(info->sc, &error,
117 				&info->sc->sa.rmap_cur))
118 			return false;
119 		goto out;
120 	}
121 
122 	/*
123 	 * Otherwise, use the (faster) regular lookup.
124 	 */
125 	error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno, 0, owner,
126 			offset, rflags, &has_rmap);
127 	if (!xchk_should_check_xref(info->sc, &error,
128 			&info->sc->sa.rmap_cur))
129 		return false;
130 	if (!has_rmap)
131 		goto out;
132 
133 	error = xfs_rmap_get_rec(info->sc->sa.rmap_cur, rmap, &has_rmap);
134 	if (!xchk_should_check_xref(info->sc, &error,
135 			&info->sc->sa.rmap_cur))
136 		return false;
137 
138 out:
139 	if (!has_rmap)
140 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
141 			irec->br_startoff);
142 	return has_rmap;
143 }
144 
145 /* Make sure that we have rmapbt records for this extent. */
146 STATIC void
xchk_bmap_xref_rmap(struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec,xfs_agblock_t agbno)147 xchk_bmap_xref_rmap(
148 	struct xchk_bmap_info	*info,
149 	struct xfs_bmbt_irec	*irec,
150 	xfs_agblock_t		agbno)
151 {
152 	struct xfs_rmap_irec	rmap;
153 	unsigned long long	rmap_end;
154 	uint64_t		owner;
155 
156 	if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm))
157 		return;
158 
159 	if (info->whichfork == XFS_COW_FORK)
160 		owner = XFS_RMAP_OWN_COW;
161 	else
162 		owner = info->sc->ip->i_ino;
163 
164 	/* Find the rmap record for this irec. */
165 	if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap))
166 		return;
167 
168 	/* Check the rmap. */
169 	rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
170 	if (rmap.rm_startblock > agbno ||
171 	    agbno + irec->br_blockcount > rmap_end)
172 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
173 				irec->br_startoff);
174 
175 	/*
176 	 * Check the logical offsets if applicable.  CoW staging extents
177 	 * don't track logical offsets since the mappings only exist in
178 	 * memory.
179 	 */
180 	if (info->whichfork != XFS_COW_FORK) {
181 		rmap_end = (unsigned long long)rmap.rm_offset +
182 				rmap.rm_blockcount;
183 		if (rmap.rm_offset > irec->br_startoff ||
184 		    irec->br_startoff + irec->br_blockcount > rmap_end)
185 			xchk_fblock_xref_set_corrupt(info->sc,
186 					info->whichfork, irec->br_startoff);
187 	}
188 
189 	if (rmap.rm_owner != owner)
190 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
191 				irec->br_startoff);
192 
193 	/*
194 	 * Check for discrepancies between the unwritten flag in the irec and
195 	 * the rmap.  Note that the (in-memory) CoW fork distinguishes between
196 	 * unwritten and written extents, but we don't track that in the rmap
197 	 * records because the blocks are owned (on-disk) by the refcountbt,
198 	 * which doesn't track unwritten state.
199 	 */
200 	if (owner != XFS_RMAP_OWN_COW &&
201 	    irec->br_state == XFS_EXT_UNWRITTEN &&
202 	    !(rmap.rm_flags & XFS_RMAP_UNWRITTEN))
203 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
204 				irec->br_startoff);
205 
206 	if (info->whichfork == XFS_ATTR_FORK &&
207 	    !(rmap.rm_flags & XFS_RMAP_ATTR_FORK))
208 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
209 				irec->br_startoff);
210 	if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK)
211 		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
212 				irec->br_startoff);
213 }
214 
215 /* Cross-reference a single rtdev extent record. */
216 STATIC void
xchk_bmap_rt_iextent_xref(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)217 xchk_bmap_rt_iextent_xref(
218 	struct xfs_inode	*ip,
219 	struct xchk_bmap_info	*info,
220 	struct xfs_bmbt_irec	*irec)
221 {
222 	xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
223 			irec->br_blockcount);
224 }
225 
226 /* Cross-reference a single datadev extent record. */
227 STATIC void
xchk_bmap_iextent_xref(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)228 xchk_bmap_iextent_xref(
229 	struct xfs_inode	*ip,
230 	struct xchk_bmap_info	*info,
231 	struct xfs_bmbt_irec	*irec)
232 {
233 	struct xfs_mount	*mp = info->sc->mp;
234 	xfs_agnumber_t		agno;
235 	xfs_agblock_t		agbno;
236 	xfs_extlen_t		len;
237 	int			error;
238 
239 	agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
240 	agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
241 	len = irec->br_blockcount;
242 
243 	error = xchk_ag_init(info->sc, agno, &info->sc->sa);
244 	if (!xchk_fblock_process_error(info->sc, info->whichfork,
245 			irec->br_startoff, &error))
246 		return;
247 
248 	xchk_xref_is_used_space(info->sc, agbno, len);
249 	xchk_xref_is_not_inode_chunk(info->sc, agbno, len);
250 	xchk_bmap_xref_rmap(info, irec, agbno);
251 	switch (info->whichfork) {
252 	case XFS_DATA_FORK:
253 		if (xfs_is_reflink_inode(info->sc->ip))
254 			break;
255 		/* fall through */
256 	case XFS_ATTR_FORK:
257 		xchk_xref_is_not_shared(info->sc, agbno,
258 				irec->br_blockcount);
259 		break;
260 	case XFS_COW_FORK:
261 		xchk_xref_is_cow_staging(info->sc, agbno,
262 				irec->br_blockcount);
263 		break;
264 	}
265 
266 	xchk_ag_free(info->sc, &info->sc->sa);
267 }
268 
269 /*
270  * Directories and attr forks should never have blocks that can't be addressed
271  * by a xfs_dablk_t.
