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_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_rtalloc.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_priv.h"
24 #include "xfs_log_priv.h"
25 #include "xfs_error.h"
26 #include "xfs_quota.h"
27 #include "xfs_fsops.h"
28 #include "xfs_icache.h"
29 #include "xfs_sysfs.h"
30 #include "xfs_rmap_btree.h"
31 #include "xfs_refcount_btree.h"
32 #include "xfs_reflink.h"
33 #include "xfs_extent_busy.h"
34 #include "xfs_health.h"
35 #include "xfs_trace.h"
37 #include "xfs_rtbitmap.h"
38 #include "xfs_metafile.h"
39 #include "xfs_rtgroup.h"
40 #include "scrub/stats.h"
42 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
43 static int xfs_uuid_table_size
;
44 static uuid_t
*xfs_uuid_table
;
47 xfs_uuid_table_free(void)
49 if (xfs_uuid_table_size
== 0)
51 kfree(xfs_uuid_table
);
52 xfs_uuid_table
= NULL
;
53 xfs_uuid_table_size
= 0;
57 * See if the UUID is unique among mounted XFS filesystems.
58 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
64 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
67 /* Publish UUID in struct super_block */
68 super_set_uuid(mp
->m_super
, uuid
->b
, sizeof(*uuid
));
70 if (xfs_has_nouuid(mp
))
73 if (uuid_is_null(uuid
)) {
74 xfs_warn(mp
, "Filesystem has null UUID - can't mount");
78 mutex_lock(&xfs_uuid_table_mutex
);
79 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
80 if (uuid_is_null(&xfs_uuid_table
[i
])) {
84 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
89 xfs_uuid_table
= krealloc(xfs_uuid_table
,
90 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
91 GFP_KERNEL
| __GFP_NOFAIL
);
92 hole
= xfs_uuid_table_size
++;
94 xfs_uuid_table
[hole
] = *uuid
;
95 mutex_unlock(&xfs_uuid_table_mutex
);
100 mutex_unlock(&xfs_uuid_table_mutex
);
101 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
107 struct xfs_mount
*mp
)
109 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
112 if (xfs_has_nouuid(mp
))
115 mutex_lock(&xfs_uuid_table_mutex
);
116 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
117 if (uuid_is_null(&xfs_uuid_table
[i
]))
119 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
121 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
124 ASSERT(i
< xfs_uuid_table_size
);
125 mutex_unlock(&xfs_uuid_table_mutex
);
129 * Check size of device based on the (data/realtime) block count.
130 * Note: this check is used by the growfs code as well as mount.
133 xfs_sb_validate_fsb_count(
139 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
141 if (check_shl_overflow(nblocks
, sbp
->sb_blocklog
, &max_bytes
))
144 /* Limited by ULONG_MAX of page cache index */
145 if (max_bytes
>> PAGE_SHIFT
> ULONG_MAX
)
153 * Does the initial read of the superblock.
157 struct xfs_mount
*mp
,
160 unsigned int sector_size
;
162 struct xfs_sb
*sbp
= &mp
->m_sb
;
164 int loud
= !(flags
& XFS_MFSI_QUIET
);
165 const struct xfs_buf_ops
*buf_ops
;
167 ASSERT(mp
->m_sb_bp
== NULL
);
168 ASSERT(mp
->m_ddev_targp
!= NULL
);
171 * For the initial read, we must guess at the sector
172 * size based on the block device. It's enough to
173 * get the sb_sectsize out of the superblock and
174 * then reread with the proper length.
175 * We don't verify it yet, because it may not be complete.
177 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
181 * Allocate a (locked) buffer to hold the superblock. This will be kept
182 * around at all times to optimize access to the superblock. Therefore,
183 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
187 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
188 BTOBB(sector_size
), XBF_NO_IOACCT
, &bp
,
192 xfs_warn(mp
, "SB validate failed with error %d.", error
);
193 /* bad CRC means corrupted metadata */
194 if (error
== -EFSBADCRC
)
195 error
= -EFSCORRUPTED
;
200 * Initialize the mount structure from the superblock.
202 xfs_sb_from_disk(sbp
, bp
->b_addr
);
205 * If we haven't validated the superblock, do so now before we try
206 * to check the sector size and reread the superblock appropriately.
208 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
210 xfs_warn(mp
, "Invalid superblock magic number");
216 * We must be able to do sector-sized and sector-aligned IO.
218 if (sector_size
> sbp
->sb_sectsize
) {
220 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
221 sector_size
, sbp
->sb_sectsize
);
226 if (buf_ops
== NULL
) {
228 * Re-read the superblock so the buffer is correctly sized,
229 * and properly verified.
232 sector_size
= sbp
->sb_sectsize
;
233 buf_ops
= loud
? &xfs_sb_buf_ops
: &xfs_sb_quiet_buf_ops
;
237 mp
->m_features
|= xfs_sb_version_to_features(sbp
);
238 xfs_reinit_percpu_counters(mp
);
241 * If logged xattrs are enabled after log recovery finishes, then set
242 * the opstate so that log recovery will work properly.
244 if (xfs_sb_version_haslogxattrs(&mp
->m_sb
))
245 xfs_set_using_logged_xattrs(mp
);
247 /* no need to be quiet anymore, so reset the buf ops */
248 bp
->b_ops
= &xfs_sb_buf_ops
;
260 * If the sunit/swidth change would move the precomputed root inode value, we
261 * must reject the ondisk change because repair will stumble over that.
