xref: /openbmc/linux/fs/xfs/libxfs/xfs_inode_buf.c (revision e330fb14)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_errortag.h"
15 #include "xfs_error.h"
16 #include "xfs_icache.h"
17 #include "xfs_trans.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_dir2.h"
20 
21 #include <linux/iversion.h>
22 
23 /*
24  * If we are doing readahead on an inode buffer, we might be in log recovery
25  * reading an inode allocation buffer that hasn't yet been replayed, and hence
26  * has not had the inode cores stamped into it. Hence for readahead, the buffer
27  * may be potentially invalid.
28  *
29  * If the readahead buffer is invalid, we need to mark it with an error and
30  * clear the DONE status of the buffer so that a followup read will re-read it
31  * from disk. We don't report the error otherwise to avoid warnings during log
32  * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
33  * because all we want to do is say readahead failed; there is no-one to report
34  * the error to, so this will distinguish it from a non-ra verifier failure.
35  * Changes to this readahead error behaviour also need to be reflected in
36  * xfs_dquot_buf_readahead_verify().
37  */
38 static void
39 xfs_inode_buf_verify(
40 	struct xfs_buf	*bp,
41 	bool		readahead)
42 {
43 	struct xfs_mount *mp = bp->b_mount;
44 	xfs_agnumber_t	agno;
45 	int		i;
46 	int		ni;
47 
48 	/*
49 	 * Validate the magic number and version of every inode in the buffer
50 	 */
51 	agno = xfs_daddr_to_agno(mp, xfs_buf_daddr(bp));
52 	ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
53 	for (i = 0; i < ni; i++) {
54 		int		di_ok;
55 		xfs_dinode_t	*dip;
56 		xfs_agino_t	unlinked_ino;
57 
58 		dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
59 		unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
60 		di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
61 			xfs_dinode_good_version(mp, dip->di_version) &&
62 			xfs_verify_agino_or_null(mp, agno, unlinked_ino);
63 		if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
64 						XFS_ERRTAG_ITOBP_INOTOBP))) {
65 			if (readahead) {
66 				bp->b_flags &= ~XBF_DONE;
67 				xfs_buf_ioerror(bp, -EIO);
68 				return;
69 			}
70 
71 #ifdef DEBUG
72 			xfs_alert(mp,
73 				"bad inode magic/vsn daddr %lld #%d (magic=%x)",
74 				(unsigned long long)xfs_buf_daddr(bp), i,
75 				be16_to_cpu(dip->di_magic));
76 #endif
77 			xfs_buf_verifier_error(bp, -EFSCORRUPTED,
78 					__func__, dip, sizeof(*dip),
79 					NULL);
80 			return;
81 		}
82 	}
83 }
84 
85 
86 static void
87 xfs_inode_buf_read_verify(
88 	struct xfs_buf	*bp)
89 {
90 	xfs_inode_buf_verify(bp, false);
91 }
92 
93 static void
94 xfs_inode_buf_readahead_verify(
95 	struct xfs_buf	*bp)
96 {
97 	xfs_inode_buf_verify(bp, true);
98 }
99 
100 static void
101 xfs_inode_buf_write_verify(
102 	struct xfs_buf	*bp)
103 {
104 	xfs_inode_buf_verify(bp, false);
105 }
106 
107 const struct xfs_buf_ops xfs_inode_buf_ops = {
108 	.name = "xfs_inode",
109 	.magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
110 		     cpu_to_be16(XFS_DINODE_MAGIC) },
111 	.verify_read = xfs_inode_buf_read_verify,
112 	.verify_write = xfs_inode_buf_write_verify,
113 };
114 
115 const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
116 	.name = "xfs_inode_ra",
117 	.magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
118 		     cpu_to_be16(XFS_DINODE_MAGIC) },
119 	.verify_read = xfs_inode_buf_readahead_verify,
120 	.verify_write = xfs_inode_buf_write_verify,
121 };
122 
123 
124 /*
125  * This routine is called to map an inode to the buffer containing the on-disk
126  * version of the inode.  It returns a pointer to the buffer containing the
127  * on-disk inode in the bpp parameter.
