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"
38 static DEFINE_MUTEX(xfs_uuid_table_mutex
);
39 static int xfs_uuid_table_size
;
40 static uuid_t
*xfs_uuid_table
;
43 xfs_uuid_table_free(void)
45 if (xfs_uuid_table_size
== 0)
47 kmem_free(xfs_uuid_table
);
48 xfs_uuid_table
= NULL
;
49 xfs_uuid_table_size
= 0;
53 * See if the UUID is unique among mounted XFS filesystems.
54 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
60 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
63 /* Publish UUID in struct super_block */
64 uuid_copy(&mp
->m_super
->s_uuid
, uuid
);
66 if (xfs_has_nouuid(mp
))
69 if (uuid_is_null(uuid
)) {
70 xfs_warn(mp
, "Filesystem has null UUID - can't mount");
74 mutex_lock(&xfs_uuid_table_mutex
);
75 for (i
= 0, hole
= -1; i
< xfs_uuid_table_size
; i
++) {
76 if (uuid_is_null(&xfs_uuid_table
[i
])) {
80 if (uuid_equal(uuid
, &xfs_uuid_table
[i
]))
85 xfs_uuid_table
= krealloc(xfs_uuid_table
,
86 (xfs_uuid_table_size
+ 1) * sizeof(*xfs_uuid_table
),
87 GFP_KERNEL
| __GFP_NOFAIL
);
88 hole
= xfs_uuid_table_size
++;
90 xfs_uuid_table
[hole
] = *uuid
;
91 mutex_unlock(&xfs_uuid_table_mutex
);
96 mutex_unlock(&xfs_uuid_table_mutex
);
97 xfs_warn(mp
, "Filesystem has duplicate UUID %pU - can't mount", uuid
);
103 struct xfs_mount
*mp
)
105 uuid_t
*uuid
= &mp
->m_sb
.sb_uuid
;
108 if (xfs_has_nouuid(mp
))
111 mutex_lock(&xfs_uuid_table_mutex
);
112 for (i
= 0; i
< xfs_uuid_table_size
; i
++) {
113 if (uuid_is_null(&xfs_uuid_table
[i
]))
115 if (!uuid_equal(uuid
, &xfs_uuid_table
[i
]))
117 memset(&xfs_uuid_table
[i
], 0, sizeof(uuid_t
));
120 ASSERT(i
< xfs_uuid_table_size
);
121 mutex_unlock(&xfs_uuid_table_mutex
);
125 * Check size of device based on the (data/realtime) block count.
126 * Note: this check is used by the growfs code as well as mount.
129 xfs_sb_validate_fsb_count(
133 ASSERT(PAGE_SHIFT
>= sbp
->sb_blocklog
);
134 ASSERT(sbp
->sb_blocklog
>= BBSHIFT
);
136 /* Limited by ULONG_MAX of page cache index */
137 if (nblocks
>> (PAGE_SHIFT
- sbp
->sb_blocklog
) > ULONG_MAX
)
145 * Does the initial read of the superblock.
149 struct xfs_mount
*mp
,
152 unsigned int sector_size
;
154 struct xfs_sb
*sbp
= &mp
->m_sb
;
156 int loud
= !(flags
& XFS_MFSI_QUIET
);
157 const struct xfs_buf_ops
*buf_ops
;
159 ASSERT(mp
->m_sb_bp
== NULL
);
160 ASSERT(mp
->m_ddev_targp
!= NULL
);
163 * For the initial read, we must guess at the sector
164 * size based on the block device. It's enough to
165 * get the sb_sectsize out of the superblock and
166 * then reread with the proper length.
167 * We don't verify it yet, because it may not be complete.
169 sector_size
= xfs_getsize_buftarg(mp
->m_ddev_targp
);
173 * Allocate a (locked) buffer to hold the superblock. This will be kept
174 * around at all times to optimize access to the superblock. Therefore,
175 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
179 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
, XFS_SB_DADDR
,
180 BTOBB(sector_size
), XBF_NO_IOACCT
, &bp
,
184 xfs_warn(mp
, "SB validate failed with error %d.", error
);
185 /* bad CRC means corrupted metadata */
186 if (error
== -EFSBADCRC
)
187 error
= -EFSCORRUPTED
;
192 * Initialize the mount structure from the superblock.
194 xfs_sb_from_disk(sbp
, bp
->b_addr
);
197 * If we haven't validated the superblock, do so now before we try
198 * to check the sector size and reread the superblock appropriately.
200 if (sbp
->sb_magicnum
!= XFS_SB_MAGIC
) {
202 xfs_warn(mp
, "Invalid superblock magic number");
208 * We must be able to do sector-sized and sector-aligned IO.
210 if (sector_size
> sbp
->sb_sectsize
) {
212 xfs_warn(mp
, "device supports %u byte sectors (not %u)",
213 sector_size
, sbp
->sb_sectsize
);
218 if (buf_ops
== NULL
) {
220 * Re-read the superblock so the buffer is correctly sized,
221 * and properly verified.