272  */
273 STATIC void
xchk_bmap_dirattr_extent(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)274 xchk_bmap_dirattr_extent(
275 	struct xfs_inode	*ip,
276 	struct xchk_bmap_info	*info,
277 	struct xfs_bmbt_irec	*irec)
278 {
279 	struct xfs_mount	*mp = ip->i_mount;
280 	xfs_fileoff_t		off;
281 
282 	if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK)
283 		return;
284 
285 	if (!xfs_verify_dablk(mp, irec->br_startoff))
286 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
287 				irec->br_startoff);
288 
289 	off = irec->br_startoff + irec->br_blockcount - 1;
290 	if (!xfs_verify_dablk(mp, off))
291 		xchk_fblock_set_corrupt(info->sc, info->whichfork, off);
292 }
293 
294 /* Scrub a single extent record. */
295 STATIC int
xchk_bmap_iextent(struct xfs_inode * ip,struct xchk_bmap_info * info,struct xfs_bmbt_irec * irec)296 xchk_bmap_iextent(
297 	struct xfs_inode	*ip,
298 	struct xchk_bmap_info	*info,
299 	struct xfs_bmbt_irec	*irec)
300 {
301 	struct xfs_mount	*mp = info->sc->mp;
302 	xfs_filblks_t		end;
303 	int			error = 0;
304 
305 	/*
306 	 * Check for out-of-order extents.  This record could have come
307 	 * from the incore list, for which there is no ordering check.
308 	 */
309 	if (irec->br_startoff < info->lastoff)
310 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
311 				irec->br_startoff);
312 
313 	xchk_bmap_dirattr_extent(ip, info, irec);
314 
315 	/* There should never be a "hole" extent in either extent list. */
316 	if (irec->br_startblock == HOLESTARTBLOCK)
317 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
318 				irec->br_startoff);
319 
320 	/*
321 	 * Check for delalloc extents.  We never iterate the ones in the
322 	 * in-core extent scan, and we should never see these in the bmbt.
323 	 */
324 	if (isnullstartblock(irec->br_startblock))
325 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
326 				irec->br_startoff);
327 
328 	/* Make sure the extent points to a valid place. */
329 	if (irec->br_blockcount > MAXEXTLEN)
330 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
331 				irec->br_startoff);
332 	if (irec->br_startblock + irec->br_blockcount <= irec->br_startblock)
333 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
334 				irec->br_startoff);
335 	end = irec->br_startblock + irec->br_blockcount - 1;
336 	if (info->is_rt &&
337 	    (!xfs_verify_rtbno(mp, irec->br_startblock) ||
338 	     !xfs_verify_rtbno(mp, end)))
339 		xchk_fblock_set_corrupt(info->sc, info->whichfork,
340 				irec->br_startoff);
341 	if (!info->is_rt &&
342 	    (!xfs_verify_fsbno(mp, irec->br_startblock) ||
343 	     !xfs_verify_fsbno(mp, end) ||
344 	     XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
345 				XFS_FSB_TO_AGNO(mp, end)))
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,union xfs_btree_rec * rec)369 xchk_bmapbt_rec(
370 	struct xchk_btree	*bs,
371 	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_private.b.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_sb_version_hascrc(&bs->cur->bc_mp->m_sb) &&
389 	    bs->cur->bc_ptrs[0] == 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 = ifp->if_flags & XFS_IFEXTENTS;
437 	if (!info->was_loaded) {
438 		error = xfs_iread_extents(sc->tp, ip, whichfork);
439 		if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
440 			goto out;
441 	}
442 
443 	/* Check the btree structure. */
444 	cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
445 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
446 	error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
447 	xfs_btree_del_cursor(cur, error);
448 out:
449 	return error;
450 }
451 
452 struct xchk_bmap_check_rmap_info {
453 	struct xfs_scrub	*sc;
454 	int			whichfork;
455 	struct xfs_iext_cursor	icur;
456 };
457 
458 /* Can we find bmaps that fit this rmap? */
459 STATIC int
xchk_bmap_check_rmap(struct xfs_btree_cur * cur,struct xfs_rmap_irec * rec,void * priv)460 xchk_bmap_check_rmap(
