1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
6 #include "libxfs_priv.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_trace.h"
20 #include "xfs_cksum.h"
21 #include "xfs_trans.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_alloc_btree.h"
24 #include "xfs_ialloc_btree.h"
25 #include "xfs_rmap_btree.h"
27 #include "xfs_refcount_btree.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
32 * Physical superblock buffer manipulations. Shared with libxfs in userspace.
36 * Reference counting access wrappers to the perag structures.
37 * Because we never free per-ag structures, the only thing we
38 * have to protect against changes is the tree structure itself.
45 struct xfs_perag
*pag
;
49 pag
= radix_tree_lookup(&mp
->m_perag_tree
, agno
);
51 ASSERT(atomic_read(&pag
->pag_ref
) >= 0);
52 ref
= atomic_inc_return(&pag
->pag_ref
);
55 trace_xfs_perag_get(mp
, agno
, ref
, _RET_IP_
);
60 * search from @first to find the next perag with the given tag set.
68 struct xfs_perag
*pag
;
73 found
= radix_tree_gang_lookup_tag(&mp
->m_perag_tree
,
74 (void **)&pag
, first
, 1, tag
);
79 ref
= atomic_inc_return(&pag
->pag_ref
);
81 trace_xfs_perag_get_tag(mp
, pag
->pag_agno
, ref
, _RET_IP_
);
87 struct xfs_perag
*pag
)
91 ASSERT(atomic_read(&pag
->pag_ref
) > 0);
92 ref
= atomic_dec_return(&pag
->pag_ref
);
93 trace_xfs_perag_put(pag
->pag_mount
, pag
->pag_agno
, ref
, _RET_IP_
);
96 /* Check all the superblock fields we care about when reading one in. */
102 if (XFS_SB_VERSION_NUM(sbp
) != XFS_SB_VERSION_5
)
106 * Version 5 superblock feature mask validation. Reject combinations
107 * the kernel cannot support up front before checking anything else.
109 if (xfs_sb_has_compat_feature(sbp
, XFS_SB_FEAT_COMPAT_UNKNOWN
)) {
111 "Superblock has unknown compatible features (0x%x) enabled.",
112 (sbp
->sb_features_compat
& XFS_SB_FEAT_COMPAT_UNKNOWN
));
114 "Using a more recent kernel is recommended.");
117 if (xfs_sb_has_ro_compat_feature(sbp
, XFS_SB_FEAT_RO_COMPAT_UNKNOWN
)) {
119 "Superblock has unknown read-only compatible features (0x%x) enabled.",
120 (sbp
->sb_features_ro_compat
&
121 XFS_SB_FEAT_RO_COMPAT_UNKNOWN
));
122 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
124 "Attempted to mount read-only compatible filesystem read-write.");
126 "Filesystem can only be safely mounted read only.");
131 if (xfs_sb_has_incompat_feature(sbp
, XFS_SB_FEAT_INCOMPAT_UNKNOWN
)) {
133 "Superblock has unknown incompatible features (0x%x) enabled.",
134 (sbp
->sb_features_incompat
&
135 XFS_SB_FEAT_INCOMPAT_UNKNOWN
));
137 "Filesystem cannot be safely mounted by this kernel.");
144 /* Check all the superblock fields we care about when writing one out. */
146 xfs_validate_sb_write(
147 struct xfs_mount
*mp
,
152 * Carry out additional sb summary counter sanity checks when we write
153 * the superblock. We skip this in the read validator because there
154 * could be newer superblocks in the log and if the values are garbage
155 * even after replay we'll recalculate them at the end of log mount.
157 * mkfs has traditionally written zeroed counters to inprogress and
158 * secondary superblocks, so allow this usage to continue because
159 * we never read counters from such superblocks.
161 if (XFS_BUF_ADDR(bp
) == XFS_SB_DADDR
&& !sbp
->sb_inprogress
&&
162 (sbp
->sb_fdblocks
> sbp
->sb_dblocks
||
163 !xfs_verify_icount(mp
, sbp
->sb_icount
) ||
164 sbp
->sb_ifree
> sbp
->sb_icount
)) {
165 xfs_warn(mp
, "SB summary counter sanity check failed");
166 return -EFSCORRUPTED
;
169 if (XFS_SB_VERSION_NUM(sbp
) != XFS_SB_VERSION_5
)
173 * Version 5 superblock feature mask validation. Reject combinations
174 * the kernel cannot support since we checked for unsupported bits in
175 * the read verifier, which means that memory is corrupt.
177 if (xfs_sb_has_compat_feature(sbp
, XFS_SB_FEAT_COMPAT_UNKNOWN
)) {
179 "Corruption detected in superblock compatible features (0x%x)!",
180 (sbp
->sb_features_compat
& XFS_SB_FEAT_COMPAT_UNKNOWN
));
181 return -EFSCORRUPTED
;
184 if (xfs_sb_has_ro_compat_feature(sbp
, XFS_SB_FEAT_RO_COMPAT_UNKNOWN
)) {
186 "Corruption detected in superblock read-only compatible features (0x%x)!",
187 (sbp
->sb_features_ro_compat
&
188 XFS_SB_FEAT_RO_COMPAT_UNKNOWN
));
189 return -EFSCORRUPTED
;
191 if (xfs_sb_has_incompat_feature(sbp
, XFS_SB_FEAT_INCOMPAT_UNKNOWN
)) {
193 "Corruption detected in superblock incompatible features (0x%x)!",
194 (sbp
->sb_features_incompat
&
195 XFS_SB_FEAT_INCOMPAT_UNKNOWN
));
196 return -EFSCORRUPTED
;
198 if (xfs_sb_has_incompat_log_feature(sbp
,
199 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
201 "Corruption detected in superblock incompatible log features (0x%x)!",
202 (sbp
->sb_features_log_incompat
&
203 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
204 return -EFSCORRUPTED
;
208 * We can't read verify the sb LSN because the read verifier is called
209 * before the log is allocated and processed. We know the log is set up
210 * before write verifier calls, so check it here.
