xref: /linux/fs/xfs/scrub/quota_repair.c (revision 46e6acfe3501fa938af9c5bd730f0020235b08a2)
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_defer.h"
13 #include "xfs_btree.h"
14 #include "xfs_bit.h"
15 #include "xfs_log_format.h"
16 #include "xfs_trans.h"
17 #include "xfs_sb.h"
18 #include "xfs_inode.h"
19 #include "xfs_inode_fork.h"
20 #include "xfs_alloc.h"
21 #include "xfs_bmap.h"
22 #include "xfs_quota.h"
23 #include "xfs_qm.h"
24 #include "xfs_dquot.h"
25 #include "xfs_dquot_item.h"
26 #include "xfs_reflink.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_trans_space.h"
29 #include "scrub/xfs_scrub.h"
30 #include "scrub/scrub.h"
31 #include "scrub/common.h"
32 #include "scrub/quota.h"
33 #include "scrub/trace.h"
34 #include "scrub/repair.h"
35 
36 /*
37  * Quota Repair
38  * ============
39  *
40  * Quota repairs are fairly simplistic; we fix everything that the dquot
41  * verifiers complain about, cap any counters or limits that make no sense,
42  * and schedule a quotacheck if we had to fix anything.  We also repair any
43  * data fork extent records that don't apply to metadata files.
44  */
45 
46 struct xrep_quota_info {
47 	struct xfs_scrub	*sc;
48 	bool			need_quotacheck;
49 };
50 
51 /*
52  * Allocate a new block into a sparse hole in the quota file backing this
53  * dquot, initialize the block, and commit the whole mess.
54  */
55 STATIC int
56 xrep_quota_item_fill_bmap_hole(
57 	struct xfs_scrub	*sc,
58 	struct xfs_dquot	*dq,
59 	struct xfs_bmbt_irec	*irec)
60 {
61 	struct xfs_buf		*bp;
62 	struct xfs_mount	*mp = sc->mp;
63 	int			nmaps = 1;
64 	int			error;
65 
66 	xfs_trans_ijoin(sc->tp, sc->ip, 0);
67 
68 	/* Map a block into the file. */
69 	error = xfs_trans_reserve_more(sc->tp, XFS_QM_DQALLOC_SPACE_RES(mp),
70 			0);
71 	if (error)
72 		return error;
73 
74 	error = xfs_bmapi_write(sc->tp, sc->ip, dq->q_fileoffset,
75 			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0,
76 			irec, &nmaps);
77 	if (error)
78 		return error;
79 
80 	dq->q_blkno = XFS_FSB_TO_DADDR(mp, irec->br_startblock);
81 
82 	trace_xrep_dquot_item_fill_bmap_hole(sc->mp, dq->q_type, dq->q_id);
83 
84 	/* Initialize the new block. */
85 	error = xfs_trans_get_buf(sc->tp, mp->m_ddev_targp, dq->q_blkno,
86 			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
87 	if (error)
88 		return error;
89 	bp->b_ops = &xfs_dquot_buf_ops;
90 
91 	xfs_qm_init_dquot_blk(sc->tp, dq->q_id, dq->q_type, bp);
92 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
93 
94 	/*
95 	 * Finish the mapping transactions and roll one more time to
96 	 * disconnect sc->ip from sc->tp.
97 	 */
98 	error = xrep_defer_finish(sc);
99 	if (error)
100 		return error;
101 	return xfs_trans_roll(&sc->tp);
102 }
103 
104 /* Make sure there's a written block backing this dquot */
105 STATIC int
106 xrep_quota_item_bmap(
107 	struct xfs_scrub	*sc,
108 	struct xfs_dquot	*dq,
109 	bool			*dirty)
110 {
111 	struct xfs_bmbt_irec	irec;
112 	struct xfs_mount	*mp = sc->mp;
113 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
114 	xfs_fileoff_t		offset = dq->q_id / qi->qi_dqperchunk;
115 	int			nmaps = 1;
116 	int			error;
117 
118 	/* The computed file offset should always be valid. */
119 	if (!xfs_verify_fileoff(mp, offset)) {
120 		ASSERT(xfs_verify_fileoff(mp, offset));
121 		return -EFSCORRUPTED;
122 	}
123 	dq->q_fileoffset = offset;
124 
125 	error = xfs_bmapi_read(sc->ip, offset, 1, &irec, &nmaps, 0);
126 	if (error)
127 		return error;
128 
129 	if (nmaps < 1 || !xfs_bmap_is_real_extent(&irec)) {
130 		/* Hole/delalloc extent; allocate a real block. */
131 		error = xrep_quota_item_fill_bmap_hole(sc, dq, &irec);
132 		if (error)
133 			return error;
134 	} else if (irec.br_state != XFS_EXT_NORM) {
135 		/* Unwritten extent, which we already took care of? */
136 		ASSERT(irec.br_state == XFS_EXT_NORM);
137 		return -EFSCORRUPTED;
138 	} else if (dq->q_blkno != XFS_FSB_TO_DADDR(mp, irec.br_startblock)) {
139 		/*
140 		 * If the cached daddr is incorrect, repair probably punched a
141 		 * hole out of the quota file and filled it back in with a new
142 		 * block.  Update the block mapping in the dquot.