262 * However, we allow the mount to proceed because we never rejected this
263 * combination before. Returns true to update the sb, false otherwise.
266 xfs_check_new_dalign(
267 struct xfs_mount
*mp
,
271 struct xfs_sb
*sbp
= &mp
->m_sb
;
274 calc_ino
= xfs_ialloc_calc_rootino(mp
, new_dalign
);
275 trace_xfs_check_new_dalign(mp
, new_dalign
, calc_ino
);
277 if (sbp
->sb_rootino
== calc_ino
) {
283 "Cannot change stripe alignment; would require moving root inode.");
286 * XXX: Next time we add a new incompat feature, this should start
287 * returning -EINVAL to fail the mount. Until then, spit out a warning
288 * that we're ignoring the administrator's instructions.
290 xfs_warn(mp
, "Skipping superblock stripe alignment update.");
296 * If we were provided with new sunit/swidth values as mount options, make sure
297 * that they pass basic alignment and superblock feature checks, and convert
298 * them into the same units (FSB) that everything else expects. This step
299 * /must/ be done before computing the inode geometry.
302 xfs_validate_new_dalign(
303 struct xfs_mount
*mp
)
305 if (mp
->m_dalign
== 0)
309 * If stripe unit and stripe width are not multiples
310 * of the fs blocksize turn off alignment.
312 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
313 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
315 "alignment check failed: sunit/swidth vs. blocksize(%d)",
316 mp
->m_sb
.sb_blocksize
);
321 * Convert the stripe unit and width to FSBs.
323 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
324 if (mp
->m_dalign
&& (mp
->m_sb
.sb_agblocks
% mp
->m_dalign
)) {
326 "alignment check failed: sunit/swidth vs. agsize(%d)",
327 mp
->m_sb
.sb_agblocks
);
333 "alignment check failed: sunit(%d) less than bsize(%d)",
334 mp
->m_dalign
, mp
->m_sb
.sb_blocksize
);
338 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
340 if (!xfs_has_dalign(mp
)) {
342 "cannot change alignment: superblock does not support data alignment");
349 /* Update alignment values based on mount options and sb values. */
351 xfs_update_alignment(
352 struct xfs_mount
*mp
)
354 struct xfs_sb
*sbp
= &mp
->m_sb
;
360 if (sbp
->sb_unit
== mp
->m_dalign
&&
361 sbp
->sb_width
== mp
->m_swidth
)
364 error
= xfs_check_new_dalign(mp
, mp
->m_dalign
, &update_sb
);
365 if (error
|| !update_sb
)
368 sbp
->sb_unit
= mp
->m_dalign
;
369 sbp
->sb_width
= mp
->m_swidth
;
370 mp
->m_update_sb
= true;
371 } else if (!xfs_has_noalign(mp
) && xfs_has_dalign(mp
)) {
372 mp
->m_dalign
= sbp
->sb_unit
;
373 mp
->m_swidth
= sbp
->sb_width
;
380 * precalculate the low space thresholds for dynamic speculative preallocation.
383 xfs_set_low_space_thresholds(
384 struct xfs_mount
*mp
)
386 uint64_t dblocks
= mp
->m_sb
.sb_dblocks
;
387 uint64_t rtexts
= mp
->m_sb
.sb_rextents
;
390 do_div(dblocks
, 100);
393 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
394 mp
->m_low_space
[i
] = dblocks
* (i
+ 1);
395 mp
->m_low_rtexts
[i
] = rtexts
* (i
+ 1);
400 * Check that the data (and log if separate) is an ok size.
404 struct xfs_mount
*mp
)
410 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
411 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
412 xfs_warn(mp
, "filesystem size mismatch detected");
415 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
416 d
- XFS_FSS_TO_BB(mp
, 1),
417 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, NULL
);
419 xfs_warn(mp
, "last sector read failed");
424 if (mp
->m_logdev_targp
== mp
->m_ddev_targp
)
427 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
428 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
429 xfs_warn(mp
, "log size mismatch detected");
432 error
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
433 d
- XFS_FSB_TO_BB(mp
, 1),
434 XFS_FSB_TO_BB(mp
, 1), 0, &bp
, NULL
);
436 xfs_warn(mp
, "log device read failed");
444 * Clear the quotaflags in memory and in the superblock.
447 xfs_mount_reset_sbqflags(
448 struct xfs_mount
*mp
)
452 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
453 if (mp
->m_sb
.sb_qflags
== 0)
455 spin_lock(&mp
->m_sb_lock
);
456 mp
->m_sb
.sb_qflags
= 0;
457 spin_unlock(&mp
->m_sb_lock
);
459 if (!xfs_fs_writable(mp
, SB_FREEZE_WRITE
))
462 return xfs_sync_sb(mp
, false);
466 xfs_default_resblks(xfs_mount_t
*mp
)
471 * We default to 5% or 8192 fsbs of space reserved, whichever is
472 * smaller. This is intended to cover concurrent allocation
473 * transactions when we initially hit enospc. These each require a 4
474 * block reservation. Hence by default we cover roughly 2000 concurrent
475 * allocation reservations.
