1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 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"
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_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_cksum.h"
22 #include "xfs_trans.h"
23 #include "xfs_buf_item.h"
24 #include "xfs_bmap_btree.h"
25 #include "xfs_alloc_btree.h"
26 #include "xfs_ialloc_btree.h"
28 #include "xfs_rmap_btree.h"
30 #include "xfs_refcount_btree.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_health.h"
36 * Physical superblock buffer manipulations. Shared with libxfs in userspace.
40 * Reference counting access wrappers to the perag structures.
41 * Because we never free per-ag structures, the only thing we
42 * have to protect against changes is the tree structure itself.
49 struct xfs_perag *pag;
53 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
55 ASSERT(atomic_read(&pag->pag_ref) >= 0);
56 ref = atomic_inc_return(&pag->pag_ref);
59 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
64 * search from @first to find the next perag with the given tag set.
72 struct xfs_perag *pag;
77 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
78 (void **)&pag, first, 1, tag);
83 ref = atomic_inc_return(&pag->pag_ref);
85 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
91 struct xfs_perag *pag)
95 ASSERT(atomic_read(&pag->pag_ref) > 0);
96 ref = atomic_dec_return(&pag->pag_ref);
97 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
100 /* Check all the superblock fields we care about when reading one in. */
102 xfs_validate_sb_read(
103 struct xfs_mount *mp,
106 if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_5)
110 * Version 5 superblock feature mask validation. Reject combinations
111 * the kernel cannot support up front before checking anything else.
113 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
115 "Superblock has unknown compatible features (0x%x) enabled.",
116 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
118 "Using a more recent kernel is recommended.");
121 if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
123 "Superblock has unknown read-only compatible features (0x%x) enabled.",
124 (sbp->sb_features_ro_compat &
125 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
126 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
128 "Attempted to mount read-only compatible filesystem read-write.");
130 "Filesystem can only be safely mounted read only.");
135 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
137 "Superblock has unknown incompatible features (0x%x) enabled.",
138 (sbp->sb_features_incompat &
139 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
141 "Filesystem cannot be safely mounted by this kernel.");
148 /* Check all the superblock fields we care about when writing one out. */
150 xfs_validate_sb_write(
151 struct xfs_mount *mp,
156 * Carry out additional sb summary counter sanity checks when we write
157 * the superblock. We skip this in the read validator because there
158 * could be newer superblocks in the log and if the values are garbage
159 * even after replay we'll recalculate them at the end of log mount.
161 * mkfs has traditionally written zeroed counters to inprogress and
162 * secondary superblocks, so allow this usage to continue because
163 * we never read counters from such superblocks.
165 if (XFS_BUF_ADDR(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
166 (sbp->sb_fdblocks > sbp->sb_dblocks ||
167 !xfs_verify_icount(mp, sbp->sb_icount) ||
168 sbp->sb_ifree > sbp->sb_icount)) {
169 xfs_warn(mp, "SB summary counter sanity check failed");
170 return -EFSCORRUPTED;
173 if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_5)
177 * Version 5 superblock feature mask validation. Reject combinations
178 * the kernel cannot support since we checked for unsupported bits in
179 * the read verifier, which means that memory is corrupt.
181 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
183 "Corruption detected in superblock compatible features (0x%x)!",
184 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
185 return -EFSCORRUPTED;
188 if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
190 "Corruption detected in superblock read-only compatible features (0x%x)!",
191 (sbp->sb_features_ro_compat &
192 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
193 return -EFSCORRUPTED;
195 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
197 "Corruption detected in superblock incompatible features (0x%x)!",
198 (sbp->sb_features_incompat &
199 XFS_SB_FEAT_INCOMPAT_UNKNOWN));
200 return -EFSCORRUPTED;
202 if (xfs_sb_has_incompat_log_feature(sbp,
203 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
205 "Corruption detected in superblock incompatible log features (0x%x)!",
206 (sbp->sb_features_log_incompat &
207 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
208 return -EFSCORRUPTED;
212 * We can't read verify the sb LSN because the read verifier is called
213 * before the log is allocated and processed. We know the log is set up
214 * before write verifier calls, so check it here.
