1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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"
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"
22 #include <linux/iversion.h>
25 * If we are doing readahead on an inode buffer, we might be in log recovery
26 * reading an inode allocation buffer that hasn't yet been replayed, and hence
27 * has not had the inode cores stamped into it. Hence for readahead, the buffer
28 * may be potentially invalid.
30 * If the readahead buffer is invalid, we need to mark it with an error and
31 * clear the DONE status of the buffer so that a followup read will re-read it
32 * from disk. We don't report the error otherwise to avoid warnings during log
33 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
34 * because all we want to do is say readahead failed; there is no-one to report
35 * the error to, so this will distinguish it from a non-ra verifier failure.
36 * Changes to this readahead error behaviour also need to be reflected in
37 * xfs_dquot_buf_readahead_verify().
44 struct xfs_mount *mp = bp->b_mount;
49 * Validate the magic number and version of every inode in the buffer
51 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
52 for (i = 0; i < ni; i++) {
53 struct xfs_dinode *dip;
54 xfs_agino_t unlinked_ino;
57 dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
58 unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
59 di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
60 xfs_dinode_good_version(mp, dip->di_version) &&
61 xfs_verify_agino_or_null(bp->b_pag, unlinked_ino);
62 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
63 XFS_ERRTAG_ITOBP_INOTOBP))) {
65 bp->b_flags &= ~XBF_DONE;
66 xfs_buf_ioerror(bp, -EIO);
72 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
73 (unsigned long long)xfs_buf_daddr(bp), i,
74 be16_to_cpu(dip->di_magic));
76 xfs_buf_verifier_error(bp, -EFSCORRUPTED,
77 __func__, dip, sizeof(*dip),
86 xfs_inode_buf_read_verify(
89 xfs_inode_buf_verify(bp, false);
93 xfs_inode_buf_readahead_verify(
96 xfs_inode_buf_verify(bp, true);
100 xfs_inode_buf_write_verify(
103 xfs_inode_buf_verify(bp, false);
106 const struct xfs_buf_ops xfs_inode_buf_ops = {
108 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
109 cpu_to_be16(XFS_DINODE_MAGIC) },
110 .verify_read = xfs_inode_buf_read_verify,
111 .verify_write = xfs_inode_buf_write_verify,
114 const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
115 .name = "xfs_inode_ra",
116 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
117 cpu_to_be16(XFS_DINODE_MAGIC) },
118 .verify_read = xfs_inode_buf_readahead_verify,
119 .verify_write = xfs_inode_buf_write_verify,
124 * This routine is called to map an inode to the buffer containing the on-disk
125 * version of the inode. It returns a pointer to the buffer containing the
126 * on-disk inode in the bpp parameter.
130 struct xfs_mount *mp,
131 struct xfs_trans *tp,
132 struct xfs_imap *imap,
133 struct xfs_buf **bpp)
135 return xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
136 imap->im_len, XBF_UNMAPPED, bpp,
140 static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
142 struct timespec64 tv;
145 tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
151 /* Convert an ondisk timestamp to an incore timestamp. */
153 xfs_inode_from_disk_ts(
154 struct xfs_dinode *dip,
155 const xfs_timestamp_t ts)
157 struct timespec64 tv;
158 struct xfs_legacy_timestamp *lts;
160 if (xfs_dinode_has_bigtime(dip))
161 return xfs_inode_decode_bigtime(be64_to_cpu(ts));
163 lts = (struct xfs_legacy_timestamp *)&ts;
164 tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
165 tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
172 struct xfs_inode *ip,
173 struct xfs_dinode *from)
175 struct inode *inode = VFS_I(ip);
179 ASSERT(ip->i_cowfp == NULL);
181 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
183 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
185 return -EFSCORRUPTED;
189 * First get the permanent information that is needed to allocate an
190 * inode. If the inode is unused, mode is zero and we shouldn't mess
191 * with the uninitialized part of it.