143 		 */
144 		dq->q_blkno = XFS_FSB_TO_DADDR(mp, irec.br_startblock);
145 	}
146 
147 	*dirty = true;
148 	return 0;
149 }
150 
151 /* Reset quota timers if incorrectly set. */
152 static inline void
153 xrep_quota_item_timer(
154 	struct xfs_scrub		*sc,
155 	const struct xfs_dquot_res	*res,
156 	bool				*dirty)
157 {
158 	if ((res->softlimit && res->count > res->softlimit) ||
159 	    (res->hardlimit && res->count > res->hardlimit)) {
160 		if (!res->timer)
161 			*dirty = true;
162 	} else {
163 		if (res->timer)
164 			*dirty = true;
165 	}
166 }
167 
168 /* Scrub the fields in an individual quota item. */
169 STATIC int
170 xrep_quota_item(
171 	struct xrep_quota_info	*rqi,
172 	struct xfs_dquot	*dq)
173 {
174 	struct xfs_scrub	*sc = rqi->sc;
175 	struct xfs_mount	*mp = sc->mp;
176 	xfs_ino_t		fs_icount;
177 	bool			dirty = false;
178 	int			error = 0;
179 
180 	/* Last chance to abort before we start committing fixes. */
181 	if (xchk_should_terminate(sc, &error))
182 		return error;
183 
184 	/*
185 	 * We might need to fix holes in the bmap record for the storage
186 	 * backing this dquot, so we need to lock the dquot and the quota file.
187 	 * dqiterate gave us a locked dquot, so drop the dquot lock to get the
188 	 * ILOCK_EXCL.
189 	 */
190 	xfs_dqunlock(dq);
191 	xchk_ilock(sc, XFS_ILOCK_EXCL);
192 	xfs_dqlock(dq);
193 
194 	error = xrep_quota_item_bmap(sc, dq, &dirty);
195 	xchk_iunlock(sc, XFS_ILOCK_EXCL);
196 	if (error)
197 		return error;
198 
199 	/* Check the limits. */
200 	if (dq->q_blk.softlimit > dq->q_blk.hardlimit) {
201 		dq->q_blk.softlimit = dq->q_blk.hardlimit;
202 		dirty = true;
203 	}
204 
205 	if (dq->q_ino.softlimit > dq->q_ino.hardlimit) {
206 		dq->q_ino.softlimit = dq->q_ino.hardlimit;
207 		dirty = true;
208 	}
209 
210 	if (dq->q_rtb.softlimit > dq->q_rtb.hardlimit) {
211 		dq->q_rtb.softlimit = dq->q_rtb.hardlimit;
212 		dirty = true;
213 	}
214 
215 	/*
216 	 * Check that usage doesn't exceed physical limits.  However, on
217 	 * a reflink filesystem we're allowed to exceed physical space
218 	 * if there are no quota limits.  We don't know what the real number
219 	 * is, but we can make quotacheck find out for us.