216 if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
217 return -EFSCORRUPTED;
222 /* Check the validity of the SB. */
224 xfs_validate_sb_common(
225 struct xfs_mount *mp,
229 struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp);
230 uint32_t agcount = 0;
233 if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
234 xfs_warn(mp, "bad magic number");
238 if (!xfs_sb_good_version(sbp)) {
239 xfs_warn(mp, "bad version");
243 if (xfs_sb_version_has_pquotino(sbp)) {
244 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
246 "Version 5 of Super block has XFS_OQUOTA bits.");
247 return -EFSCORRUPTED;
249 } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
250 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
252 "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.");
253 return -EFSCORRUPTED;
257 * Full inode chunks must be aligned to inode chunk size when
258 * sparse inodes are enabled to support the sparse chunk
259 * allocation algorithm and prevent overlapping inode records.
261 if (xfs_sb_version_hassparseinodes(sbp)) {
264 align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
266 if (sbp->sb_inoalignmt != align) {
268 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
269 sbp->sb_inoalignmt, align);
275 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
277 "filesystem is marked as having an external log; "
278 "specify logdev on the mount command line.");
283 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
285 "filesystem is marked as having an internal log; "
286 "do not specify logdev on the mount command line.");
290 /* Compute agcount for this number of dblocks and agblocks */
291 if (sbp->sb_agblocks) {
292 agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
298 * More sanity checking. Most of these were stolen directly from
302 sbp->sb_agcount <= 0 ||
303 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
304 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
305 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
306 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
307 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
308 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
309 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
310 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
311 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
312 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
313 sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
314 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
315 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
316 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
317 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
318 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
319 sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE ||
320 sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
321 XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES ||
322 XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES ||
323 sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1 ||
324 agcount == 0 || agcount != sbp->sb_agcount ||
325 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
326 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
327 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
328 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
329 sbp->sb_dblocks == 0 ||
330 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
331 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp) ||
332 sbp->sb_shared_vn != 0)) {
333 xfs_notice(mp, "SB sanity check failed");
334 return -EFSCORRUPTED;
338 if (!xfs_sb_version_hasdalign(sbp) ||
339 sbp->sb_unit > sbp->sb_width ||
340 (sbp->sb_width % sbp->sb_unit) != 0) {
341 xfs_notice(mp, "SB stripe unit sanity check failed");
342 return -EFSCORRUPTED;
344 } else if (xfs_sb_version_hasdalign(sbp)) {
345 xfs_notice(mp, "SB stripe alignment sanity check failed");
346 return -EFSCORRUPTED;
347 } else if (sbp->sb_width) {
348 xfs_notice(mp, "SB stripe width sanity check failed");
349 return -EFSCORRUPTED;
353 if (xfs_sb_version_hascrc(&mp->m_sb) &&
354 sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
355 xfs_notice(mp, "v5 SB sanity check failed");
356 return -EFSCORRUPTED;
360 * Until this is fixed only page-sized or smaller data blocks work.
362 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
364 "File system with blocksize %d bytes. "
365 "Only pagesize (%ld) or less will currently work.",
366 sbp->sb_blocksize, PAGE_SIZE);
371 * Currently only very few inode sizes are supported.
373 switch (sbp->sb_inodesize) {
380 xfs_warn(mp, "inode size of %d bytes not supported",
385 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
386 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
388 "file system too large to be mounted on this system.");
393 * Don't touch the filesystem if a user tool thinks it owns the primary
394 * superblock. mkfs doesn't clear the flag from secondary supers, so
395 * we don't check them at all.
397 if (XFS_BUF_ADDR(bp) == XFS_SB_DADDR && sbp->sb_inprogress) {
398 xfs_warn(mp, "Offline file system operation in progress!");
399 return -EFSCORRUPTED;
405 xfs_sb_quota_from_disk(struct xfs_sb *sbp)
408 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
409 * leads to in-core values having two different values for a quota
410 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
413 * Note that this change affect only the in-core values. These
414 * values are not written back to disk unless any quota information
415 * is written to the disk. Even in that case, sb_pquotino field is
416 * not written to disk unless the superblock supports pquotino.
