xfs_inode_buf.c

  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_ag.h"
 14#include "xfs_inode.h"
 15#include "xfs_errortag.h"
 16#include "xfs_error.h"
 17#include "xfs_icache.h"
 18#include "xfs_trans.h"
 19#include "xfs_ialloc.h"
 20#include "xfs_dir2.h"
 21#include "xfs_health.h"
 22#include "xfs_metafile.h"
 23
 24#include <linux/iversion.h>
 25
 26/*
 27 * If we are doing readahead on an inode buffer, we might be in log recovery
 28 * reading an inode allocation buffer that hasn't yet been replayed, and hence
 29 * has not had the inode cores stamped into it. Hence for readahead, the buffer
 30 * may be potentially invalid.
 31 *
 32 * If the readahead buffer is invalid, we need to mark it with an error and
 33 * clear the DONE status of the buffer so that a followup read will re-read it
 34 * from disk. We don't report the error otherwise to avoid warnings during log
 35 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
 36 * because all we want to do is say readahead failed; there is no-one to report
 37 * the error to, so this will distinguish it from a non-ra verifier failure.
 38 * Changes to this readahead error behaviour also need to be reflected in
 39 * xfs_dquot_buf_readahead_verify().
 40 */
 41static void
 42xfs_inode_buf_verify(
 43	struct xfs_buf	*bp,
 44	bool		readahead)
 45{
 46	struct xfs_mount *mp = bp->b_mount;
 47	int		i;
 48	int		ni;
 49
 50	/*
 51	 * Validate the magic number and version of every inode in the buffer
 52	 */
 53	ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
 54	for (i = 0; i < ni; i++) {
 55		struct xfs_dinode	*dip;
 56		xfs_agino_t		unlinked_ino;
 57		int			di_ok;
 58
 59		dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
 60		unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
 61		di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
 62			xfs_dinode_good_version(mp, dip->di_version) &&
 63			xfs_verify_agino_or_null(bp->b_pag, unlinked_ino);
 64		if (unlikely(!di_ok ||
 65				XFS_TEST_ERROR(mp, XFS_ERRTAG_ITOBP_INOTOBP))) {
 66			if (readahead) {
 67				bp->b_flags &= ~XBF_DONE;
 68				xfs_buf_ioerror(bp, -EIO);
 69				return;
 70			}
 71
 72#ifdef DEBUG
 73			xfs_alert(mp,
 74				"bad inode magic/vsn daddr %lld #%d (magic=%x)",
 75				(unsigned long long)xfs_buf_daddr(bp), i,
 76				be16_to_cpu(dip->di_magic));
 77#endif
 78			xfs_buf_verifier_error(bp, -EFSCORRUPTED,
 79					__func__, dip, sizeof(*dip),
 80					NULL);
 81			return;
 82		}
 83	}
 84}
 85
 86
 87static void
 88xfs_inode_buf_read_verify(
 89	struct xfs_buf	*bp)
 90{
 91	xfs_inode_buf_verify(bp, false);
 92}
 93
 94static void
 95xfs_inode_buf_readahead_verify(
 96	struct xfs_buf	*bp)
 97{
 98	xfs_inode_buf_verify(bp, true);
 99}
100
101static void
102xfs_inode_buf_write_verify(
103	struct xfs_buf	*bp)
104{
105	xfs_inode_buf_verify(bp, false);
106}
107
108const struct xfs_buf_ops xfs_inode_buf_ops = {
109	.name = "xfs_inode",
110	.magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
111		     cpu_to_be16(XFS_DINODE_MAGIC) },
112	.verify_read = xfs_inode_buf_read_verify,
113	.verify_write = xfs_inode_buf_write_verify,
114};
115
116const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
117	.name = "xfs_inode_ra",
118	.magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
119		     cpu_to_be16(XFS_DINODE_MAGIC) },
120	.verify_read = xfs_inode_buf_readahead_verify,
121	.verify_write = xfs_inode_buf_write_verify,
122};
123
124
125/*
126 * This routine is called to map an inode to the buffer containing the on-disk
127 * version of the inode.  It returns a pointer to the buffer containing the
128 * on-disk inode in the bpp parameter.
