lkml.org 
[lkml]   [2014]   [Apr]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [for-next][PATCH 14/14] tracing: Get trace_array ref counts when accessing trace files
On 04/08/2014 12:36 PM, Steven Rostedt wrote:
> On Sat, 05 Apr 2014 10:59:10 -0400
> Sasha Levin <sasha.levin@oracle.com> wrote:
>
>> On 07/02/2013 04:22 PM, Steven Rostedt wrote:
>>> From: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>
>>>
>>> When a trace file is opened that may access a trace array, it must
>>> increment its ref count to prevent it from being deleted.
>>>
>>> Cc: stable@vger.kernel.org # 3.10
>>> Reported-by: Alexander Lam <azl@google.com>
>>> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
>>
>> Hi Steven,
>>
>> This patch seems to cause the following lockdep warning:
>>
>> [ 5644.288019] ======================================================
>> [ 5644.288771] [ INFO: possible circular locking dependency detected ]
>> [ 5644.289657] 3.14.0-next-20140403-sasha-00019-g7474aa9-dirty #376 Not tainted
>> [ 5644.290568] -------------------------------------------------------
>> [ 5644.290783] trinity-c17/19105 is trying to acquire lock:
>> [ 5644.290783] (trace_types_lock){+.+.+.}, at: trace_array_get (kernel/trace/trace.c:225)
>> [ 5644.290783]
>> [ 5644.290783] but task is already holding lock:
>> [ 5644.290783] (&sig->cred_guard_mutex){+.+.+.}, at: prepare_bprm_creds (fs/exec.c:1165)
>> [ 5644.290783]
>> [ 5644.290783] which lock already depends on the new lock.
>> [ 5644.290783]
>> [ 5644.290783]
>> [ 5644.290783] the existing dependency chain (in reverse order) is:
>> [ 5644.290783]
>> -> #2 (&sig->cred_guard_mutex){+.+.+.}:
>> [ 5644.290783] lock_acquire (arch/x86/include/asm/current.h:14 kernel/locking/lockdep.c:3602)
>> [ 5644.290783] mutex_lock_interruptible_nested (kernel/locking/mutex.c:486 kernel/locking/mutex.c:616)
>> [ 5644.290783] proc_pid_attr_write (fs/proc/base.c:2250)
>> [ 5644.290783] __kernel_write (fs/read_write.c:457)
>> [ 5644.290783] write_pipe_buf (fs/splice.c:1072)
>> [ 5644.290783] splice_from_pipe_feed (fs/splice.c:834)
>> [ 5644.290783] __splice_from_pipe (fs/splice.c:955)
>> [ 5644.290783] splice_from_pipe (fs/splice.c:990)
>> [ 5644.290783] default_file_splice_write (fs/splice.c:1084)
>> [ 5644.290783] SyS_splice (include/linux/fs.h:2333 fs/splice.c:1391 fs/splice.c:1764 fs/splice.c:1749)
>> [ 5644.290783] tracesys (arch/x86/kernel/entry_64.S:749)
>> [ 5644.290783]
>> -> #1 (&pipe->mutex/1){+.+.+.}:
>> [ 5644.290783] lock_acquire (arch/x86/include/asm/current.h:14 kernel/locking/lockdep.c:3602)
>> [ 5644.290783] mutex_lock_nested (kernel/locking/mutex.c:486 kernel/locking/mutex.c:587)
>> [ 5644.290783] pipe_lock_nested (fs/pipe.c:61)
>> [ 5644.290783] pipe_lock (fs/pipe.c:69)
>> [ 5644.290783] splice_to_pipe (include/linux/wait.h:103 fs/splice.c:247)