418 if (sbp->sb_uquotino == 0)
419 sbp->sb_uquotino = NULLFSINO;
420 if (sbp->sb_gquotino == 0)
421 sbp->sb_gquotino = NULLFSINO;
422 if (sbp->sb_pquotino == 0)
423 sbp->sb_pquotino = NULLFSINO;
426 * We need to do these manipilations only if we are working
427 * with an older version of on-disk superblock.
429 if (xfs_sb_version_has_pquotino(sbp))
432 if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
433 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
434 XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
435 if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
436 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
437 XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
438 sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
440 if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
441 sbp->sb_gquotino != NULLFSINO) {
443 * In older version of superblock, on-disk superblock only
444 * has sb_gquotino, and in-core superblock has both sb_gquotino
445 * and sb_pquotino. But, only one of them is supported at any
446 * point of time. So, if PQUOTA is set in disk superblock,
447 * copy over sb_gquotino to sb_pquotino. The NULLFSINO test
448 * above is to make sure we don't do this twice and wipe them
451 sbp->sb_pquotino = sbp->sb_gquotino;
452 sbp->sb_gquotino = NULLFSINO;
462 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
463 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
464 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
465 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
466 to->sb_rextents = be64_to_cpu(from->sb_rextents);
467 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
468 to->sb_logstart = be64_to_cpu(from->sb_logstart);
469 to->sb_rootino = be64_to_cpu(from->sb_rootino);
470 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
471 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
472 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
473 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
474 to->sb_agcount = be32_to_cpu(from->sb_agcount);
475 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
476 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
477 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
478 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
479 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
480 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
481 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
482 to->sb_blocklog = from->sb_blocklog;
483 to->sb_sectlog = from->sb_sectlog;
484 to->sb_inodelog = from->sb_inodelog;
485 to->sb_inopblog = from->sb_inopblog;
486 to->sb_agblklog = from->sb_agblklog;
487 to->sb_rextslog = from->sb_rextslog;
488 to->sb_inprogress = from->sb_inprogress;
489 to->sb_imax_pct = from->sb_imax_pct;
490 to->sb_icount = be64_to_cpu(from->sb_icount);
491 to->sb_ifree = be64_to_cpu(from->sb_ifree);
492 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
493 to->sb_frextents = be64_to_cpu(from->sb_frextents);
494 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
495 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
496 to->sb_qflags = be16_to_cpu(from->sb_qflags);
497 to->sb_flags = from->sb_flags;
498 to->sb_shared_vn = from->sb_shared_vn;
499 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
500 to->sb_unit = be32_to_cpu(from->sb_unit);
501 to->sb_width = be32_to_cpu(from->sb_width);
502 to->sb_dirblklog = from->sb_dirblklog;
503 to->sb_logsectlog = from->sb_logsectlog;
504 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
505 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
506 to->sb_features2 = be32_to_cpu(from->sb_features2);
507 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
508 to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
509 to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
510 to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
511 to->sb_features_log_incompat =
512 be32_to_cpu(from->sb_features_log_incompat);
513 /* crc is only used on disk, not in memory; just init to 0 here. */
515 to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
516 to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
517 to->sb_lsn = be64_to_cpu(from->sb_lsn);
519 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
520 * feature flag is set; if not set we keep it only in memory.
522 if (xfs_sb_version_hasmetauuid(to))
523 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
525 uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
526 /* Convert on-disk flags to in-memory flags? */
528 xfs_sb_quota_from_disk(to);
536 __xfs_sb_from_disk(to, from, true);
540 xfs_sb_quota_to_disk(
544 uint16_t qflags = from->sb_qflags;
546 to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
547 if (xfs_sb_version_has_pquotino(from)) {
548 to->sb_qflags = cpu_to_be16(from->sb_qflags);
549 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
550 to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
555 * The in-core version of sb_qflags do not have XFS_OQUOTA_*
556 * flags, whereas the on-disk version does. So, convert incore
557 * XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
559 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
560 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
562 if (from->sb_qflags &
563 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
564 qflags |= XFS_OQUOTA_ENFD;
565 if (from->sb_qflags &
566 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
567 qflags |= XFS_OQUOTA_CHKD;
568 to->sb_qflags = cpu_to_be16(qflags);
571 * GQUOTINO and PQUOTINO cannot be used together in versions
572 * of superblock that do not have pquotino. from->sb_flags
573 * tells us which quota is active and should be copied to
574 * disk. If neither are active, we should NULL the inode.
