zfs: suppress reclaim lockdep in zfs_inactive

[Gality369]

Motivation and Context

A possible circular locking dependency involving kswapd was reported
in the zfs_inactive/zfs_zinactive path. The warning occurs when reclaim
context reaches the synchronous unlinked-deletion path.

WARNING: possible circular locking dependency detected
------------------------------------------------------
kswapd0/55 is trying to acquire lock:
ffff88801d089f18 (&zfsvfs->z_teardown_inactive_lock){.+.+}-{4:4}, at: zfs_inactive+0x194/0x1300 fs/zfs/os/linux/zfs/zfs_vnops_os.c:4084

but task is already holding lock:
ffffffff87abcaa0 (fs_reclaim){+.+.}-{0:0}, at: balance_pgdat+0xc06/0x18b0 mm/vmscan.c:7137

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #3 (fs_reclaim){+.+.}-{0:0}:
       __fs_reclaim_acquire mm/page_alloc.c:4264 [inline]
       fs_reclaim_acquire+0xe6/0x120 mm/page_alloc.c:4278
       might_alloc include/linux/sched/mm.h:318 [inline]
       slab_pre_alloc_hook mm/slub.c:4929 [inline]
       slab_alloc_node mm/slub.c:5264 [inline]
       __do_kmalloc_node mm/slub.c:5649 [inline]
       __kvmalloc_node_noprof+0xef/0xa20 mm/slub.c:7112
       spl_kvmalloc+0x102/0x140 fs/zfs/os/linux/spl/spl-kmem.c:146
       spl_kmem_alloc_impl fs/zfs/os/linux/spl/spl-kmem.c:260 [inline]
       spl_kmem_alloc+0x240/0x260 fs/zfs/os/linux/spl/spl-kmem.c:467
       rrn_add fs/zfs/zfs/rrwlock.c:107 [inline]
       rrw_enter_read_impl+0x2c8/0x970 fs/zfs/zfs/rrwlock.c:186
       rrw_enter_read fs/zfs/zfs/rrwlock.c:198 [inline]
       rrw_enter+0x57/0x70 fs/zfs/zfs/rrwlock.c:235
       dsl_pool_config_enter fs/zfs/zfs/dsl_pool.c:1431 [inline]
       dsl_pool_hold+0x105/0x130 fs/zfs/zfs/dsl_pool.c:1403
       dmu_objset_hold_flags+0xae/0x1b0 fs/zfs/zfs/dmu_objset.c:751
       dmu_objset_hold+0x2c/0x40 fs/zfs/zfs/dmu_objset.c:772
       zpl_mount_impl fs/zfs/os/linux/zfs/zpl_super.c:394 [inline]
       zpl_mount+0xb6/0x770 fs/zfs/os/linux/zfs/zpl_super.c:465
       legacy_get_tree+0x112/0x220 fs/fs_context.c:663
       vfs_get_tree+0x9a/0x370 fs/super.c:1758
       fc_mount fs/namespace.c:1199 [inline]
       do_new_mount_fc fs/namespace.c:3642 [inline]
       do_new_mount fs/namespace.c:3718 [inline]
       path_mount+0x5b8/0x1ea0 fs/namespace.c:4028
       do_mount fs/namespace.c:4041 [inline]
       __do_sys_mount fs/namespace.c:4229 [inline]
       __se_sys_mount fs/namespace.c:4206 [inline]
       __x64_sys_mount+0x282/0x320 fs/namespace.c:4206
       x64_sys_call+0x1a7d/0x26a0 arch/x86/include/generated/asm/syscalls_64.h:166
       do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
       do_syscall_64+0x93/0xf80 arch/x86/entry/syscall_64.c:94
       entry_SYSCALL_64_after_hwframe+0x76/0x7e