128  */
129 int
130 xfs_imap_to_bp(
131 	struct xfs_mount	*mp,
132 	struct xfs_trans	*tp,
133 	struct xfs_imap		*imap,
134 	struct xfs_buf		**bpp)
135 {
136 	return xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
137 				   imap->im_len, XBF_UNMAPPED, bpp,
138 				   &xfs_inode_buf_ops);
139 }
140 
141 static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
142 {
143 	struct timespec64	tv;
144 	uint32_t		n;
145 
146 	tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
147 	tv.tv_nsec = n;
148 
149 	return tv;
150 }
151 
152 /* Convert an ondisk timestamp to an incore timestamp. */
153 struct timespec64
154 xfs_inode_from_disk_ts(
155 	struct xfs_dinode		*dip,
156 	const xfs_timestamp_t		ts)
157 {
158 	struct timespec64		tv;
159 	struct xfs_legacy_timestamp	*lts;
160 
161 	if (xfs_dinode_has_bigtime(dip))
162 		return xfs_inode_decode_bigtime(be64_to_cpu(ts));
163 
164 	lts = (struct xfs_legacy_timestamp *)&ts;
165 	tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
166 	tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
167 
168 	return tv;
169 }
170 
171 int
172 xfs_inode_from_disk(
173 	struct xfs_inode	*ip,
174 	struct xfs_dinode	*from)
175 {
176 	struct inode		*inode = VFS_I(ip);
177 	int			error;
178 	xfs_failaddr_t		fa;
179 
180 	ASSERT(ip->i_cowfp == NULL);
181 	ASSERT(ip->i_afp == NULL);
182 
183 	fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
184 	if (fa) {
185 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
186 				sizeof(*from), fa);
187 		return -EFSCORRUPTED;
188 	}
189 
190 	/*
191 	 * First get the permanent information that is needed to allocate an
192 	 * inode. If the inode is unused, mode is zero and we shouldn't mess
193 	 * with the uninitialized part of it.
194 	 */
195 	if (!xfs_has_v3inodes(ip->i_mount))
196 		ip->i_flushiter = be16_to_cpu(from->di_flushiter);
197 	inode->i_generation = be32_to_cpu(from->di_gen);
198 	inode->i_mode = be16_to_cpu(from->di_mode);
199 	if (!inode->i_mode)
200 		return 0;
201 
202 	/*
203 	 * Convert v1 inodes immediately to v2 inode format as this is the
204 	 * minimum inode version format we support in the rest of the code.
205 	 * They will also be unconditionally written back to disk as v2 inodes.
206 	 */
207 	if (unlikely(from->di_version == 1)) {
208 		set_nlink(inode, be16_to_cpu(from->di_onlink));
209 		ip->i_projid = 0;
210 	} else {
211 		set_nlink(inode, be32_to_cpu(from->di_nlink));
212 		ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
213 					be16_to_cpu(from->di_projid_lo);
214 	}
215 
216 	i_uid_write(inode, be32_to_cpu(from->di_uid));
217 	i_gid_write(inode, be32_to_cpu(from->di_gid));
218 
219 	/*
220 	 * Time is signed, so need to convert to signed 32 bit before
221 	 * storing in inode timestamp which may be 64 bit. Otherwise
222 	 * a time before epoch is converted to a time long after epoch
223 	 * on 64 bit systems.