212 if (!xfs_log_check_lsn(mp
, sbp
->sb_lsn
))
213 return -EFSCORRUPTED
;
218 /* Check the validity of the SB. */
220 xfs_validate_sb_common(
221 struct xfs_mount
*mp
,
225 uint32_t agcount
= 0;
228 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
229 xfs_warn(mp
, "bad magic number");
233 if (!xfs_sb_good_version(sbp
)) {
234 xfs_warn(mp
, "bad version");
238 if (xfs_sb_version_has_pquotino(sbp
)) {
239 if (sbp
->sb_qflags
& (XFS_OQUOTA_ENFD
| XFS_OQUOTA_CHKD
)) {
241 "Version 5 of Super block has XFS_OQUOTA bits.");
242 return -EFSCORRUPTED
;
244 } else if (sbp
->sb_qflags
& (XFS_PQUOTA_ENFD
| XFS_GQUOTA_ENFD
|
245 XFS_PQUOTA_CHKD
| XFS_GQUOTA_CHKD
)) {
247 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
248 return -EFSCORRUPTED
;
252 * Full inode chunks must be aligned to inode chunk size when
253 * sparse inodes are enabled to support the sparse chunk
254 * allocation algorithm and prevent overlapping inode records.
256 if (xfs_sb_version_hassparseinodes(sbp
)) {
259 align
= XFS_INODES_PER_CHUNK
* sbp
->sb_inodesize
261 if (sbp
->sb_inoalignmt
!= align
) {
263 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
264 sbp
->sb_inoalignmt
, align
);
270 sbp
->sb_logstart
== 0 && mp
->m_logdev_targp
== mp
->m_ddev_targp
)) {
272 "filesystem is marked as having an external log; "
273 "specify logdev on the mount command line.");
278 sbp
->sb_logstart
!= 0 && mp
->m_logdev_targp
!= mp
->m_ddev_targp
)) {
280 "filesystem is marked as having an internal log; "
281 "do not specify logdev on the mount command line.");
285 /* Compute agcount for this number of dblocks and agblocks */
286 if (sbp
->sb_agblocks
) {
287 agcount
= div_u64_rem(sbp
->sb_dblocks
, sbp
->sb_agblocks
, &rem
);
293 * More sanity checking. Most of these were stolen directly from
297 sbp
->sb_agcount
<= 0 ||
298 sbp
->sb_sectsize
< XFS_MIN_SECTORSIZE
||
299 sbp
->sb_sectsize
> XFS_MAX_SECTORSIZE
||
300 sbp
->sb_sectlog
< XFS_MIN_SECTORSIZE_LOG
||
301 sbp
->sb_sectlog
> XFS_MAX_SECTORSIZE_LOG
||
302 sbp
->sb_sectsize
!= (1 << sbp
->sb_sectlog
) ||
303 sbp
->sb_blocksize
< XFS_MIN_BLOCKSIZE
||
304 sbp
->sb_blocksize
> XFS_MAX_BLOCKSIZE
||
305 sbp
->sb_blocklog
< XFS_MIN_BLOCKSIZE_LOG
||
306 sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
307 sbp
->sb_blocksize
!= (1 << sbp
->sb_blocklog
) ||
308 sbp
->sb_dirblklog
+ sbp
->sb_blocklog
> XFS_MAX_BLOCKSIZE_LOG
||
309 sbp
->sb_inodesize
< XFS_DINODE_MIN_SIZE
||
310 sbp
->sb_inodesize
> XFS_DINODE_MAX_SIZE
||
311 sbp
->sb_inodelog
< XFS_DINODE_MIN_LOG
||
312 sbp
->sb_inodelog
> XFS_DINODE_MAX_LOG
||
313 sbp
->sb_inodesize
!= (1 << sbp
->sb_inodelog
) ||
314 sbp
->sb_logsunit
> XLOG_MAX_RECORD_BSIZE
||
315 sbp
->sb_inopblock
!= howmany(sbp
->sb_blocksize
,sbp
->sb_inodesize
) ||
316 XFS_FSB_TO_B(mp
, sbp
->sb_agblocks
) < XFS_MIN_AG_BYTES
||
317 XFS_FSB_TO_B(mp
, sbp
->sb_agblocks
) > XFS_MAX_AG_BYTES
||
318 sbp
->sb_agblklog
!= xfs_highbit32(sbp
->sb_agblocks
- 1) + 1 ||
319 agcount
== 0 || agcount
!= sbp
->sb_agcount
||
320 (sbp
->sb_blocklog
- sbp
->sb_inodelog
!= sbp
->sb_inopblog
) ||
321 (sbp
->sb_rextsize
* sbp
->sb_blocksize
> XFS_MAX_RTEXTSIZE
) ||
322 (sbp
->sb_rextsize
* sbp
->sb_blocksize
< XFS_MIN_RTEXTSIZE
) ||
323 (sbp
->sb_imax_pct
> 100 /* zero sb_imax_pct is valid */) ||
324 sbp
->sb_dblocks
== 0 ||
325 sbp
->sb_dblocks
> XFS_MAX_DBLOCKS(sbp
) ||
326 sbp
->sb_dblocks
< XFS_MIN_DBLOCKS(sbp
) ||
327 sbp
->sb_shared_vn
!= 0)) {
328 xfs_notice(mp
, "SB sanity check failed");
329 return -EFSCORRUPTED
;
333 if (!xfs_sb_version_hasdalign(sbp
) ||
334 sbp
->sb_unit
> sbp
->sb_width
||
335 (sbp
->sb_width
% sbp
->sb_unit
) != 0) {
336 xfs_notice(mp
, "SB stripe unit sanity check failed");
337 return -EFSCORRUPTED
;
339 } else if (xfs_sb_version_hasdalign(sbp
)) {
340 xfs_notice(mp
, "SB stripe alignment sanity check failed");
341 return -EFSCORRUPTED
;
342 } else if (sbp
->sb_width
) {
343 xfs_notice(mp
, "SB stripe width sanity check failed");
344 return -EFSCORRUPTED
;
348 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
349 sbp
->sb_blocksize
< XFS_MIN_CRC_BLOCKSIZE
) {
350 xfs_notice(mp
, "v5 SB sanity check failed");
351 return -EFSCORRUPTED
;
355 * Currently only very few inode sizes are supported.