193 if (!xfs_has_v3inodes(ip->i_mount))
194 ip->i_flushiter = be16_to_cpu(from->di_flushiter);
195 inode->i_generation = be32_to_cpu(from->di_gen);
196 inode->i_mode = be16_to_cpu(from->di_mode);
201 * Convert v1 inodes immediately to v2 inode format as this is the
202 * minimum inode version format we support in the rest of the code.
203 * They will also be unconditionally written back to disk as v2 inodes.
205 if (unlikely(from->di_version == 1)) {
206 set_nlink(inode, be16_to_cpu(from->di_onlink));
209 set_nlink(inode, be32_to_cpu(from->di_nlink));
210 ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
211 be16_to_cpu(from->di_projid_lo);
214 i_uid_write(inode, be32_to_cpu(from->di_uid));
215 i_gid_write(inode, be32_to_cpu(from->di_gid));
218 * Time is signed, so need to convert to signed 32 bit before
219 * storing in inode timestamp which may be 64 bit. Otherwise
220 * a time before epoch is converted to a time long after epoch
223 inode->i_atime = xfs_inode_from_disk_ts(from, from->di_atime);
224 inode->i_mtime = xfs_inode_from_disk_ts(from, from->di_mtime);
225 inode->i_ctime = xfs_inode_from_disk_ts(from, from->di_ctime);
227 ip->i_disk_size = be64_to_cpu(from->di_size);
228 ip->i_nblocks = be64_to_cpu(from->di_nblocks);
229 ip->i_extsize = be32_to_cpu(from->di_extsize);
230 ip->i_forkoff = from->di_forkoff;
231 ip->i_diflags = be16_to_cpu(from->di_flags);
232 ip->i_next_unlinked = be32_to_cpu(from->di_next_unlinked);
234 if (from->di_dmevmask || from->di_dmstate)
235 xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
237 if (xfs_has_v3inodes(ip->i_mount)) {
238 inode_set_iversion_queried(inode,
239 be64_to_cpu(from->di_changecount));
240 ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
241 ip->i_diflags2 = be64_to_cpu(from->di_flags2);
242 ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
245 error = xfs_iformat_data_fork(ip, from);
248 if (from->di_forkoff) {
249 error = xfs_iformat_attr_fork(ip, from);
251 goto out_destroy_data_fork;
253 if (xfs_is_reflink_inode(ip))
254 xfs_ifork_init_cow(ip);
257 out_destroy_data_fork:
258 xfs_idestroy_fork(&ip->i_df);
262 /* Convert an incore timestamp to an ondisk timestamp. */
263 static inline xfs_timestamp_t
264 xfs_inode_to_disk_ts(
265 struct xfs_inode *ip,
266 const struct timespec64 tv)
268 struct xfs_legacy_timestamp *lts;
271 if (xfs_inode_has_bigtime(ip))
272 return cpu_to_be64(xfs_inode_encode_bigtime(tv));
274 lts = (struct xfs_legacy_timestamp *)&ts;
275 lts->t_sec = cpu_to_be32(tv.tv_sec);
276 lts->t_nsec = cpu_to_be32(tv.tv_nsec);
282 xfs_inode_to_disk_iext_counters(
283 struct xfs_inode *ip,
284 struct xfs_dinode *to)
286 if (xfs_inode_has_large_extent_counts(ip)) {
287 to->di_big_nextents = cpu_to_be64(xfs_ifork_nextents(&ip->i_df));
288 to->di_big_anextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_af));
290 * We might be upgrading the inode to use larger extent counters
291 * than was previously used. Hence zero the unused field.