477 resblks
= mp
->m_sb
.sb_dblocks
;
479 resblks
= min_t(uint64_t, resblks
, 8192);
483 /* Ensure the summary counts are correct. */
485 xfs_check_summary_counts(
486 struct xfs_mount
*mp
)
491 * The AG0 superblock verifier rejects in-progress filesystems,
492 * so we should never see the flag set this far into mounting.
494 if (mp
->m_sb
.sb_inprogress
) {
495 xfs_err(mp
, "sb_inprogress set after log recovery??");
497 return -EFSCORRUPTED
;
501 * Now the log is mounted, we know if it was an unclean shutdown or
502 * not. If it was, with the first phase of recovery has completed, we
503 * have consistent AG blocks on disk. We have not recovered EFIs yet,
504 * but they are recovered transactionally in the second recovery phase
507 * If the log was clean when we mounted, we can check the summary
508 * counters. If any of them are obviously incorrect, we can recompute
509 * them from the AGF headers in the next step.
511 if (xfs_is_clean(mp
) &&
512 (mp
->m_sb
.sb_fdblocks
> mp
->m_sb
.sb_dblocks
||
513 !xfs_verify_icount(mp
, mp
->m_sb
.sb_icount
) ||
514 mp
->m_sb
.sb_ifree
> mp
->m_sb
.sb_icount
))
515 xfs_fs_mark_sick(mp
, XFS_SICK_FS_COUNTERS
);
518 * We can safely re-initialise incore superblock counters from the
519 * per-ag data. These may not be correct if the filesystem was not
520 * cleanly unmounted, so we waited for recovery to finish before doing
523 * If the filesystem was cleanly unmounted or the previous check did
524 * not flag anything weird, then we can trust the values in the
525 * superblock to be correct and we don't need to do anything here.
526 * Otherwise, recalculate the summary counters.
528 if ((xfs_has_lazysbcount(mp
) && !xfs_is_clean(mp
)) ||
529 xfs_fs_has_sickness(mp
, XFS_SICK_FS_COUNTERS
)) {
530 error
= xfs_initialize_perag_data(mp
, mp
->m_sb
.sb_agcount
);
536 * Older kernels misused sb_frextents to reflect both incore
537 * reservations made by running transactions and the actual count of
538 * free rt extents in the ondisk metadata. Transactions committed
539 * during runtime can therefore contain a superblock update that
540 * undercounts the number of free rt extents tracked in the rt bitmap.
541 * A clean unmount record will have the correct frextents value since
542 * there can be no other transactions running at that point.
544 * If we're mounting the rt volume after recovering the log, recompute
545 * frextents from the rtbitmap file to fix the inconsistency.
547 if (xfs_has_realtime(mp
) && !xfs_is_clean(mp
)) {
548 error
= xfs_rtalloc_reinit_frextents(mp
);
558 struct xfs_mount
*mp
)
560 if (xfs_is_shutdown(mp
))
563 if (percpu_counter_sum(&mp
->m_ifree
) >
564 percpu_counter_sum(&mp
->m_icount
)) {
565 xfs_alert(mp
, "ifree/icount mismatch at unmount");
566 xfs_fs_mark_sick(mp
, XFS_SICK_FS_COUNTERS
);
571 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
572 * internal inode structures can be sitting in the CIL and AIL at this point,
573 * so we need to unpin them, write them back and/or reclaim them before unmount
574 * can proceed. In other words, callers are required to have inactivated all
577 * An inode cluster that has been freed can have its buffer still pinned in
578 * memory because the transaction is still sitting in a iclog. The stale inodes
579 * on that buffer will be pinned to the buffer until the transaction hits the
580 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
581 * may never see the pinned buffer, so nothing will push out the iclog and
584 * Hence we need to force the log to unpin everything first. However, log
585 * forces don't wait for the discards they issue to complete, so we have to
586 * explicitly wait for them to complete here as well.
588 * Then we can tell the world we are unmounting so that error handling knows
589 * that the filesystem is going away and we should error out anything that we
590 * have been retrying in the background. This will prevent never-ending
591 * retries in AIL pushing from hanging the unmount.
593 * Finally, we can push the AIL to clean all the remaining dirty objects, then
594 * reclaim the remaining inodes that are still in memory at this point in time.