357 switch (sbp
->sb_inodesize
) {
364 xfs_warn(mp
, "inode size of %d bytes not supported",
369 if (xfs_sb_validate_fsb_count(sbp
, sbp
->sb_dblocks
) ||
370 xfs_sb_validate_fsb_count(sbp
, sbp
->sb_rblocks
)) {
372 "file system too large to be mounted on this system.");
380 xfs_sb_quota_from_disk(struct xfs_sb
*sbp
)
383 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
384 * leads to in-core values having two different values for a quota
385 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
388 * Note that this change affect only the in-core values. These
389 * values are not written back to disk unless any quota information
390 * is written to the disk. Even in that case, sb_pquotino field is
391 * not written to disk unless the superblock supports pquotino.
393 if (sbp
->sb_uquotino
== 0)
394 sbp
->sb_uquotino
= NULLFSINO
;
395 if (sbp
->sb_gquotino
== 0)
396 sbp
->sb_gquotino
= NULLFSINO
;
397 if (sbp
->sb_pquotino
== 0)
398 sbp
->sb_pquotino
= NULLFSINO
;
401 * We need to do these manipilations only if we are working
402 * with an older version of on-disk superblock.
404 if (xfs_sb_version_has_pquotino(sbp
))
407 if (sbp
->sb_qflags
& XFS_OQUOTA_ENFD
)
408 sbp
->sb_qflags
|= (sbp
->sb_qflags
& XFS_PQUOTA_ACCT
) ?
409 XFS_PQUOTA_ENFD
: XFS_GQUOTA_ENFD
;
410 if (sbp
->sb_qflags
& XFS_OQUOTA_CHKD
)
411 sbp
->sb_qflags
|= (sbp
->sb_qflags
& XFS_PQUOTA_ACCT
) ?
412 XFS_PQUOTA_CHKD
: XFS_GQUOTA_CHKD
;
413 sbp
->sb_qflags
&= ~(XFS_OQUOTA_ENFD
| XFS_OQUOTA_CHKD
);
415 if (sbp
->sb_qflags
& XFS_PQUOTA_ACCT
&&
416 sbp
->sb_gquotino
!= NULLFSINO
) {
418 * In older version of superblock, on-disk superblock only
419 * has sb_gquotino, and in-core superblock has both sb_gquotino
420 * and sb_pquotino. But, only one of them is supported at any
421 * point of time. So, if PQUOTA is set in disk superblock,
422 * copy over sb_gquotino to sb_pquotino. The NULLFSINO test
423 * above is to make sure we don't do this twice and wipe them
426 sbp
->sb_pquotino
= sbp
->sb_gquotino
;
427 sbp
->sb_gquotino
= NULLFSINO
;
437 to
->sb_magicnum
= be32_to_cpu(from
->sb_magicnum
);
438 to
->sb_blocksize
= be32_to_cpu(from
->sb_blocksize
);
439 to
->sb_dblocks
= be64_to_cpu(from
->sb_dblocks
);
440 to
->sb_rblocks
= be64_to_cpu(from
->sb_rblocks
);
441 to
->sb_rextents
= be64_to_cpu(from
->sb_rextents
);
442 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
443 to
->sb_logstart
= be64_to_cpu(from
->sb_logstart
);
444 to
->sb_rootino
= be64_to_cpu(from
->sb_rootino
);
445 to
->sb_rbmino
= be64_to_cpu(from
->sb_rbmino
);
446 to
->sb_rsumino
= be64_to_cpu(from
->sb_rsumino
);
447 to
->sb_rextsize
= be32_to_cpu(from
->sb_rextsize
);
448 to
->sb_agblocks
= be32_to_cpu(from
->sb_agblocks
);
449 to
->sb_agcount
= be32_to_cpu(from
->sb_agcount
);
450 to
->sb_rbmblocks
= be32_to_cpu(from
->sb_rbmblocks
);
451 to
->sb_logblocks
= be32_to_cpu(from
->sb_logblocks
);
452 to
->sb_versionnum
= be16_to_cpu(from
->sb_versionnum
);
453 to
->sb_sectsize
= be16_to_cpu(from
->sb_sectsize
);
454 to
->sb_inodesize
= be16_to_cpu(from
->sb_inodesize
);
455 to
->sb_inopblock
= be16_to_cpu(from
->sb_inopblock
);
456 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
457 to
->sb_blocklog
= from
->sb_blocklog
;
458 to
->sb_sectlog
= from
->sb_sectlog
;
459 to
->sb_inodelog
= from
->sb_inodelog
;
460 to
->sb_inopblog
= from
->sb_inopblog
;
461 to
->sb_agblklog
= from
->sb_agblklog
;
462 to
->sb_rextslog
= from
->sb_rextslog
;
463 to
->sb_inprogress
= from
->sb_inprogress
;
464 to
->sb_imax_pct
= from
->sb_imax_pct
;
465 to
->sb_icount
= be64_to_cpu(from
