xref: /openbmc/linux/fs/xfs/scrub/agheader_repair.c (revision a634c0a6)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2018-2023 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
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_log_format.h"
14 #include "xfs_trans.h"
15 #include "xfs_sb.h"
16 #include "xfs_alloc.h"
17 #include "xfs_alloc_btree.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_ialloc_btree.h"
20 #include "xfs_rmap.h"
21 #include "xfs_rmap_btree.h"
22 #include "xfs_refcount_btree.h"
23 #include "xfs_ag.h"
24 #include "scrub/scrub.h"
25 #include "scrub/common.h"
26 #include "scrub/trace.h"
27 #include "scrub/repair.h"
28 #include "scrub/bitmap.h"
29 #include "scrub/reap.h"
30 
31 /* Superblock */
32 
33 /* Repair the superblock. */
34 int
xrep_superblock(struct xfs_scrub * sc)35 xrep_superblock(
36 	struct xfs_scrub	*sc)
37 {
38 	struct xfs_mount	*mp = sc->mp;
39 	struct xfs_buf		*bp;
40 	xfs_agnumber_t		agno;
41 	int			error;
42 
43 	/* Don't try to repair AG 0's sb; let xfs_repair deal with it. */
44 	agno = sc->sm->sm_agno;
45 	if (agno == 0)
46 		return -EOPNOTSUPP;
47 
48 	error = xfs_sb_get_secondary(mp, sc->tp, agno, &bp);
49 	if (error)
50 		return error;
51 
52 	/* Last chance to abort before we start committing fixes. */
53 	if (xchk_should_terminate(sc, &error))
54 		return error;
55 
56 	/* Copy AG 0's superblock to this one. */
57 	xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
58 	xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
59 
60 	/*
61 	 * Don't write out a secondary super with NEEDSREPAIR or log incompat
62 	 * features set, since both are ignored when set on a secondary.
63 	 */
64 	if (xfs_has_crc(mp)) {
65 		struct xfs_dsb		*sb = bp->b_addr;
66 
67 		sb->sb_features_incompat &=
68 				~cpu_to_be32(XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR);
69 		sb->sb_features_log_incompat = 0;
70 	}
71 
72 	/* Write this to disk. */
73 	xfs_trans_buf_set_type(sc->tp, bp, XFS_BLFT_SB_BUF);
74 	xfs_trans_log_buf(sc->tp, bp, 0, BBTOB(bp->b_length) - 1);
75 	return error;
76 }
77 
78 /* AGF */
79 
80 struct xrep_agf_allocbt {
81 	struct xfs_scrub	*sc;
82 	xfs_agblock_t		freeblks;
83 	xfs_agblock_t		longest;
84 };
85 
86 /* Record free space shape information. */
87 STATIC int
xrep_agf_walk_allocbt(struct xfs_btree_cur * cur,const struct xfs_alloc_rec_incore * rec,void * priv)88 xrep_agf_walk_allocbt(
89 	struct xfs_btree_cur		*cur,
90 	const struct xfs_alloc_rec_incore *rec,
91 	void				*priv)
92 {
93 	struct xrep_agf_allocbt		*raa = priv;
94 	int				error = 0;
95 
96 	if (xchk_should_terminate(raa->sc, &error))
97 		return error;
98 
99 	raa->freeblks += rec->ar_blockcount;
100 	if (rec->ar_blockcount > raa->longest)
101 		raa->longest = rec->ar_blockcount;
102 	return error;
103 }
104 
105 /* Does this AGFL block look sane? */
106 STATIC int
xrep_agf_check_agfl_block(struct xfs_mount * mp,xfs_agblock_t agbno,void * priv)107 xrep_agf_check_agfl_block(
108 	struct xfs_mount	*mp,
109 	xfs_agblock_t		agbno,
110 	void			*priv)
111 {
112 	struct xfs_scrub	*sc = priv;
113 
114 	if (!xfs_verify_agbno(sc->sa.pag, agbno))
115 		return -EFSCORRUPTED;
116 	return 0;
117 }
118 
119 /*
120  * Offset within the xrep_find_ag_btree array for each btree type.  Avoid the
121  * XFS_BTNUM_ names here to avoid creating a sparse array.
122  */
123 enum {
124 	XREP_AGF_BNOBT = 0,
125 	XREP_AGF_CNTBT,
126 	XREP_AGF_RMAPBT,
127 	XREP_AGF_REFCOUNTBT,
128 	XREP_AGF_END,
129 	XREP_AGF_MAX
130 };
131 
132 /* Check a btree root candidate. */
133 static inline bool
xrep_check_btree_root(struct xfs_scrub * sc,struct xrep_find_ag_btree * fab)134 xrep_check_btree_root(
135 	struct xfs_scrub		*sc,
136 	struct xrep_find_ag_btree	*fab)
137 {
138 	return xfs_verify_agbno(sc->sa.pag, fab->root) &&
139 	       fab->height <= fab->maxlevels;
140 }
141 
142 /*
143  * Given the btree roots described by *fab, find the roots, check them for
144  * sanity, and pass the root data back out via *fab.
145  *
146  * This is /also/ a chicken and egg problem because we have to use the rmapbt
147  * (rooted in the AGF) to find the btrees rooted in the AGF.  We also have no
148  * idea if the btrees make any sense.  If we hit obvious corruptions in those
149  * btrees we'll bail out.