597 xfs_unmount_flush_inodes(
598 struct xfs_mount
*mp
)
600 xfs_log_force(mp
, XFS_LOG_SYNC
);
601 xfs_extent_busy_wait_all(mp
);
602 flush_workqueue(xfs_discard_wq
);
604 xfs_set_unmounting(mp
);
606 xfs_ail_push_all_sync(mp
->m_ail
);
607 xfs_inodegc_stop(mp
);
608 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
609 xfs_reclaim_inodes(mp
);
610 xfs_health_unmount(mp
);
614 xfs_mount_setup_inode_geom(
615 struct xfs_mount
*mp
)
617 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
619 igeo
->attr_fork_offset
= xfs_bmap_compute_attr_offset(mp
);
620 ASSERT(igeo
->attr_fork_offset
< XFS_LITINO(mp
));
622 xfs_ialloc_setup_geometry(mp
);
625 /* Mount the metadata directory tree root. */
627 xfs_mount_setup_metadir(
628 struct xfs_mount
*mp
)
632 /* Load the metadata directory root inode into memory. */
633 error
= xfs_metafile_iget(mp
, mp
->m_sb
.sb_metadirino
, XFS_METAFILE_DIR
,
636 xfs_warn(mp
, "Failed to load metadir root directory, error %d",
641 /* Compute maximum possible height for per-AG btree types for this fs. */
643 xfs_agbtree_compute_maxlevels(
644 struct xfs_mount
*mp
)
648 levels
= max(mp
->m_alloc_maxlevels
, M_IGEO(mp
)->inobt_maxlevels
);
649 levels
= max(levels
, mp
->m_rmap_maxlevels
);
650 mp
->m_agbtree_maxlevels
= max(levels
, mp
->m_refc_maxlevels
);
654 * This function does the following on an initial mount of a file system:
655 * - reads the superblock from disk and init the mount struct
656 * - if we're a 32-bit kernel, do a size check on the superblock
657 * so we don't mount terabyte filesystems
658 * - init mount struct realtime fields
659 * - allocate inode hash table for fs
660 * - init directory manager
661 * - perform recovery and init the log manager
665 struct xfs_mount
*mp
)
667 struct xfs_sb
*sbp
= &(mp
->m_sb
);
668 struct xfs_inode
*rip
;
669 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
674 xfs_sb_mount_common(mp
, sbp
);
677 * Check for a mismatched features2 values. Older kernels read & wrote
678 * into the wrong sb offset for sb_features2 on some platforms due to
679 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
680 * which made older superblock reading/writing routines swap it as a
683 * For backwards compatibility, we make both slots equal.
685 * If we detect a mismatched field, we OR the set bits into the existing
686 * features2 field in case it has already been modified; we don't want
687 * to lose any features. We then update the bad location with the ORed
688 * value so that older kernels will see any features2 flags. The
689 * superblock writeback code ensures the new sb_features2 is copied to
690 * sb_bad_features2 before it is logged or written to disk.
692 if (xfs_sb_has_mismatched_features2(sbp
)) {
693 xfs_warn(mp
, "correcting sb_features alignment problem");
694 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
695 mp
->m_update_sb
= true;
699 /* always use v2 inodes by default now */
700 if (!(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_NLINKBIT
)) {
701 mp
->m_sb
.sb_versionnum
|= XFS_SB_VERSION_NLINKBIT
;
702 mp
->m_features
|= XFS_FEAT_NLINK
;
703 mp
->m_update_sb
= true;
707 * If we were given new sunit/swidth options, do some basic validation
708 * checks and convert the incore dalign and swidth values to the
709 * same units (FSB) that everything else uses. This /must/ happen
710 * before computing the inode geometry.
712 error
= xfs_validate_new_dalign(mp
);
716 xfs_alloc_compute_maxlevels(mp
);
717 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
718 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
719 xfs_mount_setup_inode_geom(mp
);
720 xfs_rmapbt_compute_maxlevels(mp
);
721 xfs_refcountbt_compute_maxlevels(mp
);
723 xfs_agbtree_compute_maxlevels(mp
);
726 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
727 * is NOT aligned turn off m_dalign since allocator alignment is within
728 * an ag, therefore ag has to be aligned at stripe boundary. Note that
729 * we must compute the free space and rmap btree geometry before doing
732 error
= xfs_update_alignment(mp
);
736 /* enable fail_at_unmount as default */
737 mp
->m_fail_unmount
= true;
739 super_set_sysfs_name_id(mp
->m_super
);
741 error
= xfs_sysfs_init(&mp
->m_kobj
, &xfs_mp_ktype
,
742 NULL
, mp
->m_super
->s_id
);
746 error
= xfs_sysfs_init(&mp
->m_stats
.xs_kobj
, &xfs_stats_ktype
,
747 &mp
->m_kobj
, "stats");
749 goto out_remove_sysfs
;
751 xchk_stats_register(mp
->m_scrub_stats
, mp
->m_debugfs
);
753 error
= xfs_error_sysfs_init(mp
);
755 goto out_remove_scrub_stats
;
757 error
= xfs_errortag_init(mp
);
759 goto out_remove_error_sysfs
;
761 error
= xfs_uuid_mount(mp
);
763 goto out_remove_errortag
;
766 * Update the preferred write size based on the information from the
767 * on-disk superblock.
769 mp
->m_allocsize_log
=
770 max_t(uint32_t, sbp
->sb_blocklog
, mp
->m_allocsize_log
);
771 mp
->m_allocsize_blocks
= 1U << (mp
->m_allocsize_log
- sbp
->sb_blocklog
);
773 /* set the low space thresholds for dynamic preallocation */
774 xfs_set_low_space_thresholds(mp
);
777 * If enabled, sparse inode chunk alignment is expected to match the
778 * cluster size. Full inode chunk alignment must match the chunk size,
779 * but that is checked on sb read verification...
781 if (xfs_has_sparseinodes(mp
) &&
782 mp
->m_sb
.sb_spino_align
!=
783 XFS_B_TO_FSBT(mp
, igeo
->inode_cluster_size_raw
)) {
785 "Sparse inode block alignment (%u) must match cluster size (%llu).",
786 mp
->m_sb
.sb_spino_align
,
787 XFS_B_TO_FSBT(mp
, igeo
->inode_cluster_size_raw
));
789 goto out_remove_uuid
;
793 * Check that the data (and log if separate) is an ok size.
