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_ADDR(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->m_sb, 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)bp->b_bn, 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_sb_version_has_v3inode(&ip->i_mount->m_sb))
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_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
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_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
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);
341 switch (XFS_DFORK_FORMAT(dip, whichfork)) {
342 case XFS_DINODE_FMT_LOCAL:
344 * no local regular files yet
346 if (whichfork == XFS_DATA_FORK) {
347 if (S_ISREG(be16_to_cpu(dip->di_mode)))
348 return __this_address;
349 if (be64_to_cpu(dip->di_size) >
350 XFS_DFORK_SIZE(dip, mp, whichfork))
351 return __this_address;
354 return __this_address;
356 case XFS_DINODE_FMT_EXTENTS:
357 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
358 return __this_address;
360 case XFS_DINODE_FMT_BTREE:
361 if (whichfork == XFS_ATTR_FORK) {
362 if (di_nextents > MAXAEXTNUM)
363 return __this_address;
364 } else if (di_nextents > MAXEXTNUM) {
365 return __this_address;
369 return __this_address;
374 static xfs_failaddr_t
375 xfs_dinode_verify_forkoff(
376 struct xfs_dinode *dip,
377 struct xfs_mount *mp)
379 if (!dip->di_forkoff)
382 switch (dip->di_format) {
383 case XFS_DINODE_FMT_DEV:
384 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
385 return __this_address;
387 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
388 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
389 case XFS_DINODE_FMT_BTREE:
390 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
391 return __this_address;
394 return __this_address;
401 struct xfs_mount *mp,
403 struct xfs_dinode *dip)
411 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
412 return __this_address;
414 /* Verify v3 integrity information first */
415 if (dip->di_version >= 3) {
416 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
417 return __this_address;
418 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
420 return __this_address;
421 if (be64_to_cpu(dip->di_ino) != ino)
422 return __this_address;
423 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
424 return __this_address;
427 /* don't allow invalid i_size */
428 di_size = be64_to_cpu(dip->di_size);
429 if (di_size & (1ULL << 63))
430 return __this_address;
432 mode = be16_to_cpu(dip->di_mode);
433 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
434 return __this_address;
436 /* No zero-length symlinks/dirs. */
437 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
438 return __this_address;
440 /* Fork checks carried over from xfs_iformat_fork */
442 be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
443 be64_to_cpu(dip->di_nblocks))
444 return __this_address;
446 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
447 return __this_address;
449 flags = be16_to_cpu(dip->di_flags);
451 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
452 return __this_address;
454 /* check for illegal values of forkoff */
455 fa = xfs_dinode_verify_forkoff(dip, mp);
459 /* Do we have appropriate data fork formats for the mode? */
460 switch (mode & S_IFMT) {
465 if (dip->di_format != XFS_DINODE_FMT_DEV)
466 return __this_address;
471 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
476 /* Uninitialized inode ok. */
479 return __this_address;
482 if (dip->di_forkoff) {
483 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
488 * If there is no fork offset, this may be a freshly-made inode
489 * in a new disk cluster, in which case di_aformat is zeroed.
490 * Otherwise, such an inode must be in EXTENTS format; this goes
491 * for freed inodes as well.
493 switch (dip->di_aformat) {
495 case XFS_DINODE_FMT_EXTENTS:
498 return __this_address;
500 if (dip->di_anextents)
501 return __this_address;
504 /* extent size hint validation */
505 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
510 /* only version 3 or greater inodes are extensively verified here */
511 if (dip->di_version < 3)
514 flags2 = be64_to_cpu(dip->di_flags2);
516 /* don't allow reflink/cowextsize if we don't have reflink */
517 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
518 !xfs_sb_version_hasreflink(&mp->m_sb))
519 return __this_address;
521 /* only regular files get reflink */
522 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
523 return __this_address;
525 /* don't let reflink and realtime mix */
526 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
527 return __this_address;
529 /* COW extent size hint validation */
530 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
531 mode, flags, flags2);
535 /* bigtime iflag can only happen on bigtime filesystems */
536 if (xfs_dinode_has_bigtime(dip) &&
537 !xfs_sb_version_hasbigtime(&mp->m_sb))
538 return __this_address;
545 struct xfs_mount *mp,
546 struct xfs_dinode *dip)
550 if (dip->di_version < 3)
553 ASSERT(xfs_sb_version_hascrc(&mp->m_sb));
554 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
556 dip->di_crc = xfs_end_cksum(crc);
560 * Validate di_extsize hint.