150  */
151 STATIC int
xrep_agf_find_btrees(struct xfs_scrub * sc,struct xfs_buf * agf_bp,struct xrep_find_ag_btree * fab,struct xfs_buf * agfl_bp)152 xrep_agf_find_btrees(
153 	struct xfs_scrub		*sc,
154 	struct xfs_buf			*agf_bp,
155 	struct xrep_find_ag_btree	*fab,
156 	struct xfs_buf			*agfl_bp)
157 {
158 	struct xfs_agf			*old_agf = agf_bp->b_addr;
159 	int				error;
160 
161 	/* Go find the root data. */
162 	error = xrep_find_ag_btree_roots(sc, agf_bp, fab, agfl_bp);
163 	if (error)
164 		return error;
165 
166 	/* We must find the bnobt, cntbt, and rmapbt roots. */
167 	if (!xrep_check_btree_root(sc, &fab[XREP_AGF_BNOBT]) ||
168 	    !xrep_check_btree_root(sc, &fab[XREP_AGF_CNTBT]) ||
169 	    !xrep_check_btree_root(sc, &fab[XREP_AGF_RMAPBT]))
170 		return -EFSCORRUPTED;
171 
172 	/*
173 	 * We relied on the rmapbt to reconstruct the AGF.  If we get a
174 	 * different root then something's seriously wrong.
175 	 */
176 	if (fab[XREP_AGF_RMAPBT].root !=
177 	    be32_to_cpu(old_agf->agf_roots[XFS_BTNUM_RMAPi]))
178 		return -EFSCORRUPTED;
179 
180 	/* We must find the refcountbt root if that feature is enabled. */
181 	if (xfs_has_reflink(sc->mp) &&
182 	    !xrep_check_btree_root(sc, &fab[XREP_AGF_REFCOUNTBT]))
183 		return -EFSCORRUPTED;
184 
185 	return 0;
186 }
187 
188 /*
189  * Reinitialize the AGF header, making an in-core copy of the old contents so
190  * that we know which in-core state needs to be reinitialized.
191  */
192 STATIC void
xrep_agf_init_header(struct xfs_scrub * sc,struct xfs_buf * agf_bp,struct xfs_agf * old_agf)193 xrep_agf_init_header(
194 	struct xfs_scrub	*sc,
195 	struct xfs_buf		*agf_bp,
196 	struct xfs_agf		*old_agf)
197 {
198 	struct xfs_mount	*mp = sc->mp;
199 	struct xfs_perag	*pag = sc->sa.pag;
200 	struct xfs_agf		*agf = agf_bp->b_addr;
201 
202 	memcpy(old_agf, agf, sizeof(*old_agf));
203 	memset(agf, 0, BBTOB(agf_bp->b_length));
204 	agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
205 	agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
206 	agf->agf_seqno = cpu_to_be32(pag->pag_agno);
207 	agf->agf_length = cpu_to_be32(pag->block_count);
208 	agf->agf_flfirst = old_agf->agf_flfirst;
209 	agf->agf_fllast = old_agf->agf_fllast;
210 	agf->agf_flcount = old_agf->agf_flcount;
211 	if (xfs_has_crc(mp))
212 		uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
213 
214 	/* Mark the incore AGF data stale until we're done fixing things. */
215 	ASSERT(xfs_perag_initialised_agf(pag));
216 	clear_bit(XFS_AGSTATE_AGF_INIT, &pag->pag_opstate);
217 }
218 
219 /* Set btree root information in an AGF. */
220 STATIC void
xrep_agf_set_roots(struct xfs_scrub * sc,struct xfs_agf * agf,struct xrep_find_ag_btree * fab)221 xrep_agf_set_roots(
222 	struct xfs_scrub		*sc,
223 	struct xfs_agf			*agf,
224 	struct xrep_find_ag_btree	*fab)
225 {
226 	agf->agf_roots[XFS_BTNUM_BNOi] =
227 			cpu_to_be32(fab[XREP_AGF_BNOBT].root);
228 	agf->agf_levels[XFS_BTNUM_BNOi] =
229 			cpu_to_be32(fab[XREP_AGF_BNOBT].height);
230 
231 	agf->agf_roots[XFS_BTNUM_CNTi] =
232 			cpu_to_be32(fab[XREP_AGF_CNTBT].root);
233 	agf->agf_levels[XFS_BTNUM_CNTi] =
234 			cpu_to_be32(fab[XREP_AGF_CNTBT].height);
235 
236 	agf->agf_roots[XFS_BTNUM_RMAPi] =
237 			cpu_to_be32(fab[XREP_AGF_RMAPBT].root);
238 	agf->agf_levels[XFS_BTNUM_RMAPi] =
239 			cpu_to_be32(fab[XREP_AGF_RMAPBT].height);
240 
241 	if (xfs_has_reflink(sc->mp)) {
242 		agf->agf_refcount_root =
243 				cpu_to_be32(fab[XREP_AGF_REFCOUNTBT].root);
244 		agf->agf_refcount_level =
245 				cpu_to_be32(fab[XREP_AGF_REFCOUNTBT].height);
246 	}
247 }
248 
249 /* Update all AGF fields which derive from btree contents. */
250 STATIC int
xrep_agf_calc_from_btrees(struct xfs_scrub * sc,struct xfs_buf * agf_bp)251 xrep_agf_calc_from_btrees(
252 	struct xfs_scrub	*sc,
253 	struct xfs_buf		*agf_bp)
254 {
255 	struct xrep_agf_allocbt	raa = { .sc = sc };
256 	struct xfs_btree_cur	*cur = NULL;
257 	struct xfs_agf		*agf = agf_bp->b_addr;
258 	struct xfs_mount	*mp = sc->mp;
259 	xfs_agblock_t		btreeblks;
260 	xfs_agblock_t		blocks;
261 	int			error;
262 
263 	/* Update the AGF counters from the bnobt. */
264 	cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp,