224 	 */
225 	inode->i_atime = xfs_inode_from_disk_ts(from, from->di_atime);
226 	inode->i_mtime = xfs_inode_from_disk_ts(from, from->di_mtime);
227 	inode->i_ctime = xfs_inode_from_disk_ts(from, from->di_ctime);
228 
229 	ip->i_disk_size = be64_to_cpu(from->di_size);
230 	ip->i_nblocks = be64_to_cpu(from->di_nblocks);
231 	ip->i_extsize = be32_to_cpu(from->di_extsize);
232 	ip->i_forkoff = from->di_forkoff;
233 	ip->i_diflags	= be16_to_cpu(from->di_flags);
234 
235 	if (from->di_dmevmask || from->di_dmstate)
236 		xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
237 
238 	if (xfs_has_v3inodes(ip->i_mount)) {
239 		inode_set_iversion_queried(inode,
240 					   be64_to_cpu(from->di_changecount));
241 		ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
242 		ip->i_diflags2 = be64_to_cpu(from->di_flags2);
243 		ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
244 	}
245 
246 	error = xfs_iformat_data_fork(ip, from);
247 	if (error)
248 		return error;
249 	if (from->di_forkoff) {
250 		error = xfs_iformat_attr_fork(ip, from);
251 		if (error)
252 			goto out_destroy_data_fork;
253 	}
254 	if (xfs_is_reflink_inode(ip))
255 		xfs_ifork_init_cow(ip);
256 	return 0;
257 
258 out_destroy_data_fork:
259 	xfs_idestroy_fork(&ip->i_df);
260 	return error;
261 }
262 
263 /* Convert an incore timestamp to an ondisk timestamp. */
264 static inline xfs_timestamp_t
265 xfs_inode_to_disk_ts(
266 	struct xfs_inode		*ip,
267 	const struct timespec64		tv)
268 {
269 	struct xfs_legacy_timestamp	*lts;
270 	xfs_timestamp_t			ts;
271 
272 	if (xfs_inode_has_bigtime(ip))
273 		return cpu_to_be64(xfs_inode_encode_bigtime(tv));
274 
275 	lts = (struct xfs_legacy_timestamp *)&ts;
276 	lts->t_sec = cpu_to_be32(tv.tv_sec);
277 	lts->t_nsec = cpu_to_be32(tv.tv_nsec);
278 
279 	return ts;
280 }
281 
282 void
283 xfs_inode_to_disk(
284 	struct xfs_inode	*ip,
285 	struct xfs_dinode	*to,
286 	xfs_lsn_t		lsn)
287 {
288 	struct inode		*inode = VFS_I(ip);
289 
290 	to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
291 	to->di_onlink = 0;
292 
293 	to->di_format = xfs_ifork_format(&ip->i_df);
294 	to->di_uid = cpu_to_be32(i_uid_read(inode));
295 	to->di_gid = cpu_to_be32(i_gid_read(inode));
296 	to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
297 	to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
298 
299 	memset(to->di_pad, 0, sizeof(to->di_pad));
300 	to->di_atime = xfs_inode_to_disk_ts(ip, inode->i_atime);
301 	to->di_mtime = xfs_inode_to_disk_ts(ip, inode->i_mtime);
302 	to->di_ctime = xfs_inode_to_disk_ts(ip, inode->i_ctime);
303 	to->di_nlink = cpu_to_be32(inode->i_nlink);
304 	to->di_gen = cpu_to_be32(inode->i_generation);
305 	to->di_mode = cpu_to_be16(inode->i_mode);
306 
307 	to->di_size = cpu_to_be64(ip->i_disk_size);
308 	to->di_nblocks = cpu_to_be64(ip->i_nblocks);
309 	to->di_extsize = cpu_to_be32(ip->i_extsize);
310 	to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
311 	to->di_anextents = cpu_to_be16(xfs_ifork_nextents(ip->i_afp));
312 	to->di_forkoff = ip->i_forkoff;
313 	to->di_aformat = xfs_ifork_format(ip->i_afp);
314 	to->di_flags = cpu_to_be16(ip->i_diflags);
315 
316 	if (xfs_has_v3inodes(ip->i_mount)) {
317 		to->di_version = 3;
318 		to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
319 		to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
320 		to->di_flags2 = cpu_to_be64(ip->i_diflags2);
321 		to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
322 		to->di_ino = cpu_to_be64(ip->i_ino);
323 		to->di_lsn = cpu_to_be64(lsn);
324 		memset(to->di_pad2, 0, sizeof(to->di_pad2));
325 		uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
326 		to->di_flushiter = 0;
327 	} else {
328 		to->di_version = 2;
329 		to->di_flushiter = cpu_to_be16(ip->i_flushiter);
330 	}
331 }
332 
333 static xfs_failaddr_t
334 xfs_dinode_verify_fork(
335 	struct xfs_dinode	*dip,
336 	struct xfs_mount	*mp,
337 	int			whichfork)
338 {
339 	uint32_t		di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
340 
341 	switch (XFS_DFORK_FORMAT(dip, whichfork)) {
342 	case XFS_DINODE_FMT_LOCAL:
343 		/*