224 sector_size
= sbp
->sb_sectsize
;
225 buf_ops
= loud
? &xfs_sb_buf_ops
: &xfs_sb_quiet_buf_ops
;
229 mp
->m_features
|= xfs_sb_version_to_features(sbp
);
230 xfs_reinit_percpu_counters(mp
);
232 /* no need to be quiet anymore, so reset the buf ops */
233 bp
->b_ops
= &xfs_sb_buf_ops
;
245 * If the sunit/swidth change would move the precomputed root inode value, we
246 * must reject the ondisk change because repair will stumble over that.
247 * However, we allow the mount to proceed because we never rejected this
248 * combination before. Returns true to update the sb, false otherwise.
251 xfs_check_new_dalign(
252 struct xfs_mount
*mp
,
256 struct xfs_sb
*sbp
= &mp
->m_sb
;
259 calc_ino
= xfs_ialloc_calc_rootino(mp
, new_dalign
);
260 trace_xfs_check_new_dalign(mp
, new_dalign
, calc_ino
);
262 if (sbp
->sb_rootino
== calc_ino
) {
268 "Cannot change stripe alignment; would require moving root inode.");
271 * XXX: Next time we add a new incompat feature, this should start
272 * returning -EINVAL to fail the mount. Until then, spit out a warning
273 * that we're ignoring the administrator's instructions.
275 xfs_warn(mp
, "Skipping superblock stripe alignment update.");
281 * If we were provided with new sunit/swidth values as mount options, make sure
282 * that they pass basic alignment and superblock feature checks, and convert
283 * them into the same units (FSB) that everything else expects. This step
284 * /must/ be done before computing the inode geometry.
287 xfs_validate_new_dalign(
288 struct xfs_mount
*mp
)
290 if (mp
->m_dalign
== 0)
294 * If stripe unit and stripe width are not multiples
295 * of the fs blocksize turn off alignment.
297 if ((BBTOB(mp
->m_dalign
) & mp
->m_blockmask
) ||
298 (BBTOB(mp
->m_swidth
) & mp
->m_blockmask
)) {
300 "alignment check failed: sunit/swidth vs. blocksize(%d)",
301 mp
->m_sb
.sb_blocksize
);
305 * Convert the stripe unit and width to FSBs.
307 mp
->m_dalign
= XFS_BB_TO_FSBT(mp
, mp
->m_dalign
);
308 if (mp
->m_dalign
&& (mp
->m_sb
.sb_agblocks
% mp
->m_dalign
)) {
310 "alignment check failed: sunit/swidth vs. agsize(%d)",
311 mp
->m_sb
.sb_agblocks
);
313 } else if (mp
->m_dalign
) {
314 mp
->m_swidth
= XFS_BB_TO_FSBT(mp
, mp
->m_swidth
);
317 "alignment check failed: sunit(%d) less than bsize(%d)",
318 mp
->m_dalign
, mp
->m_sb
.sb_blocksize
);
323 if (!xfs_has_dalign(mp
)) {
325 "cannot change alignment: superblock does not support data alignment");
332 /* Update alignment values based on mount options and sb values. */
334 xfs_update_alignment(
335 struct xfs_mount
*mp
)
337 struct xfs_sb
*sbp
= &mp
->m_sb
;
343 if (sbp
->sb_unit
== mp
->m_dalign
&&
344 sbp
->sb_width
== mp
->m_swidth
)
347 error
= xfs_check_new_dalign(mp
, mp
->m_dalign
, &update_sb
);
348 if (error
|| !update_sb
)
351 sbp
->sb_unit
= mp
->m_dalign
;
352 sbp
->sb_width
= mp
->m_swidth
;
353 mp
->m_update_sb
= true;
354 } else if (!xfs_has_noalign(mp
) && xfs_has_dalign(mp
)) {
355 mp
->m_dalign
= sbp
->sb_unit
;
356 mp
->m_swidth
= sbp
->sb_width
;
363 * precalculate the low space thresholds for dynamic speculative preallocation.
366 xfs_set_low_space_thresholds(
367 struct xfs_mount
*mp
)
369 uint64_t dblocks
= mp
->m_sb
.sb_dblocks
;
370 uint64_t rtexts
= mp
->m_sb
.sb_rextents
;
373 do_div(dblocks
, 100);
376 for (i
= 0; i
< XFS_LOWSP_MAX
; i
++) {
377 mp
->m_low_space
[i
] = dblocks
* (i
+ 1);
378 mp
->m_low_rtexts
[i
] = rtexts
* (i
+ 1);
383 * Check that the data (and log if separate) is an ok size.