562 * 1. Extent size hint is only valid for directories and regular files.
563 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
564 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
565 * 4. Hint cannot be larger than MAXTEXTLEN.
566 * 5. Can be changed on directories at any time.
567 * 6. Hint value of 0 turns off hints, clears inode flags.
568 * 7. Extent size must be a multiple of the appropriate block size.
569 * For realtime files, this is the rt extent size.
570 * 8. For non-realtime files, the extent size hint must be limited
571 * to half the AG size to avoid alignment extending the extent beyond the
575 xfs_inode_validate_extsize(
576 struct xfs_mount *mp,
584 uint32_t extsize_bytes;
585 uint32_t blocksize_bytes;
587 rt_flag = (flags & XFS_DIFLAG_REALTIME);
588 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
589 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
590 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
593 * This comment describes a historic gap in this verifier function.
595 * On older kernels, the extent size hint verifier doesn't check that
596 * the extent size hint is an integer multiple of the realtime extent
597 * size on a directory with both RTINHERIT and EXTSZINHERIT flags set.
598 * The verifier has always enforced the alignment rule for regular
599 * files with the REALTIME flag set.
601 * If a directory with a misaligned extent size hint is allowed to
602 * propagate that hint into a new regular realtime file, the result
603 * is that the inode cluster buffer verifier will trigger a corruption
604 * shutdown the next time it is run.
606 * Unfortunately, there could be filesystems with these misconfigured
607 * directories in the wild, so we cannot add a check to this verifier
608 * at this time because that will result a new source of directory
609 * corruption errors when reading an existing filesystem. Instead, we
610 * permit the misconfiguration to pass through the verifiers so that
611 * callers of this function can correct and mitigate externally.
615 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
617 blocksize_bytes = mp->m_sb.sb_blocksize;
619 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
620 return __this_address;
622 if (hint_flag && !S_ISREG(mode))
623 return __this_address;
625 if (inherit_flag && !S_ISDIR(mode))
626 return __this_address;
628 if ((hint_flag || inherit_flag) && extsize == 0)
629 return __this_address;
631 /* free inodes get flags set to zero but extsize remains */
632 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
633 return __this_address;
635 if (extsize_bytes % blocksize_bytes)
636 return __this_address;
638 if (extsize > MAXEXTLEN)
639 return __this_address;
641 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
642 return __this_address;
648 * Validate di_cowextsize hint.
650 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
651 * The inode does not have to have any shared blocks, but it must be a v3.
652 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
653 * for a directory, the hint is propagated to new files.
654 * 3. Can be changed on files & directories at any time.
655 * 4. Hint value of 0 turns off hints, clears inode flags.
656 * 5. Extent size must be a multiple of the appropriate block size.
657 * 6. The extent size hint must be limited to half the AG size to avoid
658 * alignment extending the extent beyond the limits of the AG.
661 xfs_inode_validate_cowextsize(
662 struct xfs_mount *mp,
670 uint32_t cowextsize_bytes;
672 rt_flag = (flags & XFS_DIFLAG_REALTIME);
673 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
674 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
676 if (hint_flag && !xfs_sb_version_hasreflink(&mp->m_sb))
677 return __this_address;
679 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
680 return __this_address;
682 if (hint_flag && cowextsize == 0)
683 return __this_address;
685 /* free inodes get flags set to zero but cowextsize remains */
686 if (mode && !hint_flag && cowextsize != 0)
687 return __this_address;
689 if (hint_flag && rt_flag)
690 return __this_address;
692 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
693 return __this_address;
695 if (cowextsize > MAXEXTLEN)
696 return __this_address;
698 if (cowextsize > mp->m_sb.sb_agblocks / 2)
699 return __this_address;