Re: [PATCH] ubifs: Kill unneeded locking in ubifs_init_security

[Richard Weinberger <richard@nod.at>]

Am 28.07.2015 um 13:21 schrieb Boris Brezillon:
> Hi Richard,
> 
> On Wed,  8 Jul 2015 11:46:36 +0200
> Richard Weinberger <richard@nod.at> wrote:
> 
>> Fixes the following lockdep splat:
>> [    1.244527] =============================================
>> [    1.245193] [ INFO: possible recursive locking detected ]
>> [    1.245193] 4.2.0-rc1+ #37 Not tainted
>> [    1.245193] ---------------------------------------------
>> [    1.245193] cp/742 is trying to acquire lock:
>> [    1.245193]  (&sb->s_type->i_mutex_key#9){+.+.+.}, at: [<ffffffff812b3f69>] ubifs_init_security+0x29/0xb0
>> [    1.245193]
>> [    1.245193] but task is already holding lock:
>> [    1.245193]  (&sb->s_type->i_mutex_key#9){+.+.+.}, at: [<ffffffff81198e7f>] path_openat+0x3af/0x1280
>> [    1.245193]
>> [    1.245193] other info that might help us debug this:
>> [    1.245193]  Possible unsafe locking scenario:
>> [    1.245193]
>> [    1.245193]        CPU0
>> [    1.245193]        ----
>> [    1.245193]   lock(&sb->s_type->i_mutex_key#9);
>> [    1.245193]   lock(&sb->s_type->i_mutex_key#9);
>> [    1.245193]
>> [    1.245193]  *** DEADLOCK ***
>> [    1.245193]
>> [    1.245193]  May be due to missing lock nesting notation
>> [    1.245193]
>> [    1.245193] 2 locks held by cp/742:
>> [    1.245193]  #0:  (sb_writers#5){.+.+.+}, at: [<ffffffff811ad37f>] mnt_want_write+0x1f/0x50
>> [    1.245193]  #1:  (&sb->s_type->i_mutex_key#9){+.+.+.}, at: [<ffffffff81198e7f>] path_openat+0x3af/0x1280
>> [    1.245193]
>> [    1.245193] stack backtrace:
>> [    1.245193] CPU: 2 PID: 742 Comm: cp Not tainted 4.2.0-rc1+ #37
>> [    1.245193] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5-0-ge51488c-20140816_022509-build35 04/01/2014
>> [    1.245193]  ffffffff8252d530 ffff88007b023a38 ffffffff814f6f49 ffffffff810b56c5
>> [    1.245193]  ffff88007c30cc80 ffff88007b023af8 ffffffff810a150d ffff88007b023a68
>> [    1.245193]  000000008101302a ffff880000000000 00000008f447e23f ffffffff8252d500
>> [    1.245193] Call Trace:
>> [    1.245193]  [<ffffffff814f6f49>] dump_stack+0x4c/0x65
>> [    1.245193]  [<ffffffff810b56c5>] ? console_unlock+0x1c5/0x510
>> [    1.245193]  [<ffffffff810a150d>] __lock_acquire+0x1a6d/0x1ea0
>> [    1.245193]  [<ffffffff8109fa78>] ? __lock_is_held+0x58/0x80
>> [    1.245193]  [<ffffffff810a1a93>] lock_acquire+0xd3/0x270
>> [    1.245193]  [<ffffffff812b3f69>] ? ubifs_init_security+0x29/0xb0
>> [    1.245193]  [<ffffffff814fc83b>] mutex_lock_nested+0x6b/0x3a0
>> [    1.245193]  [<ffffffff812b3f69>] ? ubifs_init_security+0x29/0xb0
>> [    1.245193]  [<ffffffff812b3f69>] ? ubifs_init_security+0x29/0xb0
>> [    1.245193]  [<ffffffff812b3f69>] ubifs_init_security+0x29/0xb0
>> [    1.245193]  [<ffffffff8128e286>] ubifs_create+0xa6/0x1f0
>> [    1.245193]  [<ffffffff81198e7f>] ? path_openat+0x3af/0x1280
>> [    1.245193]  [<ffffffff81195d15>] vfs_create+0x95/0xc0
>> [    1.245193]  [<ffffffff8119929c>] path_openat+0x7cc/0x1280
>> [    1.245193]  [<ffffffff8109ffe3>] ? __lock_acquire+0x543/0x1ea0
>> [    1.245193]  [<ffffffff81088f20>] ? sched_clock_cpu+0x90/0xc0
>> [    1.245193]  [<ffffffff81088c00>] ? calc_global_load_tick+0x60/0x90
>> [    1.245193]  [<ffffffff81088f20>] ? sched_clock_cpu+0x90/0xc0
>> [    1.245193]  [<ffffffff811a9cef>] ? __alloc_fd+0xaf/0x180
>> [    1.245193]  [<ffffffff8119ac55>] do_filp_open+0x75/0xd0
>> [    1.245193]  [<ffffffff814ffd86>] ? _raw_spin_unlock+0x26/0x40
>> [    1.245193]  [<ffffffff811a9cef>] ? __alloc_fd+0xaf/0x180
>> [    1.245193]  [<ffffffff81189bd9>] do_sys_open+0x129/0x200
>> [    1.245193]  [<ffffffff81189cc9>] SyS_open+0x19/0x20
>> [    1.245193]  [<ffffffff81500717>] entry_SYSCALL_64_fastpath+0x12/0x6f
>>
>> While the lockdep splat is a false positive, becuase path_openat holds i_mutex
>> of the parent directory and ubifs_init_security() tries to acquire i_mutex
>> of a new inode, it reveals that taking i_mutex in ubifs_init_security() is
>> in vain because it is only being called in the inode allocation path
>> and therefore nobody else can see the inode yet.
>>
>> Reported-by: Boris Brezillon <boris.brezillon@free-electrons.com>
>> Signed-off-by: Richard Weinberger <richard@nod.at>
> 
> It might be too late, but if it's not you can add my
> 
> Tested-by: Boris Brezillon <boris.brezillon@free-electrons.com>

Applied and pushed!

Thanks,
//richard