344 		 * no local regular files yet
345 		 */
346 		if (whichfork == XFS_DATA_FORK) {
347 			if (S_ISREG(be16_to_cpu(dip->di_mode)))
348 				return __this_address;
349 			if (be64_to_cpu(dip->di_size) >
350 					XFS_DFORK_SIZE(dip, mp, whichfork))
351 				return __this_address;
352 		}
353 		if (di_nextents)
354 			return __this_address;
355 		break;
356 	case XFS_DINODE_FMT_EXTENTS:
357 		if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
358 			return __this_address;
359 		break;
360 	case XFS_DINODE_FMT_BTREE:
361 		if (whichfork == XFS_ATTR_FORK) {
362 			if (di_nextents > MAXAEXTNUM)
363 				return __this_address;
364 		} else if (di_nextents > MAXEXTNUM) {
365 			return __this_address;
366 		}
367 		break;
368 	default:
369 		return __this_address;
370 	}
371 	return NULL;
372 }
373 
374 static xfs_failaddr_t
375 xfs_dinode_verify_forkoff(
376 	struct xfs_dinode	*dip,
377 	struct xfs_mount	*mp)
378 {
379 	if (!dip->di_forkoff)
380 		return NULL;
381 
382 	switch (dip->di_format)  {
383 	case XFS_DINODE_FMT_DEV:
384 		if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
385 			return __this_address;
386 		break;
387 	case XFS_DINODE_FMT_LOCAL:	/* fall through ... */
388 	case XFS_DINODE_FMT_EXTENTS:    /* fall through ... */
389 	case XFS_DINODE_FMT_BTREE:
390 		if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
391 			return __this_address;
392 		break;
393 	default:
394 		return __this_address;
395 	}
396 	return NULL;
397 }
398 
399 xfs_failaddr_t
400 xfs_dinode_verify(
401 	struct xfs_mount	*mp,
402 	xfs_ino_t		ino,
403 	struct xfs_dinode	*dip)
404 {
405 	xfs_failaddr_t		fa;
406 	uint16_t		mode;
407 	uint16_t		flags;
408 	uint64_t		flags2;
409 	uint64_t		di_size;
410 
411 	if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
412 		return __this_address;
413 
414 	/* Verify v3 integrity information first */
415 	if (dip->di_version >= 3) {
416 		if (!xfs_has_v3inodes(mp))
417 			return __this_address;
418 		if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
419 				      XFS_DINODE_CRC_OFF))
420 			return __this_address;
421 		if (be64_to_cpu(dip->di_ino) != ino)
422 			return __this_address;
423 		if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
424 			return __this_address;
425 	}
426 
427 	/* don't allow invalid i_size */
428 	di_size = be64_to_cpu(dip->di_size);
429 	if (di_size & (1ULL << 63))
430 		return __this_address;
431 
432 	mode = be16_to_cpu(dip->di_mode);
433 	if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
434 		return __this_address;
435 
436 	/* No zero-length symlinks/dirs. */
437 	if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
438 		return __this_address;
439 
440 	/* Fork checks carried over from xfs_iformat_fork */
441 	if (mode &&
442 	    be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
443 			be64_to_cpu(dip->di_nblocks))
444 		return __this_address;
445 
446 	if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
447 		return __this_address;
448 
449 	flags = be16_to_cpu(dip->di_flags);
450 
451 	if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
452 		return __this_address;
453 
454 	/* check for illegal values of forkoff */
455 	fa = xfs_dinode_verify_forkoff(dip, mp);
456 	if (fa)
457 		return fa;
458 
459 	/* Do we have appropriate data fork formats for the mode? */
460 	switch (mode & S_IFMT) {
461 	case S_IFIFO:
462 	case S_IFCHR:
463 	case S_IFBLK:
464 	case S_IFSOCK:
465 		if (dip->di_format != XFS_DINODE_FMT_DEV)
466 			return __this_address;
467 		break;
468 	case S_IFREG:
469 	case S_IFLNK:
470 	case S_IFDIR:
471 		fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
472 		if (fa)
473 			return fa;
474 		break;
475 	case 0:
476 		/* Uninitialized inode ok. */
477 		break;
478 	default:
479 		return __this_address;
480 	}
481 
482 	if (dip->di_forkoff) {
483 		fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
484 		if (fa)
485 			return fa;
486 	} else {
487 		/*
488 		 * If there is no fork offset, this may be a freshly-made inode
489 		 * in a new disk cluster, in which case di_aformat is zeroed.
