Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'md-6.17-20250819' of gitolite.kernel.org:pub/scm/linux/kernel/git/mdraid/linux into block-6.17
Pull MD fixes from Yu:
"- Add a legacy_async_del_gendisk mode, to prevent a user tools
regression. New user tools releases will not use such a mode, the old
release with a new kernel now will have warning about deprecated
behavior, and we prepare to remove this legacy mode after about a
year later.
- The rename in kernel causing user tools build failure, revert the
rename in mdp_superblock_s.
- Fix a regression that interrupted resync can be shown as recover from
mdstat or sysfs."
* tag 'md-6.17-20250819' of gitolite.kernel.org:pub/scm/linux/kernel/git/mdraid/linux:
md: fix sync_action incorrect display during resync
md: add helper rdev_needs_recovery()
md: keep recovery_cp in mdp_superblock_s
md: add legacy_async_del_gendisk mode
+93
-31
+92
-30
drivers/md/md.c
+92
-30
drivers/md/md.c
···
339
339
* so all the races disappear.
340
340
*/
341
341
static bool create_on_open = true;
342
+
static bool legacy_async_del_gendisk = true;
342
343
343
344
/*
344
345
* We have a system wide 'event count' that is incremented
···
878
877
export_rdev(rdev, mddev);
879
878
}
880
879
881
-
/* Call del_gendisk after release reconfig_mutex to avoid
882
-
* deadlock (e.g. call del_gendisk under the lock and an
883
-
* access to sysfs files waits the lock)
884
-
* And MD_DELETED is only used for md raid which is set in
885
-
* do_md_stop. dm raid only uses md_stop to stop. So dm raid
886
-
* doesn't need to check MD_DELETED when getting reconfig lock
887
-
*/
888
-
if (test_bit(MD_DELETED, &mddev->flags))
889
-
del_gendisk(mddev->gendisk);
880
+
if (!legacy_async_del_gendisk) {
881
+
/*
882
+
* Call del_gendisk after release reconfig_mutex to avoid
883
+
* deadlock (e.g. call del_gendisk under the lock and an
884
+
* access to sysfs files waits the lock)
885
+
* And MD_DELETED is only used for md raid which is set in
886
+
* do_md_stop. dm raid only uses md_stop to stop. So dm raid
887
+
* doesn't need to check MD_DELETED when getting reconfig lock
888
+
*/
889
+
if (test_bit(MD_DELETED, &mddev->flags))
890
+
del_gendisk(mddev->gendisk);
891
+
}
890
892
}
891
893
EXPORT_SYMBOL_GPL(mddev_unlock);
892
894
···
1423
1419
else {
1424
1420
if (sb->events_hi == sb->cp_events_hi &&
1425
1421
sb->events_lo == sb->cp_events_lo) {
1426
-
mddev->resync_offset = sb->resync_offset;
1422
+
mddev->resync_offset = sb->recovery_cp;
1427
1423
} else
1428
1424
mddev->resync_offset = 0;
1429
1425
}
···
1551
1547
mddev->minor_version = sb->minor_version;
1552
1548
if (mddev->in_sync)
1553
1549
{
1554
-
sb->resync_offset = mddev->resync_offset;
1550
+
sb->recovery_cp = mddev->resync_offset;
1555
1551
sb->cp_events_hi = (mddev->events>>32);
1556
1552
sb->cp_events_lo = (u32)mddev->events;
1557
1553
if (mddev->resync_offset == MaxSector)
1558
1554
sb->state = (1<< MD_SB_CLEAN);
1559
1555
} else
1560
-
sb->resync_offset = 0;
1556
+
sb->recovery_cp = 0;
1561
1557
1562
1558
sb->layout = mddev->layout;
1563
1559
sb->chunk_size = mddev->chunk_sectors << 9;
···
4839
4835
static struct md_sysfs_entry md_metadata =
4840
4836
__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4841
4837
4838
+
static bool rdev_needs_recovery(struct md_rdev *rdev, sector_t sectors)
4839
+
{
4840
+
return rdev->raid_disk >= 0 &&
4841
+
!test_bit(Journal, &rdev->flags) &&
4842
+
!test_bit(Faulty, &rdev->flags) &&
4843
+
!test_bit(In_sync, &rdev->flags) &&
4844
+
rdev->recovery_offset < sectors;
4845
+
}
4846
+
4847
+
static enum sync_action md_get_active_sync_action(struct mddev *mddev)
4848
+
{
4849
+
struct md_rdev *rdev;
4850
+
bool is_recover = false;
4851
+
4852
+
if (mddev->resync_offset < MaxSector)
4853
+
return ACTION_RESYNC;
4854
+
4855
+
if (mddev->reshape_position != MaxSector)
4856
+
return ACTION_RESHAPE;
4857
+
4858
+
rcu_read_lock();
4859
+
rdev_for_each_rcu(rdev, mddev) {
4860
+
if (rdev_needs_recovery(rdev, MaxSector)) {
4861
+
is_recover = true;
4862
+
break;
4863
+
}
4864
+
}
4865
+
rcu_read_unlock();
4866
+
4867
+
return is_recover ? ACTION_RECOVER : ACTION_IDLE;
4868
+
}
4869
+
4842
4870
enum sync_action md_sync_action(struct mddev *mddev)
4843
4871
{
4844
4872
unsigned long recovery = mddev->recovery;
4873
+
enum sync_action active_action;
4845
4874
4846
4875
/*
4847
4876
* frozen has the highest priority, means running sync_thread will be
···
4898
4861
!test_bit(MD_RECOVERY_NEEDED, &recovery))
4899
4862
return ACTION_IDLE;
4900
4863
4901
-
if (test_bit(MD_RECOVERY_RESHAPE, &recovery) ||
4902
-
mddev->reshape_position != MaxSector)
4864
+
/*
4865
+
* Check if any sync operation (resync/recover/reshape) is
4866
+
* currently active. This ensures that only one sync operation
4867
+
* can run at a time. Returns the type of active operation, or
4868
+
* ACTION_IDLE if none are active.
