xref: /linux/fs/xfs/scrub/ialloc.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  * Copyright (C) 2017 Oracle.  All Rights Reserved.
3  *
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version 2
9  * of the License, or (at your option) any later version.
10  *
11  * This program is distributed in the hope that it would be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write the Free Software Foundation,
18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
19  */
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_btree.h"
28 #include "xfs_bit.h"
29 #include "xfs_log_format.h"
30 #include "xfs_trans.h"
31 #include "xfs_sb.h"
32 #include "xfs_inode.h"
33 #include "xfs_alloc.h"
34 #include "xfs_ialloc.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_icache.h"
37 #include "xfs_rmap.h"
38 #include "xfs_log.h"
39 #include "xfs_trans_priv.h"
40 #include "scrub/xfs_scrub.h"
41 #include "scrub/scrub.h"
42 #include "scrub/common.h"
43 #include "scrub/btree.h"
44 #include "scrub/trace.h"
45 
46 /*
47  * Set us up to scrub inode btrees.
48  * If we detect a discrepancy between the inobt and the inode,
49  * try again after forcing logged inode cores out to disk.
50  */
51 int
52 xfs_scrub_setup_ag_iallocbt(
53 	struct xfs_scrub_context	*sc,
54 	struct xfs_inode		*ip)
55 {
56 	return xfs_scrub_setup_ag_btree(sc, ip, sc->try_harder);
57 }
58 
59 /* Inode btree scrubber. */
60 
61 /* Is this chunk worth checking? */
62 STATIC bool
63 xfs_scrub_iallocbt_chunk(
64 	struct xfs_scrub_btree		*bs,
65 	struct xfs_inobt_rec_incore	*irec,
66 	xfs_agino_t			agino,
67 	xfs_extlen_t			len)
68 {
69 	struct xfs_mount		*mp = bs->cur->bc_mp;
70 	xfs_agnumber_t			agno = bs->cur->bc_private.a.agno;
71 	xfs_agblock_t			bno;
72 
73 	bno = XFS_AGINO_TO_AGBNO(mp, agino);
74 	if (bno + len <= bno ||
75 	    !xfs_verify_agbno(mp, agno, bno) ||
76 	    !xfs_verify_agbno(mp, agno, bno + len - 1))
77 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
78 
79 	return true;
80 }
81 
82 /* Count the number of free inodes. */
83 static unsigned int
84 xfs_scrub_iallocbt_freecount(
85 	xfs_inofree_t			freemask)
86 {
87 	BUILD_BUG_ON(sizeof(freemask) != sizeof(__u64));
88 	return hweight64(freemask);
89 }
90 
91 /* Check a particular inode with ir_free. */
92 STATIC int
93 xfs_scrub_iallocbt_check_cluster_freemask(
94 	struct xfs_scrub_btree		*bs,
95 	xfs_ino_t			fsino,
96 	xfs_agino_t			chunkino,
97 	xfs_agino_t			clusterino,
98 	struct xfs_inobt_rec_incore	*irec,
99 	struct xfs_buf			*bp)
100 {
101 	struct xfs_dinode		*dip;
102 	struct xfs_mount		*mp = bs->cur->bc_mp;
103 	bool				inode_is_free = false;
104 	bool				freemask_ok;
105 	bool				inuse;
106 	int				error = 0;
107 
108 	if (xfs_scrub_should_terminate(bs->sc, &error))
109 		return error;
110 
111 	dip = xfs_buf_offset(bp, clusterino * mp->m_sb.sb_inodesize);
112 	if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
113 	    (dip->di_version >= 3 &&
114 	     be64_to_cpu(dip->di_ino) != fsino + clusterino)) {
115 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
116 		goto out;
117 	}
118 
119 	if (irec->ir_free & XFS_INOBT_MASK(chunkino + clusterino))
120 		inode_is_free = true;
121 	error = xfs_icache_inode_is_allocated(mp, bs->cur->bc_tp,
122 			fsino + clusterino, &inuse);
123 	if (error == -ENODATA) {
124 		/* Not cached, just read the disk buffer */
125 		freemask_ok = inode_is_free ^ !!(dip->di_mode);
126 		if (!bs->sc->try_harder && !freemask_ok)
127 			return -EDEADLOCK;
128 	} else if (error < 0) {
129 		/*
130 		 * Inode is only half assembled, or there was an IO error,
131 		 * or the verifier failed, so don't bother trying to check.