387 struct xfs_mount
*mp
)
393 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
);
394 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_dblocks
) {
395 xfs_warn(mp
, "filesystem size mismatch detected");
398 error
= xfs_buf_read_uncached(mp
->m_ddev_targp
,
399 d
- XFS_FSS_TO_BB(mp
, 1),
400 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, NULL
);
402 xfs_warn(mp
, "last sector read failed");
407 if (mp
->m_logdev_targp
== mp
->m_ddev_targp
)
410 d
= (xfs_daddr_t
)XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_logblocks
);
411 if (XFS_BB_TO_FSB(mp
, d
) != mp
->m_sb
.sb_logblocks
) {
412 xfs_warn(mp
, "log size mismatch detected");
415 error
= xfs_buf_read_uncached(mp
->m_logdev_targp
,
416 d
- XFS_FSB_TO_BB(mp
, 1),
417 XFS_FSB_TO_BB(mp
, 1), 0, &bp
, NULL
);
419 xfs_warn(mp
, "log device read failed");
427 * Clear the quotaflags in memory and in the superblock.
430 xfs_mount_reset_sbqflags(
431 struct xfs_mount
*mp
)
435 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
436 if (mp
->m_sb
.sb_qflags
== 0)
438 spin_lock(&mp
->m_sb_lock
);
439 mp
->m_sb
.sb_qflags
= 0;
440 spin_unlock(&mp
->m_sb_lock
);
442 if (!xfs_fs_writable(mp
, SB_FREEZE_WRITE
))
445 return xfs_sync_sb(mp
, false);
449 xfs_default_resblks(xfs_mount_t
*mp
)
454 * We default to 5% or 8192 fsbs of space reserved, whichever is
455 * smaller. This is intended to cover concurrent allocation
456 * transactions when we initially hit enospc. These each require a 4
457 * block reservation. Hence by default we cover roughly 2000 concurrent
458 * allocation reservations.
460 resblks
= mp
->m_sb
.sb_dblocks
;
462 resblks
= min_t(uint64_t, resblks
, 8192);
466 /* Ensure the summary counts are correct. */
468 xfs_check_summary_counts(
469 struct xfs_mount
*mp
)
474 * The AG0 superblock verifier rejects in-progress filesystems,
475 * so we should never see the flag set this far into mounting.
477 if (mp
->m_sb
.sb_inprogress
) {
478 xfs_err(mp
, "sb_inprogress set after log recovery??");
480 return -EFSCORRUPTED
;
484 * Now the log is mounted, we know if it was an unclean shutdown or
485 * not. If it was, with the first phase of recovery has completed, we
486 * have consistent AG blocks on disk. We have not recovered EFIs yet,
487 * but they are recovered transactionally in the second recovery phase
490 * If the log was clean when we mounted, we can check the summary
491 * counters. If any of them are obviously incorrect, we can recompute
492 * them from the AGF headers in the next step.
494 if (xfs_is_clean(mp
) &&
495 (mp
->m_sb
.sb_fdblocks
> mp
->m_sb
.sb_dblocks
||
496 !xfs_verify_icount(mp
, mp
->m_sb
.sb_icount
) ||
497 mp
->m_sb
.sb_ifree
> mp
->m_sb
.sb_icount
))
498 xfs_fs_mark_sick(mp
, XFS_SICK_FS_COUNTERS
);
501 * We can safely re-initialise incore superblock counters from the
502 * per-ag data. These may not be correct if the filesystem was not
503 * cleanly unmounted, so we waited for recovery to finish before doing
506 * If the filesystem was cleanly unmounted or the previous check did
507 * not flag anything weird, then we can trust the values in the
508 * superblock to be correct and we don't need to do anything here.
509 * Otherwise, recalculate the summary counters.
511 if ((xfs_has_lazysbcount(mp
) && !xfs_is_clean(mp
)) ||
512 xfs_fs_has_sickness(mp
, XFS_SICK_FS_COUNTERS
)) {
513 error
= xfs_initialize_perag_data(mp
, mp
->m_sb
.sb_agcount
);
519 * Older kernels misused sb_frextents to reflect both incore
520 * reservations made by running transactions and the actual count of
521 * free rt extents in the ondisk metadata. Transactions committed
522 * during runtime can therefore contain a superblock update that
523 * undercounts the number of free rt extents tracked in the rt bitmap.
524 * A clean unmount record will have the correct frextents value since
525 * there can be no other transactions running at that point.
527 * If we're mounting the rt volume after recovering the log, recompute
528 * frextents from the rtbitmap file to fix the inconsistency.
530 if (xfs_has_realtime(mp
) && !xfs_is_clean(mp
)) {
531 error
= xfs_rtalloc_reinit_frextents(mp
);
540 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
541 * internal inode structures can be sitting in the CIL and AIL at this point,
542 * so we need to unpin them, write them back and/or reclaim them before unmount
543 * can proceed. In other words, callers are required to have inactivated all
546 * An inode cluster that has been freed can have its buffer still pinned in
547 * memory because the transaction is still sitting in a iclog. The stale inodes
548 * on that buffer will be pinned to the buffer until the transaction hits the
549 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
550 * may never see the pinned buffer, so nothing will push out the iclog and
553 * Hence we need to force the log to unpin everything first. However, log
554 * forces don't wait for the discards they issue to complete, so we have to
555 * explicitly wait for them to complete here as well.
557 * Then we can tell the world we are unmounting so that error handling knows
558 * that the filesystem is going away and we should error out anything that we
559 * have been retrying in the background. This will prevent never-ending
560 * retries in AIL pushing from hanging the unmount.