795 error
= xfs_check_sizes(mp
);
797 goto out_remove_uuid
;
800 * Initialize realtime fields in the mount structure
802 error
= xfs_rtmount_init(mp
);
804 xfs_warn(mp
, "RT mount failed");
805 goto out_remove_uuid
;
809 * Copies the low order bits of the timestamp and the randomly
810 * set "sequence" number out of a UUID.
813 (get_unaligned_be16(&sbp
->sb_uuid
.b
[8]) << 16) |
814 get_unaligned_be16(&sbp
->sb_uuid
.b
[4]);
815 mp
->m_fixedfsid
[1] = get_unaligned_be32(&sbp
->sb_uuid
.b
[0]);
817 error
= xfs_da_mount(mp
);
819 xfs_warn(mp
, "Failed dir/attr init: %d", error
);
820 goto out_remove_uuid
;
824 * Initialize the precomputed transaction reservations values.
829 * Allocate and initialize the per-ag data.
831 error
= xfs_initialize_perag(mp
, 0, sbp
->sb_agcount
,
832 mp
->m_sb
.sb_dblocks
, &mp
->m_maxagi
);
834 xfs_warn(mp
, "Failed per-ag init: %d", error
);
838 error
= xfs_initialize_rtgroups(mp
, 0, sbp
->sb_rgcount
,
839 mp
->m_sb
.sb_rextents
);
841 xfs_warn(mp
, "Failed rtgroup init: %d", error
);
845 if (XFS_IS_CORRUPT(mp
, !sbp
->sb_logblocks
)) {
846 xfs_warn(mp
, "no log defined");
847 error
= -EFSCORRUPTED
;
848 goto out_free_rtgroup
;
851 error
= xfs_inodegc_register_shrinker(mp
);
856 * If we're resuming quota status, pick up the preliminary qflags from
857 * the ondisk superblock so that we know if we should recover dquots.
859 if (xfs_is_resuming_quotaon(mp
))
860 xfs_qm_resume_quotaon(mp
);
863 * Log's mount-time initialization. The first part of recovery can place
864 * some items on the AIL, to be handled when recovery is finished or
867 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
868 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
869 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
871 xfs_warn(mp
, "log mount failed");
872 goto out_inodegc_shrinker
;
876 * If we're resuming quota status and recovered the log, re-sample the
877 * qflags from the ondisk superblock now that we've recovered it, just
878 * in case someone shut down enforcement just before a crash.
880 if (xfs_clear_resuming_quotaon(mp
) && xlog_recovery_needed(mp
->m_log
))
881 xfs_qm_resume_quotaon(mp
);
884 * If logged xattrs are still enabled after log recovery finishes, then
885 * they'll be available until unmount. Otherwise, turn them off.
887 if (xfs_sb_version_haslogxattrs(&mp
->m_sb
))
888 xfs_set_using_logged_xattrs(mp
);
890 xfs_clear_using_logged_xattrs(mp
);
892 /* Enable background inode inactivation workers. */
893 xfs_inodegc_start(mp
);
894 xfs_blockgc_start(mp
);
897 * Now that we've recovered any pending superblock feature bit
898 * additions, we can finish setting up the attr2 behaviour for the
899 * mount. The noattr2 option overrides the superblock flag, so only
900 * check the superblock feature flag if the mount option is not set.
902 if (xfs_has_noattr2(mp
)) {
903 mp
->m_features
&= ~XFS_FEAT_ATTR2
;
904 } else if (!xfs_has_attr2(mp
) &&
905 (mp
->m_sb
.sb_features2
& XFS_SB_VERSION2_ATTR2BIT
)) {
906 mp
->m_features
|= XFS_FEAT_ATTR2
;
909 if (xfs_has_metadir(mp
)) {
910 error
= xfs_mount_setup_metadir(mp
);
912 goto out_free_metadir
;
916 * Get and sanity-check the root inode.
917 * Save the pointer to it in the mount structure.
919 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, XFS_IGET_UNTRUSTED
,
920 XFS_ILOCK_EXCL
, &rip
);
923 "Failed to read root inode 0x%llx, error %d",
924 sbp
->sb_rootino
, -error
);
925 goto out_free_metadir
;
930 if (XFS_IS_CORRUPT(mp
, !S_ISDIR(VFS_I(rip
)->i_mode
))) {
931 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
932 (unsigned long long)rip
->i_ino
);
933 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
934 error
= -EFSCORRUPTED
;
937 mp
->m_rootip
= rip
; /* save it */
939 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
942 * Initialize realtime inode pointers in the mount structure
944 error
= xfs_rtmount_inodes(mp
);
947 * Free up the root inode.
949 xfs_warn(mp
, "failed to read RT inodes");
953 /* Make sure the summary counts are ok. */
954 error
= xfs_check_summary_counts(mp
);
959 * If this is a read-only mount defer the superblock updates until
960 * the next remount into writeable mode. Otherwise we would never
961 * perform the update e.g. for the root filesystem.
963 if (mp
->m_update_sb
&& !xfs_is_readonly(mp
)) {
964 error
= xfs_sync_sb(mp
, false);
966 xfs_warn(mp
, "failed to write sb changes");
972 * Initialise the XFS quota management subsystem for this mount
974 if (XFS_IS_QUOTA_ON(mp
)) {
975 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
980 * If a file system had quotas running earlier, but decided to
981 * mount without -o uquota/pquota/gquota options, revoke the
982 * quotachecked license.