->sb_icount
);
466 to
->sb_ifree
= be64_to_cpu(from
->sb_ifree
);
467 to
->sb_fdblocks
= be64_to_cpu(from
->sb_fdblocks
);
468 to
->sb_frextents
= be64_to_cpu(from
->sb_frextents
);
469 to
->sb_uquotino
= be64_to_cpu(from
->sb_uquotino
);
470 to
->sb_gquotino
= be64_to_cpu(from
->sb_gquotino
);
471 to
->sb_qflags
= be16_to_cpu(from
->sb_qflags
);
472 to
->sb_flags
= from
->sb_flags
;
473 to
->sb_shared_vn
= from
->sb_shared_vn
;
474 to
->sb_inoalignmt
= be32_to_cpu(from
->sb_inoalignmt
);
475 to
->sb_unit
= be32_to_cpu(from
->sb_unit
);
476 to
->sb_width
= be32_to_cpu(from
->sb_width
);
477 to
->sb_dirblklog
= from
->sb_dirblklog
;
478 to
->sb_logsectlog
= from
->sb_logsectlog
;
479 to
->sb_logsectsize
= be16_to_cpu(from
->sb_logsectsize
);
480 to
->sb_logsunit
= be32_to_cpu(from
->sb_logsunit
);
481 to
->sb_features2
= be32_to_cpu(from
->sb_features2
);
482 to
->sb_bad_features2
= be32_to_cpu(from
->sb_bad_features2
);
483 to
->sb_features_compat
= be32_to_cpu(from
->sb_features_compat
);
484 to
->sb_features_ro_compat
= be32_to_cpu(from
->sb_features_ro_compat
);
485 to
->sb_features_incompat
= be32_to_cpu(from
->sb_features_incompat
);
486 to
->sb_features_log_incompat
=
487 be32_to_cpu(from
->sb_features_log_incompat
);
488 /* crc is only used on disk, not in memory; just init to 0 here. */
490 to
->sb_spino_align
= be32_to_cpu(from
->sb_spino_align
);
491 to
->sb_pquotino
= be64_to_cpu(from
->sb_pquotino
);
492 to
->sb_lsn
= be64_to_cpu(from
->sb_lsn
);
494 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
495 * feature flag is set; if not set we keep it only in memory.
497 if (xfs_sb_version_hasmetauuid(to
))
498 uuid_copy(&to
->sb_meta_uuid
, &from
->sb_meta_uuid
);
500 uuid_copy(&to
->sb_meta_uuid
, &from
->sb_uuid
);
501 /* Convert on-disk flags to in-memory flags? */
503 xfs_sb_quota_from_disk(to
);
511 __xfs_sb_from_disk(to
, from
, true);
515 xfs_sb_quota_to_disk(
519 uint16_t qflags
= from
->sb_qflags
;
521 to
->sb_uquotino
= cpu_to_be64(from
->sb_uquotino
);
522 if (xfs_sb_version_has_pquotino(from
)) {
523 to
->sb_qflags
= cpu_to_be16(from
->sb_qflags
);
524 to
->sb_gquotino
= cpu_to_be64(from
->sb_gquotino
);
525 to
->sb_pquotino
= cpu_to_be64(from
->sb_pquotino
);
530 * The in-core version of sb_qflags do not have XFS_OQUOTA_*
531 * flags, whereas the on-disk version does. So, convert incore
532 * XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
534 qflags
&= ~(XFS_PQUOTA_ENFD
| XFS_PQUOTA_CHKD
|
535 XFS_GQUOTA_ENFD
| XFS_GQUOTA_CHKD
);
537 if (from
->sb_qflags
&
538 (XFS_PQUOTA_ENFD
| XFS_GQUOTA_ENFD
))
539 qflags
|= XFS_OQUOTA_ENFD
;
540 if (from
->sb_qflags
&
541 (XFS_PQUOTA_CHKD
| XFS_GQUOTA_CHKD
))
542 qflags
|= XFS_OQUOTA_CHKD
;
543 to
->sb_qflags
= cpu_to_be16(qflags
);
546 * GQUOTINO and PQUOTINO cannot be used together in versions
547 * of superblock that do not have pquotino. from->sb_flags
548 * tells us which quota is active and should be copied to
549 * disk. If neither are active, we should NULL the inode.
551 * In all cases, the separate pquotino must remain 0 because it
552 * it beyond the "end" of the valid non-pquotino superblock.
554 if (from
->sb_qflags
& XFS_GQUOTA_ACCT
)
555 to
->sb_gquotino
= cpu_to_be64(from
->sb_gquotino
);
556 else if (from
->sb_qflags
& XFS_PQUOTA_ACCT
)
557 to
->sb_gquotino
= cpu_to_be64(from
->sb_pquotino
);
560 * We can't rely on just the fields being logged to tell us
561 * that it is safe to write NULLFSINO - we should only do that
562 * if quotas are not actually enabled. Hence only write
563 * NULLFSINO if both in-core quota inodes are NULL.