562 * Finally, we can push the AIL to clean all the remaining dirty objects, then
563 * reclaim the remaining inodes that are still in memory at this point in time.
566 xfs_unmount_flush_inodes(
567 struct xfs_mount
*mp
)
569 xfs_log_force(mp
, XFS_LOG_SYNC
);
570 xfs_extent_busy_wait_all(mp
);
571 flush_workqueue(xfs_discard_wq
);
573 set_bit(XFS_OPSTATE_UNMOUNTING
, &mp
->m_opstate
);
575 xfs_ail_push_all_sync(mp
->m_ail
);
576 xfs_inodegc_stop(mp
);
577 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
578 xfs_reclaim_inodes(mp
);
579 xfs_health_unmount(mp
);
583 xfs_mount_setup_inode_geom(
584 struct xfs_mount
*mp
)
586 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
588 igeo
->attr_fork_offset
= xfs_bmap_compute_attr_offset(mp
);
589 ASSERT(igeo
->attr_fork_offset
< XFS_LITINO(mp
));
591 xfs_ialloc_setup_geometry(mp
);
594 /* Compute maximum possible height for per-AG btree types for this fs. */
596 xfs_agbtree_compute_maxlevels(
597 struct xfs_mount
*mp
)
601 levels
= max(mp
->m_alloc_maxlevels
, M_IGEO(mp
)->inobt_maxlevels
);
602 levels
= max(levels
, mp
->m_rmap_maxlevels
);
603 mp
->m_agbtree_maxlevels
= max(levels
, mp
->m_refc_maxlevels
);
607 * This function does the following on an initial mount of a file system:
608 * - reads the superblock from disk and init the mount struct
609 * - if we're a 32-bit kernel, do a size check on the superblock
610 * so we don't mount terabyte filesystems
611 * - init mount struct realtime fields
612 * - allocate inode hash table for fs
613 * - init directory manager
614 * - perform recovery and init the log manager
618 struct xfs_mount
*mp
)
620 struct xfs_sb
*sbp
= &(mp
->m_sb
);
621 struct xfs_inode
*rip
;
622 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
628 xfs_sb_mount_common(mp
, sbp
);
631 * Check for a mismatched features2 values. Older kernels read & wrote
632 * into the wrong sb offset for sb_features2 on some platforms due to
633 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
634 * which made older superblock reading/writing routines swap it as a
637 * For backwards compatibility, we make both slots equal.
639 * If we detect a mismatched field, we OR the set bits into the existing
640 * features2 field in case it has already been modified; we don't want
641 * to lose any features. We then update the bad location with the ORed
642 * value so that older kernels will see any features2 flags. The
643 * superblock writeback code ensures the new sb_features2 is copied to
644 * sb_bad_features2 before it is logged or written to disk.
646 if (xfs_sb_has_mismatched_features2(sbp
)) {
647 xfs_warn(mp
, "correcting sb_features alignment problem");
648 sbp
->sb_features2
|= sbp
->sb_bad_features2
;
649 mp
->m_update_sb
= true;
653 /* always use v2 inodes by default now */
654 if (!(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_NLINKBIT
)) {
655 mp
->m_sb
.sb_versionnum
|= XFS_SB_VERSION_NLINKBIT
;
656 mp
->m_features
|= XFS_FEAT_NLINK
;
657 mp
->m_update_sb
= true;
661 * If we were given new sunit/swidth options, do some basic validation
662 * checks and convert the incore dalign and swidth values to the
663 * same units (FSB) that everything else uses. This /must/ happen
664 * before computing the inode geometry.
666 error
= xfs_validate_new_dalign(mp
);
670 xfs_alloc_compute_maxlevels(mp
);
671 xfs_bmap_compute_maxlevels(mp
, XFS_DATA_FORK
);
672 xfs_bmap_compute_maxlevels(mp
, XFS_ATTR_FORK
);
673 xfs_mount_setup_inode_geom(mp
);
674 xfs_rmapbt_compute_maxlevels(mp
);
675 xfs_refcountbt_compute_maxlevels(mp
);
677 xfs_agbtree_compute_maxlevels(mp
);
680 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
681 * is NOT aligned turn off m_dalign since allocator alignment is within
682 * an ag, therefore ag has to be aligned at stripe boundary. Note that
683 * we must compute the free space and rmap btree geometry before doing
686 error
= xfs_update_alignment(mp
);
690 /* enable fail_at_unmount as default */
691 mp
->m_fail_unmount
= true;
693 error
= xfs_sysfs_init(&mp
->m_kobj
, &xfs_mp_ktype
,
694 NULL
, mp
->m_super
->s_id
);
698 error
= xfs_sysfs_init(&mp
->m_stats
.xs_kobj
, &xfs_stats_ktype
,
699 &mp
->m_kobj
, "stats");
701 goto out_remove_sysfs
;
703 error
= xfs_error_sysfs_init(mp
);
707 error
= xfs_errortag_init(mp
);
709 goto out_remove_error_sysfs
;
711 error
= xfs_uuid_mount(mp
);
713 goto out_remove_errortag
;
716 * Update the preferred write size based on the information from the
717 * on-disk superblock.