293 to->di_nrext64_pad = cpu_to_be16(0);
295 to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
296 to->di_anextents = cpu_to_be16(xfs_ifork_nextents(&ip->i_af));
302 struct xfs_inode *ip,
303 struct xfs_dinode *to,
306 struct inode *inode = VFS_I(ip);
308 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
311 to->di_format = xfs_ifork_format(&ip->i_df);
312 to->di_uid = cpu_to_be32(i_uid_read(inode));
313 to->di_gid = cpu_to_be32(i_gid_read(inode));
314 to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
315 to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
317 to->di_atime = xfs_inode_to_disk_ts(ip, inode->i_atime);
318 to->di_mtime = xfs_inode_to_disk_ts(ip, inode->i_mtime);
319 to->di_ctime = xfs_inode_to_disk_ts(ip, inode->i_ctime);
320 to->di_nlink = cpu_to_be32(inode->i_nlink);
321 to->di_gen = cpu_to_be32(inode->i_generation);
322 to->di_mode = cpu_to_be16(inode->i_mode);
324 to->di_size = cpu_to_be64(ip->i_disk_size);
325 to->di_nblocks = cpu_to_be64(ip->i_nblocks);
326 to->di_extsize = cpu_to_be32(ip->i_extsize);
327 to->di_forkoff = ip->i_forkoff;
328 to->di_aformat = xfs_ifork_format(&ip->i_af);
329 to->di_flags = cpu_to_be16(ip->i_diflags);
331 if (xfs_has_v3inodes(ip->i_mount)) {
333 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
334 to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
335 to->di_flags2 = cpu_to_be64(ip->i_diflags2);
336 to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
337 to->di_ino = cpu_to_be64(ip->i_ino);
338 to->di_lsn = cpu_to_be64(lsn);
339 memset(to->di_pad2, 0, sizeof(to->di_pad2));
340 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
344 to->di_flushiter = cpu_to_be16(ip->i_flushiter);
345 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
348 xfs_inode_to_disk_iext_counters(ip, to);
351 static xfs_failaddr_t
352 xfs_dinode_verify_fork(
353 struct xfs_dinode *dip,
354 struct xfs_mount *mp,
357 xfs_extnum_t di_nextents;
358 xfs_extnum_t max_extents;
359 mode_t mode = be16_to_cpu(dip->di_mode);
360 uint32_t fork_size = XFS_DFORK_SIZE(dip, mp, whichfork);
361 uint32_t fork_format = XFS_DFORK_FORMAT(dip, whichfork);
363 di_nextents = xfs_dfork_nextents(dip, whichfork);
366 * For fork types that can contain local data, check that the fork
367 * format matches the size of local data contained within the fork.
369 * For all types, check that when the size says the should be in extent
370 * or btree format, the inode isn't claiming it is in local format.
372 if (whichfork == XFS_DATA_FORK) {
373 if (S_ISDIR(mode) || S_ISLNK(mode)) {
374 if (be64_to_cpu(dip->di_size) <= fork_size &&
375 fork_format != XFS_DINODE_FMT_LOCAL)
376 return __this_address;
379 if (be64_to_cpu(dip->di_size) > fork_size &&
380 fork_format == XFS_DINODE_FMT_LOCAL)
381 return __this_address;
384 switch (fork_format) {
385 case XFS_DINODE_FMT_LOCAL:
387 * No local regular files yet.