565 if (from
->sb_gquotino
== NULLFSINO
&&
566 from
->sb_pquotino
== NULLFSINO
)
567 to
->sb_gquotino
= cpu_to_be64(NULLFSINO
);
578 xfs_sb_quota_to_disk(to
, from
);
580 to
->sb_magicnum
= cpu_to_be32(from
->sb_magicnum
);
581 to
->sb_blocksize
= cpu_to_be32(from
->sb_blocksize
);
582 to
->sb_dblocks
= cpu_to_be64(from
->sb_dblocks
);
583 to
->sb_rblocks
= cpu_to_be64(from
->sb_rblocks
);
584 to
->sb_rextents
= cpu_to_be64(from
->sb_rextents
);
585 memcpy(&to
->sb_uuid
, &from
->sb_uuid
, sizeof(to
->sb_uuid
));
586 to
->sb_logstart
= cpu_to_be64(from
->sb_logstart
);
587 to
->sb_rootino
= cpu_to_be64(from
->sb_rootino
);
588 to
->sb_rbmino
= cpu_to_be64(from
->sb_rbmino
);
589 to
->sb_rsumino
= cpu_to_be64(from
->sb_rsumino
);
590 to
->sb_rextsize
= cpu_to_be32(from
->sb_rextsize
);
591 to
->sb_agblocks
= cpu_to_be32(from
->sb_agblocks
);
592 to
->sb_agcount
= cpu_to_be32(from
->sb_agcount
);
593 to
->sb_rbmblocks
= cpu_to_be32(from
->sb_rbmblocks
);
594 to
->sb_logblocks
= cpu_to_be32(from
->sb_logblocks
);
595 to
->sb_versionnum
= cpu_to_be16(from
->sb_versionnum
);
596 to
->sb_sectsize
= cpu_to_be16(from
->sb_sectsize
);
597 to
->sb_inodesize
= cpu_to_be16(from
->sb_inodesize
);
598 to
->sb_inopblock
= cpu_to_be16(from
->sb_inopblock
);
599 memcpy(&to
->sb_fname
, &from
->sb_fname
, sizeof(to
->sb_fname
));
600 to
->sb_blocklog
= from
->sb_blocklog
;
601 to
->sb_sectlog
= from
->sb_sectlog
;
602 to
->sb_inodelog
= from
->sb_inodelog
;
603 to
->sb_inopblog
= from
->sb_inopblog
;
604 to
->sb_agblklog
= from
->sb_agblklog
;
605 to
->sb_rextslog
= from
->sb_rextslog
;
606 to
->sb_inprogress
= from
->sb_inprogress
;
607 to
->sb_imax_pct
= from
->sb_imax_pct
;
608 to
->sb_icount
= cpu_to_be64(from
->sb_icount
);
609 to
->sb_ifree
= cpu_to_be64(from
->sb_ifree
);
610 to
->sb_fdblocks
= cpu_to_be64(from
->sb_fdblocks
);
611 to
->sb_frextents
= cpu_to_be64(from
->sb_frextents
);
613 to
->sb_flags
= from
->sb_flags
;
614 to
->sb_shared_vn
= from
->sb_shared_vn
;
615 to
->sb_inoalignmt
= cpu_to_be32(from
->sb_inoalignmt
);
616 to
->sb_unit
= cpu_to_be32(from
->sb_unit
);
617 to
->sb_width
= cpu_to_be32(from
->sb_width
);
618 to
->sb_dirblklog
= from
->sb_dirblklog
;
619 to
->sb_logsectlog
= from
->sb_logsectlog
;
620 to
->sb_logsectsize
= cpu_to_be16(from
->sb_logsectsize
);
621 to
->sb_logsunit
= cpu_to_be32(from
->sb_logsunit
);
624 * We need to ensure that bad_features2 always matches features2.
625 * Hence we enforce that here rather than having to remember to do it
626 * everywhere else that updates features2.
628 from
->sb_bad_features2
= from
->sb_features2
;
629 to
->sb_features2
= cpu_to_be32(from
->sb_features2
);
630 to
->sb_bad_features2
= cpu_to_be32(from
->sb_bad_features2
);
632 if (xfs_sb_version_hascrc(from
)) {
633 to
->sb_features_compat
= cpu_to_be32(from
->sb_features_compat
);
634 to
->sb_features_ro_compat
=
635 cpu_to_be32(from
->sb_features_ro_compat
);
636 to
->sb_features_incompat
=
637 cpu_to_be32(from
->sb_features_incompat
);
638 to
->sb_features_log_incompat
=
639 cpu_to_be32(from
->sb_features_log_incompat
);
640 to
->sb_spino_align
= cpu_to_be32(from
->sb_spino_align
);
641 to
->sb_lsn
= cpu_to_be64(from
->sb_lsn
);
642 if (xfs_sb_version_hasmetauuid(from
))
643 uuid_copy(&to
->sb_meta_uuid
, &from
->sb_meta_uuid
);
648 * If the superblock has the CRC feature bit set or the CRC field is non-null,
649 * check that the CRC is valid. We check the CRC field is non-null because a
650 * single bit error could clear the feature bit and unused parts of the
651 * superblock are supposed to be zero. Hence a non-null crc field indicates that
652 * we've potentially lost a feature bit and we should check it anyway.
654 * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
655 * last field in V4 secondary superblocks. So for secondary superblocks,
656 * we are more forgiving, and ignore CRC failures if the primary doesn't
657 * indicate that the fs version is V5.
664 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
665 struct xfs_dsb
*dsb
= XFS_BUF_TO_SBP(bp
);
669 * open code the version check to avoid needing to convert the entire
670 * superblock from disk order just to check the version number
672 if (dsb
->sb_magicnum
== cpu_to_be32(XFS_SB_MAGIC
) &&
673 (((be16_to_cpu(dsb
->sb_versionnum
) & XFS_SB_VERSION_NUMBITS
) ==
677 if (!xfs_buf_verify_cksum(bp
, XFS_SB_CRC_OFF
)) {
678 /* Only fail bad secondaries on a known V5 filesystem */
679 if (bp
->b_bn
== XFS_SB_DADDR
||
680 xfs_sb_version_hascrc(&mp
->m_sb
)) {
688 * Check all the superblock fields. Don't byteswap the xquota flags
689 * because _verify_common checks the on-disk values.