490 		 * Otherwise, such an inode must be in EXTENTS format; this goes
491 		 * for freed inodes as well.
492 		 */
493 		switch (dip->di_aformat) {
494 		case 0:
495 		case XFS_DINODE_FMT_EXTENTS:
496 			break;
497 		default:
498 			return __this_address;
499 		}
500 		if (dip->di_anextents)
501 			return __this_address;
502 	}
503 
504 	/* extent size hint validation */
505 	fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
506 			mode, flags);
507 	if (fa)
508 		return fa;
509 
510 	/* only version 3 or greater inodes are extensively verified here */
511 	if (dip->di_version < 3)
512 		return NULL;
513 
514 	flags2 = be64_to_cpu(dip->di_flags2);
515 
516 	/* don't allow reflink/cowextsize if we don't have reflink */
517 	if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
518 	     !xfs_has_reflink(mp))
519 		return __this_address;
520 
521 	/* only regular files get reflink */
522 	if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
523 		return __this_address;
524 
525 	/* don't let reflink and realtime mix */
526 	if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
527 		return __this_address;
528 
529 	/* COW extent size hint validation */
530 	fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
531 			mode, flags, flags2);
532 	if (fa)
533 		return fa;
534 
535 	/* bigtime iflag can only happen on bigtime filesystems */
536 	if (xfs_dinode_has_bigtime(dip) &&
537 	    !xfs_has_bigtime(mp))
538 		return __this_address;
539 
540 	return NULL;
541 }
542 
543 void
544 xfs_dinode_calc_crc(
545 	struct xfs_mount	*mp,
546 	struct xfs_dinode	*dip)
547 {
548 	uint32_t		crc;
549 
550 	if (dip->di_version < 3)
551 		return;
552 
553 	ASSERT(xfs_has_crc(mp));
554 	crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
555 			      XFS_DINODE_CRC_OFF);
556 	dip->di_crc = xfs_end_cksum(crc);
557 }
558 
559 /*
560  * Validate di_extsize hint.
561  *
562  * 1. Extent size hint is only valid for directories and regular files.
563  * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
564  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
565  * 4. Hint cannot be larger than MAXTEXTLEN.
566  * 5. Can be changed on directories at any time.
567  * 6. Hint value of 0 turns off hints, clears inode flags.
568  * 7. Extent size must be a multiple of the appropriate block size.
569  *    For realtime files, this is the rt extent size.
570  * 8. For non-realtime files, the extent size hint must be limited
571  *    to half the AG size to avoid alignment extending the extent beyond the
572  *    limits of the AG.
573  */
574 xfs_failaddr_t
575 xfs_inode_validate_extsize(
576 	struct xfs_mount		*mp,
577 	uint32_t			extsize,
578 	uint16_t			mode,
579 	uint16_t			flags)
580 {
581 	bool				rt_flag;
582 	bool				hint_flag;
583 	bool				inherit_flag;
584 	uint32_t			extsize_bytes;
585 	uint32_t			blocksize_bytes;
586 
587 	rt_flag = (flags & XFS_DIFLAG_REALTIME);
588 	hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
589 	inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
590 	extsize_bytes = XFS_FSB_TO_B(mp, extsize);
591 
592 	/*
593 	 * This comment describes a historic gap in this verifier function.