132 		 * The inode scrubber can deal with this.
133 		 */
134 		goto out;
135 	} else {
136 		/* Inode is all there. */
137 		freemask_ok = inode_is_free ^ inuse;
138 	}
139 	if (!freemask_ok)
140 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
141 out:
142 	return 0;
143 }
144 
145 /* Make sure the free mask is consistent with what the inodes think. */
146 STATIC int
147 xfs_scrub_iallocbt_check_freemask(
148 	struct xfs_scrub_btree		*bs,
149 	struct xfs_inobt_rec_incore	*irec)
150 {
151 	struct xfs_owner_info		oinfo;
152 	struct xfs_imap			imap;
153 	struct xfs_mount		*mp = bs->cur->bc_mp;
154 	struct xfs_dinode		*dip;
155 	struct xfs_buf			*bp;
156 	xfs_ino_t			fsino;
157 	xfs_agino_t			nr_inodes;
158 	xfs_agino_t			agino;
159 	xfs_agino_t			chunkino;
160 	xfs_agino_t			clusterino;
161 	xfs_agblock_t			agbno;
162 	int				blks_per_cluster;
163 	uint16_t			holemask;
164 	uint16_t			ir_holemask;
165 	int				error = 0;
166 
167 	/* Make sure the freemask matches the inode records. */
168 	blks_per_cluster = xfs_icluster_size_fsb(mp);
169 	nr_inodes = XFS_OFFBNO_TO_AGINO(mp, blks_per_cluster, 0);
170 	xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES);
171 
172 	for (agino = irec->ir_startino;
173 	     agino < irec->ir_startino + XFS_INODES_PER_CHUNK;
174 	     agino += blks_per_cluster * mp->m_sb.sb_inopblock) {
175 		fsino = XFS_AGINO_TO_INO(mp, bs->cur->bc_private.a.agno, agino);
176 		chunkino = agino - irec->ir_startino;
177 		agbno = XFS_AGINO_TO_AGBNO(mp, agino);
178 
179 		/* Compute the holemask mask for this cluster. */
180 		for (clusterino = 0, holemask = 0; clusterino < nr_inodes;
181 		     clusterino += XFS_INODES_PER_HOLEMASK_BIT)
182 			holemask |= XFS_INOBT_MASK((chunkino + clusterino) /
183 					XFS_INODES_PER_HOLEMASK_BIT);
184 
185 		/* The whole cluster must be a hole or not a hole. */
186 		ir_holemask = (irec->ir_holemask & holemask);
187 		if (ir_holemask != holemask && ir_holemask != 0) {
188 			xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
189 			continue;
190 		}
191 
192 		/* If any part of this is a hole, skip it. */
193 		if (ir_holemask)
194 			continue;
195 
196 		/* Grab the inode cluster buffer. */
197 		imap.im_blkno = XFS_AGB_TO_DADDR(mp, bs->cur->bc_private.a.agno,
198 				agbno);
199 		imap.im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
200 		imap.im_boffset = 0;
201 
202 		error = xfs_imap_to_bp(mp, bs->cur->bc_tp, &imap,
203 				&dip, &bp, 0, 0);
204 		if (!xfs_scrub_btree_process_error(bs->sc, bs->cur, 0, &error))
205 			continue;
206 
207 		/* Which inodes are free? */
208 		for (clusterino = 0; clusterino < nr_inodes; clusterino++) {
209 			error = xfs_scrub_iallocbt_check_cluster_freemask(bs,
210 					fsino, chunkino, clusterino, irec, bp);
211 			if (error) {
212 				xfs_trans_brelse(bs->cur->bc_tp, bp);
213 				return error;
214 			}
215 		}
216 
217 		xfs_trans_brelse(bs->cur->bc_tp, bp);
218 	}
219 
220 	return error;
221 }
222 
223 /* Scrub an inobt/finobt record. */
224 STATIC int
225 xfs_scrub_iallocbt_rec(
226 	struct xfs_scrub_btree		*bs,
227 	union xfs_btree_rec		*rec)
228 {
229 	struct xfs_mount		*mp = bs->cur->bc_mp;
230 	struct xfs_inobt_rec_incore	irec;
231 	uint64_t			holes;
232 	xfs_agnumber_t			agno = bs->cur->bc_private.a.agno;
233 	xfs_agino_t			agino;
234 	xfs_agblock_t			agbno;