691 __xfs_sb_from_disk(&sb
, XFS_BUF_TO_SBP(bp
), false);
692 error
= xfs_validate_sb_common(mp
, bp
, &sb
);
695 error
= xfs_validate_sb_read(mp
, &sb
);
698 if (error
== -EFSCORRUPTED
|| error
== -EFSBADCRC
)
699 xfs_verifier_error(bp
, error
, __this_address
);
701 xfs_buf_ioerror(bp
, error
);
705 * We may be probed for a filesystem match, so we may not want to emit
706 * messages when the superblock buffer is not actually an XFS superblock.
707 * If we find an XFS superblock, then run a normal, noisy mount because we are
708 * really going to mount it and want to know about errors.
711 xfs_sb_quiet_read_verify(
714 struct xfs_dsb
*dsb
= XFS_BUF_TO_SBP(bp
);
716 if (dsb
->sb_magicnum
== cpu_to_be32(XFS_SB_MAGIC
)) {
717 /* XFS filesystem, verify noisily! */
718 xfs_sb_read_verify(bp
);
722 xfs_buf_ioerror(bp
, -EWRONGFS
);
730 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
731 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
735 * Check all the superblock fields. Don't byteswap the xquota flags
736 * because _verify_common checks the on-disk values.
738 __xfs_sb_from_disk(&sb
, XFS_BUF_TO_SBP(bp
), false);
739 error
= xfs_validate_sb_common(mp
, bp
, &sb
);
742 error
= xfs_validate_sb_write(mp
, bp
, &sb
);
746 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
750 XFS_BUF_TO_SBP(bp
)->sb_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
752 xfs_buf_update_cksum(bp
, XFS_SB_CRC_OFF
);
756 xfs_verifier_error(bp
, error
, __this_address
);
759 const struct xfs_buf_ops xfs_sb_buf_ops
= {
761 .verify_read
= xfs_sb_read_verify
,
762 .verify_write
= xfs_sb_write_verify
,
765 const struct xfs_buf_ops xfs_sb_quiet_buf_ops
= {
766 .name
= "xfs_sb_quiet",
767 .verify_read
= xfs_sb_quiet_read_verify
,
768 .verify_write
= xfs_sb_write_verify
,
774 * Mount initialization code establishing various mount
775 * fields from the superblock associated with the given
780 struct xfs_mount
*mp
,
783 mp
->m_agfrotor
= mp
->m_agirotor
= 0;
784 mp
->m_maxagi
= mp
->m_sb
.sb_agcount
;
785 mp
->m_blkbit_log
= sbp
->sb_blocklog
+ XFS_NBBYLOG
;
786 mp
->m_blkbb_log
= sbp
->sb_blocklog
- BBSHIFT
;
787 mp
->m_sectbb_log
= sbp
->sb_sectlog
- BBSHIFT
;
788 mp
->m_agno_log
= xfs_highbit32(sbp
->sb_agcount
- 1) + 1;
789 mp
->m_agino_log
= sbp
->sb_inopblog
+ sbp
->sb_agblklog
;
790 mp
->m_blockmask
= sbp
->sb_blocksize
- 1;
791 mp
->m_blockwsize
= sbp
->sb_blocksize
>> XFS_WORDLOG
;
792 mp
->m_blockwmask
= mp
->m_blockwsize
- 1;
794 mp
->m_alloc_mxr
[0] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
795 mp
->m_alloc_mxr
[1] = xfs_allocbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
796 mp
->m_alloc_mnr
[0] = mp
->m_alloc_mxr
[0] / 2;
797 mp
->m_alloc_mnr
[1] = mp
->m_alloc_mxr
[1] / 2;
799 mp
->m_inobt_mxr
[0] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
800 mp
->m_inobt_mxr
[1] = xfs_inobt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
801 mp
->m_inobt_mnr
[0] = mp
->m_inobt_mxr
[0] / 2;
802 mp
->m_inobt_mnr
[1] = mp
->m_inobt_mxr
[1] / 2;
804 mp
->m_bmap_dmxr
[0] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 1);
805 mp
->m_bmap_dmxr
[1] = xfs_bmbt_maxrecs(mp
, sbp
->sb_blocksize
, 0);
806 mp
->m_bmap_dmnr
[0] = mp
->m_bmap_dmxr
[0] / 2;
807 mp
->m_bmap_dmnr
[1] = mp
->m_bmap_dmxr
[1] / 2;
809 mp
->m_rmap_mxr
[0] = xfs_rmapbt_maxrecs(sbp
->sb_blocksize
, 1);
810 mp
->m_rmap_mxr
[1] = xfs_rmapbt_maxrecs(sbp
->sb_blocksize
, 0);
811 mp
->m_rmap_mnr
[0] = mp
->m_rmap_mxr
[0] / 2;
812 mp
->m_rmap_mnr
[1] = mp
->m_rmap_mxr
[1] / 2;
814 mp
->m_refc_mxr
[0] = xfs_refcountbt_maxrecs(sbp
->sb_blocksize
, true);
815 mp
->m_refc_mxr
[1] = xfs_refcountbt_maxrecs(sbp
->sb_blocksize
, false);
816 mp
->m_refc_mnr
[0] = mp
->m_refc_mxr
[0] / 2;
817 mp
->m_refc_mnr
[1] = mp
->m_refc_mxr
[1] / 2;
819 mp
->m_bsize
= XFS_FSB_TO_BB(mp
, 1);
820 mp
->m_ialloc_inos
= max_t(uint16_t, XFS_INODES_PER_CHUNK
,
822 mp
->m_ialloc_blks
= mp
->m_ialloc_inos
>> sbp
->sb_inopblog
;
824 if (sbp
->sb_spino_align
)
825 mp
->m_ialloc_min_blks
= sbp
->sb_spino_align
;
827 mp
->m_ialloc_min_blks
= mp
->m_ialloc_blks
;
828 mp
->m_alloc_set_aside
= xfs_alloc_set_aside(mp
);
829 mp
->m_ag_max_usable
= xfs_alloc_ag_max_usable(mp
);
833 * xfs_initialize_perag_data
835 * Read in each per-ag structure so we can count up the number of
836 * allocated inodes, free inodes and used filesystem blocks as this
837 * information is no longer persistent in the superblock. Once we have
838 * this information, write it into the in-core superblock structure.