129 */
130int
131xfs_imap_to_bp(
132	struct xfs_mount	*mp,
133	struct xfs_trans	*tp,
134	struct xfs_imap		*imap,
135	struct xfs_buf		**bpp)
136{
137	int			error;
138
139	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
140			imap->im_len, 0, bpp, &xfs_inode_buf_ops);
141	if (xfs_metadata_is_sick(error))
142		xfs_agno_mark_sick(mp, xfs_daddr_to_agno(mp, imap->im_blkno),
143				XFS_SICK_AG_INODES);
144	return error;
145}
146
147static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
148{
149	struct timespec64	tv;
150	uint32_t		n;
151
152	tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
153	tv.tv_nsec = n;
154
155	return tv;
156}
157
158/* Convert an ondisk timestamp to an incore timestamp. */
159struct timespec64
160xfs_inode_from_disk_ts(
161	struct xfs_dinode		*dip,
162	const xfs_timestamp_t		ts)
163{
164	struct timespec64		tv;
165	struct xfs_legacy_timestamp	*lts;
166
167	if (xfs_dinode_has_bigtime(dip))
168		return xfs_inode_decode_bigtime(be64_to_cpu(ts));
169
170	lts = (struct xfs_legacy_timestamp *)&ts;
171	tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
172	tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
173
174	return tv;
175}
176
177int
178xfs_inode_from_disk(
179	struct xfs_inode	*ip,
180	struct xfs_dinode	*from)
181{
182	struct inode		*inode = VFS_I(ip);
183	int			error;
184	xfs_failaddr_t		fa;
185
186	ASSERT(ip->i_cowfp == NULL);
187
188	fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
189	if (fa) {
190		xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
191				sizeof(*from), fa);
192		return -EFSCORRUPTED;
193	}
194
195	/*
196	 * First get the permanent information that is needed to allocate an
197	 * inode. If the inode is unused, mode is zero and we shouldn't mess
198	 * with the uninitialized part of it.
199	 */
200	if (!xfs_has_v3inodes(ip->i_mount))
201		ip->i_flushiter = be16_to_cpu(from->di_flushiter);
202	inode->i_generation = be32_to_cpu(from->di_gen);
203	inode->i_mode = be16_to_cpu(from->di_mode);
204	if (!inode->i_mode)
205		return 0;
206
207	/*
208	 * Convert v1 inodes immediately to v2 inode format as this is the
209	 * minimum inode version format we support in the rest of the code.
210	 * They will also be unconditionally written back to disk as v2 inodes.
211	 */
212	if (unlikely(from->di_version == 1)) {
213		/* di_metatype used to be di_onlink */
214		set_nlink(inode, be16_to_cpu(from->di_metatype));
215		ip->i_projid = 0;
216	} else {
217		set_nlink(inode, be32_to_cpu(from->di_nlink));
218		ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
219					be16_to_cpu(from->di_projid_lo);
220		if (xfs_dinode_is_metadir(from))
221			ip->i_metatype = be16_to_cpu(from->di_metatype);
222	}
223
224	i_uid_write(inode, be32_to_cpu(from->di_uid));
225	i_gid_write(inode, be32_to_cpu(from->di_gid));
226
227	/*
228	 * Time is signed, so need to convert to signed 32 bit before
229	 * storing in inode timestamp which may be 64 bit. Otherwise
230	 * a time before epoch is converted to a time long after epoch
231	 * on 64 bit systems.