984 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
985 xfs_notice(mp
, "resetting quota flags");
986 error
= xfs_mount_reset_sbqflags(mp
);
993 * Finish recovering the file system. This part needed to be delayed
994 * until after the root and real-time bitmap inodes were consistently
995 * read in. Temporarily create per-AG space reservations for metadata
996 * btree shape changes because space freeing transactions (for inode
997 * inactivation) require the per-AG reservation in lieu of reserving
1000 error
= xfs_fs_reserve_ag_blocks(mp
);
1001 if (error
&& error
== -ENOSPC
)
1003 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
1004 error
= xfs_log_mount_finish(mp
);
1005 xfs_fs_unreserve_ag_blocks(mp
);
1007 xfs_warn(mp
, "log mount finish failed");
1012 * Now the log is fully replayed, we can transition to full read-only
1013 * mode for read-only mounts. This will sync all the metadata and clean
1014 * the log so that the recovery we just performed does not have to be
1015 * replayed again on the next mount.
1017 * We use the same quiesce mechanism as the rw->ro remount, as they are
1018 * semantically identical operations.
1020 if (xfs_is_readonly(mp
) && !xfs_has_norecovery(mp
))
1024 * Complete the quota initialisation, post-log-replay component.
1027 ASSERT(mp
->m_qflags
== 0);
1028 mp
->m_qflags
= quotaflags
;
1030 xfs_qm_mount_quotas(mp
);
1034 * Now we are mounted, reserve a small amount of unused space for
1035 * privileged transactions. This is needed so that transaction
1036 * space required for critical operations can dip into this pool
1037 * when at ENOSPC. This is needed for operations like create with
1038 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1039 * are not allowed to use this reserved space.
1041 * This may drive us straight to ENOSPC on mount, but that implies
1042 * we were already there on the last unmount. Warn if this occurs.
1044 if (!xfs_is_readonly(mp
)) {
1045 error
= xfs_reserve_blocks(mp
, xfs_default_resblks(mp
));
1048 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1050 /* Reserve AG blocks for future btree expansion. */
1051 error
= xfs_fs_reserve_ag_blocks(mp
);
1052 if (error
&& error
!= -ENOSPC
)
1059 xfs_fs_unreserve_ag_blocks(mp
);
1060 xfs_qm_unmount_quotas(mp
);
1062 xfs_rtunmount_inodes(mp
);
1065 /* Clean out dquots that might be in memory after quotacheck. */
1068 if (mp
->m_metadirip
)
1069 xfs_irele(mp
->m_metadirip
);
1072 * Inactivate all inodes that might still be in memory after a log
1073 * intent recovery failure so that reclaim can free them. Metadata
1074 * inodes and the root directory shouldn't need inactivation, but the
1075 * mount failed for some reason, so pull down all the state and flee.
1077 xfs_inodegc_flush(mp
);
1080 * Flush all inode reclamation work and flush the log.
1081 * We have to do this /after/ rtunmount and qm_unmount because those
1082 * two will have scheduled delayed reclaim for the rt/quota inodes.
1084 * This is slightly different from the unmountfs call sequence
1085 * because we could be tearing down a partially set up mount. In
1086 * particular, if log_mount_finish fails we bail out without calling
1087 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1090 xfs_unmount_flush_inodes(mp
);
1091 xfs_log_mount_cancel(mp
);
1092 out_inodegc_shrinker
:
1093 shrinker_free(mp
->m_inodegc_shrinker
);
1095 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1096 xfs_buftarg_drain(mp
->m_logdev_targp
);
1097 xfs_buftarg_drain(mp
->m_ddev_targp
);
1099 xfs_free_rtgroups(mp
, 0, mp
->m_sb
.sb_rgcount
);
1101 xfs_free_perag_range(mp
, 0, mp
->m_sb
.sb_agcount
);
1105 xfs_uuid_unmount(mp
);
1106 out_remove_errortag
:
1107 xfs_errortag_del(mp
);
1108 out_remove_error_sysfs
:
1109 xfs_error_sysfs_del(mp
);
1110 out_remove_scrub_stats
:
1111 xchk_stats_unregister(mp
->m_scrub_stats
);
1112 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1114 xfs_sysfs_del(&mp
->m_kobj
);
1120 * This flushes out the inodes,dquots and the superblock, unmounts the
1121 * log and makes sure that incore structures are freed.
1125 struct xfs_mount
*mp
)
1130 * Perform all on-disk metadata updates required to inactivate inodes
1131 * that the VFS evicted earlier in the unmount process. Freeing inodes
1132 * and discarding CoW fork preallocations can cause shape changes to
1133 * the free inode and refcount btrees, respectively, so we must finish
1134 * this before we discard the metadata space reservations. Metadata
1135 * inodes and the root directory do not require inactivation.
1137 xfs_inodegc_flush(mp
);
1139 xfs_blockgc_stop(mp
);
1140 xfs_fs_unreserve_ag_blocks(mp
);
1141 xfs_qm_unmount_quotas(mp
);
1142 xfs_rtunmount_inodes(mp
);
1143 xfs_irele(mp
->m_rootip
);
1144 if (mp
->m_metadirip
)
1145 xfs_irele(mp
->m_metadirip
);
1147 xfs_unmount_flush_inodes(mp
);
1152 * Unreserve any blocks we have so that when we unmount we don't account
1153 * the reserved free space as used. This is really only necessary for
1154 * lazy superblock counting because it trusts the incore superblock
1155 * counters to be absolutely correct on clean unmount.
