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"
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"
21 #include <linux/iversion.h>
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.
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().
43 struct xfs_mount *mp = bp->b_mount;
49 * Validate the magic number and version of every inode in the buffer
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++) {
56 xfs_agino_t unlinked_ino;
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))) {
66 bp->b_flags &= ~XBF_DONE;
67 xfs_buf_ioerror(bp, -EIO);
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));
77 xfs_buf_verifier_error(bp, -EFSCORRUPTED,
78 __func__, dip, sizeof(*dip),
87 xfs_inode_buf_read_verify(
90 xfs_inode_buf_verify(bp, false);
94 xfs_inode_buf_readahead_verify(
97 xfs_inode_buf_verify(bp, true);
101 xfs_inode_buf_write_verify(
104 xfs_inode_buf_verify(bp, false);
107 const struct xfs_buf_ops xfs_inode_buf_ops = {
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,
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,
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.
131 struct xfs_mount *mp,
132 struct xfs_trans *tp,
133 struct xfs_imap *imap,
134 struct xfs_buf **bpp)
136 return xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
137 imap->im_len, XBF_UNMAPPED, bpp,
141 static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
143 struct timespec64 tv;
146 tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
152 /* Convert an ondisk timestamp to an incore timestamp. */
154 xfs_inode_from_disk_ts(
155 struct xfs_dinode *dip,
156 const xfs_timestamp_t ts)
158 struct timespec64 tv;
159 struct xfs_legacy_timestamp *lts;
161 if (xfs_dinode_has_bigtime(dip))
162 return xfs_inode_decode_bigtime(be64_to_cpu(ts));
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);
173 struct xfs_inode *ip,
174 struct xfs_dinode *from)
176 struct inode *inode = VFS_I(ip);
180 ASSERT(ip->i_cowfp == NULL);
181 ASSERT(ip->i_afp == NULL);
183 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
185 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
187 return -EFSCORRUPTED;
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.
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);
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.
207 if (unlikely(from->di_version == 1)) {
208 set_nlink(inode, be16_to_cpu(from->di_onlink));
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);
216 i_uid_write(inode, be32_to_cpu(from->di_uid));
217 i_gid_write(inode, be32_to_cpu(from->di_gid));
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
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);
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);
235 if (from->di_dmevmask || from->di_dmstate)
236 xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
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);
246 error = xfs_iformat_data_fork(ip, from);
249 if (from->di_forkoff) {
250 error = xfs_iformat_attr_fork(ip, from);
252 goto out_destroy_data_fork;
254 if (xfs_is_reflink_inode(ip))
255 xfs_ifork_init_cow(ip);
258 out_destroy_data_fork:
259 xfs_idestroy_fork(&ip->i_df);
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)
269 struct xfs_legacy_timestamp *lts;
272 if (xfs_inode_has_bigtime(ip))
273 return cpu_to_be64(xfs_inode_encode_bigtime(tv));
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);
284 struct xfs_inode *ip,
285 struct xfs_dinode *to,
288 struct inode *inode = VFS_I(ip);
290 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
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);
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);
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);
316 if (xfs_has_v3inodes(ip->i_mount)) {
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;
329 to->di_flushiter = cpu_to_be16(ip->i_flushiter);
333 static xfs_failaddr_t
334 xfs_dinode_verify_fork(
335 struct xfs_dinode *dip,
336 struct xfs_mount *mp,
339 uint32_t di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
340 mode_t mode = be16_to_cpu(dip->di_mode);
341 uint32_t fork_size = XFS_DFORK_SIZE(dip, mp, whichfork);
342 uint32_t fork_format = XFS_DFORK_FORMAT(dip, whichfork);
345 * For fork types that can contain local data, check that the fork
346 * format matches the size of local data contained within the fork.
348 * For all types, check that when the size says the should be in extent
349 * or btree format, the inode isn't claiming it is in local format.