232	 */
233	inode_set_atime_to_ts(inode,
234			      xfs_inode_from_disk_ts(from, from->di_atime));
235	inode_set_mtime_to_ts(inode,
236			      xfs_inode_from_disk_ts(from, from->di_mtime));
237	inode_set_ctime_to_ts(inode,
238			      xfs_inode_from_disk_ts(from, from->di_ctime));
239
240	ip->i_disk_size = be64_to_cpu(from->di_size);
241	ip->i_nblocks = be64_to_cpu(from->di_nblocks);
242	ip->i_extsize = be32_to_cpu(from->di_extsize);
243	ip->i_forkoff = from->di_forkoff;
244	ip->i_diflags = be16_to_cpu(from->di_flags);
245	ip->i_next_unlinked = be32_to_cpu(from->di_next_unlinked);
246
247	if (from->di_dmevmask || from->di_dmstate)
248		xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
249
250	if (xfs_has_v3inodes(ip->i_mount)) {
251		inode_set_iversion_queried(inode,
252					   be64_to_cpu(from->di_changecount));
253		ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
254		ip->i_diflags2 = be64_to_cpu(from->di_flags2);
255		/* also covers the di_used_blocks union arm: */
256		ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
257		BUILD_BUG_ON(sizeof(from->di_cowextsize) !=
258			     sizeof(from->di_used_blocks));
259	}
260
261	error = xfs_iformat_data_fork(ip, from);
262	if (error)
263		return error;
264	if (from->di_forkoff) {
265		error = xfs_iformat_attr_fork(ip, from);
266		if (error)
267			goto out_destroy_data_fork;
268	}
269	if (xfs_is_reflink_inode(ip))
270		xfs_ifork_init_cow(ip);
271	return 0;
272
273out_destroy_data_fork:
274	xfs_idestroy_fork(&ip->i_df);
275	return error;
276}
277
278/* Convert an incore timestamp to an ondisk timestamp. */
279static inline xfs_timestamp_t
280xfs_inode_to_disk_ts(
281	struct xfs_inode		*ip,
282	const struct timespec64		tv)
283{
284	struct xfs_legacy_timestamp	*lts;
285	xfs_timestamp_t			ts;
286
287	if (xfs_inode_has_bigtime(ip))
288		return cpu_to_be64(xfs_inode_encode_bigtime(tv));
289
290	lts = (struct xfs_legacy_timestamp *)&ts;
291	lts->t_sec = cpu_to_be32(tv.tv_sec);
292	lts->t_nsec = cpu_to_be32(tv.tv_nsec);
293
294	return ts;
295}
296
297static inline void
298xfs_inode_to_disk_iext_counters(
299	struct xfs_inode	*ip,
300	struct xfs_dinode	*to)
301{
302	if (xfs_inode_has_large_extent_counts(ip)) {
303		to->di_big_nextents = cpu_to_be64(xfs_ifork_nextents(&ip->i_df));
304		to->di_big_anextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_af));
305		/*
306		 * We might be upgrading the inode to use larger extent counters
307		 * than was previously used. Hence zero the unused field.
308		 */
309		to->di_nrext64_pad = cpu_to_be16(0);
310	} else {
311		to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
312		to->di_anextents = cpu_to_be16(xfs_ifork_nextents(&ip->i_af));
313	}
314}
315
316void
317xfs_inode_to_disk(
318	struct xfs_inode	*ip,
319	struct xfs_dinode	*to,
320	xfs_lsn_t		lsn)
321{
322	struct inode		*inode = VFS_I(ip);
323
324	to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
325	if (xfs_is_metadir_inode(ip))
326		to->di_metatype = cpu_to_be16(ip->i_metatype);
327	else
328		to->di_metatype = 0;
329
330	to->di_format = xfs_ifork_format(&ip->i_df);
331	to->di_uid = cpu_to_be32(i_uid_read(inode));
332	to->di_gid = cpu_to_be32(i_gid_read(inode));
333	to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
334	to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
335
336	to->di_atime = xfs_inode_to_disk_ts(ip, inode_get_atime(inode));
337	to->di_mtime = xfs_inode_to_disk_ts(ip, inode_get_mtime(inode));
338	to->di_ctime = xfs_inode_to_disk_ts(ip, inode_get_ctime(inode));
339	to->di_nlink = cpu_to_be32(inode->i_nlink);
340	to->di_gen = cpu_to_be32(inode->i_generation);
341	to->di_mode = cpu_to_be16(inode->i_mode);
342
343	to->di_size = cpu_to_be64(ip->i_disk_size);
344	to->di_nblocks = cpu_to_be64(ip->i_nblocks);
345	to->di_extsize = cpu_to_be32(ip->i_extsize);
346	to->di_forkoff = ip->i_forkoff;