220 	 */
221 	if (!xfs_has_reflink(mp) && dq->q_blk.count > mp->m_sb.sb_dblocks) {
222 		dq->q_blk.reserved -= dq->q_blk.count;
223 		dq->q_blk.reserved += mp->m_sb.sb_dblocks;
224 		dq->q_blk.count = mp->m_sb.sb_dblocks;
225 		rqi->need_quotacheck = true;
226 		dirty = true;
227 	}
228 	fs_icount = percpu_counter_sum(&mp->m_icount);
229 	if (dq->q_ino.count > fs_icount) {
230 		dq->q_ino.reserved -= dq->q_ino.count;
231 		dq->q_ino.reserved += fs_icount;
232 		dq->q_ino.count = fs_icount;
233 		rqi->need_quotacheck = true;
234 		dirty = true;
235 	}
236 	if (dq->q_rtb.count > mp->m_sb.sb_rblocks) {
237 		dq->q_rtb.reserved -= dq->q_rtb.count;
238 		dq->q_rtb.reserved += mp->m_sb.sb_rblocks;
239 		dq->q_rtb.count = mp->m_sb.sb_rblocks;
240 		rqi->need_quotacheck = true;
241 		dirty = true;
242 	}
243 
244 	xrep_quota_item_timer(sc, &dq->q_blk, &dirty);
245 	xrep_quota_item_timer(sc, &dq->q_ino, &dirty);
246 	xrep_quota_item_timer(sc, &dq->q_rtb, &dirty);
247 
248 	if (!dirty)
249 		return 0;
250 
251 	trace_xrep_dquot_item(sc->mp, dq->q_type, dq->q_id);
252 
253 	dq->q_flags |= XFS_DQFLAG_DIRTY;
254 	xfs_trans_dqjoin(sc->tp, dq);
255 	if (dq->q_id) {
256 		xfs_qm_adjust_dqlimits(dq);
257 		xfs_qm_adjust_dqtimers(dq);
258 	}
259 	xfs_trans_log_dquot(sc->tp, dq);
260 	error = xfs_trans_roll(&sc->tp);
261 	xfs_dqlock(dq);
262 	return error;
263 }
264 
265 /* Fix a quota timer so that we can pass the verifier. */
266 STATIC void
267 xrep_quota_fix_timer(
268 	struct xfs_mount	*mp,
269 	const struct xfs_disk_dquot *ddq,
270 	__be64			softlimit,
271 	__be64			countnow,
272 	__be32			*timer,
273 	time64_t		timelimit)
274 {
275 	uint64_t		soft = be64_to_cpu(softlimit);
276 	uint64_t		count = be64_to_cpu(countnow);
277 	time64_t		new_timer;
278 	uint32_t		t;
279 
280 	if (!soft || count <= soft || *timer != 0)
281 		return;
282 
283 	new_timer = xfs_dquot_set_timeout(mp,
284 				ktime_get_real_seconds() + timelimit);
285 	if (ddq->d_type & XFS_DQTYPE_BIGTIME)
286 		t = xfs_dq_unix_to_bigtime(new_timer);
287 	else
288 		t = new_timer;
289 
290 	*timer = cpu_to_be32(t);
291 }
292 
293 /* Fix anything the verifiers complain about. */
294 STATIC int
295 xrep_quota_block(
296 	struct xfs_scrub	*sc,
297 	xfs_daddr_t		daddr,
298 	xfs_dqtype_t		dqtype,
299 	xfs_dqid_t		id)
300 {
301 	struct xfs_dqblk	*dqblk;
302 	struct xfs_disk_dquot	*ddq;
303 	struct xfs_quotainfo	*qi = sc->mp->m_quotainfo;
304 	struct xfs_def_quota	*defq = xfs_get_defquota(qi, dqtype);
305 	struct xfs_buf		*bp = NULL;
306 	enum xfs_blft		buftype = 0;
307 	int			i;
308 	int			error;
309 
310 	error = xfs_trans_read_buf(sc->mp, sc->tp, sc->mp->m_ddev_targp, daddr,
311 			qi->qi_dqchunklen, 0, &bp, &xfs_dquot_buf_ops);
312 	switch (error) {
313 	case -EFSBADCRC:
314 	case -EFSCORRUPTED:
315 		/* Failed verifier, retry read with no ops. */
316 		error = xfs_trans_read_buf(sc->mp, sc->tp,
317 				sc->mp->m_ddev_targp, daddr, qi->qi_dqchunklen,
318 				0, &bp, NULL);
319 		if (error)
320 			return error;
321 		break;
322 	case 0:
323 		dqblk = bp->b_addr;
324 		ddq = &dqblk[0].dd_diskdq;
325 
326 		/*
327 		 * If there's nothing that would impede a dqiterate, we're
328 		 * done.