719 mp
->m_allocsize_log
=
720 max_t(uint32_t, sbp
->sb_blocklog
, mp
->m_allocsize_log
);
721 mp
->m_allocsize_blocks
= 1U << (mp
->m_allocsize_log
- sbp
->sb_blocklog
);
723 /* set the low space thresholds for dynamic preallocation */
724 xfs_set_low_space_thresholds(mp
);
727 * If enabled, sparse inode chunk alignment is expected to match the
728 * cluster size. Full inode chunk alignment must match the chunk size,
729 * but that is checked on sb read verification...
731 if (xfs_has_sparseinodes(mp
) &&
732 mp
->m_sb
.sb_spino_align
!=
733 XFS_B_TO_FSBT(mp
, igeo
->inode_cluster_size_raw
)) {
735 "Sparse inode block alignment (%u) must match cluster size (%llu).",
736 mp
->m_sb
.sb_spino_align
,
737 XFS_B_TO_FSBT(mp
, igeo
->inode_cluster_size_raw
));
739 goto out_remove_uuid
;
743 * Check that the data (and log if separate) is an ok size.
745 error
= xfs_check_sizes(mp
);
747 goto out_remove_uuid
;
750 * Initialize realtime fields in the mount structure
752 error
= xfs_rtmount_init(mp
);
754 xfs_warn(mp
, "RT mount failed");
755 goto out_remove_uuid
;
759 * Copies the low order bits of the timestamp and the randomly
760 * set "sequence" number out of a UUID.
763 (get_unaligned_be16(&sbp
->sb_uuid
.b
[8]) << 16) |
764 get_unaligned_be16(&sbp
->sb_uuid
.b
[4]);
765 mp
->m_fixedfsid
[1] = get_unaligned_be32(&sbp
->sb_uuid
.b
[0]);
767 error
= xfs_da_mount(mp
);
769 xfs_warn(mp
, "Failed dir/attr init: %d", error
);
770 goto out_remove_uuid
;
774 * Initialize the precomputed transaction reservations values.
779 * Allocate and initialize the per-ag data.
781 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, mp
->m_sb
.sb_dblocks
,
784 xfs_warn(mp
, "Failed per-ag init: %d", error
);
788 if (XFS_IS_CORRUPT(mp
, !sbp
->sb_logblocks
)) {
789 xfs_warn(mp
, "no log defined");
790 error
= -EFSCORRUPTED
;
794 error
= xfs_inodegc_register_shrinker(mp
);
799 * Log's mount-time initialization. The first part of recovery can place
800 * some items on the AIL, to be handled when recovery is finished or
803 error
= xfs_log_mount(mp
, mp
->m_logdev_targp
,
804 XFS_FSB_TO_DADDR(mp
, sbp
->sb_logstart
),
805 XFS_FSB_TO_BB(mp
, sbp
->sb_logblocks
));
807 xfs_warn(mp
, "log mount failed");
808 goto out_inodegc_shrinker
;
811 /* Enable background inode inactivation workers. */
812 xfs_inodegc_start(mp
);
813 xfs_blockgc_start(mp
);
816 * Now that we've recovered any pending superblock feature bit
817 * additions, we can finish setting up the attr2 behaviour for the
818 * mount. The noattr2 option overrides the superblock flag, so only
819 * check the superblock feature flag if the mount option is not set.
821 if (xfs_has_noattr2(mp
)) {
822 mp
->m_features
&= ~XFS_FEAT_ATTR2
;
823 } else if (!xfs_has_attr2(mp
) &&
824 (mp
->m_sb
.sb_features2
& XFS_SB_VERSION2_ATTR2BIT
)) {
825 mp
->m_features
|= XFS_FEAT_ATTR2
;
829 * Get and sanity-check the root inode.
830 * Save the pointer to it in the mount structure.
832 error
= xfs_iget(mp
, NULL
, sbp
->sb_rootino
, XFS_IGET_UNTRUSTED
,
833 XFS_ILOCK_EXCL
, &rip
);
836 "Failed to read root inode 0x%llx, error %d",
837 sbp
->sb_rootino
, -error
);
838 goto out_log_dealloc
;
843 if (XFS_IS_CORRUPT(mp
, !S_ISDIR(VFS_I(rip
)->i_mode
))) {
844 xfs_warn(mp
, "corrupted root inode %llu: not a directory",
845 (unsigned long long)rip
->i_ino
);
846 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
847 error
= -EFSCORRUPTED
;
850 mp
->m_rootip
= rip
; /* save it */
852 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
855 * Initialize realtime inode pointers in the mount structure
857 error
= xfs_rtmount_inodes(mp
);
860 * Free up the root inode.
862 xfs_warn(mp
, "failed to read RT inodes");
866 /* Make sure the summary counts are ok. */
867 error
= xfs_check_summary_counts(mp
);
872 * If this is a read-only mount defer the superblock updates until
873 * the next remount into writeable mode. Otherwise we would never
874 * perform the update e.g. for the root filesystem.