4869
+
*/
4870
+
active_action = md_get_active_sync_action(mddev);
4871
+
if (active_action != ACTION_IDLE)
4872
+
return active_action;
4873
+
4874
+
if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4903
4875
return ACTION_RESHAPE;
4904
4876
4905
4877
if (test_bit(MD_RECOVERY_RECOVER, &recovery))
···
5864
5818
{
5865
5819
struct mddev *mddev = container_of(ko, struct mddev, kobj);
5866
5820
5821
+
if (legacy_async_del_gendisk) {
5822
+
if (mddev->sysfs_state)
5823
+
sysfs_put(mddev->sysfs_state);
5824
+
if (mddev->sysfs_level)
5825
+
sysfs_put(mddev->sysfs_level);
5826
+
del_gendisk(mddev->gendisk);
5827
+
}
5867
5828
put_disk(mddev->gendisk);
5868
5829
}
5869
5830
···
6073
6020
static int md_alloc_and_put(dev_t dev, char *name)
6074
6021
{
6075
6022
struct mddev *mddev = md_alloc(dev, name);
6023
+
6024
+
if (legacy_async_del_gendisk)
6025
+
pr_warn("md: async del_gendisk mode will be removed in future, please upgrade to mdadm-4.5+\n");
6076
6026
6077
6027
if (IS_ERR(mddev))
6078
6028
return PTR_ERR(mddev);
···
6487
6431
mddev->persistent = 0;
6488
6432
mddev->level = LEVEL_NONE;
6489
6433
mddev->clevel[0] = 0;
6490
-
/* if UNTIL_STOP is set, it's cleared here */
6491
-
mddev->hold_active = 0;
6492
-
/* Don't clear MD_CLOSING, or mddev can be opened again. */
6493
-
mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6434
+
6435
+
/*
6436
+
* For legacy_async_del_gendisk mode, it can stop the array in the
6437
+
* middle of assembling it, then it still can access the array. So
6438
+
* it needs to clear MD_CLOSING. If not legacy_async_del_gendisk,
6439
+
* it can't open the array again after stopping it. So it doesn't
6440
+
* clear MD_CLOSING.
6441
+
*/
6442
+
if (legacy_async_del_gendisk && mddev->hold_active) {
6443
+
clear_bit(MD_CLOSING, &mddev->flags);
6444
+
} else {
6445
+
/* if UNTIL_STOP is set, it's cleared here */
6446
+
mddev->hold_active = 0;
6447
+
/* Don't clear MD_CLOSING, or mddev can be opened again. */
6448
+
mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6449
+
}
6494
6450
mddev->sb_flags = 0;
6495
6451
mddev->ro = MD_RDWR;
6496
6452
mddev->metadata_type[0] = 0;
···
6726
6658
6727
6659
export_array(mddev);
6728
6660
md_clean(mddev);
6729
-
set_bit(MD_DELETED, &mddev->flags);
6661
+
if (!legacy_async_del_gendisk)
6662
+
set_bit(MD_DELETED, &mddev->flags);
6730
6663
}
6731
6664
md_new_event();
6732
6665
sysfs_notify_dirent_safe(mddev->sysfs_state);
···
9037
8968
start = MaxSector;
9038
8969
rcu_read_lock();
9039
8970
rdev_for_each_rcu(rdev, mddev)
9040
-
if (rdev->raid_disk >= 0 &&
9041
-
!test_bit(Journal, &rdev->flags) &&
9042
-
!test_bit(Faulty, &rdev->flags) &&
9043
-
!test_bit(In_sync, &rdev->flags) &&
9044
-
rdev->recovery_offset < start)
8971
+
if (rdev_needs_recovery(rdev, start))
9045
8972
start = rdev->recovery_offset;
9046
8973
rcu_read_unlock();
9047
8974
···
9396
9331
test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9397
9332
rcu_read_lock();
9398
9333
rdev_for_each_rcu(rdev, mddev)
9399
-
if (rdev->raid_disk >= 0 &&
9400
-
mddev->delta_disks >= 0 &&
9401
-
!test_bit(Journal, &rdev->flags) &&
9402
-
!test_bit(Faulty, &rdev->flags) &&
9403
-
!test_bit(In_sync, &rdev->flags) &&
9404
-
rdev->recovery_offset < mddev->curr_resync)
9334
+
if (mddev->delta_disks >= 0 &&
9335
+
rdev_needs_recovery(rdev, mddev->curr_resync))
9405
9336
rdev->recovery_offset = mddev->curr_resync;
9406
9337
rcu_read_unlock();
9407
9338
}
···
10453
10392
module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10454
10393
module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10455
10394
module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10395
+
module_param(legacy_async_del_gendisk, bool, 0600);
10456
10396
10457
10397
MODULE_LICENSE("GPL");
10458
10398
MODULE_DESCRIPTION("MD RAID framework");
+1
-1
include/uapi/linux/raid/md_p.h
+1
-1
include/uapi/linux/raid/md_p.h
···
173
173
#else
174
174
#error unspecified endianness
175
175
#endif
176
-
__u32 resync_offset; /* 11 resync checkpoint sector count */
176
+
__u32 recovery_cp; /* 11 resync checkpoint sector count */
177
177
/* There are only valid for minor_version > 90 */
178
178
__u64 reshape_position; /* 12,13 next address in array-space for reshape */
179
179
__u32 new_level; /* 14 new level we are reshaping to */