329 		 */
330 		if ((ddq->d_type & XFS_DQTYPE_REC_MASK) != dqtype ||
331 		    id == be32_to_cpu(ddq->d_id)) {
332 			xfs_trans_brelse(sc->tp, bp);
333 			return 0;
334 		}
335 		break;
336 	default:
337 		return error;
338 	}
339 
340 	/* Something's wrong with the block, fix the whole thing. */
341 	dqblk = bp->b_addr;
342 	bp->b_ops = &xfs_dquot_buf_ops;
343 	for (i = 0; i < qi->qi_dqperchunk; i++, dqblk++) {
344 		ddq = &dqblk->dd_diskdq;
345 
346 		trace_xrep_disk_dquot(sc->mp, dqtype, id + i);
347 
348 		ddq->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
349 		ddq->d_version = XFS_DQUOT_VERSION;
350 		ddq->d_type = dqtype;
351 		ddq->d_id = cpu_to_be32(id + i);
352 
353 		if (xfs_has_bigtime(sc->mp) && ddq->d_id)
354 			ddq->d_type |= XFS_DQTYPE_BIGTIME;
355 
356 		xrep_quota_fix_timer(sc->mp, ddq, ddq->d_blk_softlimit,
357 				ddq->d_bcount, &ddq->d_btimer,
358 				defq->blk.time);
359 
360 		xrep_quota_fix_timer(sc->mp, ddq, ddq->d_ino_softlimit,
361 				ddq->d_icount, &ddq->d_itimer,
362 				defq->ino.time);
363 
364 		xrep_quota_fix_timer(sc->mp, ddq, ddq->d_rtb_softlimit,
365 				ddq->d_rtbcount, &ddq->d_rtbtimer,
366 				defq->rtb.time);
367 
368 		/* We only support v5 filesystems so always set these. */
369 		uuid_copy(&dqblk->dd_uuid, &sc->mp->m_sb.sb_meta_uuid);
370 		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
371 				 XFS_DQUOT_CRC_OFF);
372 		dqblk->dd_lsn = 0;
373 	}
374 	switch (dqtype) {
375 	case XFS_DQTYPE_USER:
376 		buftype = XFS_BLFT_UDQUOT_BUF;
377 		break;
378 	case XFS_DQTYPE_GROUP:
379 		buftype = XFS_BLFT_GDQUOT_BUF;
380 		break;
381 	case XFS_DQTYPE_PROJ:
382 		buftype = XFS_BLFT_PDQUOT_BUF;
383 		break;
384 	}
385 	xfs_trans_buf_set_type(sc->tp, bp, buftype);
386 	xfs_trans_log_buf(sc->tp, bp, 0, BBTOB(bp->b_length) - 1);
387 	return xrep_roll_trans(sc);
388 }
389 
390 /*
391  * Repair a quota file's data fork.  The function returns with the inode
392  * joined.
393  */
394 STATIC int
395 xrep_quota_data_fork(
396 	struct xfs_scrub	*sc,
397 	xfs_dqtype_t		dqtype)
398 {
399 	struct xfs_bmbt_irec	irec = { 0 };
400 	struct xfs_iext_cursor	icur;
401 	struct xfs_quotainfo	*qi = sc->mp->m_quotainfo;
402 	struct xfs_ifork	*ifp;
403 	xfs_fileoff_t		max_dqid_off;
404 	xfs_fileoff_t		off;
405 	xfs_fsblock_t		fsbno;
406 	bool			truncate = false;
407 	bool			joined = false;
408 	int			error = 0;
409 
410 	error = xrep_metadata_inode_forks(sc);
411 	if (error)
412 		goto out;
413 
414 	/* Check for data fork problems that apply only to quota files. */
415 	max_dqid_off = XFS_DQ_ID_MAX / qi->qi_dqperchunk;
416 	ifp = xfs_ifork_ptr(sc->ip, XFS_DATA_FORK);
417 	for_each_xfs_iext(ifp, &icur, &irec) {
418 		if (isnullstartblock(irec.br_startblock)) {
419 			error = -EFSCORRUPTED;
420 			goto out;
421 		}
422 
423 		if (irec.br_startoff > max_dqid_off ||
424 		    irec.br_startoff + irec.br_blockcount - 1 > max_dqid_off) {
425 			truncate = true;
426 			break;
427 		}
428 
429 		/* Convert unwritten extents to real ones. */
430 		if (irec.br_state == XFS_EXT_UNWRITTEN) {
431 			struct xfs_bmbt_irec	nrec;
432 			int			nmap = 1;
433 
434 			if (!joined) {
435 				xfs_trans_ijoin(sc->tp, sc->ip, 0);
436 				joined = true;
437 			}
438 
439 			error = xfs_bmapi_write(sc->tp, sc->ip,
440 					irec.br_startoff, irec.br_blockcount,
441 					XFS_BMAPI_CONVERT, 0, &nrec, &nmap);
442 			if (error)
443 				goto out;
444 			ASSERT(nrec.br_startoff == irec.br_startoff);
445 			ASSERT(nrec.br_blockcount == irec.br_blockcount);
446 
447 			error = xfs_defer_finish(&sc->tp);
448 			if (error)
449 				goto out;
450 		}
451 	}
452 
453 	if (!joined) {
454 		xfs_trans_ijoin(sc->tp, sc->ip, 0);
455 		joined = true;
456 	}
457 
458 	if (truncate) {
459 		/* Erase everything after the block containing the max dquot */
460 		error = xfs_bunmapi_range(&sc->tp, sc->ip, 0,
461 				max_dqid_off * sc->mp->m_sb.sb_blocksize,
462 				XFS_MAX_FILEOFF);
463 		if (error)
464 			goto out;
465 
466 		/* Remove all CoW reservations. */
467 		error = xfs_reflink_cancel_cow_blocks(sc->ip, &sc->tp, 0,
468 				XFS_MAX_FILEOFF, true);
469 		if (error)
470 			goto out;
471 		sc->ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
472 
473 		/*
474 		 * Always re-log the inode so that our permanent transaction
475 		 * can keep on rolling it forward in the log.