389 if (S_ISREG(mode) && whichfork == XFS_DATA_FORK)
390 return __this_address;
392 return __this_address;
394 case XFS_DINODE_FMT_EXTENTS:
395 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
396 return __this_address;
398 case XFS_DINODE_FMT_BTREE:
399 max_extents = xfs_iext_max_nextents(
400 xfs_dinode_has_large_extent_counts(dip),
402 if (di_nextents > max_extents)
403 return __this_address;
406 return __this_address;
411 static xfs_failaddr_t
412 xfs_dinode_verify_forkoff(
413 struct xfs_dinode *dip,
414 struct xfs_mount *mp)
416 if (!dip->di_forkoff)
419 switch (dip->di_format) {
420 case XFS_DINODE_FMT_DEV:
421 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
422 return __this_address;
424 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
425 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
426 case XFS_DINODE_FMT_BTREE:
427 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
428 return __this_address;
431 return __this_address;
436 static xfs_failaddr_t
437 xfs_dinode_verify_nrext64(
438 struct xfs_mount *mp,
439 struct xfs_dinode *dip)
441 if (xfs_dinode_has_large_extent_counts(dip)) {
442 if (!xfs_has_large_extent_counts(mp))
443 return __this_address;
444 if (dip->di_nrext64_pad != 0)
445 return __this_address;
446 } else if (dip->di_version >= 3) {
447 if (dip->di_v3_pad != 0)
448 return __this_address;
456 struct xfs_mount *mp,
458 struct xfs_dinode *dip)
465 xfs_extnum_t nextents;
466 xfs_extnum_t naextents;
467 xfs_filblks_t nblocks;
469 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
470 return __this_address;
472 /* Verify v3 integrity information first */
473 if (dip->di_version >= 3) {
474 if (!xfs_has_v3inodes(mp))
475 return __this_address;
476 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
478 return __this_address;
479 if (be64_to_cpu(dip->di_ino) != ino)
480 return __this_address;
481 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
482 return __this_address;
485 /* don't allow invalid i_size */
486 di_size = be64_to_cpu(dip->di_size);
487 if (di_size & (1ULL << 63))
488 return __this_address;
490 mode = be16_to_cpu(dip->di_mode);
491 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
492 return __this_address;
494 /* No zero-length symlinks/dirs. */
495 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
496 return __this_address;
498 fa = xfs_dinode_verify_nrext64(mp, dip);
502 nextents = xfs_dfork_data_extents(dip);
503 naextents = xfs_dfork_attr_extents(dip);
504 nblocks = be64_to_cpu(dip->di_nblocks);
506 /* Fork checks carried over from xfs_iformat_fork */
507 if (mode && nextents + naextents > nblocks)
508 return __this_address;
510 if (S_ISDIR(mode) && nextents > mp->m_dir_geo->max_extents)
511 return __this_address;
513 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
514 return __this_address;
516 flags = be16_to_cpu(dip->di_flags);
518 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
519 return __this_address;
521 /* check for illegal values of forkoff */
522 fa = xfs_dinode_verify_forkoff(dip, mp);
526 /* Do we have appropriate data fork formats for the mode? */
527 switch (mode & S_IFMT) {
532 if (dip->di_format != XFS_DINODE_FMT_DEV)
533 return __this_address;
538 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
543 /* Uninitialized inode ok. */
546 return __this_address;
549 if (dip->di_forkoff) {
550 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
555 * If there is no fork offset, this may be a freshly-made inode
556 * in a new disk cluster, in which case di_aformat is zeroed.
557 * Otherwise, such an inode must be in EXTENTS format; this goes
558 * for freed inodes as well.
560 switch (dip->di_aformat) {
562 case XFS_DINODE_FMT_EXTENTS:
565 return __this_address;
568 return __this_address;
571 /* extent size hint validation */
572 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
577 /* only version 3 or greater inodes are extensively verified here */
578 if (dip->di_version < 3)
581 flags2 = be64_to_cpu(dip->di_flags2);
583 /* don't allow reflink/cowextsize if we don't have reflink */
584 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
585 !xfs_has_reflink(mp))
586 return __this_address;
588 /* only regular files get reflink */
589 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
590 return __this_address;
592 /* don't let reflink and realtime mix */
593 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
594 return __this_address;
596 /* COW extent size hint validation */
597 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
598 mode, flags, flags2);
602 /* bigtime iflag can only happen on bigtime filesystems */
603 if (xfs_dinode_has_bigtime(dip) &&
604 !xfs_has_bigtime(mp))
605 return __this_address;
612 struct xfs_mount *mp,
613 struct xfs_dinode *dip)
617 if (dip->di_version < 3)
620 ASSERT(xfs_has_crc(mp));
621 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
623 dip->di_crc = xfs_end_cksum(crc);
627 * Validate di_extsize hint.
