lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250612042005.99602-1-lance.yang@linux.dev>
Date: Thu, 12 Jun 2025 12:19:23 +0800
From: Lance Yang <ioworker0@...il.com>
To: akpm@...ux-foundation.org
Cc: zi.li@...ux.dev,
	anna.schumaker@...cle.com,
	boqun.feng@...il.com,
	joel.granados@...nel.org,
	jstultz@...gle.com,
	kent.overstreet@...ux.dev,
	leonylgao@...cent.com,
	linux-kernel@...r.kernel.org,
	longman@...hat.com,
	mhiramat@...nel.org,
	mingo@...hat.com,
	mingzhe.yang@...com,
	peterz@...radead.org,
	rostedt@...dmis.org,
	senozhatsky@...omium.org,
	tfiga@...omium.org,
	will@...nel.org
Subject: [PATCH RFC 0/3] extend hung task blocker tracking to rwsems

Hi all,

Inspired by mutex blocker tracking[1], and having already extended it to
semaphores, let's now add support for reader-writer semaphores (rwsems).

The approach is simple: when a task enters TASK_UNINTERRUPTIBLE while
waiting for an rwsem, we just call hung_task_set_blocker(). The hung task
detector can then query the rwsem's owner to identify the lock holder.

Tracking works reliably for writers, as there can only be a single writer
holding the lock, and its task struct is stored in the owner field.

The main challenge lies with readers. The owner field points to only one
of many concurrent readers, so we might lose track of the blocker if that
specific reader unlocks, even while others remain. This is not a
significant issue, however. In practice, long-lasting lock contention is
almost always caused by a writer. Therefore, reliably tracking the writer
is the primary goal of this patch series ;)

With this change, the hung task detector can now show blocker task's info
like below:

[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 blocked for more than 122 seconds.
[Thu Jun 12 11:01:33 2025]       Tainted: G S         O        6.16.0-rc1 #1
[Thu Jun 12 11:01:33 2025] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[Thu Jun 12 11:01:33 2025] task:rw_sem_thread2  state:D stack:0     pid:36526 tgid:36526 ppid:2      task_flags:0x208040 flags:0x00004000
[Thu Jun 12 11:01:33 2025] Call Trace:
[Thu Jun 12 11:01:33 2025]  <TASK>
[Thu Jun 12 11:01:33 2025]  __schedule+0x7c7/0x1930
[Thu Jun 12 11:01:33 2025]  ? __pfx___schedule+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? _raw_spin_lock_irq+0x8a/0xe0
[Thu Jun 12 11:01:33 2025]  ? __pfx__raw_spin_lock_irq+0x10/0x10
[Thu Jun 12 11:01:33 2025]  schedule+0x6a/0x180
[Thu Jun 12 11:01:33 2025]  schedule_preempt_disabled+0x15/0x30
[Thu Jun 12 11:01:33 2025]  rwsem_down_write_slowpath+0x447/0x1090
[Thu Jun 12 11:01:33 2025]  ? __pfx_rwsem_down_write_slowpath+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx___schedule+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx___might_resched+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx_thread2_func+0x10/0x10 [rw_sem_test_2]
[Thu Jun 12 11:01:33 2025]  down_write+0x125/0x140
[Thu Jun 12 11:01:33 2025]  ? __pfx_down_write+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? msleep+0x91/0xf0
[Thu Jun 12 11:01:33 2025]  ? __raw_spin_lock_irqsave+0x8c/0xf0
[Thu Jun 12 11:01:33 2025]  thread2_func+0x37/0x70 [rw_sem_test_2]
[Thu Jun 12 11:01:33 2025]  kthread+0x39f/0x750
[Thu Jun 12 11:01:33 2025]  ? __pfx_kthread+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx__raw_spin_lock_irq+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx_kthread+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ret_from_fork+0x25d/0x320
[Thu Jun 12 11:01:33 2025]  ? __pfx_kthread+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ret_from_fork_asm+0x1a/0x30
[Thu Jun 12 11:01:33 2025]  </TASK>
[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 <writer> blocked on an rw-semaphore likely owned by task rw_sem_thread1:36525 <writer>
[Thu Jun 12 11:01:33 2025] task:rw_sem_thread1  state:S stack:0     pid:36525 tgid:36525 ppid:2      task_flags:0x208040 flags:0x00004000
[Thu Jun 12 11:01:33 2025] Call Trace:
[Thu Jun 12 11:01:33 2025]  <TASK>
[Thu Jun 12 11:01:33 2025]  __schedule+0x7c7/0x1930
[Thu Jun 12 11:01:33 2025]  ? __pfx___schedule+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __mod_timer+0x304/0xa80
[Thu Jun 12 11:01:33 2025]  ? irq_work_queue+0x6a/0xa0
[Thu Jun 12 11:01:33 2025]  ? __pfx_vprintk_emit+0x10/0x10
[Thu Jun 12 11:01:33 2025]  schedule+0x6a/0x180
[Thu Jun 12 11:01:33 2025]  schedule_timeout+0xfb/0x230
[Thu Jun 12 11:01:33 2025]  ? __pfx_schedule_timeout+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx_process_timeout+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? down_write+0xc4/0x140
[Thu Jun 12 11:01:33 2025]  msleep_interruptible+0xbe/0x150
[Thu Jun 12 11:01:33 2025]  ? __pfx_thread1_func+0x10/0x10 [rw_sem_test_2]
[Thu Jun 12 11:01:33 2025]  thread1_func+0x37/0x60 [rw_sem_test_2]
[Thu Jun 12 11:01:33 2025]  kthread+0x39f/0x750
[Thu Jun 12 11:01:33 2025]  ? __pfx_kthread+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx__raw_spin_lock_irq+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ? __pfx_kthread+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ret_from_fork+0x25d/0x320
[Thu Jun 12 11:01:33 2025]  ? __pfx_kthread+0x10/0x10
[Thu Jun 12 11:01:33 2025]  ret_from_fork_asm+0x1a/0x30
[Thu Jun 12 11:01:33 2025]  </TASK>

[1] https://lore.kernel.org/all/174046694331.2194069.15472952050240807469.stgit@mhiramat.tok.corp.google.com/

Thanks,
Lance

Lance Yang (3):
  locking/rwsem: make owner helpers globally available
  locking/rwsem: clear reader-owner on unlock to reduce false positives
  hung_task: extend hung task blocker tracking to rwsems

 include/linux/hung_task.h | 18 +++++++++---------
 include/linux/rwsem.h     | 12 ++++++++++++
 kernel/hung_task.c        | 29 +++++++++++++++++++++++++----
 kernel/locking/rwsem.c    | 31 +++++++++++++++++++++++--------
 4 files changed, 69 insertions(+), 21 deletions(-)

-- 
2.49.0


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