876 if (mp
->m_update_sb
&& !xfs_is_readonly(mp
)) {
877 error
= xfs_sync_sb(mp
, false);
879 xfs_warn(mp
, "failed to write sb changes");
885 * Initialise the XFS quota management subsystem for this mount
887 if (XFS_IS_QUOTA_ON(mp
)) {
888 error
= xfs_qm_newmount(mp
, "amount
, "aflags
);
893 * If a file system had quotas running earlier, but decided to
894 * mount without -o uquota/pquota/gquota options, revoke the
895 * quotachecked license.
897 if (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_ACCT
) {
898 xfs_notice(mp
, "resetting quota flags");
899 error
= xfs_mount_reset_sbqflags(mp
);
906 * Finish recovering the file system. This part needed to be delayed
907 * until after the root and real-time bitmap inodes were consistently
908 * read in. Temporarily create per-AG space reservations for metadata
909 * btree shape changes because space freeing transactions (for inode
910 * inactivation) require the per-AG reservation in lieu of reserving
913 error
= xfs_fs_reserve_ag_blocks(mp
);
914 if (error
&& error
== -ENOSPC
)
916 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
917 error
= xfs_log_mount_finish(mp
);
918 xfs_fs_unreserve_ag_blocks(mp
);
920 xfs_warn(mp
, "log mount finish failed");
925 * Now the log is fully replayed, we can transition to full read-only
926 * mode for read-only mounts. This will sync all the metadata and clean
927 * the log so that the recovery we just performed does not have to be
928 * replayed again on the next mount.
930 * We use the same quiesce mechanism as the rw->ro remount, as they are
931 * semantically identical operations.
933 if (xfs_is_readonly(mp
) && !xfs_has_norecovery(mp
))
937 * Complete the quota initialisation, post-log-replay component.
940 ASSERT(mp
->m_qflags
== 0);
941 mp
->m_qflags
= quotaflags
;
943 xfs_qm_mount_quotas(mp
);
947 * Now we are mounted, reserve a small amount of unused space for
948 * privileged transactions. This is needed so that transaction
949 * space required for critical operations can dip into this pool
950 * when at ENOSPC. This is needed for operations like create with
951 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
952 * are not allowed to use this reserved space.
954 * This may drive us straight to ENOSPC on mount, but that implies
955 * we were already there on the last unmount. Warn if this occurs.
957 if (!xfs_is_readonly(mp
)) {
958 resblks
= xfs_default_resblks(mp
);
959 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
962 "Unable to allocate reserve blocks. Continuing without reserve pool.");
964 /* Reserve AG blocks for future btree expansion. */
965 error
= xfs_fs_reserve_ag_blocks(mp
);
966 if (error
&& error
!= -ENOSPC
)
973 xfs_fs_unreserve_ag_blocks(mp
);
974 xfs_qm_unmount_quotas(mp
);
976 xfs_rtunmount_inodes(mp
);
979 /* Clean out dquots that might be in memory after quotacheck. */
983 * Inactivate all inodes that might still be in memory after a log
984 * intent recovery failure so that reclaim can free them. Metadata
985 * inodes and the root directory shouldn't need inactivation, but the
986 * mount failed for some reason, so pull down all the state and flee.
988 xfs_inodegc_flush(mp
);
991 * Flush all inode reclamation work and flush the log.
992 * We have to do this /after/ rtunmount and qm_unmount because those
993 * two will have scheduled delayed reclaim for the rt/quota inodes.
995 * This is slightly different from the unmountfs call sequence
996 * because we could be tearing down a partially set up mount. In
997 * particular, if log_mount_finish fails we bail out without calling
998 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1001 xfs_unmount_flush_inodes(mp
);
1003 xfs_log_mount_cancel(mp
);
1004 out_inodegc_shrinker
:
1005 unregister_shrinker(&mp
->m_inodegc_shrinker
);
1007 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
1008 xfs_buftarg_drain(mp
->m_logdev_targp
);
1009 xfs_buftarg_drain(mp
->m_ddev_targp
);
1015 xfs_uuid_unmount(mp
);
1016 out_remove_errortag
:
1017 xfs_errortag_del(mp
);
1018 out_remove_error_sysfs
:
1019 xfs_error_sysfs_del(mp
);
1021 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1023 xfs_sysfs_del(&mp
->m_kobj
);
1029 * This flushes out the inodes,dquots and the superblock, unmounts the
1030 * log and makes sure that incore structures are freed.
1034 struct xfs_mount
*mp
)
1040 * Perform all on-disk metadata updates required to inactivate inodes
1041 * that the VFS evicted earlier in the unmount process. Freeing inodes
1042 * and discarding CoW fork preallocations can cause shape changes to
1043 * the free inode and refcount btrees, respectively, so we must finish
1044 * this before we discard the metadata space reservations. Metadata
1045 * inodes and the root directory do not require inactivation.
