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(XFS_TEST_ERROR(!di_ok, mp,
 65						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, XBF_UNMAPPED, 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		ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
256	}
257
258	error = xfs_iformat_data_fork(ip, from);
259	if (error)
260		return error;
261	if (from->di_forkoff) {
262		error = xfs_iformat_attr_fork(ip, from);
263		if (error)
264			goto out_destroy_data_fork;
265	}
266	if (xfs_is_reflink_inode(ip))
267		xfs_ifork_init_cow(ip);
268	return 0;
269
270out_destroy_data_fork:
271	xfs_idestroy_fork(&ip->i_df);
272	return error;
273}
274
275/* Convert an incore timestamp to an ondisk timestamp. */
276static inline xfs_timestamp_t
277xfs_inode_to_disk_ts(
278	struct xfs_inode		*ip,
279	const struct timespec64		tv)
280{
281	struct xfs_legacy_timestamp	*lts;
282	xfs_timestamp_t			ts;
283
284	if (xfs_inode_has_bigtime(ip))
285		return cpu_to_be64(xfs_inode_encode_bigtime(tv));
286
287	lts = (struct xfs_legacy_timestamp *)&ts;
288	lts->t_sec = cpu_to_be32(tv.tv_sec);
289	lts->t_nsec = cpu_to_be32(tv.tv_nsec);
290
291	return ts;
292}
293
294static inline void
295xfs_inode_to_disk_iext_counters(
296	struct xfs_inode	*ip,
297	struct xfs_dinode	*to)
298{
299	if (xfs_inode_has_large_extent_counts(ip)) {
300		to->di_big_nextents = cpu_to_be64(xfs_ifork_nextents(&ip->i_df));
301		to->di_big_anextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_af));
302		/*
303		 * We might be upgrading the inode to use larger extent counters
304		 * than was previously used. Hence zero the unused field.
305		 */
306		to->di_nrext64_pad = cpu_to_be16(0);
307	} else {
308		to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
309		to->di_anextents = cpu_to_be16(xfs_ifork_nextents(&ip->i_af));
310	}
311}
312
313void
314xfs_inode_to_disk(
315	struct xfs_inode	*ip,
316	struct xfs_dinode	*to,
317	xfs_lsn_t		lsn)
318{
319	struct inode		*inode = VFS_I(ip);
320
321	to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
322	if (xfs_is_metadir_inode(ip))
323		to->di_metatype = cpu_to_be16(ip->i_metatype);
324	else
325		to->di_metatype = 0;
326
327	to->di_format = xfs_ifork_format(&ip->i_df);
328	to->di_uid = cpu_to_be32(i_uid_read(inode));
329	to->di_gid = cpu_to_be32(i_gid_read(inode));
330	to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
331	to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
332
333	to->di_atime = xfs_inode_to_disk_ts(ip, inode_get_atime(inode));
334	to->di_mtime = xfs_inode_to_disk_ts(ip, inode_get_mtime(inode));
335	to->di_ctime = xfs_inode_to_disk_ts(ip, inode_get_ctime(inode));
336	to->di_nlink = cpu_to_be32(inode->i_nlink);
337	to->di_gen = cpu_to_be32(inode->i_generation);
338	to->di_mode = cpu_to_be16(inode->i_mode);
339
340	to->di_size = cpu_to_be64(ip->i_disk_size);
341	to->di_nblocks = cpu_to_be64(ip->i_nblocks);
342	to->di_extsize = cpu_to_be32(ip->i_extsize);
343	to->di_forkoff = ip->i_forkoff;
344	to->di_aformat = xfs_ifork_format(&ip->i_af);
345	to->di_flags = cpu_to_be16(ip->i_diflags);
346
347	if (xfs_has_v3inodes(ip->i_mount)) {
348		to->di_version = 3;
349		to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
350		to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
351		to->di_flags2 = cpu_to_be64(ip->i_diflags2);
352		to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
353		to->di_ino = cpu_to_be64(ip->i_ino);
354		to->di_lsn = cpu_to_be64(lsn);
355		memset(to->di_pad2, 0, sizeof(to->di_pad2));
356		uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
357		to->di_v3_pad = 0;
358	} else {
359		to->di_version = 2;
360		to->di_flushiter = cpu_to_be16(ip->i_flushiter);
361		memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
362	}
363
364	xfs_inode_to_disk_iext_counters(ip, to);
365}
366
367static xfs_failaddr_t
368xfs_dinode_verify_fork(
369	struct xfs_dinode	*dip,
370	struct xfs_mount	*mp,
371	int			whichfork)
372{
373	xfs_extnum_t		di_nextents;
374	xfs_extnum_t		max_extents;
375	mode_t			mode = be16_to_cpu(dip->di_mode);
376	uint32_t		fork_size = XFS_DFORK_SIZE(dip, mp, whichfork);
377	uint32_t		fork_format = XFS_DFORK_FORMAT(dip, whichfork);
378
379	di_nextents = xfs_dfork_nextents(dip, whichfork);
380
381	/*
382	 * For fork types that can contain local data, check that the fork
383	 * format matches the size of local data contained within the fork.