>> [ 5644.290783] tracing_buffers_splice_read (kernel/trace/trace.c:5423)
>> [ 5644.290783] do_splice_to (fs/splice.c:1151)
>> [ 5644.290783] SyS_splice (fs/splice.c:1416 fs/splice.c:1764 fs/splice.c:1749)
>> [ 5644.290783] ia32_sysret (arch/x86/ia32/ia32entry.S:430)
>> [ 5644.290783]
>> -> #0 (trace_types_lock){+.+.+.}:
>> [ 5644.290783] __lock_acquire (kernel/locking/lockdep.c:1840 kernel/locking/lockdep.c:1945 kernel/locking/lockdep.c:2131 kernel/locking/lockdep.c:3182)
>> [ 5644.290783] lock_acquire (arch/x86/include/asm/current.h:14 kernel/locking/lockdep.c:3602)
>> [ 5644.290783] mutex_lock_nested (kernel/locking/mutex.c:486 kernel/locking/mutex.c:587)
>> [ 5644.290783] trace_array_get (kernel/trace/trace.c:225)
>> [ 5644.290783] tracing_open_generic_tr (kernel/trace/trace.c:3053)
>> [ 5644.290783] do_dentry_open (fs/open.c:753)
>> [ 5644.290783] finish_open (fs/open.c:818)
>> [ 5644.290783] do_last (fs/namei.c:3040)
>> [ 5644.290783] path_openat (fs/namei.c:3182)
>> [ 5644.290783] do_filp_open (fs/namei.c:3231)
>> [ 5644.290783] do_open_exec (fs/exec.c:766)
>> [ 5644.290783] do_execve_common.isra.19 (fs/exec.c:1491)
>> [ 5644.290783] compat_SyS_execve (fs/exec.c:1627)
>> [ 5644.290783] ia32_ptregs_common (arch/x86/ia32/ia32entry.S:495)
>> [ 5644.290783]
>> [ 5644.290783] other info that might help us debug this:
>> [ 5644.290783]
>> [ 5644.290783] Chain exists of:
>> trace_types_lock --> &pipe->mutex/1 --> &sig->cred_guard_mutex
>>
>> [ 5644.290783] Possible unsafe locking scenario:
>> [ 5644.290783]
>> [ 5644.290783] CPU0 CPU1
>> [ 5644.290783] ---- ----
>> [ 5644.290783] lock(&sig->cred_guard_mutex);
>> [ 5644.290783] lock(&pipe->mutex/1);
>> [ 5644.290783] lock(&sig->cred_guard_mutex);
>> [ 5644.290783] lock(trace_types_lock);
>> [ 5644.290783]
>> [ 5644.290783] *** DEADLOCK ***
>> [ 5644.290783]
>> [ 5644.290783] 1 lock held by trinity-c17/19105:
>> [ 5644.290783] #0: (&sig->cred_guard_mutex){+.+.+.}, at: prepare_bprm_creds (fs/exec.c:1165)
>> [ 5644.290783]
>> [ 5644.290783] stack backtrace:
>> [ 5644.290783] CPU: 10 PID: 19105 Comm: trinity-c17 Not tainted 3.14.0-next-20140403-sasha-00019-g7474aa9-dirty #376
>> [ 5644.290783] ffffffffb4a1a1e0 ffff88071a7738f8 ffffffffb14bfb2f 0000000000000000
>> [ 5644.290783] ffffffffb49a9dd0 ffff88071a773948 ffffffffb14b2527 0000000000000001
>> [ 5644.290783] ffff88071a7739d8 ffff88071a773948 ffff8805d98cbcf0 ffff8805d98cbd28
>> [ 5644.290783] Call Trace:
>> [ 5644.290783] dump_stack (lib/dump_stack.c:52)
>> [ 5644.290783] print_circular_bug (kernel/locking/lockdep.c:1214)
>> [ 5644.290783] __lock_acquire (kernel/locking/lockdep.c:1840 kernel/locking/lockdep.c:1945 kernel/locking/lockdep.c:2131 kernel/locking/lockdep.c:3182)
>> [ 5644.290783] ? _raw_spin_unlock_irqrestore (arch/x86/include/asm/paravirt.h:809 include/linux/spinlock_api_smp.h:160 kernel/locking/spinlock.c:191)