629 * 1. Extent size hint is only valid for directories and regular files.
630 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
631 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
632 * 4. Hint cannot be larger than MAXTEXTLEN.
633 * 5. Can be changed on directories at any time.
634 * 6. Hint value of 0 turns off hints, clears inode flags.
635 * 7. Extent size must be a multiple of the appropriate block size.
636 * For realtime files, this is the rt extent size.
637 * 8. For non-realtime files, the extent size hint must be limited
638 * to half the AG size to avoid alignment extending the extent beyond the
642 xfs_inode_validate_extsize(
643 struct xfs_mount *mp,
651 uint32_t extsize_bytes;
652 uint32_t blocksize_bytes;
654 rt_flag = (flags & XFS_DIFLAG_REALTIME);
655 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
656 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
657 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
660 * This comment describes a historic gap in this verifier function.
662 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
663 * function has never checked that the extent size hint is an integer
664 * multiple of the realtime extent size. Since we allow users to set
665 * this combination on non-rt filesystems /and/ to change the rt
666 * extent size when adding a rt device to a filesystem, the net effect
667 * is that users can configure a filesystem anticipating one rt
668 * geometry and change their minds later. Directories do not use the
669 * extent size hint, so this is harmless for them.
671 * If a directory with a misaligned extent size hint is allowed to
672 * propagate that hint into a new regular realtime file, the result
673 * is that the inode cluster buffer verifier will trigger a corruption
674 * shutdown the next time it is run, because the verifier has always
675 * enforced the alignment rule for regular files.
677 * Because we allow administrators to set a new rt extent size when
678 * adding a rt section, we cannot add a check to this verifier because
679 * that will result a new source of directory corruption errors when
680 * reading an existing filesystem. Instead, we rely on callers to
681 * decide when alignment checks are appropriate, and fix things up as
686 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
688 blocksize_bytes = mp->m_sb.sb_blocksize;
690 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
691 return __this_address;
693 if (hint_flag && !S_ISREG(mode))
694 return __this_address;
696 if (inherit_flag && !S_ISDIR(mode))
697 return __this_address;
699 if ((hint_flag || inherit_flag) && extsize == 0)
700 return __this_address;
702 /* free inodes get flags set to zero but extsize remains */
703 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
704 return __this_address;
706 if (extsize_bytes % blocksize_bytes)
707 return __this_address;
709 if (extsize > XFS_MAX_BMBT_EXTLEN)
710 return __this_address;
712 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
713 return __this_address;
719 * Validate di_cowextsize hint.
721 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
722 * The inode does not have to have any shared blocks, but it must be a v3.
723 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
724 * for a directory, the hint is propagated to new files.
725 * 3. Can be changed on files & directories at any time.
726 * 4. Hint value of 0 turns off hints, clears inode flags.
727 * 5. Extent size must be a multiple of the appropriate block size.
728 * 6. The extent size hint must be limited to half the AG size to avoid
729 * alignment extending the extent beyond the limits of the AG.
732 xfs_inode_validate_cowextsize(
733 struct xfs_mount *mp,
741 uint32_t cowextsize_bytes;
743 rt_flag = (flags & XFS_DIFLAG_REALTIME);
744 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
745 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
747 if (hint_flag && !xfs_has_reflink(mp))
748 return __this_address;
750 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
751 return __this_address;
753 if (hint_flag && cowextsize == 0)
754 return __this_address;
756 /* free inodes get flags set to zero but cowextsize remains */
757 if (mode && !hint_flag && cowextsize != 0)
758 return __this_address;
760 if (hint_flag && rt_flag)
761 return __this_address;
763 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
764 return __this_address;
766 if (cowextsize > XFS_MAX_BMBT_EXTLEN)
767 return __this_address;
769 if (cowextsize > mp->m_sb.sb_agblocks / 2)
770 return __this_address;