 
 
 
384	 */
385	if (whichfork == XFS_DATA_FORK) {
386		/*
387		 * A directory small enough to fit in the inode must be stored
388		 * in local format.  The directory sf <-> extents conversion
389		 * code updates the directory size accordingly.  Directories
390		 * being truncated have zero size and are not subject to this
391		 * check.
392		 */
393		if (S_ISDIR(mode)) {
394			if (dip->di_size &&
395			    be64_to_cpu(dip->di_size) <= fork_size &&
396			    fork_format != XFS_DINODE_FMT_LOCAL)
397				return __this_address;
398		}
399
400		/*
401		 * A symlink with a target small enough to fit in the inode can
402		 * be stored in extents format if xattrs were added (thus
403		 * converting the data fork from shortform to remote format)
404		 * and then removed.
405		 */
406		if (S_ISLNK(mode)) {
407			if (be64_to_cpu(dip->di_size) <= fork_size &&
408			    fork_format != XFS_DINODE_FMT_EXTENTS &&
409			    fork_format != XFS_DINODE_FMT_LOCAL)
410				return __this_address;
411		}
412
413		/*
414		 * For all types, check that when the size says the fork should
415		 * be in extent or btree format, the inode isn't claiming to be
416		 * in local format.
417		 */
418		if (be64_to_cpu(dip->di_size) > fork_size &&
419		    fork_format == XFS_DINODE_FMT_LOCAL)
420			return __this_address;
421	}
422
423	switch (fork_format) {
424	case XFS_DINODE_FMT_LOCAL:
425		/*
426		 * No local regular files yet.
427		 */
428		if (S_ISREG(mode) && whichfork == XFS_DATA_FORK)
429			return __this_address;
430		if (di_nextents)
431			return __this_address;
432		break;
433	case XFS_DINODE_FMT_EXTENTS:
434		if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
435			return __this_address;
436		break;
437	case XFS_DINODE_FMT_BTREE:
438		max_extents = xfs_iext_max_nextents(
439					xfs_dinode_has_large_extent_counts(dip),
440					whichfork);
441		if (di_nextents > max_extents)
442			return __this_address;
443		break;
444	default:
445		return __this_address;
446	}
447	return NULL;
448}
449
450static xfs_failaddr_t
451xfs_dinode_verify_forkoff(
452	struct xfs_dinode	*dip,
453	struct xfs_mount	*mp)
454{
455	if (!dip->di_forkoff)
456		return NULL;
457
458	switch (dip->di_format)  {
459	case XFS_DINODE_FMT_DEV:
460		if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
461			return __this_address;
462		break;
463	case XFS_DINODE_FMT_LOCAL:	/* fall through ... */
464	case XFS_DINODE_FMT_EXTENTS:    /* fall through ... */
465	case XFS_DINODE_FMT_BTREE:
466		if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
467			return __this_address;
468		break;
469	default:
470		return __this_address;
471	}
472	return NULL;
473}
474
475static xfs_failaddr_t
476xfs_dinode_verify_nrext64(
477	struct xfs_mount	*mp,
478	struct xfs_dinode	*dip)
479{
480	if (xfs_dinode_has_large_extent_counts(dip)) {
481		if (!xfs_has_large_extent_counts(mp))
482			return __this_address;
483		if (dip->di_nrext64_pad != 0)
484			return __this_address;
485	} else if (dip->di_version >= 3) {
486		if (dip->di_v3_pad != 0)
487			return __this_address;
488	}
489
490	return NULL;
491}
492
493/*
494 * Validate all the picky requirements we have for a file that claims to be
495 * filesystem metadata.
