fs: proc: lockdep spew and questions

fs: proc: lockdep spew and questions

[Sasha Levin]

Hi all,

I've been looking at permissions in procfs, and noticed that directories are usually added with 
execute permission.

As far as I know, there's nothing executable by default inside procfs, and on top of that, many of 
the files there don't deal well with being executed so it's easy to cause lockdep spews (one below).

Is there a reason execute bit is set for directories?


sh-4.1# chmod +x /proc/10/net/igmp
sh-4.1# /proc/10/net/igmp
/proc/10/net/igmp: line 1: Idx: command not found
sh-4.1# cat /proc/10/maps
[   64.535651]
[   64.535898] ======================================================
[   64.536674] [ INFO: possible circular locking dependency detected ]
[   64.537288] 3.13.0-rc2-next-20131206-sasha-00005-g8be2375-dirty #4052 Not tainted
[   64.538211] -------------------------------------------------------
[   64.538831] cat/7896 is trying to acquire lock:
[   64.539307]  (&sig->cred_guard_mutex){+.+.+.}, at: [<ffffffff8112e684>] mm_access+0x34/0xb0
[   64.540349]
[   64.540349] but task is already holding lock:
[   64.540906]  (&p->lock){+.+.+.}, at: [<ffffffff813000df>] seq_read+0x3f/0x430
[   64.541877]
[   64.541877] which lock already depends on the new lock.
[   64.541877]
[   64.542449]
[   64.542449] the existing dependency chain (in reverse order) is:
[   64.542449]
-> #1 (&p->lock){+.+.+.}:
[   64.542449]        [<ffffffff81194833>] validate_chain+0x6c3/0x7b0
[   64.542449]        [<ffffffff81194dcd>] __lock_acquire+0x4ad/0x580
[   64.542449]        [<ffffffff81195022>] lock_acquire+0x182/0x1d0
[   64.542449]        [<ffffffff8439825f>] mutex_lock_nested+0x6f/0x520
[   64.542449]        [<ffffffff813000df>] seq_read+0x3f/0x430
[   64.542449]        [<ffffffff8134e62f>] proc_reg_read+0x5f/0x80
[   64.542449]        [<ffffffff812d7177>] vfs_read+0xc7/0x180
[   64.542449]        [<ffffffff812dce66>] kernel_read+0x46/0x60
[   64.542449]        [<ffffffff812dcfd3>] prepare_binprm+0x153/0x170
[   64.542449]        [<ffffffff812de2d5>] do_execve_common+0x1a5/0x360
[   64.542449]        [<ffffffff812de557>] do_execve+0x37/0x40
[   64.542449]        [<ffffffff812de5a6>] SyS_execve+0x46/0x70
[   64.542449]        [<ffffffff843a6c19>] stub_execve+0x69/0xa0
[   64.542449]
-> #0 (&sig->cred_guard_mutex){+.+.+.}:
[   64.542449]        [<ffffffff81193d5f>] check_prev_add+0x13f/0x550
[   64.542449]        [<ffffffff81194833>] validate_chain+0x6c3/0x7b0
[   64.542449]        [<ffffffff81194dcd>] __lock_acquire+0x4ad/0x580
[   64.542449]        [<ffffffff81195022>] lock_acquire+0x182/0x1d0
[   64.542449]        [<ffffffff8439967f>] mutex_lock_killable_nested+0x6f/0x620
[   64.542449]        [<ffffffff8112e684>] mm_access+0x34/0xb0
[   64.542449]        [<ffffffff8134bf49>] m_start+0x79/0x1d0
[   64.542449]        [<ffffffff81300228>] seq_read+0x188/0x430
[   64.542449]        [<ffffffff812d7177>] vfs_read+0xc7/0x180
[   64.542449]        [<ffffffff812d7332>] SyS_read+0x62/0xa0
[   64.542449]        [<ffffffff843a6810>] tracesys+0xdd/0xe2
[   64.542449]
[   64.542449] other info that might help us debug this:
[   64.542449]
[   64.542449]  Possible unsafe locking scenario:
[   64.542449]
[   64.542449]        CPU0                    CPU1
[   64.542449]        ----                    ----
[   64.542449]   lock(&p->lock);
[   64.542449]                                lock(&sig->cred_guard_mutex);
[   64.542449]                                lock(&p->lock);
[   64.542449]   lock(&sig->cred_guard_mutex);
[   64.542449]
[   64.542449]  *** DEADLOCK ***
[   64.542449]
[   64.542449] 1 lock held by cat/7896:
[   64.542449]  #0:  (&p->lock){+.+.+.}, at: [<ffffffff813000df>] seq_read+0x3f/0x430
[   64.542449]
[   64.542449] stack backtrace:
[   64.542449] CPU: 0 PID: 7896 Comm: cat Not tainted 
3.13.0-rc2-next-20131206-sasha-00005-g8be2375-dirty #4052
[   64.542449]  0000000000000000 ffff880fbc169af8 ffffffff843956f7 0000000000000000
[   64.542449]  0000000000000000 ffff880fbc169b48 ffffffff81191909 ffff880fbc169b68
[   64.542449]  ffff880fbc169b68 ffff880fbc169b48 ffff880fb89fbbd8 ffff880fb89fbc10
[   64.542449] Call Trace:
[   64.542449]  [<ffffffff843956f7>] dump_stack+0x52/0x7f
[   64.542449]  [<ffffffff81191909>] print_circular_bug+0x129/0x160
[   64.542449]  [<ffffffff81193d5f>] check_prev_add+0x13f/0x550
[   64.542449]  [<ffffffff812aec6f>] ? deactivate_slab+0x8cf/0x920
[   64.542449]  [<ffffffff81194833>] validate_chain+0x6c3/0x7b0
[   64.542449]  [<ffffffff811755b8>] ? sched_clock_cpu+0x108/0x120
[   64.542449]  [<ffffffff81194dcd>] __lock_acquire+0x4ad/0x580
[   64.542449]  [<ffffffff81195022>] lock_acquire+0x182/0x1d0
[   64.542449]  [<ffffffff8112e684>] ? mm_access+0x34/0xb0
[   64.542449]  [<ffffffff8439967f>] mutex_lock_killable_nested+0x6f/0x620
[   64.542449]  [<ffffffff8112e684>] ? mm_access+0x34/0xb0
[   64.542449]  [<ffffffff81159480>] ? get_pid_task+0xd0/0x100
[   64.542449]  [<ffffffff8112e684>] ? mm_access+0x34/0xb0
[   64.542449]  [<ffffffff8112e684>] mm_access+0x34/0xb0
[   64.542449]  [<ffffffff811593b0>] ? alloc_pid+0x260/0x260
[   64.542449]  [<ffffffff8134bf49>] m_start+0x79/0x1d0
[   64.542449]  [<ffffffff81176376>] ? vtime_account_user+0x96/0xb0
[   64.542449]  [<ffffffff81300228>] seq_read+0x188/0x430
[   64.542449]  [<ffffffff812d7177>] vfs_read+0xc7/0x180
[   64.542449]  [<ffffffff812d7332>] SyS_read+0x62/0xa0
[   64.542449]  [<ffffffff843a6810>] tracesys+0xdd/0xe2