461 	struct xfs_btree_cur		*cur,
462 	struct xfs_rmap_irec		*rec,
463 	void				*priv)
464 {
465 	struct xfs_bmbt_irec		irec;
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 	while (have_map) {
500 		if (irec.br_startoff != rec->rm_offset)
501 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
502 					rec->rm_offset);
503 		if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp,
504 				cur->bc_private.a.agno, rec->rm_startblock))
505 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
506 					rec->rm_offset);
507 		if (irec.br_blockcount > rec->rm_blockcount)
508 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
509 					rec->rm_offset);
510 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
511 			break;
512 		rec->rm_startblock += irec.br_blockcount;
513 		rec->rm_offset += irec.br_blockcount;
514 		rec->rm_blockcount -= irec.br_blockcount;
515 		if (rec->rm_blockcount == 0)
516 			break;
517 		have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec);
518 		if (!have_map)
519 			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
520 					rec->rm_offset);
521 	}
522 
523 out:
524 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
525 		return -ECANCELED;
526 	return 0;
527 }
528 
529 /* Make sure each rmap has a corresponding bmbt entry. */
530 STATIC int
xchk_bmap_check_ag_rmaps(struct xfs_scrub * sc,int whichfork,xfs_agnumber_t agno)531 xchk_bmap_check_ag_rmaps(
532 	struct xfs_scrub		*sc,
533 	int				whichfork,
534 	xfs_agnumber_t			agno)
535 {
536 	struct xchk_bmap_check_rmap_info	sbcri;
537 	struct xfs_btree_cur		*cur;
538 	struct xfs_buf			*agf;
539 	int				error;
540 
541 	error = xfs_alloc_read_agf(sc->mp, sc->tp, agno, 0, &agf);
542 	if (error)
543 		return error;
544 
545 	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, agno);
546 	if (!cur) {
547 		error = -ENOMEM;
548 		goto out_agf;
549 	}
550 
551 	sbcri.sc = sc;
552 	sbcri.whichfork = whichfork;
553 	error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri);
554 	if (error == -ECANCELED)
555 		error = 0;
556 
557 	xfs_btree_del_cursor(cur, error);
558 out_agf:
559 	xfs_trans_brelse(sc->tp, agf);
560 	return error;
561 }
562 
563 /* Make sure each rmap has a corresponding bmbt entry. */
564 STATIC int
xchk_bmap_check_rmaps(struct xfs_scrub * sc,int whichfork)565 xchk_bmap_check_rmaps(
566 	struct xfs_scrub	*sc,
567 	int			whichfork)
568 {
569 	loff_t			size;
570 	xfs_agnumber_t		agno;
571 	int			error;
572 
573 	if (!xfs_sb_version_hasrmapbt(&sc->mp->m_sb) ||
574 	    whichfork == XFS_COW_FORK ||
575 	    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
576 		return 0;
577 
578 	/* Don't support realtime rmap checks yet. */
579 	if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK)
580 		return 0;
581 
582 	/*
583 	 * Only do this for complex maps that are in btree format, or for
584 	 * situations where we would seem to have a size but zero extents.
585 	 * The inode repair code can zap broken iforks, which means we have
586 	 * to flag this bmap as corrupt if there are rmaps that need to be
587 	 * reattached.
588 	 */
589 	switch (whichfork) {
590 	case XFS_DATA_FORK:
591 		size = i_size_read(VFS_I(sc->ip));
592 		break;
593 	case XFS_ATTR_FORK:
594 		size = XFS_IFORK_Q(sc->ip);
595 		break;
596 	default:
597 		size = 0;
598 		break;
599 	}
600 	if (XFS_IFORK_FORMAT(sc->ip, whichfork) != XFS_DINODE_FMT_BTREE &&
601 	    (size == 0 || XFS_IFORK_NEXTENTS(sc->ip, whichfork) > 0))
602 		return 0;
603 
604 	for (agno = 0; agno < sc->mp->m_sb.sb_agcount; agno++) {
605 		error = xchk_bmap_check_ag_rmaps(sc, whichfork, agno);
606 		if (error)
607 			return error;
608 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
609 			break;
610 	}
611 
612 	return 0;
613 }
614 
615 /*
616  * Scrub an inode fork's block mappings.
617  *
618  * First we scan every record in every btree block, if applicable.