496 */
497xfs_failaddr_t
498xfs_dinode_verify_metadir(
499	struct xfs_mount	*mp,
500	struct xfs_dinode	*dip,
501	uint16_t		mode,
502	uint16_t		flags,
503	uint64_t		flags2)
504{
505	if (!xfs_has_metadir(mp))
506		return __this_address;
507
508	/* V5 filesystem only */
509	if (dip->di_version < 3)
510		return __this_address;
511
512	if (be16_to_cpu(dip->di_metatype) >= XFS_METAFILE_MAX)
513		return __this_address;
514
515	/* V3 inode fields that are always zero */
516	if ((flags2 & XFS_DIFLAG2_NREXT64) && dip->di_nrext64_pad)
517		return __this_address;
518	if (!(flags2 & XFS_DIFLAG2_NREXT64) && dip->di_flushiter)
519		return __this_address;
520
521	/* Metadata files can only be directories or regular files */
522	if (!S_ISDIR(mode) && !S_ISREG(mode))
523		return __this_address;
524
525	/* They must have zero access permissions */
526	if (mode & 0777)
527		return __this_address;
528
529	/* DMAPI event and state masks are zero */
530	if (dip->di_dmevmask || dip->di_dmstate)
531		return __this_address;
532
533	/*
534	 * User and group IDs must be zero.  The project ID is used for
535	 * grouping inodes.  Metadata inodes are never accounted to quotas.
536	 */
537	if (dip->di_uid || dip->di_gid)
538		return __this_address;
539
540	/* Mandatory inode flags must be set */
541	if (S_ISDIR(mode)) {
542		if ((flags & XFS_METADIR_DIFLAGS) != XFS_METADIR_DIFLAGS)
543			return __this_address;
544	} else {
545		if ((flags & XFS_METAFILE_DIFLAGS) != XFS_METAFILE_DIFLAGS)
546			return __this_address;
547	}
548
549	/* dax flags2 must not be set */
550	if (flags2 & XFS_DIFLAG2_DAX)
551		return __this_address;
552
553	return NULL;
554}
555
556xfs_failaddr_t
557xfs_dinode_verify(
558	struct xfs_mount	*mp,
559	xfs_ino_t		ino,
560	struct xfs_dinode	*dip)
561{
562	xfs_failaddr_t		fa;
563	uint16_t		mode;
564	uint16_t		flags;
565	uint64_t		flags2;
566	uint64_t		di_size;
567	xfs_extnum_t		nextents;
568	xfs_extnum_t		naextents;
569	xfs_filblks_t		nblocks;
570
571	if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
572		return __this_address;
573
574	/* Verify v3 integrity information first */
575	if (dip->di_version >= 3) {
576		if (!xfs_has_v3inodes(mp))
577			return __this_address;
578		if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
579				      XFS_DINODE_CRC_OFF))
580			return __this_address;
581		if (be64_to_cpu(dip->di_ino) != ino)
582			return __this_address;
583		if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
584			return __this_address;
585	}
586
587	/*
588	 * Historical note: xfsprogs in the 3.2 era set up its incore inodes to
589	 * have di_nlink track the link count, even if the actual filesystem
590	 * only supported V1 inodes (i.e. di_onlink).  When writing out the
591	 * ondisk inode, it would set both the ondisk di_nlink and di_onlink to
592	 * the the incore di_nlink value, which is why we cannot check for
593	 * di_nlink==0 on a V1 inode.  V2/3 inodes would get written out with
594	 * di_onlink==0, so we can check that.
595	 */
596	if (dip->di_version == 2) {
597		if (dip->di_metatype)
598			return __this_address;
599	} else if (dip->di_version >= 3) {
600		if (!xfs_dinode_is_metadir(dip) && dip->di_metatype)
601			return __this_address;
602	}
603
604	/* don't allow invalid i_size */
605	di_size = be64_to_cpu(dip->di_size);
606	if (di_size & (1ULL << 63))
607		return __this_address;
608
609	mode = be16_to_cpu(dip->di_mode);
610	if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
611		return __this_address;
612
613	/*
614	 * No zero-length symlinks/dirs unless they're unlinked and hence being
615	 * inactivated.