265 			sc->sa.pag, XFS_BTNUM_BNO);
266 	error = xfs_alloc_query_all(cur, xrep_agf_walk_allocbt, &raa);
267 	if (error)
268 		goto err;
269 	error = xfs_btree_count_blocks(cur, &blocks);
270 	if (error)
271 		goto err;
272 	xfs_btree_del_cursor(cur, error);
273 	btreeblks = blocks - 1;
274 	agf->agf_freeblks = cpu_to_be32(raa.freeblks);
275 	agf->agf_longest = cpu_to_be32(raa.longest);
276 
277 	/* Update the AGF counters from the cntbt. */
278 	cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp,
279 			sc->sa.pag, XFS_BTNUM_CNT);
280 	error = xfs_btree_count_blocks(cur, &blocks);
281 	if (error)
282 		goto err;
283 	xfs_btree_del_cursor(cur, error);
284 	btreeblks += blocks - 1;
285 
286 	/* Update the AGF counters from the rmapbt. */
287 	cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag);
288 	error = xfs_btree_count_blocks(cur, &blocks);
289 	if (error)
290 		goto err;
291 	xfs_btree_del_cursor(cur, error);
292 	agf->agf_rmap_blocks = cpu_to_be32(blocks);
293 	btreeblks += blocks - 1;
294 
295 	agf->agf_btreeblks = cpu_to_be32(btreeblks);
296 
297 	/* Update the AGF counters from the refcountbt. */
298 	if (xfs_has_reflink(mp)) {
299 		cur = xfs_refcountbt_init_cursor(mp, sc->tp, agf_bp,
300 				sc->sa.pag);
301 		error = xfs_btree_count_blocks(cur, &blocks);
302 		if (error)
303 			goto err;
304 		xfs_btree_del_cursor(cur, error);
305 		agf->agf_refcount_blocks = cpu_to_be32(blocks);
306 	}
307 
308 	return 0;
309 err:
310 	xfs_btree_del_cursor(cur, error);
311 	return error;
312 }
313 
314 /* Commit the new AGF and reinitialize the incore state. */
315 STATIC int
xrep_agf_commit_new(struct xfs_scrub * sc,struct xfs_buf * agf_bp)316 xrep_agf_commit_new(
317 	struct xfs_scrub	*sc,
318 	struct xfs_buf		*agf_bp)
319 {
320 	struct xfs_perag	*pag;
321 	struct xfs_agf		*agf = agf_bp->b_addr;
322 
323 	/* Trigger fdblocks recalculation */
324 	xfs_force_summary_recalc(sc->mp);
325 
326 	/* Write this to disk. */
327 	xfs_trans_buf_set_type(sc->tp, agf_bp, XFS_BLFT_AGF_BUF);
328 	xfs_trans_log_buf(sc->tp, agf_bp, 0, BBTOB(agf_bp->b_length) - 1);
329 
330 	/* Now reinitialize the in-core counters we changed. */
331 	pag = sc->sa.pag;
332 	pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
333 	pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
334 	pag->pagf_longest = be32_to_cpu(agf->agf_longest);
335 	pag->pagf_levels[XFS_BTNUM_BNOi] =
336 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
337 	pag->pagf_levels[XFS_BTNUM_CNTi] =
338 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
339 	pag->pagf_levels[XFS_BTNUM_RMAPi] =
340 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]);
341 	pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level);
342 	set_bit(XFS_AGSTATE_AGF_INIT, &pag->pag_opstate);
343 
344 	return 0;
345 }
346 
347 /* Repair the AGF. v5 filesystems only. */
348 int
xrep_agf(struct xfs_scrub * sc)349 xrep_agf(
350 	struct xfs_scrub		*sc)
351 {
352 	struct xrep_find_ag_btree	fab[XREP_AGF_MAX] = {
353 		[XREP_AGF_BNOBT] = {
354 			.rmap_owner = XFS_RMAP_OWN_AG,
355 			.buf_ops = &xfs_bnobt_buf_ops,
356 			.maxlevels = sc->mp->m_alloc_maxlevels,
357 		},
358 		[XREP_AGF_CNTBT] = {
359 			.rmap_owner = XFS_RMAP_OWN_AG,
360 			.buf_ops = &xfs_cntbt_buf_ops,
361 			.maxlevels = sc->mp->m_alloc_maxlevels,
362 		},
363 		[XREP_AGF_RMAPBT] = {
364 			.rmap_owner = XFS_RMAP_OWN_AG,
365 			.buf_ops = &xfs_rmapbt_buf_ops,
366 			.maxlevels = sc->mp->m_rmap_maxlevels,
367 		},
368 		[XREP_AGF_REFCOUNTBT] = {
369 			.rmap_owner = XFS_RMAP_OWN_REFC,
370 			.buf_ops = &xfs_refcountbt_buf_ops,
371 			.maxlevels = sc->mp->m_refc_maxlevels,
372 		},
373 		[XREP_AGF_END] = {
374 			.buf_ops = NULL,
375 		},
376 	};
377 	struct xfs_agf			old_agf;
378 	struct xfs_mount		*mp = sc->mp;
379 	struct xfs_buf			*agf_bp;
380 	struct xfs_buf			*agfl_bp;
381 	struct xfs_agf			*agf;
382 	int				error;
383 
384 	/* We require the rmapbt to rebuild anything. */
385 	if (!xfs_has_rmapbt(mp))
386 		return -EOPNOTSUPP;
387 
388 	/*
389 	 * Make sure we have the AGF buffer, as scrub might have decided it
390 	 * was corrupt after xfs_alloc_read_agf failed with -EFSCORRUPTED.
