Hi,
I got this with 2.6.25-rc3 when doing an rm -rf on a HFS+ filesystem:
[ INFO: possible recursive locking detected ]
2.6.25-rc3 #6
---------------------------------------------
rm/7564 is trying to acquire lock:
(&sb->s_type->i_mutex_key#8){--..}, at: [<ffffffff880fc4ba>] hfsplus_block_free+0x57/0x209 [hfsplus]
but task is already holding lock:
(&sb->s_type->i_mutex_key#8){--..}, at: [<ffffffff80284f1c>] vfs_unlink+0x41/0xb7
other info that might help us debug this:
2 locks held by rm/7564:
#0: (&type->i_mutex_dir_key#5/1){--..}, at: [<ffffffff80286cac>] do_unlinkat+0x6c/0x154
#1: (&sb->s_type->i_mutex_key#8){--..}, at: [<ffffffff80284f1c>] vfs_unlink+0x41/0xb7
stack backtrace:
Pid: 7564, comm: rm Not tainted 2.6.25-rc3 #6
Call Trace:
[<ffffffff802497bb>] __lock_acquire+0x849/0xbd5
[<ffffffff880fc4ba>] :hfsplus:hfsplus_block_free+0x57/0x209
[<ffffffff80249efd>] lock_acquire+0x51/0x6c
[<ffffffff880fc4ba>] :hfsplus:hfsplus_block_free+0x57/0x209
[<ffffffff80246b71>] debug_mutex_lock_common+0x16/0x23
[<ffffffff80418eb0>] mutex_lock_nested+0xd9/0x268
[<ffffffff880fc4ba>] :hfsplus:hfsplus_block_free+0x57/0x209
[<ffffffff880f647f>] :hfsplus:hfsplus_free_extents+0x54/0x9b
[<ffffffff880f6a92>] :hfsplus:hfsplus_file_truncate+0xa4/0x2ce
[<ffffffff880f52de>] :hfsplus:hfsplus_delete_inode+0x57/0x5d
[<ffffffff880f77e6>] :hfsplus:hfsplus_unlink+0xd0/0x158
[<ffffffff80284f36>] vfs_unlink+0x5b/0xb7
[<ffffffff80286cf1>] do_unlinkat+0xb1/0x154
[<ffffffff80419e4c>] trace_hardirqs_on_thunk+0x35/0x3a
[<ffffffff80248b03>] trace_hardirqs_on+0xf3/0x117
[<ffffffff80419e4c>] trace_hardirqs_on_thunk+0x35/0x3a
[<ffffffff880fb4a5>] :hfsplus:hfsplus_uni2asc+0x251/0x29f
[<ffffffff8020b0bb>] system_call_after_swapgs+0x7b/0x80
Is this merely a case for annotation?
--
Stefan Richter
-=====-==--- --=- ===--
http://arcgraph.de/sr/
On Thu, 2008-02-28 at 16:07 +0100, Stefan Richter wrote:
> Hi,
>
> I got this with 2.6.25-rc3 when doing an rm -rf on a HFS+ filesystem:
>
>
> [ INFO: possible recursive locking detected ]
> 2.6.25-rc3 #6
> ---------------------------------------------
> rm/7564 is trying to acquire lock:
> (&sb->s_type->i_mutex_key#8){--..}, at: [<ffffffff880fc4ba>] hfsplus_block_free+0x57/0x209 [hfsplus]
>
> but task is already holding lock:
> (&sb->s_type->i_mutex_key#8){--..}, at: [<ffffffff80284f1c>] vfs_unlink+0x41/0xb7
>
> other info that might help us debug this:
> 2 locks held by rm/7564:
> #0: (&type->i_mutex_dir_key#5/1){--..}, at: [<ffffffff80286cac>] do_unlinkat+0x6c/0x154
> #1: (&sb->s_type->i_mutex_key#8){--..}, at: [<ffffffff80284f1c>] vfs_unlink+0x41/0xb7
>
> stack backtrace:
> Pid: 7564, comm: rm Not tainted 2.6.25-rc3 #6
>
> Call Trace:
> [<ffffffff802497bb>] __lock_acquire+0x849/0xbd5
> [<ffffffff880fc4ba>] :hfsplus:hfsplus_block_free+0x57/0x209
> [<ffffffff80249efd>] lock_acquire+0x51/0x6c
> [<ffffffff880fc4ba>] :hfsplus:hfsplus_block_free+0x57/0x209
> [<ffffffff80246b71>] debug_mutex_lock_common+0x16/0x23
> [<ffffffff80418eb0>] mutex_lock_nested+0xd9/0x268
> [<ffffffff880fc4ba>] :hfsplus:hfsplus_block_free+0x57/0x209
> [<ffffffff880f647f>] :hfsplus:hfsplus_free_extents+0x54/0x9b
> [<ffffffff880f6a92>] :hfsplus:hfsplus_file_truncate+0xa4/0x2ce
> [<ffffffff880f52de>] :hfsplus:hfsplus_delete_inode+0x57/0x5d
> [<ffffffff880f77e6>] :hfsplus:hfsplus_unlink+0xd0/0x158
> [<ffffffff80284f36>] vfs_unlink+0x5b/0xb7
> [<ffffffff80286cf1>] do_unlinkat+0xb1/0x154
> [<ffffffff80419e4c>] trace_hardirqs_on_thunk+0x35/0x3a
> [<ffffffff80248b03>] trace_hardirqs_on+0xf3/0x117
> [<ffffffff80419e4c>] trace_hardirqs_on_thunk+0x35/0x3a
> [<ffffffff880fb4a5>] :hfsplus:hfsplus_uni2asc+0x251/0x29f
> [<ffffffff8020b0bb>] system_call_after_swapgs+0x7b/0x80
>
>
> Is this merely a case for annotation?
Being utterly clueless on HFS, and not having had a look yet, I'd say
its genuine. Esp. since the i_mutex lock class is per filesystem type.
So HFS has internal lock ordering problems, its not interaction with
another filesystem - like we used to have with ext vs the pseudo
filesystems.
Hi,
On Thu, 28 Feb 2008, Peter Zijlstra wrote:
> > Is this merely a case for annotation?
>
> Being utterly clueless on HFS, and not having had a look yet, I'd say
> its genuine. Esp. since the i_mutex lock class is per filesystem type.
>
> So HFS has internal lock ordering problems, its not interaction with
> another filesystem - like we used to have with ext vs the pseudo
> filesystems.
Please look at the code before you get to such conclusions...
The allocation bitmap is very much organized like a normal file, so HFS+
treats it like a file (e.g. its data is in the page cache), this file is
just not visible outside HFS+, so I used the inode lock to synchronize the
access to it. The lock ordering should be fine, lockdep just doesn't know
about it.
bye, Roman