235 	xfs_extlen_t			len;
236 	int				holecount;
237 	int				i;
238 	int				error = 0;
239 	unsigned int			real_freecount;
240 	uint16_t			holemask;
241 
242 	xfs_inobt_btrec_to_irec(mp, rec, &irec);
243 
244 	if (irec.ir_count > XFS_INODES_PER_CHUNK ||
245 	    irec.ir_freecount > XFS_INODES_PER_CHUNK)
246 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
247 
248 	real_freecount = irec.ir_freecount +
249 			(XFS_INODES_PER_CHUNK - irec.ir_count);
250 	if (real_freecount != xfs_scrub_iallocbt_freecount(irec.ir_free))
251 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
252 
253 	agino = irec.ir_startino;
254 	/* Record has to be properly aligned within the AG. */
255 	if (!xfs_verify_agino(mp, agno, agino) ||
256 	    !xfs_verify_agino(mp, agno, agino + XFS_INODES_PER_CHUNK - 1)) {
257 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
258 		goto out;
259 	}
260 
261 	/* Make sure this record is aligned to cluster and inoalignmnt size. */
262 	agbno = XFS_AGINO_TO_AGBNO(mp, irec.ir_startino);
263 	if ((agbno & (xfs_ialloc_cluster_alignment(mp) - 1)) ||
264 	    (agbno & (xfs_icluster_size_fsb(mp) - 1)))
265 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
266 
267 	/* Handle non-sparse inodes */
268 	if (!xfs_inobt_issparse(irec.ir_holemask)) {
269 		len = XFS_B_TO_FSB(mp,
270 				XFS_INODES_PER_CHUNK * mp->m_sb.sb_inodesize);
271 		if (irec.ir_count != XFS_INODES_PER_CHUNK)
272 			xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
273 
274 		if (!xfs_scrub_iallocbt_chunk(bs, &irec, agino, len))
275 			goto out;
276 		goto check_freemask;
277 	}
278 
279 	/* Check each chunk of a sparse inode cluster. */
280 	holemask = irec.ir_holemask;
281 	holecount = 0;
282 	len = XFS_B_TO_FSB(mp,
283 			XFS_INODES_PER_HOLEMASK_BIT * mp->m_sb.sb_inodesize);
284 	holes = ~xfs_inobt_irec_to_allocmask(&irec);
285 	if ((holes & irec.ir_free) != holes ||
286 	    irec.ir_freecount > irec.ir_count)
287 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
288 
289 	for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
290 		if (holemask & 1)
291 			holecount += XFS_INODES_PER_HOLEMASK_BIT;
292 		else if (!xfs_scrub_iallocbt_chunk(bs, &irec, agino, len))
293 			break;
294 		holemask >>= 1;
295 		agino += XFS_INODES_PER_HOLEMASK_BIT;
296 	}
297 
298 	if (holecount > XFS_INODES_PER_CHUNK ||
299 	    holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
300 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
301 
302 check_freemask:
303 	error = xfs_scrub_iallocbt_check_freemask(bs, &irec);
304 	if (error)
305 		goto out;
306 
307 out:
308 	return error;
309 }
310 
311 /* Scrub the inode btrees for some AG. */
312 STATIC int
313 xfs_scrub_iallocbt(
314 	struct xfs_scrub_context	*sc,
315 	xfs_btnum_t			which)
316 {
317 	struct xfs_btree_cur		*cur;
318 	struct xfs_owner_info		oinfo;
319 
320 	xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INOBT);
321 	cur = which == XFS_BTNUM_INO ? sc->sa.ino_cur : sc->sa.fino_cur;
322 	return xfs_scrub_btree(sc, cur, xfs_scrub_iallocbt_rec, &oinfo, NULL);
323 }
324 
325 int
326 xfs_scrub_inobt(
327 	struct xfs_scrub_context	*sc)
328 {
329 	return xfs_scrub_iallocbt(sc, XFS_BTNUM_INO);
330 }
331 
332 int
333 xfs_scrub_finobt(
334 	struct xfs_scrub_context	*sc)
335 {
336 	return xfs_scrub_iallocbt(sc, XFS_BTNUM_FINO);
337 }
338