476 		 */
477 		xfs_trans_log_inode(sc->tp, sc->ip, XFS_ILOG_CORE);
478 	}
479 
480 	/* Now go fix anything that fails the verifiers. */
481 	for_each_xfs_iext(ifp, &icur, &irec) {
482 		for (fsbno = irec.br_startblock, off = irec.br_startoff;
483 		     fsbno < irec.br_startblock + irec.br_blockcount;
484 		     fsbno += XFS_DQUOT_CLUSTER_SIZE_FSB,
485 				off += XFS_DQUOT_CLUSTER_SIZE_FSB) {
486 			error = xrep_quota_block(sc,
487 					XFS_FSB_TO_DADDR(sc->mp, fsbno),
488 					dqtype, off * qi->qi_dqperchunk);
489 			if (error)
490 				goto out;
491 		}
492 	}
493 
494 out:
495 	return error;
496 }
497 
498 /*
499  * Go fix anything in the quota items that we could have been mad about.  Now
500  * that we've checked the quota inode data fork we have to drop ILOCK_EXCL to
501  * use the regular dquot functions.
502  */
503 STATIC int
504 xrep_quota_problems(
505 	struct xfs_scrub	*sc,
506 	xfs_dqtype_t		dqtype)
507 {
508 	struct xchk_dqiter	cursor = { };
509 	struct xrep_quota_info	rqi = { .sc = sc };
510 	struct xfs_dquot	*dq;
511 	int			error;
512 
513 	xchk_dqiter_init(&cursor, sc, dqtype);
514 	while ((error = xchk_dquot_iter(&cursor, &dq)) == 1) {
515 		error = xrep_quota_item(&rqi, dq);
516 		xfs_qm_dqput(dq);
517 		if (error)
518 			break;
519 	}
520 	if (error)
521 		return error;
522 
523 	/* Make a quotacheck happen. */
524 	if (rqi.need_quotacheck)
525 		xrep_force_quotacheck(sc, dqtype);
526 	return 0;
527 }
528 
529 /* Repair all of a quota type's items. */
530 int
531 xrep_quota(
532 	struct xfs_scrub	*sc)
533 {
534 	xfs_dqtype_t		dqtype;
535 	int			error;
536 
537 	dqtype = xchk_quota_to_dqtype(sc);
538 
539 	/*
540 	 * Re-take the ILOCK so that we can fix any problems that we found
541 	 * with the data fork mappings, or with the dquot bufs themselves.
542 	 */
543 	if (!(sc->ilock_flags & XFS_ILOCK_EXCL))
544 		xchk_ilock(sc, XFS_ILOCK_EXCL);
545 	error = xrep_quota_data_fork(sc, dqtype);
546 	if (error)
547 		return error;
548 
549 	/*
550 	 * Finish deferred items and roll the transaction to unjoin the quota
551 	 * inode from transaction so that we can unlock the quota inode; we
552 	 * play only with dquots from now on.
553 	 */
554 	error = xrep_defer_finish(sc);
555 	if (error)
556 		return error;
557 	error = xfs_trans_roll(&sc->tp);
558 	if (error)
559 		return error;
560 	xchk_iunlock(sc, sc->ilock_flags);
561 
562 	/* Fix anything the dquot verifiers don't complain about. */
563 	error = xrep_quota_problems(sc, dqtype);
564 	if (error)
565 		return error;
566 
567 	return xrep_trans_commit(sc);
568 }
569