351 if (whichfork == XFS_DATA_FORK) {
352 if (S_ISDIR(mode) || S_ISLNK(mode)) {
353 if (be64_to_cpu(dip->di_size) <= fork_size &&
354 fork_format != XFS_DINODE_FMT_LOCAL)
355 return __this_address;
358 if (be64_to_cpu(dip->di_size) > fork_size &&
359 fork_format == XFS_DINODE_FMT_LOCAL)
360 return __this_address;
363 switch (fork_format) {
364 case XFS_DINODE_FMT_LOCAL:
366 * No local regular files yet.
368 if (S_ISREG(mode) && whichfork == XFS_DATA_FORK)
369 return __this_address;
371 return __this_address;
373 case XFS_DINODE_FMT_EXTENTS:
374 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
375 return __this_address;
377 case XFS_DINODE_FMT_BTREE:
378 if (whichfork == XFS_ATTR_FORK) {
379 if (di_nextents > MAXAEXTNUM)
380 return __this_address;
381 } else if (di_nextents > MAXEXTNUM) {
382 return __this_address;
386 return __this_address;
391 static xfs_failaddr_t
392 xfs_dinode_verify_forkoff(
393 struct xfs_dinode *dip,
394 struct xfs_mount *mp)
396 if (!dip->di_forkoff)
399 switch (dip->di_format) {
400 case XFS_DINODE_FMT_DEV:
401 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
402 return __this_address;
404 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
405 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
406 case XFS_DINODE_FMT_BTREE:
407 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
408 return __this_address;
411 return __this_address;
418 struct xfs_mount *mp,
420 struct xfs_dinode *dip)
428 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
429 return __this_address;
431 /* Verify v3 integrity information first */
432 if (dip->di_version >= 3) {
433 if (!xfs_has_v3inodes(mp))
434 return __this_address;
435 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
437 return __this_address;
438 if (be64_to_cpu(dip->di_ino) != ino)
439 return __this_address;
440 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
441 return __this_address;
444 /* don't allow invalid i_size */
445 di_size = be64_to_cpu(dip->di_size);
446 if (di_size & (1ULL << 63))
447 return __this_address;
449 mode = be16_to_cpu(dip->di_mode);
450 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
451 return __this_address;
453 /* No zero-length symlinks/dirs. */
454 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
455 return __this_address;
457 /* Fork checks carried over from xfs_iformat_fork */
459 be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
460 be64_to_cpu(dip->di_nblocks))
461 return __this_address;
463 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
464 return __this_address;
466 flags = be16_to_cpu(dip->di_flags);
468 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
469 return __this_address;
471 /* check for illegal values of forkoff */
472 fa = xfs_dinode_verify_forkoff(dip, mp);
476 /* Do we have appropriate data fork formats for the mode? */
477 switch (mode & S_IFMT) {
482 if (dip->di_format != XFS_DINODE_FMT_DEV)
483 return __this_address;
488 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
493 /* Uninitialized inode ok. */
496 return __this_address;
499 if (dip->di_forkoff) {
500 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
505 * If there is no fork offset, this may be a freshly-made inode
506 * in a new disk cluster, in which case di_aformat is zeroed.
507 * Otherwise, such an inode must be in EXTENTS format; this goes
508 * for freed inodes as well.
510 switch (dip->di_aformat) {
512 case XFS_DINODE_FMT_EXTENTS:
515 return __this_address;
517 if (dip->di_anextents)
518 return __this_address;
521 /* extent size hint validation */
522 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
527 /* only version 3 or greater inodes are extensively verified here */
528 if (dip->di_version < 3)
531 flags2 = be64_to_cpu(dip->di_flags2);
533 /* don't allow reflink/cowextsize if we don't have reflink */
534 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
535 !xfs_has_reflink(mp))
536 return __this_address;
538 /* only regular files get reflink */
539 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
540 return __this_address;
542 /* don't let reflink and realtime mix */
543 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
544 return __this_address;
546 /* COW extent size hint validation */
547 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
548 mode, flags, flags2);
552 /* bigtime iflag can only happen on bigtime filesystems */
553 if (xfs_dinode_has_bigtime(dip) &&
554 !xfs_has_bigtime(mp))
555 return __this_address;
562 struct xfs_mount *mp,
563 struct xfs_dinode *dip)
567 if (dip->di_version < 3)
570 ASSERT(xfs_has_crc(mp));
571 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
573 dip->di_crc = xfs_end_cksum(crc);
577 * Validate di_extsize hint.