391 	 */
392 	error = xfs_trans_read_buf(mp, sc->tp, mp->m_ddev_targp,
393 			XFS_AG_DADDR(mp, sc->sa.pag->pag_agno,
394 						XFS_AGF_DADDR(mp)),
395 			XFS_FSS_TO_BB(mp, 1), 0, &agf_bp, NULL);
396 	if (error)
397 		return error;
398 	agf_bp->b_ops = &xfs_agf_buf_ops;
399 	agf = agf_bp->b_addr;
400 
401 	/*
402 	 * Load the AGFL so that we can screen out OWN_AG blocks that are on
403 	 * the AGFL now; these blocks might have once been part of the
404 	 * bno/cnt/rmap btrees but are not now.  This is a chicken and egg
405 	 * problem: the AGF is corrupt, so we have to trust the AGFL contents
406 	 * because we can't do any serious cross-referencing with any of the
407 	 * btrees rooted in the AGF.  If the AGFL contents are obviously bad
408 	 * then we'll bail out.
409 	 */
410 	error = xfs_alloc_read_agfl(sc->sa.pag, sc->tp, &agfl_bp);
411 	if (error)
412 		return error;
413 
414 	/*
415 	 * Spot-check the AGFL blocks; if they're obviously corrupt then
416 	 * there's nothing we can do but bail out.
417 	 */
418 	error = xfs_agfl_walk(sc->mp, agf_bp->b_addr, agfl_bp,
419 			xrep_agf_check_agfl_block, sc);
420 	if (error)
421 		return error;
422 
423 	/*
424 	 * Find the AGF btree roots.  This is also a chicken-and-egg situation;
425 	 * see the function for more details.
426 	 */
427 	error = xrep_agf_find_btrees(sc, agf_bp, fab, agfl_bp);
428 	if (error)
429 		return error;
430 
431 	/* Last chance to abort before we start committing fixes. */
432 	if (xchk_should_terminate(sc, &error))
433 		return error;
434 
435 	/* Start rewriting the header and implant the btrees we found. */
436 	xrep_agf_init_header(sc, agf_bp, &old_agf);
437 	xrep_agf_set_roots(sc, agf, fab);
438 	error = xrep_agf_calc_from_btrees(sc, agf_bp);
439 	if (error)
440 		goto out_revert;
441 
442 	/* Commit the changes and reinitialize incore state. */
443 	return xrep_agf_commit_new(sc, agf_bp);
444 
445 out_revert:
446 	/* Mark the incore AGF state stale and revert the AGF. */
447 	clear_bit(XFS_AGSTATE_AGF_INIT, &sc->sa.pag->pag_opstate);
448 	memcpy(agf, &old_agf, sizeof(old_agf));
449 	return error;
450 }
451 
452 /* AGFL */
453 
454 struct xrep_agfl {
455 	/* Bitmap of alleged AGFL blocks that we're not going to add. */
456 	struct xagb_bitmap	crossed;
457 
458 	/* Bitmap of other OWN_AG metadata blocks. */
459 	struct xagb_bitmap	agmetablocks;
460 
461 	/* Bitmap of free space. */
462 	struct xagb_bitmap	*freesp;
463 
464 	/* rmapbt cursor for finding crosslinked blocks */
465 	struct xfs_btree_cur	*rmap_cur;
466 
467 	struct xfs_scrub	*sc;
468 };
469 
470 /* Record all OWN_AG (free space btree) information from the rmap data. */
471 STATIC int
xrep_agfl_walk_rmap(struct xfs_btree_cur * cur,const struct xfs_rmap_irec * rec,void * priv)472 xrep_agfl_walk_rmap(
473 	struct xfs_btree_cur	*cur,
474 	const struct xfs_rmap_irec *rec,
475 	void			*priv)
476 {
477 	struct xrep_agfl	*ra = priv;
478 	int			error = 0;
479 
480 	if (xchk_should_terminate(ra->sc, &error))
481 		return error;
482 
483 	/* Record all the OWN_AG blocks. */
484 	if (rec->rm_owner == XFS_RMAP_OWN_AG) {
485 		error = xagb_bitmap_set(ra->freesp, rec->rm_startblock,
486 				rec->rm_blockcount);
487 		if (error)
488 			return error;
489 	}
490 
491 	return xagb_bitmap_set_btcur_path(&ra->agmetablocks, cur);
492 }
493 
494 /* Strike out the blocks that are cross-linked according to the rmapbt. */
495 STATIC int
xrep_agfl_check_extent(uint64_t start,uint64_t len,void * priv)496 xrep_agfl_check_extent(
497 	uint64_t		start,
498 	uint64_t		len,
499 	void			*priv)
500 {
501 	struct xrep_agfl	*ra = priv;
502 	xfs_agblock_t		agbno = start;
503 	xfs_agblock_t		last_agbno = agbno + len - 1;
504 	int			error;
505 
506 	while (agbno <= last_agbno) {
507 		bool		other_owners;
508 
509 		error = xfs_rmap_has_other_keys(ra->rmap_cur, agbno, 1,
510 				&XFS_RMAP_OINFO_AG, &other_owners);
511 		if (error)
512 			return error;
513 
514 		if (other_owners) {
515 			error = xagb_bitmap_set(&ra->crossed, agbno, 1);
516 			if (error)
517 				return error;
518 		}
519 
520 		if (xchk_should_terminate(ra->sc, &error))
521 			return error;
522 		agbno++;
523 	}
524 
525 	return 0;
526 }
527 
528 /*
529  * Map out all the non-AGFL OWN_AG space in this AG so that we can deduce
530  * which blocks belong to the AGFL.