576 * In all cases, the separate pquotino must remain 0 because it
577 * it beyond the "end" of the valid non-pquotino superblock.
579 if (from->sb_qflags & XFS_GQUOTA_ACCT)
580 to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
581 else if (from->sb_qflags & XFS_PQUOTA_ACCT)
582 to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
585 * We can't rely on just the fields being logged to tell us
586 * that it is safe to write NULLFSINO - we should only do that
587 * if quotas are not actually enabled. Hence only write
588 * NULLFSINO if both in-core quota inodes are NULL.
590 if (from->sb_gquotino == NULLFSINO &&
591 from->sb_pquotino == NULLFSINO)
592 to->sb_gquotino = cpu_to_be64(NULLFSINO);
603 xfs_sb_quota_to_disk(to, from);
605 to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
606 to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
607 to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
608 to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
609 to->sb_rextents = cpu_to_be64(from->sb_rextents);
610 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
611 to->sb_logstart = cpu_to_be64(from->sb_logstart);
612 to->sb_rootino = cpu_to_be64(from->sb_rootino);
613 to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
614 to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
615 to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
616 to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
617 to->sb_agcount = cpu_to_be32(from->sb_agcount);
618 to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
619 to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
620 to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
621 to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
622 to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
623 to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
624 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
625 to->sb_blocklog = from->sb_blocklog;
626 to->sb_sectlog = from->sb_sectlog;
627 to->sb_inodelog = from->sb_inodelog;
628 to->sb_inopblog = from->sb_inopblog;
629 to->sb_agblklog = from->sb_agblklog;
630 to->sb_rextslog = from->sb_rextslog;
631 to->sb_inprogress = from->sb_inprogress;
632 to->sb_imax_pct = from->sb_imax_pct;
633 to->sb_icount = cpu_to_be64(from->sb_icount);
634 to->sb_ifree = cpu_to_be64(from->sb_ifree);
635 to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
636 to->sb_frextents = cpu_to_be64(from->sb_frextents);
638 to->sb_flags = from->sb_flags;
639 to->sb_shared_vn = from->sb_shared_vn;
640 to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
641 to->sb_unit = cpu_to_be32(from->sb_unit);
642 to->sb_width = cpu_to_be32(from->sb_width);
643 to->sb_dirblklog = from->sb_dirblklog;
644 to->sb_logsectlog = from->sb_logsectlog;
645 to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
646 to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
649 * We need to ensure that bad_features2 always matches features2.
650 * Hence we enforce that here rather than having to remember to do it
651 * everywhere else that updates features2.
653 from->sb_bad_features2 = from->sb_features2;
654 to->sb_features2 = cpu_to_be32(from->sb_features2);
655 to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
657 if (xfs_sb_version_hascrc(from)) {
658 to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
659 to->sb_features_ro_compat =
660 cpu_to_be32(from->sb_features_ro_compat);
661 to->sb_features_incompat =
662 cpu_to_be32(from->sb_features_incompat);
663 to->sb_features_log_incompat =
664 cpu_to_be32(from->sb_features_log_incompat);
665 to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
666 to->sb_lsn = cpu_to_be64(from->sb_lsn);
667 if (xfs_sb_version_hasmetauuid(from))
668 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
673 * If the superblock has the CRC feature bit set or the CRC field is non-null,
674 * check that the CRC is valid. We check the CRC field is non-null because a
675 * single bit error could clear the feature bit and unused parts of the
676 * superblock are supposed to be zero. Hence a non-null crc field indicates that
677 * we've potentially lost a feature bit and we should check it anyway.
679 * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
680 * last field in V4 secondary superblocks. So for secondary superblocks,
681 * we are more forgiving, and ignore CRC failures if the primary doesn't
682 * indicate that the fs version is V5.