616	 */
617	if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0) {
618		if (dip->di_version > 1) {
619			if (dip->di_nlink)
620				return __this_address;
621		} else {
622			/* di_metatype used to be di_onlink */
623			if (dip->di_metatype)
624				return __this_address;
625		}
626	}
627
628	fa = xfs_dinode_verify_nrext64(mp, dip);
629	if (fa)
630		return fa;
631
632	nextents = xfs_dfork_data_extents(dip);
633	naextents = xfs_dfork_attr_extents(dip);
634	nblocks = be64_to_cpu(dip->di_nblocks);
635
636	/* Fork checks carried over from xfs_iformat_fork */
637	if (mode && nextents + naextents > nblocks)
638		return __this_address;
639
640	if (nextents + naextents == 0 && nblocks != 0)
641		return __this_address;
642
643	if (S_ISDIR(mode) && nextents > mp->m_dir_geo->max_extents)
644		return __this_address;
645
646	if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
647		return __this_address;
648
649	flags = be16_to_cpu(dip->di_flags);
650
651	if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
652		return __this_address;
653
654	/* check for illegal values of forkoff */
655	fa = xfs_dinode_verify_forkoff(dip, mp);
656	if (fa)
657		return fa;
658
659	/* Do we have appropriate data fork formats for the mode? */
660	switch (mode & S_IFMT) {
661	case S_IFIFO:
662	case S_IFCHR:
663	case S_IFBLK:
664	case S_IFSOCK:
665		if (dip->di_format != XFS_DINODE_FMT_DEV)
666			return __this_address;
667		break;
668	case S_IFREG:
669	case S_IFLNK:
670	case S_IFDIR:
671		fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
672		if (fa)
673			return fa;
674		break;
675	case 0:
676		/* Uninitialized inode ok. */
677		break;
678	default:
679		return __this_address;
680	}
681
682	if (dip->di_forkoff) {
683		fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
684		if (fa)
685			return fa;
686	} else {
687		/*
688		 * If there is no fork offset, this may be a freshly-made inode
689		 * in a new disk cluster, in which case di_aformat is zeroed.
690		 * Otherwise, such an inode must be in EXTENTS format; this goes
691		 * for freed inodes as well.
692		 */
693		switch (dip->di_aformat) {
694		case 0:
695		case XFS_DINODE_FMT_EXTENTS:
696			break;
697		default:
698			return __this_address;
699		}
700		if (naextents)
701			return __this_address;
702	}
703
704	/* extent size hint validation */
705	fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
706			mode, flags);
707	if (fa)
708		return fa;
709
710	/* only version 3 or greater inodes are extensively verified here */
711	if (dip->di_version < 3)
712		return NULL;
713
714	flags2 = be64_to_cpu(dip->di_flags2);
715
716	/* don't allow reflink/cowextsize if we don't have reflink */
717	if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
718	     !xfs_has_reflink(mp))
719		return __this_address;
720
721	/* only regular files get reflink */
722	if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
723		return __this_address;
724
725	/* don't let reflink and realtime mix */
726	if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
727		return __this_address;
728
729	/* COW extent size hint validation */
730	fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
731			mode, flags, flags2);
732	if (fa)
733		return fa;
734
735	/* bigtime iflag can only happen on bigtime filesystems */
736	if (xfs_dinode_has_bigtime(dip) &&
737	    !xfs_has_bigtime(mp))
738		return __this_address;
739
740	if (flags2 & XFS_DIFLAG2_METADATA) {
741		fa = xfs_dinode_verify_metadir(mp, dip, mode, flags, flags2);
742		if (fa)
743			return fa;
744	}
745
746	return NULL;
747}
748
749void
750xfs_dinode_calc_crc(
751	struct xfs_mount	*mp,
752	struct xfs_dinode	*dip)
753{
754	uint32_t		crc;
755
756	if (dip->di_version < 3)
757		return;
758
759	ASSERT(xfs_has_crc(mp));
760	crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
761			      XFS_DINODE_CRC_OFF);
762	dip->di_crc = xfs_end_cksum(crc);
763}
764
765/*
766 * Validate di_extsize hint.
767 *
768 * 1. Extent size hint is only valid for directories and regular files.
769 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
770 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
771 * 4. Hint cannot be larger than MAXTEXTLEN.
772 * 5. Can be changed on directories at any time.
773 * 6. Hint value of 0 turns off hints, clears inode flags.
774 * 7. Extent size must be a multiple of the appropriate block size.
775 *    For realtime files, this is the rt extent size.
776 * 8. For non-realtime files, the extent size hint must be limited
777 *    to half the AG size to avoid alignment extending the extent beyond the
778 *    limits of the AG.