347	to->di_aformat = xfs_ifork_format(&ip->i_af);
348	to->di_flags = cpu_to_be16(ip->i_diflags);
349
350	if (xfs_has_v3inodes(ip->i_mount)) {
351		to->di_version = 3;
352		to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
353		to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
354		to->di_flags2 = cpu_to_be64(ip->i_diflags2);
355		/* also covers the di_used_blocks union arm: */
356		to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
357		to->di_ino = cpu_to_be64(ip->i_ino);
358		to->di_lsn = cpu_to_be64(lsn);
359		memset(to->di_pad2, 0, sizeof(to->di_pad2));
360		uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
361		to->di_v3_pad = 0;
362	} else {
363		to->di_version = 2;
364		to->di_flushiter = cpu_to_be16(ip->i_flushiter);
365		memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
366	}
367
368	xfs_inode_to_disk_iext_counters(ip, to);
369}
370
371static xfs_failaddr_t
372xfs_dinode_verify_fork(
373	struct xfs_dinode	*dip,
374	struct xfs_mount	*mp,
375	int			whichfork)
376{
377	xfs_extnum_t		di_nextents;
378	xfs_extnum_t		max_extents;
379	mode_t			mode = be16_to_cpu(dip->di_mode);
380	uint32_t		fork_size = XFS_DFORK_SIZE(dip, mp, whichfork);
381	uint32_t		fork_format = XFS_DFORK_FORMAT(dip, whichfork);
382
383	di_nextents = xfs_dfork_nextents(dip, whichfork);
384
385	/*
386	 * For fork types that can contain local data, check that the fork
387	 * format matches the size of local data contained within the fork.
388	 */
389	if (whichfork == XFS_DATA_FORK) {
390		/*
391		 * A directory small enough to fit in the inode must be stored
392		 * in local format.  The directory sf <-> extents conversion
393		 * code updates the directory size accordingly.  Directories
394		 * being truncated have zero size and are not subject to this
395		 * check.
396		 */
397		if (S_ISDIR(mode)) {
398			if (dip->di_size &&
399			    be64_to_cpu(dip->di_size) <= fork_size &&
400			    fork_format != XFS_DINODE_FMT_LOCAL)
401				return __this_address;
402		}
403
404		/*
405		 * A symlink with a target small enough to fit in the inode can
406		 * be stored in extents format if xattrs were added (thus
407		 * converting the data fork from shortform to remote format)
408		 * and then removed.
409		 */
410		if (S_ISLNK(mode)) {
411			if (be64_to_cpu(dip->di_size) <= fork_size &&
412			    fork_format != XFS_DINODE_FMT_EXTENTS &&
413			    fork_format != XFS_DINODE_FMT_LOCAL)
414				return __this_address;
415		}
416
417		/*
418		 * For all types, check that when the size says the fork should
419		 * be in extent or btree format, the inode isn't claiming to be
420		 * in local format.
421		 */
422		if (be64_to_cpu(dip->di_size) > fork_size &&
423		    fork_format == XFS_DINODE_FMT_LOCAL)
424			return __this_address;
425	}
426
427	switch (fork_format) {
428	case XFS_DINODE_FMT_LOCAL:
429		/*
430		 * No local regular files yet.
431		 */
432		if (S_ISREG(mode) && whichfork == XFS_DATA_FORK)
433			return __this_address;
434		if (di_nextents)
435			return __this_address;
436		break;
437	case XFS_DINODE_FMT_EXTENTS:
438		if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
439			return __this_address;
440		break;
441	case XFS_DINODE_FMT_BTREE:
442		max_extents = xfs_iext_max_nextents(
443					xfs_dinode_has_large_extent_counts(dip),
444					whichfork);
445		if (di_nextents > max_extents)
446			return __this_address;
447		break;
448	case XFS_DINODE_FMT_META_BTREE:
449		if (!xfs_has_metadir(mp))
450			return __this_address;
451		if (!(dip->di_flags2 & cpu_to_be64(XFS_DIFLAG2_METADATA)))
452			return __this_address;
453		switch (be16_to_cpu(dip->di_metatype)) {
454		case XFS_METAFILE_RTRMAP:
455			/*
456			 * growfs must create the rtrmap inodes before adding a
457			 * realtime volume to the filesystem, so we cannot use
458			 * the rtrmapbt predicate here.