579 * 1. Extent size hint is only valid for directories and regular files.
580 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
581 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
582 * 4. Hint cannot be larger than MAXTEXTLEN.
583 * 5. Can be changed on directories at any time.
584 * 6. Hint value of 0 turns off hints, clears inode flags.
585 * 7. Extent size must be a multiple of the appropriate block size.
586 * For realtime files, this is the rt extent size.
587 * 8. For non-realtime files, the extent size hint must be limited
588 * to half the AG size to avoid alignment extending the extent beyond the
592 xfs_inode_validate_extsize(
593 struct xfs_mount *mp,
601 uint32_t extsize_bytes;
602 uint32_t blocksize_bytes;
604 rt_flag = (flags & XFS_DIFLAG_REALTIME);
605 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
606 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
607 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
610 * This comment describes a historic gap in this verifier function.
612 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
613 * function has never checked that the extent size hint is an integer
614 * multiple of the realtime extent size. Since we allow users to set
615 * this combination on non-rt filesystems /and/ to change the rt
616 * extent size when adding a rt device to a filesystem, the net effect
617 * is that users can configure a filesystem anticipating one rt
618 * geometry and change their minds later. Directories do not use the
619 * extent size hint, so this is harmless for them.
621 * If a directory with a misaligned extent size hint is allowed to
622 * propagate that hint into a new regular realtime file, the result
623 * is that the inode cluster buffer verifier will trigger a corruption
624 * shutdown the next time it is run, because the verifier has always
625 * enforced the alignment rule for regular files.
627 * Because we allow administrators to set a new rt extent size when
628 * adding a rt section, we cannot add a check to this verifier because
629 * that will result a new source of directory corruption errors when
630 * reading an existing filesystem. Instead, we rely on callers to
631 * decide when alignment checks are appropriate, and fix things up as
636 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
638 blocksize_bytes = mp->m_sb.sb_blocksize;
640 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
641 return __this_address;
643 if (hint_flag && !S_ISREG(mode))
644 return __this_address;
646 if (inherit_flag && !S_ISDIR(mode))
647 return __this_address;
649 if ((hint_flag || inherit_flag) && extsize == 0)
650 return __this_address;
652 /* free inodes get flags set to zero but extsize remains */
653 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
654 return __this_address;
656 if (extsize_bytes % blocksize_bytes)
657 return __this_address;
659 if (extsize > MAXEXTLEN)
660 return __this_address;
662 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
663 return __this_address;
669 * Validate di_cowextsize hint.
671 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
672 * The inode does not have to have any shared blocks, but it must be a v3.
673 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
674 * for a directory, the hint is propagated to new files.
675 * 3. Can be changed on files & directories at any time.
676 * 4. Hint value of 0 turns off hints, clears inode flags.
677 * 5. Extent size must be a multiple of the appropriate block size.
678 * 6. The extent size hint must be limited to half the AG size to avoid
679 * alignment extending the extent beyond the limits of the AG.
682 xfs_inode_validate_cowextsize(
683 struct xfs_mount *mp,
691 uint32_t cowextsize_bytes;
693 rt_flag = (flags & XFS_DIFLAG_REALTIME);
694 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
695 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
697 if (hint_flag && !xfs_has_reflink(mp))
698 return __this_address;
700 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
701 return __this_address;
703 if (hint_flag && cowextsize == 0)
704 return __this_address;
706 /* free inodes get flags set to zero but cowextsize remains */
707 if (mode && !hint_flag && cowextsize != 0)
708 return __this_address;
710 if (hint_flag && rt_flag)
711 return __this_address;
713 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
714 return __this_address;
716 if (cowextsize > MAXEXTLEN)
717 return __this_address;
719 if (cowextsize > mp->m_sb.sb_agblocks / 2)
720 return __this_address;