531  *
532  * Compute the set of old AGFL blocks by subtracting from the list of OWN_AG
533  * blocks the list of blocks owned by all other OWN_AG metadata (bnobt, cntbt,
534  * rmapbt).  These are the old AGFL blocks, so return that list and the number
535  * of blocks we're actually going to put back on the AGFL.
536  */
537 STATIC int
xrep_agfl_collect_blocks(struct xfs_scrub * sc,struct xfs_buf * agf_bp,struct xagb_bitmap * agfl_extents,xfs_agblock_t * flcount)538 xrep_agfl_collect_blocks(
539 	struct xfs_scrub	*sc,
540 	struct xfs_buf		*agf_bp,
541 	struct xagb_bitmap	*agfl_extents,
542 	xfs_agblock_t		*flcount)
543 {
544 	struct xrep_agfl	ra;
545 	struct xfs_mount	*mp = sc->mp;
546 	struct xfs_btree_cur	*cur;
547 	int			error;
548 
549 	ra.sc = sc;
550 	ra.freesp = agfl_extents;
551 	xagb_bitmap_init(&ra.agmetablocks);
552 	xagb_bitmap_init(&ra.crossed);
553 
554 	/* Find all space used by the free space btrees & rmapbt. */
555 	cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag);
556 	error = xfs_rmap_query_all(cur, xrep_agfl_walk_rmap, &ra);
557 	xfs_btree_del_cursor(cur, error);
558 	if (error)
559 		goto out_bmp;
560 
561 	/* Find all blocks currently being used by the bnobt. */
562 	cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp,
563 			sc->sa.pag, XFS_BTNUM_BNO);
564 	error = xagb_bitmap_set_btblocks(&ra.agmetablocks, cur);
565 	xfs_btree_del_cursor(cur, error);
566 	if (error)
567 		goto out_bmp;
568 
569 	/* Find all blocks currently being used by the cntbt. */
570 	cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp,
571 			sc->sa.pag, XFS_BTNUM_CNT);
572 	error = xagb_bitmap_set_btblocks(&ra.agmetablocks, cur);
573 	xfs_btree_del_cursor(cur, error);
574 	if (error)
575 		goto out_bmp;
576 
577 	/*
578 	 * Drop the freesp meta blocks that are in use by btrees.
579 	 * The remaining blocks /should/ be AGFL blocks.
580 	 */
581 	error = xagb_bitmap_disunion(agfl_extents, &ra.agmetablocks);
582 	if (error)
583 		goto out_bmp;
584 
585 	/* Strike out the blocks that are cross-linked. */
586 	ra.rmap_cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag);
587 	error = xagb_bitmap_walk(agfl_extents, xrep_agfl_check_extent, &ra);
588 	xfs_btree_del_cursor(ra.rmap_cur, error);
589 	if (error)
590 		goto out_bmp;
591 	error = xagb_bitmap_disunion(agfl_extents, &ra.crossed);
592 	if (error)
593 		goto out_bmp;
594 
595 	/*
596 	 * Calculate the new AGFL size.  If we found more blocks than fit in
597 	 * the AGFL we'll free them later.
598 	 */
599 	*flcount = min_t(uint64_t, xagb_bitmap_hweight(agfl_extents),
600 			 xfs_agfl_size(mp));
601 
602 out_bmp:
603 	xagb_bitmap_destroy(&ra.crossed);
604 	xagb_bitmap_destroy(&ra.agmetablocks);
605 	return error;
606 }
607 
608 /* Update the AGF and reset the in-core state. */
609 STATIC void
xrep_agfl_update_agf(struct xfs_scrub * sc,struct xfs_buf * agf_bp,xfs_agblock_t flcount)610 xrep_agfl_update_agf(
611 	struct xfs_scrub	*sc,
612 	struct xfs_buf		*agf_bp,
613 	xfs_agblock_t		flcount)
614 {
615 	struct xfs_agf		*agf = agf_bp->b_addr;
616 
617 	ASSERT(flcount <= xfs_agfl_size(sc->mp));
618 
619 	/* Trigger fdblocks recalculation */
620 	xfs_force_summary_recalc(sc->mp);
621 
622 	/* Update the AGF counters. */
623 	if (xfs_perag_initialised_agf(sc->sa.pag)) {
624 		sc->sa.pag->pagf_flcount = flcount;
625 		clear_bit(XFS_AGSTATE_AGFL_NEEDS_RESET,
626 				&sc->sa.pag->pag_opstate);
627 	}
628 	agf->agf_flfirst = cpu_to_be32(0);
629 	agf->agf_flcount = cpu_to_be32(flcount);
630 	if (flcount)
631 		agf->agf_fllast = cpu_to_be32(flcount - 1);
632 	else
633 		agf->agf_fllast = cpu_to_be32(xfs_agfl_size(sc->mp) - 1);
634 
635 	xfs_alloc_log_agf(sc->tp, agf_bp,
636 			XFS_AGF_FLFIRST | XFS_AGF_FLLAST | XFS_AGF_FLCOUNT);
637 }
638 
639 struct xrep_agfl_fill {
640 	struct xagb_bitmap	used_extents;
641 	struct xfs_scrub	*sc;
642 	__be32			*agfl_bno;
643 	xfs_agblock_t		flcount;
644 	unsigned int		fl_off;
645 };
646 
647 /* Fill the AGFL with whatever blocks are in this extent. */
648 static int
xrep_agfl_fill(uint64_t start,uint64_t len,void * priv)649 xrep_agfl_fill(
650 	uint64_t		start,
651 	uint64_t		len,
652 	void			*priv)
653 {
654 	struct xrep_agfl_fill	*af = priv;
655 	struct xfs_scrub	*sc = af->sc;
656 	xfs_agblock_t		agbno = start;
657 	int			error;
658 
659 	trace_xrep_agfl_insert(sc->sa.pag, agbno, len);
660 
661 	while (agbno < start + len && af->fl_off < af->flcount)
662 		af->agfl_bno[af->fl_off++] = cpu_to_be32(agbno++);
663 
664 	error = xagb_bitmap_set(&af->used_extents, start, agbno - 1);
665 	if (error)
666 		return error;
667 
668 	if (af->fl_off == af->flcount)
669 		return -ECANCELED;
670 
671 	return 0;
672 }
673 
674 /* Write out a totally new AGFL. */
675 STATIC int
xrep_agfl_init_header(struct xfs_scrub * sc,struct xfs_buf * agfl_bp,struct xagb_bitmap * agfl_extents,xfs_agblock_t flcount)676 xrep_agfl_init_header(
677 	struct xfs_scrub	*sc,
678 	struct xfs_buf		*agfl_bp,
679 	struct xagb_bitmap	*agfl_extents,
680 	xfs_agblock_t		flcount)
681 {
682 	struct xrep_agfl_fill	af = {
683 		.sc		= sc,
684 		.flcount	= flcount,
685 	};
686 	struct xfs_mount	*mp = sc->mp;
687 	struct xfs_agfl		*agfl;
688 	int			error;
689 
690 	ASSERT(flcount <= xfs_agfl_size(mp));
691 
692 	/*
693 	 * Start rewriting the header by setting the bno[] array to
694 	 * NULLAGBLOCK, then setting AGFL header fields.