1157 * We don't bother correcting this elsewhere for lazy superblock
1158 * counting because on mount of an unclean filesystem we reconstruct the
1159 * correct counter value and this is irrelevant.
1161 * For non-lazy counter filesystems, this doesn't matter at all because
1162 * we only every apply deltas to the superblock and hence the incore
1163 * value does not matter....
1165 error
= xfs_reserve_blocks(mp
, 0);
1167 xfs_warn(mp
, "Unable to free reserved block pool. "
1168 "Freespace may not be correct on next mount.");
1169 xfs_unmount_check(mp
);
1172 * Indicate that it's ok to clear log incompat bits before cleaning
1173 * the log and writing the unmount record.
1175 xfs_set_done_with_log_incompat(mp
);
1176 xfs_log_unmount(mp
);
1178 xfs_uuid_unmount(mp
);
1181 xfs_errortag_clearall(mp
);
1183 shrinker_free(mp
->m_inodegc_shrinker
);
1184 xfs_free_rtgroups(mp
, 0, mp
->m_sb
.sb_rgcount
);
1185 xfs_free_perag_range(mp
, 0, mp
->m_sb
.sb_agcount
);
1186 xfs_errortag_del(mp
);
1187 xfs_error_sysfs_del(mp
);
1188 xchk_stats_unregister(mp
->m_scrub_stats
);
1189 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1190 xfs_sysfs_del(&mp
->m_kobj
);
1194 * Determine whether modifications can proceed. The caller specifies the minimum
1195 * freeze level for which modifications should not be allowed. This allows
1196 * certain operations to proceed while the freeze sequence is in progress, if
1201 struct xfs_mount
*mp
,
1204 ASSERT(level
> SB_UNFROZEN
);
1205 if ((mp
->m_super
->s_writers
.frozen
>= level
) ||
1206 xfs_is_shutdown(mp
) || xfs_is_readonly(mp
))
1213 xfs_add_freecounter(
1214 struct xfs_mount
*mp
,
1215 struct percpu_counter
*counter
,
1218 bool has_resv_pool
= (counter
== &mp
->m_fdblocks
);
1222 * If the reserve pool is depleted, put blocks back into it first.
1223 * Most of the time the pool is full.
1225 if (!has_resv_pool
|| mp
->m_resblks
== mp
->m_resblks_avail
) {
1226 percpu_counter_add(counter
, delta
);
1230 spin_lock(&mp
->m_sb_lock
);
1231 res_used
= mp
->m_resblks
- mp
->m_resblks_avail
;
1232 if (res_used
> delta
) {
1233 mp
->m_resblks_avail
+= delta
;
1236 mp
->m_resblks_avail
= mp
->m_resblks
;
1237 percpu_counter_add(counter
, delta
);
1239 spin_unlock(&mp
->m_sb_lock
);
1243 xfs_dec_freecounter(
1244 struct xfs_mount
*mp
,
1245 struct percpu_counter
*counter
,
1250 uint64_t set_aside
= 0;
1254 ASSERT(counter
== &mp
->m_fdblocks
|| counter
== &mp
->m_frextents
);
1255 has_resv_pool
= (counter
== &mp
->m_fdblocks
);
1257 ASSERT(has_resv_pool
);
1260 * Taking blocks away, need to be more accurate the closer we
1263 * If the counter has a value of less than 2 * max batch size,
1264 * then make everything serialise as we are real close to
1267 if (__percpu_counter_compare(counter
, 2 * XFS_FDBLOCKS_BATCH
,
1268 XFS_FDBLOCKS_BATCH
) < 0)
1271 batch
= XFS_FDBLOCKS_BATCH
;
1274 * Set aside allocbt blocks because these blocks are tracked as free
1275 * space but not available for allocation. Technically this means that a
1276 * single reservation cannot consume all remaining free space, but the
1277 * ratio of allocbt blocks to usable free blocks should be rather small.
1278 * The tradeoff without this is that filesystems that maintain high
1279 * perag block reservations can over reserve physical block availability
1280 * and fail physical allocation, which leads to much more serious
1281 * problems (i.e. transaction abort, pagecache discards, etc.) than
1282 * slightly premature -ENOSPC.
1285 set_aside
= xfs_fdblocks_unavailable(mp
);
1286 percpu_counter_add_batch(counter
, -((int64_t)delta
), batch
);
1287 if (__percpu_counter_compare(counter
, set_aside
,
1288 XFS_FDBLOCKS_BATCH
) >= 0) {
1294 * lock up the sb for dipping into reserves before releasing the space
1295 * that took us to ENOSPC.
1297 spin_lock(&mp
->m_sb_lock
);
1298 percpu_counter_add(counter
, delta
);
1299 if (!has_resv_pool
|| !rsvd
)
1300 goto fdblocks_enospc
;
1302 lcounter
= (long long)mp
->m_resblks_avail
- delta
;
1303 if (lcounter
>= 0) {
1304 mp
->m_resblks_avail
= lcounter
;
1305 spin_unlock(&mp
->m_sb_lock
);
1309 "Reserve blocks depleted! Consider increasing reserve pool size.");
1312 spin_unlock(&mp
->m_sb_lock
);
1317 * Used to free the superblock along various error paths.