619  * Then we unconditionally scan the incore extent cache.
620  */
621 STATIC int
xchk_bmap(struct xfs_scrub * sc,int whichfork)622 xchk_bmap(
623 	struct xfs_scrub	*sc,
624 	int			whichfork)
625 {
626 	struct xfs_bmbt_irec	irec;
627 	struct xchk_bmap_info	info = { NULL };
628 	struct xfs_mount	*mp = sc->mp;
629 	struct xfs_inode	*ip = sc->ip;
630 	struct xfs_ifork	*ifp;
631 	xfs_fileoff_t		endoff;
632 	struct xfs_iext_cursor	icur;
633 	int			error = 0;
634 
635 	ifp = XFS_IFORK_PTR(ip, whichfork);
636 
637 	info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
638 	info.whichfork = whichfork;
639 	info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
640 	info.sc = sc;
641 
642 	switch (whichfork) {
643 	case XFS_COW_FORK:
644 		/* Non-existent CoW forks are ignorable. */
645 		if (!ifp)
646 			goto out;
647 		/* No CoW forks on non-reflink inodes/filesystems. */
648 		if (!xfs_is_reflink_inode(ip)) {
649 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
650 			goto out;
651 		}
652 		break;
653 	case XFS_ATTR_FORK:
654 		if (!ifp)
655 			goto out_check_rmap;
656 		if (!xfs_sb_version_hasattr(&mp->m_sb) &&
657 		    !xfs_sb_version_hasattr2(&mp->m_sb))
658 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
659 		break;
660 	default:
661 		ASSERT(whichfork == XFS_DATA_FORK);
662 		break;
663 	}
664 
665 	/* Check the fork values */
666 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
667 	case XFS_DINODE_FMT_UUID:
668 	case XFS_DINODE_FMT_DEV:
669 	case XFS_DINODE_FMT_LOCAL:
670 		/* No mappings to check. */
671 		goto out;
672 	case XFS_DINODE_FMT_EXTENTS:
673 		if (!(ifp->if_flags & XFS_IFEXTENTS)) {
674 			xchk_fblock_set_corrupt(sc, whichfork, 0);
675 			goto out;
676 		}
677 		break;
678 	case XFS_DINODE_FMT_BTREE:
679 		if (whichfork == XFS_COW_FORK) {
680 			xchk_fblock_set_corrupt(sc, whichfork, 0);
681 			goto out;
682 		}
683 
684 		error = xchk_bmap_btree(sc, whichfork, &info);
685 		if (error)
686 			goto out;
687 		break;
688 	default:
689 		xchk_fblock_set_corrupt(sc, whichfork, 0);
690 		goto out;
691 	}
692 
693 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
694 		goto out;
695 
696 	/* Find the offset of the last extent in the mapping. */
697 	error = xfs_bmap_last_offset(ip, &endoff, whichfork);
698 	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
699 		goto out;
700 
701 	/* Scrub extent records. */
702 	info.lastoff = 0;
703 	ifp = XFS_IFORK_PTR(ip, whichfork);
704 	for_each_xfs_iext(ifp, &icur, &irec) {
705 		if (xchk_should_terminate(sc, &error) ||
706 		    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
707 			goto out;
708 		if (isnullstartblock(irec.br_startblock))
709 			continue;
710 		if (irec.br_startoff >= endoff) {
711 			xchk_fblock_set_corrupt(sc, whichfork,
712 					irec.br_startoff);
713 			goto out;
714 		}
715 		error = xchk_bmap_iextent(ip, &info, &irec);
716 		if (error)
717 			goto out;
718 	}
719 
720 out_check_rmap:
721 	error = xchk_bmap_check_rmaps(sc, whichfork);
722 	if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error))
723 		goto out;
724 out:
725 	return error;
726 }
727 
728 /* Scrub an inode's data fork. */
729 int
xchk_bmap_data(struct xfs_scrub * sc)730 xchk_bmap_data(
731 	struct xfs_scrub	*sc)
732 {
733 	return xchk_bmap(sc, XFS_DATA_FORK);
734 }
735 
736 /* Scrub an inode's attr fork. */
737 int
xchk_bmap_attr(struct xfs_scrub * sc)738 xchk_bmap_attr(
739 	struct xfs_scrub	*sc)
740 {
741 	return xchk_bmap(sc, XFS_ATTR_FORK);
742 }
743 
744 /* Scrub an inode's CoW fork. */
745 int
xchk_bmap_cow(struct xfs_scrub * sc)746 xchk_bmap_cow(
747 	struct xfs_scrub	*sc)
748 {
749 	if (!xfs_is_reflink_inode(sc->ip))
750 		return -ENOENT;
751 
752 	return xchk_bmap(sc, XFS_COW_FORK);
753 }
754