695 	 */
696 	agfl = XFS_BUF_TO_AGFL(agfl_bp);
697 	memset(agfl, 0xFF, BBTOB(agfl_bp->b_length));
698 	agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
699 	agfl->agfl_seqno = cpu_to_be32(sc->sa.pag->pag_agno);
700 	uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
701 
702 	/*
703 	 * Fill the AGFL with the remaining blocks.  If agfl_extents has more
704 	 * blocks than fit in the AGFL, they will be freed in a subsequent
705 	 * step.
706 	 */
707 	xagb_bitmap_init(&af.used_extents);
708 	af.agfl_bno = xfs_buf_to_agfl_bno(agfl_bp),
709 	xagb_bitmap_walk(agfl_extents, xrep_agfl_fill, &af);
710 	error = xagb_bitmap_disunion(agfl_extents, &af.used_extents);
711 	if (error)
712 		return error;
713 
714 	/* Write new AGFL to disk. */
715 	xfs_trans_buf_set_type(sc->tp, agfl_bp, XFS_BLFT_AGFL_BUF);
716 	xfs_trans_log_buf(sc->tp, agfl_bp, 0, BBTOB(agfl_bp->b_length) - 1);
717 	xagb_bitmap_destroy(&af.used_extents);
718 	return 0;
719 }
720 
721 /* Repair the AGFL. */
722 int
xrep_agfl(struct xfs_scrub * sc)723 xrep_agfl(
724 	struct xfs_scrub	*sc)
725 {
726 	struct xagb_bitmap	agfl_extents;
727 	struct xfs_mount	*mp = sc->mp;
728 	struct xfs_buf		*agf_bp;
729 	struct xfs_buf		*agfl_bp;
730 	xfs_agblock_t		flcount;
731 	int			error;
732 
733 	/* We require the rmapbt to rebuild anything. */
734 	if (!xfs_has_rmapbt(mp))
735 		return -EOPNOTSUPP;
736 
737 	xagb_bitmap_init(&agfl_extents);
738 
739 	/*
740 	 * Read the AGF so that we can query the rmapbt.  We hope that there's
741 	 * nothing wrong with the AGF, but all the AG header repair functions
742 	 * have this chicken-and-egg problem.
743 	 */
744 	error = xfs_alloc_read_agf(sc->sa.pag, sc->tp, 0, &agf_bp);
745 	if (error)
746 		return error;
747 
748 	/*
749 	 * Make sure we have the AGFL buffer, as scrub might have decided it
750 	 * was corrupt after xfs_alloc_read_agfl failed with -EFSCORRUPTED.
751 	 */
752 	error = xfs_trans_read_buf(mp, sc->tp, mp->m_ddev_targp,
753 			XFS_AG_DADDR(mp, sc->sa.pag->pag_agno,
754 						XFS_AGFL_DADDR(mp)),
755 			XFS_FSS_TO_BB(mp, 1), 0, &agfl_bp, NULL);
756 	if (error)
757 		return error;
758 	agfl_bp->b_ops = &xfs_agfl_buf_ops;
759 
760 	/* Gather all the extents we're going to put on the new AGFL. */
761 	error = xrep_agfl_collect_blocks(sc, agf_bp, &agfl_extents, &flcount);
762 	if (error)
763 		goto err;
764 
765 	/* Last chance to abort before we start committing fixes. */
766 	if (xchk_should_terminate(sc, &error))
767 		goto err;
768 
769 	/*
770 	 * Update AGF and AGFL.  We reset the global free block counter when
771 	 * we adjust the AGF flcount (which can fail) so avoid updating any
772 	 * buffers until we know that part works.
773 	 */
774 	xrep_agfl_update_agf(sc, agf_bp, flcount);
775 	error = xrep_agfl_init_header(sc, agfl_bp, &agfl_extents, flcount);
776 	if (error)
777 		goto err;
778 
779 	/*
780 	 * Ok, the AGFL should be ready to go now.  Roll the transaction to
781 	 * make the new AGFL permanent before we start using it to return
782 	 * freespace overflow to the freespace btrees.