>> [ 5644.290783] ? preempt_count_sub (kernel/sched/core.c:2527)
>> [ 5644.290783] ? __slab_free (mm/slub.c:2598)
>> [ 5644.290783] lock_acquire (arch/x86/include/asm/current.h:14 kernel/locking/lockdep.c:3602)
>> [ 5644.290783] ? trace_array_get (kernel/trace/trace.c:225)
>> [ 5644.290783] mutex_lock_nested (kernel/locking/mutex.c:486 kernel/locking/mutex.c:587)
>> [ 5644.290783] ? trace_array_get (kernel/trace/trace.c:225)
>> [ 5644.290783] ? locks_free_lock (fs/locks.c:244)
>> [ 5644.290783] ? trace_array_get (kernel/trace/trace.c:225)
>> [ 5644.290783] ? preempt_count_sub (kernel/sched/core.c:2527)
>> [ 5644.290783] trace_array_get (kernel/trace/trace.c:225)
>> [ 5644.290783] tracing_open_generic_tr (kernel/trace/trace.c:3053)
>> [ 5644.290783] do_dentry_open (fs/open.c:753)
>> [ 5644.290783] ? tracing_open_pipe (kernel/trace/trace.c:3047)
>> [ 5644.290783] finish_open (fs/open.c:818)
>> [ 5644.290783] do_last (fs/namei.c:3040)
>> [ 5644.290783] ? link_path_walk (fs/namei.c:1473 fs/namei.c:1744)
>> [ 5644.290783] ? __this_cpu_preempt_check (lib/smp_processor_id.c:63)
>> [ 5644.290783] ? trace_hardirqs_on_caller (kernel/locking/lockdep.c:2557 kernel/locking/lockdep.c:2599)
>> [ 5644.290783] path_openat (fs/namei.c:3182)
>> [ 5644.290783] ? __lock_acquire (kernel/locking/lockdep.c:3189)
>> [ 5644.290783] do_filp_open (fs/namei.c:3231)
>> [ 5644.290783] ? put_lock_stats.isra.12 (arch/x86/include/asm/preempt.h:98 kernel/locking/lockdep.c:254)
>> [ 5644.290783] ? do_execve_common.isra.19 (fs/exec.c:1489)
>> [ 5644.290783] ? get_parent_ip (kernel/sched/core.c:2472)
>> [ 5644.290783] do_open_exec (fs/exec.c:766)
>> [ 5644.290783] do_execve_common.isra.19 (fs/exec.c:1491)
>> [ 5644.290783] ? do_execve_common.isra.19 (include/linux/spinlock.h:303 fs/exec.c:1258 fs/exec.c:1486)
>> [ 5644.290783] compat_SyS_execve (fs/exec.c:1627)
>> [ 5644.290783] ia32_ptregs_common (arch/x86/ia32/ia32entry.S:495)
>>
>
> Wait a minute! The syscall is a compat_SyS_execve, which does an
> do_execve_common, which does a do_open_exec, which ends up opening a
> tracing file????
>
> How the hell did that happen. Is your tool execing files in the tracing
> directory? I wonder if that's even possible, and if so, how can we
> prevent that? The trace_pipe file (which is what uses the
> tracing_open_pipe) is set to rrr. No exec should be allowed.

It could happen easily manually:

# ls -al trace_pipe
-r--r--r-- 1 root 0 0 Apr 8 16:43 trace_pipe
# chmod 777 trace_pipe
# ls -al trace_pipe
-rwxrwxrwx 1 root 0 0 Apr 8 16:43 trace_pipe
# ./trace_pipe

Although when I mount everything for the fuzzer I do it with '-onoexec,nosuid'
and the fuzzer is banned from testing mount(), so I'm not sure how it would
do that thing on it's own.


Thanks,
Sasha


\
 
 \ /
  Last update: 2014-04-08 19:01    [W:0.088 / U:1.824 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site