459			 */
460			if (!xfs_has_rmapbt(mp))
461				return __this_address;
462			break;
463		case XFS_METAFILE_RTREFCOUNT:
464			/* same comment about growfs and rmap inodes applies */
465			if (!xfs_has_reflink(mp))
466				return __this_address;
467			break;
468		default:
469			return __this_address;
470		}
471		break;
472	default:
473		return __this_address;
474	}
475	return NULL;
476}
477
478static xfs_failaddr_t
479xfs_dinode_verify_forkoff(
480	struct xfs_dinode	*dip,
481	struct xfs_mount	*mp)
482{
483	if (!dip->di_forkoff)
484		return NULL;
485
486	switch (dip->di_format)  {
487	case XFS_DINODE_FMT_DEV:
488		if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
489			return __this_address;
490		break;
491	case XFS_DINODE_FMT_META_BTREE:
492		if (!xfs_has_metadir(mp) || !xfs_has_parent(mp))
493			return __this_address;
494		fallthrough;
495	case XFS_DINODE_FMT_LOCAL:	/* fall through ... */
496	case XFS_DINODE_FMT_EXTENTS:    /* fall through ... */
497	case XFS_DINODE_FMT_BTREE:
498		if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
499			return __this_address;
500		break;
501	default:
502		return __this_address;
503	}
504	return NULL;
505}
506
507static xfs_failaddr_t
508xfs_dinode_verify_nrext64(
509	struct xfs_mount	*mp,
510	struct xfs_dinode	*dip)
511{
512	if (xfs_dinode_has_large_extent_counts(dip)) {
513		if (!xfs_has_large_extent_counts(mp))
514			return __this_address;
515		if (dip->di_nrext64_pad != 0)
516			return __this_address;
517	} else if (dip->di_version >= 3) {
518		if (dip->di_v3_pad != 0)
519			return __this_address;
520	}
521
522	return NULL;
523}
524
525/*
526 * Validate all the picky requirements we have for a file that claims to be
527 * filesystem metadata.
528 */
529xfs_failaddr_t
530xfs_dinode_verify_metadir(
531	struct xfs_mount	*mp,
532	struct xfs_dinode	*dip,
533	uint16_t		mode,
534	uint16_t		flags,
535	uint64_t		flags2)
536{
537	if (!xfs_has_metadir(mp))
538		return __this_address;
539
540	/* V5 filesystem only */
541	if (dip->di_version < 3)
542		return __this_address;
543
544	if (be16_to_cpu(dip->di_metatype) >= XFS_METAFILE_MAX)
545		return __this_address;
546
547	/* V3 inode fields that are always zero */
548	if ((flags2 & XFS_DIFLAG2_NREXT64) && dip->di_nrext64_pad)
549		return __this_address;
550	if (!(flags2 & XFS_DIFLAG2_NREXT64) && dip->di_flushiter)
551		return __this_address;
552
553	/* Metadata files can only be directories or regular files */
554	if (!S_ISDIR(mode) && !S_ISREG(mode))
555		return __this_address;
556
557	/* They must have zero access permissions */
558	if (mode & 0777)
559		return __this_address;
560
561	/* DMAPI event and state masks are zero */
562	if (dip->di_dmevmask || dip->di_dmstate)
563		return __this_address;
564
565	/*
566	 * User and group IDs must be zero.  The project ID is used for
567	 * grouping inodes.  Metadata inodes are never accounted to quotas.