594 	 *
595 	 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
596 	 * function has never checked that the extent size hint is an integer
597 	 * multiple of the realtime extent size.  Since we allow users to set
598 	 * this combination  on non-rt filesystems /and/ to change the rt
599 	 * extent size when adding a rt device to a filesystem, the net effect
600 	 * is that users can configure a filesystem anticipating one rt
601 	 * geometry and change their minds later.  Directories do not use the
602 	 * extent size hint, so this is harmless for them.
603 	 *
604 	 * If a directory with a misaligned extent size hint is allowed to
605 	 * propagate that hint into a new regular realtime file, the result
606 	 * is that the inode cluster buffer verifier will trigger a corruption
607 	 * shutdown the next time it is run, because the verifier has always
608 	 * enforced the alignment rule for regular files.
609 	 *
610 	 * Because we allow administrators to set a new rt extent size when
611 	 * adding a rt section, we cannot add a check to this verifier because
612 	 * that will result a new source of directory corruption errors when
613 	 * reading an existing filesystem.  Instead, we rely on callers to
614 	 * decide when alignment checks are appropriate, and fix things up as
615 	 * needed.
616 	 */
617 
618 	if (rt_flag)
619 		blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
620 	else
621 		blocksize_bytes = mp->m_sb.sb_blocksize;
622 
623 	if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
624 		return __this_address;
625 
626 	if (hint_flag && !S_ISREG(mode))
627 		return __this_address;
628 
629 	if (inherit_flag && !S_ISDIR(mode))
630 		return __this_address;
631 
632 	if ((hint_flag || inherit_flag) && extsize == 0)
633 		return __this_address;
634 
635 	/* free inodes get flags set to zero but extsize remains */
636 	if (mode && !(hint_flag || inherit_flag) && extsize != 0)
637 		return __this_address;
638 
639 	if (extsize_bytes % blocksize_bytes)
640 		return __this_address;
641 
642 	if (extsize > MAXEXTLEN)
643 		return __this_address;
644 
645 	if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
646 		return __this_address;
647 
648 	return NULL;
649 }
650 
651 /*
652  * Validate di_cowextsize hint.
653  *
654  * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
655  *    The inode does not have to have any shared blocks, but it must be a v3.
656  * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
657  *    for a directory, the hint is propagated to new files.
658  * 3. Can be changed on files & directories at any time.
659  * 4. Hint value of 0 turns off hints, clears inode flags.
660  * 5. Extent size must be a multiple of the appropriate block size.
661  * 6. The extent size hint must be limited to half the AG size to avoid
662  *    alignment extending the extent beyond the limits of the AG.
663  */
664 xfs_failaddr_t
665 xfs_inode_validate_cowextsize(
666 	struct xfs_mount		*mp,
667 	uint32_t			cowextsize,
668 	uint16_t			mode,
669 	uint16_t			flags,
670 	uint64_t			flags2)
671 {
672 	bool				rt_flag;
673 	bool				hint_flag;
674 	uint32_t			cowextsize_bytes;
675 
676 	rt_flag = (flags & XFS_DIFLAG_REALTIME);
677 	hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
678 	cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
679 
680 	if (hint_flag && !xfs_has_reflink(mp))
681 		return __this_address;
682 
683 	if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
684 		return __this_address;
685 
686 	if (hint_flag && cowextsize == 0)
687 		return __this_address;
688 
689 	/* free inodes get flags set to zero but cowextsize remains */
690 	if (mode && !hint_flag && cowextsize != 0)
691 		return __this_address;
692 
693 	if (hint_flag && rt_flag)
694 		return __this_address;
695 
696 	if (cowextsize_bytes % mp->m_sb.sb_blocksize)
697 		return __this_address;
698 
699 	if (cowextsize > MAXEXTLEN)
700 		return __this_address;
701 
702 	if (cowextsize > mp->m_sb.sb_agblocks / 2)
703 		return __this_address;
704 
705 	return NULL;
706 }
707