783 	 */
784 	sc->sa.agf_bp = agf_bp;
785 	error = xrep_roll_ag_trans(sc);
786 	if (error)
787 		goto err;
788 
789 	/* Dump any AGFL overflow. */
790 	error = xrep_reap_agblocks(sc, &agfl_extents, &XFS_RMAP_OINFO_AG,
791 			XFS_AG_RESV_AGFL);
792 err:
793 	xagb_bitmap_destroy(&agfl_extents);
794 	return error;
795 }
796 
797 /* AGI */
798 
799 /*
800  * Offset within the xrep_find_ag_btree array for each btree type.  Avoid the
801  * XFS_BTNUM_ names here to avoid creating a sparse array.
802  */
803 enum {
804 	XREP_AGI_INOBT = 0,
805 	XREP_AGI_FINOBT,
806 	XREP_AGI_END,
807 	XREP_AGI_MAX
808 };
809 
810 /*
811  * Given the inode btree roots described by *fab, find the roots, check them
812  * for sanity, and pass the root data back out via *fab.
813  */
814 STATIC int
xrep_agi_find_btrees(struct xfs_scrub * sc,struct xrep_find_ag_btree * fab)815 xrep_agi_find_btrees(
816 	struct xfs_scrub		*sc,
817 	struct xrep_find_ag_btree	*fab)
818 {
819 	struct xfs_buf			*agf_bp;
820 	struct xfs_mount		*mp = sc->mp;
821 	int				error;
822 
823 	/* Read the AGF. */
824 	error = xfs_alloc_read_agf(sc->sa.pag, sc->tp, 0, &agf_bp);
825 	if (error)
826 		return error;
827 
828 	/* Find the btree roots. */
829 	error = xrep_find_ag_btree_roots(sc, agf_bp, fab, NULL);
830 	if (error)
831 		return error;
832 
833 	/* We must find the inobt root. */
834 	if (!xrep_check_btree_root(sc, &fab[XREP_AGI_INOBT]))
835 		return -EFSCORRUPTED;
836 
837 	/* We must find the finobt root if that feature is enabled. */
838 	if (xfs_has_finobt(mp) &&
839 	    !xrep_check_btree_root(sc, &fab[XREP_AGI_FINOBT]))
840 		return -EFSCORRUPTED;
841 
842 	return 0;
843 }
844 
845 /*
846  * Reinitialize the AGI header, making an in-core copy of the old contents so
847  * that we know which in-core state needs to be reinitialized.
848  */
849 STATIC void
xrep_agi_init_header(struct xfs_scrub * sc,struct xfs_buf * agi_bp,struct xfs_agi * old_agi)850 xrep_agi_init_header(
851 	struct xfs_scrub	*sc,
852 	struct xfs_buf		*agi_bp,
853 	struct xfs_agi		*old_agi)
854 {
855 	struct xfs_agi		*agi = agi_bp->b_addr;
856 	struct xfs_perag	*pag = sc->sa.pag;
857 	struct xfs_mount	*mp = sc->mp;
858 
859 	memcpy(old_agi, agi, sizeof(*old_agi));
860 	memset(agi, 0, BBTOB(agi_bp->b_length));
861 	agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
862 	agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
863 	agi->agi_seqno = cpu_to_be32(pag->pag_agno);
864 	agi->agi_length = cpu_to_be32(pag->block_count);
865 	agi->agi_newino = cpu_to_be32(NULLAGINO);
866 	agi->agi_dirino = cpu_to_be32(NULLAGINO);
867 	if (xfs_has_crc(mp))
868 		uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
869 
870 	/* We don't know how to fix the unlinked list yet. */
871 	memcpy(&agi->agi_unlinked, &old_agi->agi_unlinked,
872 			sizeof(agi->agi_unlinked));
873 
874 	/* Mark the incore AGF data stale until we're done fixing things. */
875 	ASSERT(xfs_perag_initialised_agi(pag));
876 	clear_bit(XFS_AGSTATE_AGI_INIT, &pag->pag_opstate);
877 }
878 
879 /* Set btree root information in an AGI. */
880 STATIC void
xrep_agi_set_roots(struct xfs_scrub * sc,struct xfs_agi * agi,struct xrep_find_ag_btree * fab)881 xrep_agi_set_roots(
882 	struct xfs_scrub		*sc,
883 	struct xfs_agi			*agi,
884 	struct xrep_find_ag_btree	*fab)
885 {
886 	agi->agi_root = cpu_to_be32(fab[XREP_AGI_INOBT].root);
887 	agi->agi_level = cpu_to_be32(fab[XREP_AGI_INOBT].height);
888 
889 	if (xfs_has_finobt(sc->mp)) {
890 		agi->agi_free_root = cpu_to_be32(fab[XREP_AGI_FINOBT].root);
891 		agi->agi_free_level = cpu_to_be32(fab[XREP_AGI_FINOBT].height);
892 	}
893 }
894 
895 /* Update the AGI counters. */
896 STATIC int
xrep_agi_calc_from_btrees(struct xfs_scrub * sc,struct xfs_buf * agi_bp)897 xrep_agi_calc_from_btrees(
898 	struct xfs_scrub	*sc,
899 	struct xfs_buf		*agi_bp)
900 {
901 	struct xfs_btree_cur	*cur;
902 	struct xfs_agi		*agi = agi_bp->b_addr;
903 	struct xfs_mount	*mp = sc->mp;
904 	xfs_agino_t		count;
905 	xfs_agino_t		freecount;