689 struct xfs_mount *mp = bp->b_target->bt_mount;
690 struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp);
694 * open code the version check to avoid needing to convert the entire
695 * superblock from disk order just to check the version number
697 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
698 (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
702 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
703 /* Only fail bad secondaries on a known V5 filesystem */
704 if (bp->b_bn == XFS_SB_DADDR ||
705 xfs_sb_version_hascrc(&mp->m_sb)) {
713 * Check all the superblock fields. Don't byteswap the xquota flags
714 * because _verify_common checks the on-disk values.
716 __xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
717 error = xfs_validate_sb_common(mp, bp, &sb);
720 error = xfs_validate_sb_read(mp, &sb);
723 if (error == -EFSCORRUPTED || error == -EFSBADCRC)
724 xfs_verifier_error(bp, error, __this_address);
726 xfs_buf_ioerror(bp, error);
730 * We may be probed for a filesystem match, so we may not want to emit
731 * messages when the superblock buffer is not actually an XFS superblock.
732 * If we find an XFS superblock, then run a normal, noisy mount because we are
733 * really going to mount it and want to know about errors.
736 xfs_sb_quiet_read_verify(
739 struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp);
741 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
742 /* XFS filesystem, verify noisily! */
743 xfs_sb_read_verify(bp);
747 xfs_buf_ioerror(bp, -EWRONGFS);
755 struct xfs_mount *mp = bp->b_target->bt_mount;
756 struct xfs_buf_log_item *bip = bp->b_log_item;
760 * Check all the superblock fields. Don't byteswap the xquota flags
761 * because _verify_common checks the on-disk values.
763 __xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp), false);
764 error = xfs_validate_sb_common(mp, bp, &sb);
767 error = xfs_validate_sb_write(mp, bp, &sb);
771 if (!xfs_sb_version_hascrc(&mp->m_sb))
775 XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
777 xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
781 xfs_verifier_error(bp, error, __this_address);
784 const struct xfs_buf_ops xfs_sb_buf_ops = {
786 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
787 .verify_read = xfs_sb_read_verify,
788 .verify_write = xfs_sb_write_verify,
791 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
792 .name = "xfs_sb_quiet",
793 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
794 .verify_read = xfs_sb_quiet_read_verify,
795 .verify_write = xfs_sb_write_verify,
801 * Mount initialization code establishing various mount
802 * fields from the superblock associated with the given
807 struct xfs_mount *mp,
810 mp->m_agfrotor = mp->m_agirotor = 0;
811 mp->m_maxagi = mp->m_sb.sb_agcount;
812 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
813 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
814 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
815 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
816 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
817 mp->m_blockmask = sbp->sb_blocksize - 1;
818 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
819 mp->m_blockwmask = mp->m_blockwsize - 1;
821 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
822 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
823 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
824 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
826 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
827 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
828 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
829 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
831 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
832 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
833 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
834 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
836 mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
837 mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
838 mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
839 mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
841 mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
842 mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
843 mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
844 mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
846 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
847 mp->m_ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
849 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
851 if (sbp->sb_spino_align)
852 mp->m_ialloc_min_blks = sbp->sb_spino_align;
854 mp->m_ialloc_min_blks = mp->m_ialloc_blks;
855 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
856 mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
860 * xfs_initialize_perag_data
862 * Read in each per-ag structure so we can count up the number of
863 * allocated inodes, free inodes and used filesystem blocks as this
864 * information is no longer persistent in the superblock. Once we have
865 * this information, write it into the in-core superblock structure.
868 xfs_initialize_perag_data(
869 struct xfs_mount *mp,
870 xfs_agnumber_t agcount)
872 xfs_agnumber_t index;
874 xfs_sb_t *sbp = &mp->m_sb;
878 uint64_t bfreelst = 0;
883 for (index = 0; index < agcount; index++) {
885 * read the agf, then the agi. This gets us
886 * all the information we need and populates the
887 * per-ag structures for us.
889 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
893 error = xfs_ialloc_pagi_init(mp, NULL, index);
896 pag = xfs_perag_get(mp, index);
897 ifree += pag->pagi_freecount;
898 ialloc += pag->pagi_count;
899 bfree += pag->pagf_freeblks;
900 bfreelst += pag->pagf_flcount;
901 btree += pag->pagf_btreeblks;
904 fdblocks = bfree + bfreelst + btree;
907 * If the new summary counts are obviously incorrect, fail the
908 * mount operation because that implies the AGFs are also corrupt.