841 xfs_initialize_perag_data(
842 struct xfs_mount
*mp
,
843 xfs_agnumber_t agcount
)
845 xfs_agnumber_t index
;
847 xfs_sb_t
*sbp
= &mp
->m_sb
;
851 uint64_t bfreelst
= 0;
856 for (index
= 0; index
< agcount
; index
++) {
858 * read the agf, then the agi. This gets us
859 * all the information we need and populates the
860 * per-ag structures for us.
862 error
= xfs_alloc_pagf_init(mp
, NULL
, index
, 0);
866 error
= xfs_ialloc_pagi_init(mp
, NULL
, index
);
869 pag
= xfs_perag_get(mp
, index
);
870 ifree
+= pag
->pagi_freecount
;
871 ialloc
+= pag
->pagi_count
;
872 bfree
+= pag
->pagf_freeblks
;
873 bfreelst
+= pag
->pagf_flcount
;
874 btree
+= pag
->pagf_btreeblks
;
877 fdblocks
= bfree
+ bfreelst
+ btree
;
880 * If the new summary counts are obviously incorrect, fail the
881 * mount operation because that implies the AGFs are also corrupt.
882 * Clear BAD_SUMMARY so that we don't unmount with a dirty log, which
883 * will prevent xfs_repair from fixing anything.
885 if (fdblocks
> sbp
->sb_dblocks
|| ifree
> ialloc
) {
886 xfs_alert(mp
, "AGF corruption. Please run xfs_repair.");
887 error
= -EFSCORRUPTED
;
891 /* Overwrite incore superblock counters with just-read data */
892 spin_lock(&mp
->m_sb_lock
);
893 sbp
->sb_ifree
= ifree
;
894 sbp
->sb_icount
= ialloc
;
895 sbp
->sb_fdblocks
= fdblocks
;
896 spin_unlock(&mp
->m_sb_lock
);
898 xfs_reinit_percpu_counters(mp
);
900 mp
->m_flags
&= ~XFS_MOUNT_BAD_SUMMARY
;
905 * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
906 * into the superblock buffer to be logged. It does not provide the higher
907 * level of locking that is needed to protect the in-core superblock from
912 struct xfs_trans
*tp
)
914 struct xfs_mount
*mp
= tp
->t_mountp
;
915 struct xfs_buf
*bp
= xfs_trans_getsb(tp
, mp
, 0);
917 mp
->m_sb
.sb_icount
= percpu_counter_sum(&mp
->m_icount
);
918 mp
->m_sb
.sb_ifree
= percpu_counter_sum(&mp
->m_ifree
);
919 mp
->m_sb
.sb_fdblocks
= percpu_counter_sum(&mp
->m_fdblocks
);
921 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
);
922 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_SB_BUF
);
923 xfs_trans_log_buf(tp
, bp
, 0, sizeof(struct xfs_dsb
));
929 * Sync the superblock to disk.
931 * Note that the caller is responsible for checking the frozen state of the
932 * filesystem. This procedure uses the non-blocking transaction allocator and
933 * thus will allow modifications to a frozen fs. This is required because this
934 * code can be called during the process of freezing where use of the high-level
935 * allocator would deadlock.
939 struct xfs_mount
*mp
,
942 struct xfs_trans
*tp
;
945 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_sb
, 0, 0,
946 XFS_TRANS_NO_WRITECOUNT
, &tp
);
952 xfs_trans_set_sync(tp
);
953 return xfs_trans_commit(tp
);
957 * Update all the secondary superblocks to match the new state of the primary.
958 * Because we are completely overwriting all the existing fields in the
959 * secondary superblock buffers, there is no need to read them in from disk.
960 * Just get a new buffer, stamp it and write it.
962 * The sb buffers need to be cached here so that we serialise against other
963 * operations that access the secondary superblocks, but we don't want to keep
964 * them in memory once it is written so we mark it as a one-shot buffer.
967 xfs_update_secondary_sbs(
968 struct xfs_mount
*mp
)
973 LIST_HEAD (buffer_list
);
975 /* update secondary superblocks. */
976 for (agno
= 1; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
979 bp
= xfs_buf_get(mp
->m_ddev_targp
,
980 XFS_AG_DADDR(mp
, agno
, XFS_SB_DADDR
),
981 XFS_FSS_TO_BB(mp
, 1), 0);
983 * If we get an error reading or writing alternate superblocks,
984 * continue. xfs_repair chooses the "best" superblock based
985 * on most matches; if we break early, we'll leave more
986 * superblocks un-updated than updated, and xfs_repair may
987 * pick them over the properly-updated primary.
991 "error allocating secondary superblock for ag %d",
994 saved_error
= -ENOMEM
;
998 bp
->b_ops
= &xfs_sb_buf_ops
;
1000 xfs_buf_zero(bp
, 0, BBTOB(bp
->b_length
));
1001 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp
), &mp
->m_sb
);
1002 xfs_buf_delwri_queue(bp
, &buffer_list
);
1005 /* don't hold too many buffers at once */
1009 error
= xfs_buf_delwri_submit(&buffer_list
);
1012 "write error %d updating a secondary superblock near ag %d",
1015 saved_error
= error
;
1019 error
= xfs_buf_delwri_submit(&buffer_list
);
1022 "write error %d updating a secondary superblock near ag %d",
1026 return saved_error
? saved_error
: error
;
1030 * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1031 * also writes the superblock buffer to disk sector 0 immediately.