779 */
780xfs_failaddr_t
781xfs_inode_validate_extsize(
782	struct xfs_mount		*mp,
783	uint32_t			extsize,
784	uint16_t			mode,
785	uint16_t			flags)
786{
787	bool				rt_flag;
788	bool				hint_flag;
789	bool				inherit_flag;
790	uint32_t			extsize_bytes;
791	uint32_t			blocksize_bytes;
792
793	rt_flag = (flags & XFS_DIFLAG_REALTIME);
794	hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
795	inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
796	extsize_bytes = XFS_FSB_TO_B(mp, extsize);
797
798	/*
799	 * This comment describes a historic gap in this verifier function.
800	 *
801	 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
802	 * function has never checked that the extent size hint is an integer
803	 * multiple of the realtime extent size.  Since we allow users to set
804	 * this combination  on non-rt filesystems /and/ to change the rt
805	 * extent size when adding a rt device to a filesystem, the net effect
806	 * is that users can configure a filesystem anticipating one rt
807	 * geometry and change their minds later.  Directories do not use the
808	 * extent size hint, so this is harmless for them.
809	 *
810	 * If a directory with a misaligned extent size hint is allowed to
811	 * propagate that hint into a new regular realtime file, the result
812	 * is that the inode cluster buffer verifier will trigger a corruption
813	 * shutdown the next time it is run, because the verifier has always
814	 * enforced the alignment rule for regular files.
815	 *
816	 * Because we allow administrators to set a new rt extent size when
817	 * adding a rt section, we cannot add a check to this verifier because
818	 * that will result a new source of directory corruption errors when
819	 * reading an existing filesystem.  Instead, we rely on callers to
820	 * decide when alignment checks are appropriate, and fix things up as
821	 * needed.
822	 */
823
824	if (rt_flag)
825		blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
826	else
827		blocksize_bytes = mp->m_sb.sb_blocksize;
828
829	if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
830		return __this_address;
831
832	if (hint_flag && !S_ISREG(mode))
833		return __this_address;
834
835	if (inherit_flag && !S_ISDIR(mode))
836		return __this_address;
837
838	if ((hint_flag || inherit_flag) && extsize == 0)
839		return __this_address;
840
841	/* free inodes get flags set to zero but extsize remains */
842	if (mode && !(hint_flag || inherit_flag) && extsize != 0)
843		return __this_address;
844
845	if (extsize_bytes % blocksize_bytes)
846		return __this_address;
847
848	if (extsize > XFS_MAX_BMBT_EXTLEN)
849		return __this_address;
850
851	if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
852		return __this_address;
853
854	return NULL;
855}
856
857/*
858 * Validate di_cowextsize hint.
859 *
860 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
861 *    The inode does not have to have any shared blocks, but it must be a v3.
862 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
863 *    for a directory, the hint is propagated to new files.
864 * 3. Can be changed on files & directories at any time.
865 * 4. Hint value of 0 turns off hints, clears inode flags.
866 * 5. Extent size must be a multiple of the appropriate block size.
867 * 6. The extent size hint must be limited to half the AG size to avoid
868 *    alignment extending the extent beyond the limits of the AG.
869 */
870xfs_failaddr_t
871xfs_inode_validate_cowextsize(
872	struct xfs_mount		*mp,
873	uint32_t			cowextsize,
874	uint16_t			mode,
875	uint16_t			flags,
876	uint64_t			flags2)
877{
878	bool				rt_flag;
879	bool				hint_flag;
880	uint32_t			cowextsize_bytes;
881
882	rt_flag = (flags & XFS_DIFLAG_REALTIME);
883	hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
884	cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
885
886	if (hint_flag && !xfs_has_reflink(mp))
887		return __this_address;
888
889	if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
890		return __this_address;
891
892	if (hint_flag && cowextsize == 0)
893		return __this_address;
894
895	/* free inodes get flags set to zero but cowextsize remains */
896	if (mode && !hint_flag && cowextsize != 0)
897		return __this_address;
898
899	if (hint_flag && rt_flag)
900		return __this_address;
901
902	if (cowextsize_bytes % mp->m_sb.sb_blocksize)
903		return __this_address;
904
905	if (cowextsize > XFS_MAX_BMBT_EXTLEN)
906		return __this_address;
907
908	if (cowextsize > mp->m_sb.sb_agblocks / 2)
909		return __this_address;
910
911	return NULL;
912}