909 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
910 * will prevent xfs_repair from fixing anything.
912 if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
913 xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
914 error = -EFSCORRUPTED;
918 /* Overwrite incore superblock counters with just-read data */
919 spin_lock(&mp->m_sb_lock);
920 sbp->sb_ifree = ifree;
921 sbp->sb_icount = ialloc;
922 sbp->sb_fdblocks = fdblocks;
923 spin_unlock(&mp->m_sb_lock);
925 xfs_reinit_percpu_counters(mp);
927 xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
932 * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
933 * into the superblock buffer to be logged. It does not provide the higher
934 * level of locking that is needed to protect the in-core superblock from
939 struct xfs_trans *tp)
941 struct xfs_mount *mp = tp->t_mountp;
942 struct xfs_buf *bp = xfs_trans_getsb(tp, mp, 0);
944 mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
945 mp->m_sb.sb_ifree = percpu_counter_sum(&mp->m_ifree);
946 mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
948 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
949 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
950 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb));
956 * Sync the superblock to disk.
958 * Note that the caller is responsible for checking the frozen state of the
959 * filesystem. This procedure uses the non-blocking transaction allocator and
960 * thus will allow modifications to a frozen fs. This is required because this
961 * code can be called during the process of freezing where use of the high-level
962 * allocator would deadlock.
966 struct xfs_mount *mp,
969 struct xfs_trans *tp;
972 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
973 XFS_TRANS_NO_WRITECOUNT, &tp);
979 xfs_trans_set_sync(tp);
980 return xfs_trans_commit(tp);
984 * Update all the secondary superblocks to match the new state of the primary.
985 * Because we are completely overwriting all the existing fields in the
986 * secondary superblock buffers, there is no need to read them in from disk.
987 * Just get a new buffer, stamp it and write it.
989 * The sb buffers need to be cached here so that we serialise against other
990 * operations that access the secondary superblocks, but we don't want to keep
991 * them in memory once it is written so we mark it as a one-shot buffer.
994 xfs_update_secondary_sbs(
995 struct xfs_mount *mp)
1000 LIST_HEAD (buffer_list);
1002 /* update secondary superblocks. */
1003 for (agno = 1; agno < mp->m_sb.sb_agcount; agno++) {
1006 bp = xfs_buf_get(mp->m_ddev_targp,
1007 XFS_AG_DADDR(mp, agno, XFS_SB_DADDR),
1008 XFS_FSS_TO_BB(mp, 1), 0);
1010 * If we get an error reading or writing alternate superblocks,
1011 * continue. xfs_repair chooses the "best" superblock based
1012 * on most matches; if we break early, we'll leave more
1013 * superblocks un-updated than updated, and xfs_repair may
1014 * pick them over the properly-updated primary.
1018 "error allocating secondary superblock for ag %d",
1021 saved_error = -ENOMEM;
1025 bp->b_ops = &xfs_sb_buf_ops;
1026 xfs_buf_oneshot(bp);
1027 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
1028 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
1029 xfs_buf_delwri_queue(bp, &buffer_list);
1032 /* don't hold too many buffers at once */
1036 error = xfs_buf_delwri_submit(&buffer_list);
1039 "write error %d updating a secondary superblock near ag %d",
1042 saved_error = error;
1046 error = xfs_buf_delwri_submit(&buffer_list);
1049 "write error %d updating a secondary superblock near ag %d",
1053 return saved_error ? saved_error : error;
1057 * Same behavior as xfs_sync_sb, except that it is always synchronous and it
1058 * also writes the superblock buffer to disk sector 0 immediately.