1035 struct xfs_mount
*mp
)
1037 struct xfs_trans
*tp
;
1041 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_sb
, 0, 0, 0, &tp
);
1045 bp
= xfs_trans_getsb(tp
, mp
, 0);
1047 xfs_trans_bhold(tp
, bp
);
1048 xfs_trans_set_sync(tp
);
1049 error
= xfs_trans_commit(tp
);
1053 * write out the sb buffer to get the changes to disk
1055 error
= xfs_bwrite(bp
);
1064 struct xfs_fsop_geom
*geo
,
1067 memset(geo
, 0, sizeof(struct xfs_fsop_geom
));
1069 geo
->blocksize
= sbp
->sb_blocksize
;
1070 geo
->rtextsize
= sbp
->sb_rextsize
;
1071 geo
->agblocks
= sbp
->sb_agblocks
;
1072 geo
->agcount
= sbp
->sb_agcount
;
1073 geo
->logblocks
= sbp
->sb_logblocks
;
1074 geo
->sectsize
= sbp
->sb_sectsize
;
1075 geo
->inodesize
= sbp
->sb_inodesize
;
1076 geo
->imaxpct
= sbp
->sb_imax_pct
;
1077 geo
->datablocks
= sbp
->sb_dblocks
;
1078 geo
->rtblocks
= sbp
->sb_rblocks
;
1079 geo
->rtextents
= sbp
->sb_rextents
;
1080 geo
->logstart
= sbp
->sb_logstart
;
1081 BUILD_BUG_ON(sizeof(geo
->uuid
) != sizeof(sbp
->sb_uuid
));
1082 memcpy(geo
->uuid
, &sbp
->sb_uuid
, sizeof(sbp
->sb_uuid
));
1084 if (struct_version
< 2)
1087 geo
->sunit
= sbp
->sb_unit
;
1088 geo
->swidth
= sbp
->sb_width
;
1090 if (struct_version
< 3)
1093 geo
->version
= XFS_FSOP_GEOM_VERSION
;
1094 geo
->flags
= XFS_FSOP_GEOM_FLAGS_NLINK
|
1095 XFS_FSOP_GEOM_FLAGS_DIRV2
|
1096 XFS_FSOP_GEOM_FLAGS_EXTFLG
;
1097 if (xfs_sb_version_hasattr(sbp
))
1098 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_ATTR
;
1099 if (xfs_sb_version_hasquota(sbp
))
1100 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_QUOTA
;
1101 if (xfs_sb_version_hasalign(sbp
))
1102 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_IALIGN
;
1103 if (xfs_sb_version_hasdalign(sbp
))
1104 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_DALIGN
;
1105 if (xfs_sb_version_hassector(sbp
))
1106 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_SECTOR
;
1107 if (xfs_sb_version_hasasciici(sbp
))
1108 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_DIRV2CI
;
1109 if (xfs_sb_version_haslazysbcount(sbp
))
1110 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_LAZYSB
;
1111 if (xfs_sb_version_hasattr2(sbp
))
1112 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_ATTR2
;
1113 if (xfs_sb_version_hasprojid32bit(sbp
))
1114 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_PROJID32
;
1115 if (xfs_sb_version_hascrc(sbp
))
1116 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_V5SB
;
1117 if (xfs_sb_version_hasftype(sbp
))
1118 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_FTYPE
;
1119 if (xfs_sb_version_hasfinobt(sbp
))
1120 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_FINOBT
;
1121 if (xfs_sb_version_hassparseinodes(sbp
))
1122 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_SPINODES
;
1123 if (xfs_sb_version_hasrmapbt(sbp
))
1124 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_RMAPBT
;
1125 if (xfs_sb_version_hasreflink(sbp
))
1126 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_REFLINK
;
1127 if (xfs_sb_version_hassector(sbp
))
1128 geo
->logsectsize
= sbp
->sb_logsectsize
;
1130 geo
->logsectsize
= BBSIZE
;
1131 geo
->rtsectsize
= sbp
->sb_blocksize
;
1132 geo
->dirblocksize
= xfs_dir2_dirblock_bytes(sbp
);
1134 if (struct_version
< 4)
1137 if (xfs_sb_version_haslogv2(sbp
))
1138 geo
->flags
|= XFS_FSOP_GEOM_FLAGS_LOGV2
;
1140 geo
->logsunit
= sbp
->sb_logsunit
;
1145 /* Read a secondary superblock. */
1147 xfs_sb_read_secondary(
1148 struct xfs_mount
*mp
,
1149 struct xfs_trans
*tp
,
1150 xfs_agnumber_t agno
,
1151 struct xfs_buf
**bpp
)
1156 ASSERT(agno
!= 0 && agno
!= NULLAGNUMBER
);
1157 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
1158 XFS_AG_DADDR(mp
, agno
, XFS_SB_BLOCK(mp
)),
1159 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, &xfs_sb_buf_ops
);
1162 xfs_buf_set_ref(bp
, XFS_SSB_REF
);
1167 /* Get an uninitialised secondary superblock buffer. */
1169 xfs_sb_get_secondary(
1170 struct xfs_mount
*mp
,
1171 struct xfs_trans
*tp
,
1172 xfs_agnumber_t agno
,
1173 struct xfs_buf
**bpp
)
1177 ASSERT(agno
!= 0 && agno
!= NULLAGNUMBER
);
1178 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
1179 XFS_AG_DADDR(mp
, agno
, XFS_SB_BLOCK(mp
)),
1180 XFS_FSS_TO_BB(mp
, 1), 0);
1183 bp
->b_ops
= &xfs_sb_buf_ops
;
1184 xfs_buf_oneshot(bp
);