568	 */
569	if (dip->di_uid || dip->di_gid)
570		return __this_address;
571
572	/* Mandatory inode flags must be set */
573	if (S_ISDIR(mode)) {
574		if ((flags & XFS_METADIR_DIFLAGS) != XFS_METADIR_DIFLAGS)
575			return __this_address;
576	} else {
577		if ((flags & XFS_METAFILE_DIFLAGS) != XFS_METAFILE_DIFLAGS)
578			return __this_address;
579	}
580
581	/* dax flags2 must not be set */
582	if (flags2 & XFS_DIFLAG2_DAX)
583		return __this_address;
584
585	return NULL;
586}
587
588xfs_failaddr_t
589xfs_dinode_verify(
590	struct xfs_mount	*mp,
591	xfs_ino_t		ino,
592	struct xfs_dinode	*dip)
593{
594	xfs_failaddr_t		fa;
595	uint16_t		mode;
596	uint16_t		flags;
597	uint64_t		flags2;
598	uint64_t		di_size;
599	xfs_extnum_t		nextents;
600	xfs_extnum_t		naextents;
601	xfs_filblks_t		nblocks;
602
603	if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
604		return __this_address;
605
606	/* Verify v3 integrity information first */
607	if (dip->di_version >= 3) {
608		if (!xfs_has_v3inodes(mp))
609			return __this_address;
610		if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
611				      XFS_DINODE_CRC_OFF))
612			return __this_address;
613		if (be64_to_cpu(dip->di_ino) != ino)
614			return __this_address;
615		if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
616			return __this_address;
617	}
618
619	/*
620	 * Historical note: xfsprogs in the 3.2 era set up its incore inodes to
621	 * have di_nlink track the link count, even if the actual filesystem
622	 * only supported V1 inodes (i.e. di_onlink).  When writing out the
623	 * ondisk inode, it would set both the ondisk di_nlink and di_onlink to
624	 * the the incore di_nlink value, which is why we cannot check for
625	 * di_nlink==0 on a V1 inode.  V2/3 inodes would get written out with
626	 * di_onlink==0, so we can check that.
627	 */
628	if (dip->di_version == 2) {
629		if (dip->di_metatype)
630			return __this_address;
631	} else if (dip->di_version >= 3) {
632		if (!xfs_dinode_is_metadir(dip) && dip->di_metatype)
633			return __this_address;
634	}
635
636	/* don't allow invalid i_size */
637	di_size = be64_to_cpu(dip->di_size);
638	if (di_size & (1ULL << 63))
639		return __this_address;
640
641	mode = be16_to_cpu(dip->di_mode);
642	if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
643		return __this_address;
644
645	/*
646	 * No zero-length symlinks/dirs unless they're unlinked and hence being
647	 * inactivated.
648	 */
649	if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0) {
650		if (dip->di_version > 1) {
651			if (dip->di_nlink)
652				return __this_address;
653		} else {
654			/* di_metatype used to be di_onlink */
655			if (dip->di_metatype)
656				return __this_address;
657		}
658	}
659
660	fa = xfs_dinode_verify_nrext64(mp, dip);
661	if (fa)
662		return fa;
663
664	nextents = xfs_dfork_data_extents(dip);
665	naextents = xfs_dfork_attr_extents(dip);
666	nblocks = be64_to_cpu(dip->di_nblocks);
667
668	/* Fork checks carried over from xfs_iformat_fork */
669	if (mode && nextents + naextents > nblocks)
670		return __this_address;
671
672	if (S_ISDIR(mode) && nextents > mp->m_dir_geo->max_extents)
673		return __this_address;
674
675	if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
676		return __this_address;
677
678	flags = be16_to_cpu(dip->di_flags);
679
680	if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
681		return __this_address;
682
683	/* check for illegal values of forkoff */
684	fa = xfs_dinode_verify_forkoff(dip, mp);
685	if (fa)
686		return fa;
687
688	/* Do we have appropriate data fork formats for the mode? */
689	switch (mode & S_IFMT) {
690	case S_IFIFO:
691	case S_IFCHR:
692	case S_IFBLK:
693	case S_IFSOCK:
694		if (dip->di_format != XFS_DINODE_FMT_DEV)
695			return __this_address;
696		break;
697	case S_IFREG:
698	case S_IFLNK:
699	case S_IFDIR:
700		fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
701		if (fa)
702			return fa;
703		break;
704	case 0:
705		/* Uninitialized inode ok. */
706		break;
707	default:
708		return __this_address;
709	}
710
711	if (dip->di_forkoff) {
712		fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
713		if (fa)
714			return fa;
715	} else {
716		/*
717		 * If there is no fork offset, this may be a freshly-made inode
718		 * in a new disk cluster, in which case di_aformat is zeroed.
719		 * Otherwise, such an inode must be in EXTENTS format; this goes
720		 * for freed inodes as well.