-> #2 (&rrl->rr_lock){+.+.}-{4:4}:
       __mutex_lock_common kernel/locking/mutex.c:598 [inline]
       __mutex_lock+0x1a6/0x1d80 kernel/locking/mutex.c:760
       mutex_lock_nested+0x1b/0x30 kernel/locking/mutex.c:812
       rrw_enter_read_impl+0x77/0x970 fs/zfs/zfs/rrwlock.c:166
       rrw_enter_read fs/zfs/zfs/rrwlock.c:198 [inline]
       rrw_enter+0x57/0x70 fs/zfs/zfs/rrwlock.c:235
       dmu_buf_lock_parent+0x22c/0x380 fs/zfs/zfs/dbuf.c:1350
       dbuf_read+0xb88/0x1940 fs/zfs/zfs/dbuf.c:1834
       dmu_buf_hold_by_dnode+0x8e/0xe0 fs/zfs/zfs/dmu.c:232
       mzap_create_impl+0xb5/0x470 fs/zfs/zfs/zap_micro.c:842
       zap_create_impl fs/zfs/zfs/zap_micro.c:879 [inline]
       zap_create_norm_dnsize fs/zfs/zfs/zap_micro.c:968 [inline]
       zap_create_dnsize+0xd1/0x120 fs/zfs/zfs/zap_micro.c:952
       zap_create_link_dnsize+0xae/0x1b0 fs/zfs/zfs/zap.c:1104
       zap_create_link+0x3d/0x60 fs/zfs/zfs/zap.c:1095
       sa_attr_register_sync+0x8f1/0xd30 fs/zfs/zfs/sa.c:1859
       sa_replace_all_by_template_locked fs/zfs/zfs/sa.c:1892 [inline]
       sa_replace_all_by_template+0x4cd/0x640 fs/zfs/zfs/sa.c:1903
       zfs_mknode+0x162f/0x3e20 fs/zfs/os/linux/zfs/zfs_znode_os.c:905
       zfs_create_fs+0xc46/0x11f0 fs/zfs/os/linux/zfs/zfs_znode_os.c:1952
       dsl_pool_create+0x3f8/0x840 fs/zfs/zfs/dsl_pool.c:556
       spa_create+0xd2e/0x1a00 fs/zfs/zfs/spa.c:7207
       zfs_ioc_pool_create+0x4fb/0x650 fs/zfs/zfs/zfs_ioctl.c:1522
       zfsdev_ioctl_common+0x128c/0x1600 fs/zfs/zfs/zfs_ioctl.c:8239
       zfsdev_ioctl+0x65/0x120 fs/zfs/os/linux/zfs/zfs_ioctl_os.c:145
       vfs_ioctl fs/ioctl.c:51 [inline]
       __do_sys_ioctl fs/ioctl.c:597 [inline]
       __se_sys_ioctl fs/ioctl.c:583 [inline]
       __x64_sys_ioctl+0x197/0x1e0 fs/ioctl.c:583
       x64_sys_call+0x1144/0x26a0 arch/x86/include/generated/asm/syscalls_64.h:17
       do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
       do_syscall_64+0x93/0xf80 arch/x86/entry/syscall_64.c:94
       entry_SYSCALL_64_after_hwframe+0x76/0x7e

-> #1 (&hdl->sa_lock){+.+.}-{4:4}:
       __mutex_lock_common kernel/locking/mutex.c:598 [inline]
       __mutex_lock+0x1a6/0x1d80 kernel/locking/mutex.c:760
       mutex_lock_nested+0x1b/0x30 kernel/locking/mutex.c:812
       sa_lookup+0xf8/0x560 fs/zfs/zfs/sa.c:1509
       zfs_rmnode+0x245/0x14a0 fs/zfs/os/linux/zfs/zfs_dir.c:710
       zfs_zinactive+0x612/0x7e0 fs/zfs/os/linux/zfs/zfs_znode_os.c:1354
       zfs_inactive+0x258/0x1300 fs/zfs/os/linux/zfs/zfs_vnops_os.c:4113
       zpl_evict_inode+0x7d/0xe0 fs/zfs/os/linux/zfs/zpl_super.c:144
       evict+0x38e/0x8f0 fs/inode.c:810
       iput_final fs/inode.c:1914 [inline]
       iput fs/inode.c:1966 [inline]
       iput+0x55b/0x8b0 fs/inode.c:1926
       do_unlinkat+0x4cb/0x680 fs/namei.c:4744
       __do_sys_unlinkat fs/namei.c:4778 [inline]
       __se_sys_unlinkat fs/namei.c:4771 [inline]
       __x64_sys_unlinkat+0xae/0x130 fs/namei.c:4771
       x64_sys_call+0x1066/0x26a0 arch/x86/include/generated/asm/syscalls_64.h:264
       do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
       do_syscall_64+0x93/0xf80 arch/x86/entry/syscall_64.c:94
       entry_SYSCALL_64_after_hwframe+0x76/0x7e