1321 struct xfs_mount
*mp
)
1323 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1331 * If the underlying (data/log/rt) device is readonly, there are some
1332 * operations that cannot proceed.
1335 xfs_dev_is_read_only(
1336 struct xfs_mount
*mp
,
1339 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1340 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1341 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1342 xfs_notice(mp
, "%s required on read-only device.", message
);
1343 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1349 /* Force the summary counters to be recalculated at next mount. */
1351 xfs_force_summary_recalc(
1352 struct xfs_mount
*mp
)
1354 if (!xfs_has_lazysbcount(mp
))
1357 xfs_fs_mark_sick(mp
, XFS_SICK_FS_COUNTERS
);
1361 * Enable a log incompat feature flag in the primary superblock. The caller
1362 * cannot have any other transactions in progress.
1365 xfs_add_incompat_log_feature(
1366 struct xfs_mount
*mp
,
1369 struct xfs_dsb
*dsb
;
1372 ASSERT(hweight32(feature
) == 1);
1373 ASSERT(!(feature
& XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
1376 * Force the log to disk and kick the background AIL thread to reduce
1377 * the chances that the bwrite will stall waiting for the AIL to unpin
1378 * the primary superblock buffer. This isn't a data integrity
1379 * operation, so we don't need a synchronous push.
1381 error
= xfs_log_force(mp
, XFS_LOG_SYNC
);
1384 xfs_ail_push_all(mp
->m_ail
);
1387 * Lock the primary superblock buffer to serialize all callers that
1388 * are trying to set feature bits.
1390 xfs_buf_lock(mp
->m_sb_bp
);
1391 xfs_buf_hold(mp
->m_sb_bp
);
1393 if (xfs_is_shutdown(mp
)) {
1398 if (xfs_sb_has_incompat_log_feature(&mp
->m_sb
, feature
))
1402 * Write the primary superblock to disk immediately, because we need
1403 * the log_incompat bit to be set in the primary super now to protect
1404 * the log items that we're going to commit later.
1406 dsb
= mp
->m_sb_bp
->b_addr
;
1407 xfs_sb_to_disk(dsb
, &mp
->m_sb
);
1408 dsb
->sb_features_log_incompat
|= cpu_to_be32(feature
);
1409 error
= xfs_bwrite(mp
->m_sb_bp
);
1414 * Add the feature bits to the incore superblock before we unlock the
1417 xfs_sb_add_incompat_log_features(&mp
->m_sb
, feature
);
1418 xfs_buf_relse(mp
->m_sb_bp
);
1420 /* Log the superblock to disk. */
1421 return xfs_sync_sb(mp
, false);
1423 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1425 xfs_buf_relse(mp
->m_sb_bp
);
1430 * Clear all the log incompat flags from the superblock.
1432 * The caller cannot be in a transaction, must ensure that the log does not
1433 * contain any log items protected by any log incompat bit, and must ensure
1434 * that there are no other threads that depend on the state of the log incompat
1435 * feature flags in the primary super.
1437 * Returns true if the superblock is dirty.
1440 xfs_clear_incompat_log_features(
1441 struct xfs_mount
*mp
)
1445 if (!xfs_has_crc(mp
) ||
1446 !xfs_sb_has_incompat_log_feature(&mp
->m_sb
,
1447 XFS_SB_FEAT_INCOMPAT_LOG_ALL
) ||
1448 xfs_is_shutdown(mp
) ||
1449 !xfs_is_done_with_log_incompat(mp
))
1453 * Update the incore superblock. We synchronize on the primary super
1454 * buffer lock to be consistent with the add function, though at least
1455 * in theory this shouldn't be necessary.
1457 xfs_buf_lock(mp
->m_sb_bp
);
1458 xfs_buf_hold(mp
->m_sb_bp
);
1460 if (xfs_sb_has_incompat_log_feature(&mp
->m_sb
,
1461 XFS_SB_FEAT_INCOMPAT_LOG_ALL
)) {
1462 xfs_sb_remove_incompat_log_features(&mp
->m_sb
);
1466 xfs_buf_relse(mp
->m_sb_bp
);
1471 * Update the in-core delayed block counter.
1473 * We prefer to update the counter without having to take a spinlock for every
1474 * counter update (i.e. batching). Each change to delayed allocation
1475 * reservations can change can easily exceed the default percpu counter
1476 * batching, so we use a larger batch factor here.
1478 * Note that we don't currently have any callers requiring fast summation
1479 * (e.g. percpu_counter_read) so we can use a big batch value here.
1481 #define XFS_DELALLOC_BATCH (4096)
1484 struct xfs_inode
*ip
,
1488 struct xfs_mount
*mp
= ip
->i_mount
;
1490 if (XFS_IS_REALTIME_INODE(ip
)) {
1491 percpu_counter_add_batch(&mp
->m_delalloc_rtextents
,
1492 xfs_blen_to_rtbxlen(mp
, data_delta
),
1493 XFS_DELALLOC_BATCH
);
1498 percpu_counter_add_batch(&mp
->m_delalloc_blks
, data_delta
+ ind_delta
,
1499 XFS_DELALLOC_BATCH
);