721		 */
722		switch (dip->di_aformat) {
723		case 0:
724		case XFS_DINODE_FMT_EXTENTS:
725			break;
726		default:
727			return __this_address;
728		}
729		if (naextents)
730			return __this_address;
731	}
732
733	/* extent size hint validation */
734	fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
735			mode, flags);
736	if (fa)
737		return fa;
738
739	/* only version 3 or greater inodes are extensively verified here */
740	if (dip->di_version < 3)
741		return NULL;
742
743	flags2 = be64_to_cpu(dip->di_flags2);
744
745	/* don't allow reflink/cowextsize if we don't have reflink */
746	if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
747	     !xfs_has_reflink(mp))
748		return __this_address;
749
750	/* only regular files get reflink */
751	if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
752		return __this_address;
753
754	/* don't let reflink and realtime mix */
755	if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME) &&
756	    !xfs_has_rtreflink(mp))
757		return __this_address;
758
759	if (xfs_has_zoned(mp) &&
760	    dip->di_metatype == cpu_to_be16(XFS_METAFILE_RTRMAP)) {
761		if (be32_to_cpu(dip->di_used_blocks) > mp->m_sb.sb_rgextents)
762			return __this_address;
763	} else {
764		/* COW extent size hint validation */
765		fa = xfs_inode_validate_cowextsize(mp,
766				be32_to_cpu(dip->di_cowextsize),
767				mode, flags, flags2);
768		if (fa)
769			return fa;
770	}
771
772	/* bigtime iflag can only happen on bigtime filesystems */
773	if (xfs_dinode_has_bigtime(dip) &&
774	    !xfs_has_bigtime(mp))
775		return __this_address;
776
777	if (flags2 & XFS_DIFLAG2_METADATA) {
778		fa = xfs_dinode_verify_metadir(mp, dip, mode, flags, flags2);
779		if (fa)
780			return fa;
781	}
782
783	/* metadata inodes containing btrees always have zero extent count */
784	if (XFS_DFORK_FORMAT(dip, XFS_DATA_FORK) != XFS_DINODE_FMT_META_BTREE) {
785		if (nextents + naextents == 0 && nblocks != 0)
786			return __this_address;
787	}
788
789	return NULL;
790}
791
792void
793xfs_dinode_calc_crc(
794	struct xfs_mount	*mp,
795	struct xfs_dinode	*dip)
796{
797	uint32_t		crc;
798
799	if (dip->di_version < 3)
800		return;
801
802	ASSERT(xfs_has_crc(mp));
803	crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
804			      XFS_DINODE_CRC_OFF);
805	dip->di_crc = xfs_end_cksum(crc);
806}
807
808/*
809 * Validate di_extsize hint.
810 *
811 * 1. Extent size hint is only valid for directories and regular files.
812 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
813 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
814 * 4. Hint cannot be larger than MAXTEXTLEN.
815 * 5. Can be changed on directories at any time.
816 * 6. Hint value of 0 turns off hints, clears inode flags.
817 * 7. Extent size must be a multiple of the appropriate block size.
818 *    For realtime files, this is the rt extent size.
819 * 8. For non-realtime files, the extent size hint must be limited
820 *    to half the AG size to avoid alignment extending the extent beyond the
821 *    limits of the AG.
822 */
823xfs_failaddr_t
824xfs_inode_validate_extsize(
825	struct xfs_mount		*mp,
826	uint32_t			extsize,
827	uint16_t			mode,
828	uint16_t			flags)
829{
830	bool				rt_flag;
831	bool				hint_flag;
832	bool				inherit_flag;
833	uint32_t			extsize_bytes;
834	uint32_t			blocksize_bytes;
835
836	rt_flag = (flags & XFS_DIFLAG_REALTIME);
837	hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
838	inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
839	extsize_bytes = XFS_FSB_TO_B(mp, extsize);
840
841	/*
842	 * This comment describes a historic gap in this verifier function.