1047 xfs_inodegc_flush(mp
);
1049 xfs_blockgc_stop(mp
);
1050 xfs_fs_unreserve_ag_blocks(mp
);
1051 xfs_qm_unmount_quotas(mp
);
1052 xfs_rtunmount_inodes(mp
);
1053 xfs_irele(mp
->m_rootip
);
1055 xfs_unmount_flush_inodes(mp
);
1060 * Unreserve any blocks we have so that when we unmount we don't account
1061 * the reserved free space as used. This is really only necessary for
1062 * lazy superblock counting because it trusts the incore superblock
1063 * counters to be absolutely correct on clean unmount.
1065 * We don't bother correcting this elsewhere for lazy superblock
1066 * counting because on mount of an unclean filesystem we reconstruct the
1067 * correct counter value and this is irrelevant.
1069 * For non-lazy counter filesystems, this doesn't matter at all because
1070 * we only every apply deltas to the superblock and hence the incore
1071 * value does not matter....
1074 error
= xfs_reserve_blocks(mp
, &resblks
, NULL
);
1076 xfs_warn(mp
, "Unable to free reserved block pool. "
1077 "Freespace may not be correct on next mount.");
1079 xfs_log_unmount(mp
);
1081 xfs_uuid_unmount(mp
);
1084 xfs_errortag_clearall(mp
);
1086 unregister_shrinker(&mp
->m_inodegc_shrinker
);
1089 xfs_errortag_del(mp
);
1090 xfs_error_sysfs_del(mp
);
1091 xfs_sysfs_del(&mp
->m_stats
.xs_kobj
);
1092 xfs_sysfs_del(&mp
->m_kobj
);
1096 * Determine whether modifications can proceed. The caller specifies the minimum
1097 * freeze level for which modifications should not be allowed. This allows
1098 * certain operations to proceed while the freeze sequence is in progress, if
1103 struct xfs_mount
*mp
,
1106 ASSERT(level
> SB_UNFROZEN
);
1107 if ((mp
->m_super
->s_writers
.frozen
>= level
) ||
1108 xfs_is_shutdown(mp
) || xfs_is_readonly(mp
))
1114 /* Adjust m_fdblocks or m_frextents. */
1116 xfs_mod_freecounter(
1117 struct xfs_mount
*mp
,
1118 struct percpu_counter
*counter
,
1124 uint64_t set_aside
= 0;
1128 ASSERT(counter
== &mp
->m_fdblocks
|| counter
== &mp
->m_frextents
);
1129 has_resv_pool
= (counter
== &mp
->m_fdblocks
);
1131 ASSERT(has_resv_pool
);
1135 * If the reserve pool is depleted, put blocks back into it
1136 * first. Most of the time the pool is full.
1138 if (likely(!has_resv_pool
||
1139 mp
->m_resblks
== mp
->m_resblks_avail
)) {
1140 percpu_counter_add(counter
, delta
);
1144 spin_lock(&mp
->m_sb_lock
);
1145 res_used
= (long long)(mp
->m_resblks
- mp
->m_resblks_avail
);
1147 if (res_used
> delta
) {
1148 mp
->m_resblks_avail
+= delta
;
1151 mp
->m_resblks_avail
= mp
->m_resblks
;
1152 percpu_counter_add(counter
, delta
);
1154 spin_unlock(&mp
->m_sb_lock
);
1159 * Taking blocks away, need to be more accurate the closer we
1162 * If the counter has a value of less than 2 * max batch size,
1163 * then make everything serialise as we are real close to
1166 if (__percpu_counter_compare(counter
, 2 * XFS_FDBLOCKS_BATCH
,
1167 XFS_FDBLOCKS_BATCH
) < 0)
1170 batch
= XFS_FDBLOCKS_BATCH
;
1173 * Set aside allocbt blocks because these blocks are tracked as free
1174 * space but not available for allocation. Technically this means that a
1175 * single reservation cannot consume all remaining free space, but the
1176 * ratio of allocbt blocks to usable free blocks should be rather small.
1177 * The tradeoff without this is that filesystems that maintain high
1178 * perag block reservations can over reserve physical block availability
1179 * and fail physical allocation, which leads to much more serious
1180 * problems (i.e. transaction abort, pagecache discards, etc.) than
1181 * slightly premature -ENOSPC.
1184 set_aside
= xfs_fdblocks_unavailable(mp
);
1185 percpu_counter_add_batch(counter
, delta
, batch
);
1186 if (__percpu_counter_compare(counter
, set_aside
,
1187 XFS_FDBLOCKS_BATCH
) >= 0) {
1193 * lock up the sb for dipping into reserves before releasing the space
1194 * that took us to ENOSPC.
1196 spin_lock(&mp
->m_sb_lock
);
1197 percpu_counter_add(counter
, -delta
);
1198 if (!has_resv_pool
|| !rsvd
)
1199 goto fdblocks_enospc
;
1201 lcounter
= (long long)mp
->m_resblks_avail
+ delta
;
1202 if (lcounter
>= 0) {
1203 mp
->m_resblks_avail
= lcounter
;
1204 spin_unlock(&mp
->m_sb_lock
);
1208 "Reserve blocks depleted! Consider increasing reserve pool size.");
1211 spin_unlock(&mp
->m_sb_lock
);
1216 * Used to free the superblock along various error paths.