Re: fs: proc: lockdep spew and questions

[Al Viro]

On Sun, Dec 08, 2013 at 03:57:53PM -0500, Sasha Levin wrote:
> Hi all,
> 
> I've been looking at permissions in procfs, and noticed that
> directories are usually added with execute permission.
> 
> As far as I know, there's nothing executable by default inside
> procfs, and on top of that, many of the files there don't deal well
> with being executed so it's easy to cause lockdep spews (one below).
> 
> Is there a reason execute bit is set for directories?

Take any introductory textbook on Unix and read the chapter on permissions.
Or say chmod a-x ~, if you prefer education by self-LART...

Re: fs: proc: lockdep spew and questions

[Sasha Levin]

On 12/08/2013 05:10 PM, Al Viro wrote:
> On Sun, Dec 08, 2013 at 03:57:53PM -0500, Sasha Levin wrote:
>> Hi all,
>>
>> I've been looking at permissions in procfs, and noticed that
>> directories are usually added with execute permission.
>>
>> As far as I know, there's nothing executable by default inside
>> procfs, and on top of that, many of the files there don't deal well
>> with being executed so it's easy to cause lockdep spews (one below).
>>
>> Is there a reason execute bit is set for directories?
>
> Take any introductory textbook on Unix and read the chapter on permissions.
> Or say chmod a-x ~, if you prefer education by self-LART...
>

Uh, forgot about that.

So how would you suggest to deal with the execution issue in procfs?


Thanks,
Sasha

Re: fs: proc: lockdep spew and questions

[Christian Kujau]

On Sun, 8 Dec 2013 at 22:14, Sasha Levin wrote:
> So how would you suggest to deal with the execution issue in procfs?

Files will not be executable by itsself if /proc is mounted with noexec, 
as some distributions now do by default.

C.
-- 
BOFH excuse #14:

sounds like a Windows problem, try calling Microsoft support