1062 struct xfs_mount *mp)
1064 struct xfs_trans *tp;
1068 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
1072 bp = xfs_trans_getsb(tp, mp, 0);
1074 xfs_trans_bhold(tp, bp);
1075 xfs_trans_set_sync(tp);
1076 error = xfs_trans_commit(tp);
1080 * write out the sb buffer to get the changes to disk
1082 error = xfs_bwrite(bp);
1091 struct xfs_fsop_geom *geo,
1094 memset(geo, 0, sizeof(struct xfs_fsop_geom));
1096 geo->blocksize = sbp->sb_blocksize;
1097 geo->rtextsize = sbp->sb_rextsize;
1098 geo->agblocks = sbp->sb_agblocks;
1099 geo->agcount = sbp->sb_agcount;
1100 geo->logblocks = sbp->sb_logblocks;
1101 geo->sectsize = sbp->sb_sectsize;
1102 geo->inodesize = sbp->sb_inodesize;
1103 geo->imaxpct = sbp->sb_imax_pct;
1104 geo->datablocks = sbp->sb_dblocks;
1105 geo->rtblocks = sbp->sb_rblocks;
1106 geo->rtextents = sbp->sb_rextents;
1107 geo->logstart = sbp->sb_logstart;
1108 BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
1109 memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
1111 if (struct_version < 2)
1114 geo->sunit = sbp->sb_unit;
1115 geo->swidth = sbp->sb_width;
1117 if (struct_version < 3)
1120 geo->version = XFS_FSOP_GEOM_VERSION;
1121 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
1122 XFS_FSOP_GEOM_FLAGS_DIRV2 |
1123 XFS_FSOP_GEOM_FLAGS_EXTFLG;
1124 if (xfs_sb_version_hasattr(sbp))
1125 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
1126 if (xfs_sb_version_hasquota(sbp))
1127 geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
1128 if (xfs_sb_version_hasalign(sbp))
1129 geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
1130 if (xfs_sb_version_hasdalign(sbp))
1131 geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
1132 if (xfs_sb_version_hassector(sbp))
1133 geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
1134 if (xfs_sb_version_hasasciici(sbp))
1135 geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
1136 if (xfs_sb_version_haslazysbcount(sbp))
1137 geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
1138 if (xfs_sb_version_hasattr2(sbp))
1139 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
1140 if (xfs_sb_version_hasprojid32bit(sbp))
1141 geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
1142 if (xfs_sb_version_hascrc(sbp))
1143 geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
1144 if (xfs_sb_version_hasftype(sbp))
1145 geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
1146 if (xfs_sb_version_hasfinobt(sbp))
1147 geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
1148 if (xfs_sb_version_hassparseinodes(sbp))
1149 geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
1150 if (xfs_sb_version_hasrmapbt(sbp))
1151 geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
1152 if (xfs_sb_version_hasreflink(sbp))
1153 geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
1154 if (xfs_sb_version_hassector(sbp))
1155 geo->logsectsize = sbp->sb_logsectsize;
1157 geo->logsectsize = BBSIZE;
1158 geo->rtsectsize = sbp->sb_blocksize;
1159 geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
1161 if (struct_version < 4)
1164 if (xfs_sb_version_haslogv2(sbp))
1165 geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
1167 geo->logsunit = sbp->sb_logsunit;
1169 if (struct_version < 5)
1172 geo->version = XFS_FSOP_GEOM_VERSION_V5;
1175 /* Read a secondary superblock. */
1177 xfs_sb_read_secondary(
1178 struct xfs_mount *mp,
1179 struct xfs_trans *tp,
1180 xfs_agnumber_t agno,
1181 struct xfs_buf **bpp)
1186 ASSERT(agno != 0 && agno != NULLAGNUMBER);
1187 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1188 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1189 XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
1192 xfs_buf_set_ref(bp, XFS_SSB_REF);
1197 /* Get an uninitialised secondary superblock buffer. */
1199 xfs_sb_get_secondary(
1200 struct xfs_mount *mp,
1201 struct xfs_trans *tp,
1202 xfs_agnumber_t agno,
1203 struct xfs_buf **bpp)
1207 ASSERT(agno != 0 && agno != NULLAGNUMBER);
1208 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1209 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
1210 XFS_FSS_TO_BB(mp, 1), 0);
1213 bp->b_ops = &xfs_sb_buf_ops;
1214 xfs_buf_oneshot(bp);