843	 *
844	 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
845	 * function has never checked that the extent size hint is an integer
846	 * multiple of the realtime extent size.  Since we allow users to set
847	 * this combination  on non-rt filesystems /and/ to change the rt
848	 * extent size when adding a rt device to a filesystem, the net effect
849	 * is that users can configure a filesystem anticipating one rt
850	 * geometry and change their minds later.  Directories do not use the
851	 * extent size hint, so this is harmless for them.
852	 *
853	 * If a directory with a misaligned extent size hint is allowed to
854	 * propagate that hint into a new regular realtime file, the result
855	 * is that the inode cluster buffer verifier will trigger a corruption
856	 * shutdown the next time it is run, because the verifier has always
857	 * enforced the alignment rule for regular files.
858	 *
859	 * Because we allow administrators to set a new rt extent size when
860	 * adding a rt section, we cannot add a check to this verifier because
861	 * that will result a new source of directory corruption errors when
862	 * reading an existing filesystem.  Instead, we rely on callers to
863	 * decide when alignment checks are appropriate, and fix things up as
864	 * needed.
865	 */
866
867	if (rt_flag)
868		blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
869	else
870		blocksize_bytes = mp->m_sb.sb_blocksize;
871
872	if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
873		return __this_address;
874
875	if (hint_flag && !S_ISREG(mode))
876		return __this_address;
877
878	if (inherit_flag && !S_ISDIR(mode))
879		return __this_address;
880
881	if ((hint_flag || inherit_flag) && extsize == 0)
882		return __this_address;
883
884	/* free inodes get flags set to zero but extsize remains */
885	if (mode && !(hint_flag || inherit_flag) && extsize != 0)
886		return __this_address;
887
888	if (extsize_bytes % blocksize_bytes)
889		return __this_address;
890
891	if (extsize > XFS_MAX_BMBT_EXTLEN)
892		return __this_address;
893
894	if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
895		return __this_address;
896
897	return NULL;
898}
899
900/*
901 * Validate di_cowextsize hint.
902 *
903 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
904 *    The inode does not have to have any shared blocks, but it must be a v3.
905 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
906 *    for a directory, the hint is propagated to new files.
907 * 3. Can be changed on files & directories at any time.
908 * 4. Hint value of 0 turns off hints, clears inode flags.
909 * 5. Extent size must be a multiple of the appropriate block size.
910 * 6. The extent size hint must be limited to half the AG size to avoid
911 *    alignment extending the extent beyond the limits of the AG.
912 */
913xfs_failaddr_t
914xfs_inode_validate_cowextsize(
915	struct xfs_mount		*mp,
916	uint32_t			cowextsize,
917	uint16_t			mode,
918	uint16_t			flags,
919	uint64_t			flags2)
920{
921	bool				rt_flag;
922	bool				hint_flag;
923	uint32_t			cowextsize_bytes;
924	uint32_t			blocksize_bytes;
925
926	rt_flag = (flags & XFS_DIFLAG_REALTIME);
927	hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
928	cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
929
930	/*
931	 * Similar to extent size hints, a directory can be configured to
932	 * propagate realtime status and a CoW extent size hint to newly
933	 * created files even if there is no realtime device, and the hints on
934	 * disk can become misaligned if the sysadmin changes the rt extent
935	 * size while adding the realtime device.
936	 *
937	 * Therefore, we can only enforce the rextsize alignment check against
938	 * regular realtime files, and rely on callers to decide when alignment
939	 * checks are appropriate, and fix things up as needed.
940	 */
941
942	if (rt_flag)
943		blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
944	else
945		blocksize_bytes = mp->m_sb.sb_blocksize;
946
947	if (hint_flag && !xfs_has_reflink(mp))
948		return __this_address;
949
950	if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
951		return __this_address;
952
953	if (hint_flag && cowextsize == 0)
954		return __this_address;
955
956	/* free inodes get flags set to zero but cowextsize remains */
957	if (mode && !hint_flag && cowextsize != 0)
958		return __this_address;
959
960	if (cowextsize_bytes % blocksize_bytes)
961		return __this_address;
962
963	if (cowextsize > XFS_MAX_BMBT_EXTLEN)
964		return __this_address;
965
966	if (!rt_flag && cowextsize > mp->m_sb.sb_agblocks / 2)
967		return __this_address;
968
969	return NULL;
970}