906 	int			error;
907 
908 	cur = xfs_inobt_init_cursor(sc->sa.pag, sc->tp, agi_bp, XFS_BTNUM_INO);
909 	error = xfs_ialloc_count_inodes(cur, &count, &freecount);
910 	if (error)
911 		goto err;
912 	if (xfs_has_inobtcounts(mp)) {
913 		xfs_agblock_t	blocks;
914 
915 		error = xfs_btree_count_blocks(cur, &blocks);
916 		if (error)
917 			goto err;
918 		agi->agi_iblocks = cpu_to_be32(blocks);
919 	}
920 	xfs_btree_del_cursor(cur, error);
921 
922 	agi->agi_count = cpu_to_be32(count);
923 	agi->agi_freecount = cpu_to_be32(freecount);
924 
925 	if (xfs_has_finobt(mp) && xfs_has_inobtcounts(mp)) {
926 		xfs_agblock_t	blocks;
927 
928 		cur = xfs_inobt_init_cursor(sc->sa.pag, sc->tp, agi_bp,
929 				XFS_BTNUM_FINO);
930 		error = xfs_btree_count_blocks(cur, &blocks);
931 		if (error)
932 			goto err;
933 		xfs_btree_del_cursor(cur, error);
934 		agi->agi_fblocks = cpu_to_be32(blocks);
935 	}
936 
937 	return 0;
938 err:
939 	xfs_btree_del_cursor(cur, error);
940 	return error;
941 }
942 
943 /* Trigger reinitialization of the in-core data. */
944 STATIC int
xrep_agi_commit_new(struct xfs_scrub * sc,struct xfs_buf * agi_bp)945 xrep_agi_commit_new(
946 	struct xfs_scrub	*sc,
947 	struct xfs_buf		*agi_bp)
948 {
949 	struct xfs_perag	*pag;
950 	struct xfs_agi		*agi = agi_bp->b_addr;
951 
952 	/* Trigger inode count recalculation */
953 	xfs_force_summary_recalc(sc->mp);
954 
955 	/* Write this to disk. */
956 	xfs_trans_buf_set_type(sc->tp, agi_bp, XFS_BLFT_AGI_BUF);
957 	xfs_trans_log_buf(sc->tp, agi_bp, 0, BBTOB(agi_bp->b_length) - 1);
958 
959 	/* Now reinitialize the in-core counters if necessary. */
960 	pag = sc->sa.pag;
961 	pag->pagi_count = be32_to_cpu(agi->agi_count);
962 	pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
963 	set_bit(XFS_AGSTATE_AGI_INIT, &pag->pag_opstate);
964 
965 	return 0;
966 }
967 
968 /* Repair the AGI. */
969 int
xrep_agi(struct xfs_scrub * sc)970 xrep_agi(
971 	struct xfs_scrub		*sc)
972 {
973 	struct xrep_find_ag_btree	fab[XREP_AGI_MAX] = {
974 		[XREP_AGI_INOBT] = {
975 			.rmap_owner = XFS_RMAP_OWN_INOBT,
976 			.buf_ops = &xfs_inobt_buf_ops,
977 			.maxlevels = M_IGEO(sc->mp)->inobt_maxlevels,
978 		},
979 		[XREP_AGI_FINOBT] = {
980 			.rmap_owner = XFS_RMAP_OWN_INOBT,
981 			.buf_ops = &xfs_finobt_buf_ops,
982 			.maxlevels = M_IGEO(sc->mp)->inobt_maxlevels,
983 		},
984 		[XREP_AGI_END] = {
985 			.buf_ops = NULL
986 		},
987 	};
988 	struct xfs_agi			old_agi;
989 	struct xfs_mount		*mp = sc->mp;
990 	struct xfs_buf			*agi_bp;
991 	struct xfs_agi			*agi;
992 	int				error;
993 
994 	/* We require the rmapbt to rebuild anything. */
995 	if (!xfs_has_rmapbt(mp))
996 		return -EOPNOTSUPP;
997 
998 	/*
999 	 * Make sure we have the AGI buffer, as scrub might have decided it
1000 	 * was corrupt after xfs_ialloc_read_agi failed with -EFSCORRUPTED.
1001 	 */
1002 	error = xfs_trans_read_buf(mp, sc->tp, mp->m_ddev_targp,
1003 			XFS_AG_DADDR(mp, sc->sa.pag->pag_agno,
1004 						XFS_AGI_DADDR(mp)),
1005 			XFS_FSS_TO_BB(mp, 1), 0, &agi_bp, NULL);
1006 	if (error)
1007 		return error;
1008 	agi_bp->b_ops = &xfs_agi_buf_ops;
1009 	agi = agi_bp->b_addr;
1010 
1011 	/* Find the AGI btree roots. */
1012 	error = xrep_agi_find_btrees(sc, fab);
1013 	if (error)
1014 		return error;
1015 
1016 	/* Last chance to abort before we start committing fixes. */
1017 	if (xchk_should_terminate(sc, &error))
1018 		return error;
1019 
1020 	/* Start rewriting the header and implant the btrees we found. */
1021 	xrep_agi_init_header(sc, agi_bp, &old_agi);
1022 	xrep_agi_set_roots(sc, agi, fab);
1023 	error = xrep_agi_calc_from_btrees(sc, agi_bp);
1024 	if (error)
1025 		goto out_revert;
1026 
1027 	/* Reinitialize in-core state. */
1028 	return xrep_agi_commit_new(sc, agi_bp);
1029 
1030 out_revert:
1031 	/* Mark the incore AGI state stale and revert the AGI. */
1032 	clear_bit(XFS_AGSTATE_AGI_INIT, &sc->sa.pag->pag_opstate);
1033 	memcpy(agi, &old_agi, sizeof(old_agi));
1034 	return error;
1035 }
1036