1220 struct xfs_mount
*mp
)
1222 struct xfs_buf
*bp
= mp
->m_sb_bp
;
1230 * If the underlying (data/log/rt) device is readonly, there are some
1231 * operations that cannot proceed.
1234 xfs_dev_is_read_only(
1235 struct xfs_mount
*mp
,
1238 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1239 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1240 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1241 xfs_notice(mp
, "%s required on read-only device.", message
);
1242 xfs_notice(mp
, "write access unavailable, cannot proceed.");
1248 /* Force the summary counters to be recalculated at next mount. */
1250 xfs_force_summary_recalc(
1251 struct xfs_mount
*mp
)
1253 if (!xfs_has_lazysbcount(mp
))
1256 xfs_fs_mark_sick(mp
, XFS_SICK_FS_COUNTERS
);
1260 * Enable a log incompat feature flag in the primary superblock. The caller
1261 * cannot have any other transactions in progress.
1264 xfs_add_incompat_log_feature(
1265 struct xfs_mount
*mp
,
1268 struct xfs_dsb
*dsb
;
1271 ASSERT(hweight32(feature
) == 1);
1272 ASSERT(!(feature
& XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
1275 * Force the log to disk and kick the background AIL thread to reduce
1276 * the chances that the bwrite will stall waiting for the AIL to unpin
1277 * the primary superblock buffer. This isn't a data integrity
1278 * operation, so we don't need a synchronous push.
1280 error
= xfs_log_force(mp
, XFS_LOG_SYNC
);
1283 xfs_ail_push_all(mp
->m_ail
);
1286 * Lock the primary superblock buffer to serialize all callers that
1287 * are trying to set feature bits.
1289 xfs_buf_lock(mp
->m_sb_bp
);
1290 xfs_buf_hold(mp
->m_sb_bp
);
1292 if (xfs_is_shutdown(mp
)) {
1297 if (xfs_sb_has_incompat_log_feature(&mp
->m_sb
, feature
))
1301 * Write the primary superblock to disk immediately, because we need
1302 * the log_incompat bit to be set in the primary super now to protect
1303 * the log items that we're going to commit later.
1305 dsb
= mp
->m_sb_bp
->b_addr
;
1306 xfs_sb_to_disk(dsb
, &mp
->m_sb
);
1307 dsb
->sb_features_log_incompat
|= cpu_to_be32(feature
);
1308 error
= xfs_bwrite(mp
->m_sb_bp
);
1313 * Add the feature bits to the incore superblock before we unlock the
1316 xfs_sb_add_incompat_log_features(&mp
->m_sb
, feature
);
1317 xfs_buf_relse(mp
->m_sb_bp
);
1319 /* Log the superblock to disk. */
1320 return xfs_sync_sb(mp
, false);
1322 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1324 xfs_buf_relse(mp
->m_sb_bp
);
1329 * Clear all the log incompat flags from the superblock.
1331 * The caller cannot be in a transaction, must ensure that the log does not
1332 * contain any log items protected by any log incompat bit, and must ensure
1333 * that there are no other threads that depend on the state of the log incompat
1334 * feature flags in the primary super.
1336 * Returns true if the superblock is dirty.
1339 xfs_clear_incompat_log_features(
1340 struct xfs_mount
*mp
)
1344 if (!xfs_has_crc(mp
) ||
1345 !xfs_sb_has_incompat_log_feature(&mp
->m_sb
,
1346 XFS_SB_FEAT_INCOMPAT_LOG_ALL
) ||
1347 xfs_is_shutdown(mp
))
1351 * Update the incore superblock. We synchronize on the primary super
1352 * buffer lock to be consistent with the add function, though at least
1353 * in theory this shouldn't be necessary.
1355 xfs_buf_lock(mp
->m_sb_bp
);
1356 xfs_buf_hold(mp
->m_sb_bp
);
1358 if (xfs_sb_has_incompat_log_feature(&mp
->m_sb
,
1359 XFS_SB_FEAT_INCOMPAT_LOG_ALL
)) {
1360 xfs_sb_remove_incompat_log_features(&mp
->m_sb
);
1364 xfs_buf_relse(mp
->m_sb_bp
);
1369 * Update the in-core delayed block counter.
1371 * We prefer to update the counter without having to take a spinlock for every
1372 * counter update (i.e. batching). Each change to delayed allocation
1373 * reservations can change can easily exceed the default percpu counter
1374 * batching, so we use a larger batch factor here.
1376 * Note that we don't currently have any callers requiring fast summation
1377 * (e.g. percpu_counter_read) so we can use a big batch value here.
1379 #define XFS_DELALLOC_BATCH (4096)
1382 struct xfs_mount
*mp
,
1385 percpu_counter_add_batch(&mp
->m_delalloc_blks
, delta
,
1386 XFS_DELALLOC_BATCH
);