-> #0 (&zfsvfs->z_teardown_inactive_lock){.+.+}-{4:4}:
       check_prev_add kernel/locking/lockdep.c:3165 [inline]
       check_prevs_add kernel/locking/lockdep.c:3284 [inline]
       validate_chain kernel/locking/lockdep.c:3908 [inline]
       __lock_acquire+0x14ae/0x21e0 kernel/locking/lockdep.c:5237
       lock_acquire kernel/locking/lockdep.c:5868 [inline]
       lock_acquire+0x169/0x2f0 kernel/locking/lockdep.c:5825
       down_read+0x9c/0x4a0 kernel/locking/rwsem.c:1537
       zfs_inactive+0x194/0x1300 fs/zfs/os/linux/zfs/zfs_vnops_os.c:4084
       zpl_evict_inode+0x7d/0xe0 fs/zfs/os/linux/zfs/zpl_super.c:144
       evict+0x38e/0x8f0 fs/inode.c:810
       dispose_list+0xe4/0x1d0 fs/inode.c:852
       prune_icache_sb+0xef/0x170 fs/inode.c:1000
       super_cache_scan+0x354/0x530 fs/super.c:224
       do_shrink_slab+0x37f/0xe00 mm/shrinker.c:437
       shrink_slab+0x173/0x1000 mm/shrinker.c:664
       shrink_one+0x44f/0x8a0 mm/vmscan.c:4955
       shrink_many mm/vmscan.c:5016 [inline]
       lru_gen_shrink_node mm/vmscan.c:5094 [inline]
       shrink_node+0x2428/0x3870 mm/vmscan.c:6081
       kswapd_shrink_node mm/vmscan.c:6941 [inline]
       balance_pgdat+0xb3d/0x18b0 mm/vmscan.c:7124
       kswapd+0x527/0xa20 mm/vmscan.c:7389
       kthread+0x3f0/0x850 kernel/kthread.c:463
       ret_from_fork+0x50f/0x610 arch/x86/kernel/process.c:158
       ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:245

other info that might help us debug this:

Chain exists of:
  &zfsvfs->z_teardown_inactive_lock --> &rrl->rr_lock --> fs_reclaim

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               lock(&rrl->rr_lock);
                               lock(fs_reclaim);
  rlock(&zfsvfs->z_teardown_inactive_lock);

 *** DEADLOCK ***

2 locks held by kswapd0/55:
 #0: ffffffff87abcaa0 (fs_reclaim){+.+.}-{0:0}, at: balance_pgdat+0xc06/0x18b0 mm/vmscan.c:7137
 #1: ffff88800afbc0e0 (&type->s_umount_key#60){.+.+}-{4:4}, at: super_trylock_shared fs/super.c:562 [inline]
 #1: ffff88800afbc0e0 (&type->s_umount_key#60){.+.+}-{4:4}, at: super_cache_scan+0x8c/0x530 fs/super.c:197

Call Trace:
 __dump_stack lib/dump_stack.c:94 [inline]
 dump_stack_lvl+0xbe/0x130 lib/dump_stack.c:120
 dump_stack+0x15/0x20 lib/dump_stack.c:129
 print_circular_bug+0x285/0x360 kernel/locking/lockdep.c:2043
 check_noncircular+0x14e/0x170 kernel/locking/lockdep.c:2175
 check_prev_add kernel/locking/lockdep.c:3165 [inline]
 check_prevs_add kernel/locking/lockdep.c:3284 [inline]
 validate_chain kernel/locking/lockdep.c:3908 [inline]
 __lock_acquire+0x14ae/0x21e0 kernel/locking/lockdep.c:5237
 lock_acquire kernel/locking/lockdep.c:5868 [inline]
 lock_acquire+0x169/0x2f0 kernel/locking/lockdep.c:5825
 down_read+0x9c/0x4a0 kernel/locking/rwsem.c:1537
 zfs_inactive+0x194/0x1300 fs/zfs/os/linux/zfs/zfs_vnops_os.c:4084
 zpl_evict_inode+0x7d/0xe0 fs/zfs/os/linux/zfs/zpl_super.c:144
 evict+0x38e/0x8f0 fs/inode.c:810
 dispose_list+0xe4/0x1d0 fs/inode.c:852
 prune_icache_sb+0xef/0x170 fs/inode.c:1000
 super_cache_scan+0x354/0x530 fs/super.c:224
 do_shrink_slab+0x37f/0xe00 mm/shrinker.c:437
 shrink_slab+0x173/0x1000 mm/shrinker.c:664
 shrink_one+0x44f/0x8a0 mm/vmscan.c:4955
 shrink_many mm/vmscan.c:5016 [inline]
 lru_gen_shrink_node mm/vmscan.c:5094 [inline]
 shrink_node+0x2428/0x3870 mm/vmscan.c:6081
 kswapd_shrink_node mm/vmscan.c:6941 [inline]
 balance_pgdat+0xb3d/0x18b0 mm/vmscan.c:7124
 kswapd+0x527/0xa20 mm/vmscan.c:7389
 kthread+0x3f0/0x850 kernel/kthread.c:463
 ret_from_fork+0x50f/0x610 arch/x86/kernel/process.c:158
 ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:245

Description

This keeps the existing zfs_zinactive() and unlinked cleanup semantics.

For reclaim threads only, zfs_inactive() temporarily disables lockdep
around the acquire and release of z_teardown_inactive_lock. The real
rwsem is still taken normally, so teardown serialization is unchanged.

The patch also removes the extra zfs_unlinked_drain_try() path and the
additional z_drain_cv/TASKQID_INITIAL drain state added by the previous
version.

How Has This Been Tested?

Tested on Linux.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• Quality assurance (non-breaking change which makes the code more robust against bugs)

Checklist:

• My code follows the OpenZFS code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• I have run the ZFS Test Suite with this change applied.
• All commit messages are properly formatted and contain Signed-off-by.

[Copilot]

Pull request overview

This PR addresses a reported Linux lockdep circular dependency involving kswapd entering the zfs_inactive() / zfs_zinactive() path by avoiding synchronous zfs_rmnode() work in reclaim-thread eviction, and instead attempting to defer final deletion to the unlinked-drain taskq using a non-blocking dispatch.

Changes:

• Detect reclaim-thread eviction in zfs_zinactive() and defer inline zfs_rmnode() for unlinked znodes.
• Introduce zfs_unlinked_drain_try() to attempt scheduling unlinked draining using TQ_NOSLEEP.
• Export the new helper via the Linux zfs_dir.h header.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
module/os/linux/zfs/zfs_znode_os.c	Defers inline zfs_rmnode() during reclaim-thread eviction and triggers best-effort unlinked draining.
module/os/linux/zfs/zfs_dir.c	Adds zfs_unlinked_drain_try() which dispatches the drain task using TQ_NOSLEEP.
include/os/linux/zfs/sys/zfs_dir.h	Declares zfs_unlinked_drain_try() for Linux consumers.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[behlendorf]

I haven't looked closely yet, but this change appears to have introduced a new deadlock. See the "vm1: console logs" link under the "Test Summary" step for the failed builders. For example, for Centos Stream 9:

https://github.com/openzfs/zfs/actions/runs/25496566454/job/74817848886?pr=18505

[Gality369]

Thanks for taking a look and pointing this out. The lockdep fix turned out to be more complicated than I expected, and I’m sorry for the trouble caused by the new deadlock. I’ll continue investigating this and update the PR once I have a better solution.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (2)

module/os/linux/zfs/zfs_vnops_os.c:4095

• current_is_reclaim_thread() on Linux returns true only for kswapd (current_is_kswapd()), but the same lockdep chain can be hit from any task performing direct reclaim that ends up in super_cache_scan → prune_icache_sb → evict → zpl_evict_inode → zfs_inactive. Direct-reclaim tasks hold fs_reclaim just like kswapd, so the splat (and the underlying ordering concern) will still be reported for them. If the intent really is to suppress the warning, the gating predicate should also cover direct reclaim (e.g., current->flags & PF_MEMALLOC or check current->reclaim_state).

		no_lockdep = current_is_reclaim_thread();

include/os/linux/spl/sys/rwlock.h:264

• rw_tryenter_nolockdep and rw_downgrade_nolockdep are added but never used in this PR (verified via search). If they are not needed by the immediate fix, please drop them; otherwise add the actual call sites. Adding unused API surface invites future misuse of lockdep_off().

#define	rw_tryenter_nolockdep(rwp, rw) /* CSTYLED */		\
({									\
	spl_rw_lockdep_off();						\
	int _rc_ = spl_rw_tryenter_impl(rwp, rw);			\
	spl_rw_lockdep_on();						\
	_rc_;								\
})

#define	rw_enter(rwp, rw) /* CSTYLED */				\
({									\
	spl_rw_lockdep_off_maybe(rwp);					\
	spl_rw_enter_impl(rwp, rw);					\
	spl_rw_lockdep_on_maybe(rwp);					\
})

#define	rw_enter_nolockdep(rwp, rw) /* CSTYLED */		\
({									\
	spl_rw_lockdep_off();						\
	spl_rw_enter_impl(rwp, rw);					\
	spl_rw_lockdep_on();						\
})

#define	rw_exit(rwp) /* CSTYLED */				\
({									\
	spl_rw_lockdep_off_maybe(rwp);					\
	spl_rw_exit_impl(rwp);						\
	spl_rw_lockdep_on_maybe(rwp);					\
})

#define	rw_exit_nolockdep(rwp) /* CSTYLED */			\
({									\
	spl_rw_lockdep_off();						\
	spl_rw_exit_impl(rwp);						\
	spl_rw_lockdep_on();						\
})

#define	rw_downgrade(rwp) /* CSTYLED */			\
({									\
	spl_rw_lockdep_off_maybe(rwp);					\
	spl_rw_downgrade_impl(rwp);					\
	spl_rw_lockdep_on_maybe(rwp);					\
})

#define	rw_downgrade_nolockdep(rwp) /* CSTYLED */		\
({									\
	spl_rw_lockdep_off();						\
	spl_rw_downgrade_impl(rwp);					\
	spl_rw_lockdep_on();						\
})

[behlendorf]

Much nicer. Thanks for reworking this.