eCryptfs: Fix lockdep-reported AB-BA mutex issue
authorRoland Dreier <rdreier@cisco.com>
Wed, 1 Jul 2009 22:48:18 +0000 (15:48 -0700)
committerTyler Hicks <tyhicks@linux.vnet.ibm.com>
Wed, 23 Sep 2009 14:10:30 +0000 (09:10 -0500)
commitaa06117f19944573cda0c4bee026c916b5256090
treee5cb24f500431a5201e7a50b1e9df62fbd24e1bb
parent05dafedb906425fe935199f4c92700d87285e3e9
eCryptfs: Fix lockdep-reported AB-BA mutex issue

Lockdep reports the following valid-looking possible AB-BA deadlock with
global_auth_tok_list_mutex and keysig_list_mutex:

  ecryptfs_new_file_context() ->
      ecryptfs_copy_mount_wide_sigs_to_inode_sigs() ->
          mutex_lock(&mount_crypt_stat->global_auth_tok_list_mutex);
          -> ecryptfs_add_keysig() ->
              mutex_lock(&crypt_stat->keysig_list_mutex);

vs

  ecryptfs_generate_key_packet_set() ->
      mutex_lock(&crypt_stat->keysig_list_mutex);
      -> ecryptfs_find_global_auth_tok_for_sig() ->
          mutex_lock(&mount_crypt_stat->global_auth_tok_list_mutex);

ie the two mutexes are taken in opposite orders in the two different
code paths.  I'm not sure if this is a real bug where two threads could
actually hit the two paths in parallel and deadlock, but it at least
makes lockdep impossible to use with ecryptfs since this report triggers
every time and disables future lockdep reporting.

Since ecryptfs_add_keysig() is called only from the single callsite in
ecryptfs_copy_mount_wide_sigs_to_inode_sigs(), the simplest fix seems to
be to move the lock of keysig_list_mutex back up outside of the where
global_auth_tok_list_mutex is taken.  This patch does that, and fixes
the lockdep report on my system (and ecryptfs still works OK).

The full output of lockdep fixed by this patch is:

=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.31-2-generic #14~rbd2
-------------------------------------------------------
gdm/2640 is trying to acquire lock:
 (&mount_crypt_stat->global_auth_tok_list_mutex){+.+.+.}, at: [<ffffffff8121591e>] ecryptfs_find_global_auth_tok_for_sig+0x2e/0x90

but task is already holding lock:
 (&crypt_stat->keysig_list_mutex){+.+.+.}, at: [<ffffffff81217728>] ecryptfs_generate_key_packet_set+0x58/0x2b0

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&crypt_stat->keysig_list_mutex){+.+.+.}:
       [<ffffffff8108c897>] check_prev_add+0x2a7/0x370
       [<ffffffff8108cfc1>] validate_chain+0x661/0x750
       [<ffffffff8108d2e7>] __lock_acquire+0x237/0x430
       [<ffffffff8108d585>] lock_acquire+0xa5/0x150
       [<ffffffff815526cd>] __mutex_lock_common+0x4d/0x3d0
       [<ffffffff81552b56>] mutex_lock_nested+0x46/0x60
       [<ffffffff8121526a>] ecryptfs_add_keysig+0x5a/0xb0
       [<ffffffff81213299>] ecryptfs_copy_mount_wide_sigs_to_inode_sigs+0x59/0xb0
       [<ffffffff81214b06>] ecryptfs_new_file_context+0xa6/0x1a0
       [<ffffffff8120e42a>] ecryptfs_initialize_file+0x4a/0x140
       [<ffffffff8120e54d>] ecryptfs_create+0x2d/0x60
       [<ffffffff8113a7d4>] vfs_create+0xb4/0xe0
       [<ffffffff8113a8c4>] __open_namei_create+0xc4/0x110
       [<ffffffff8113d1c1>] do_filp_open+0xa01/0xae0
       [<ffffffff8112d8d9>] do_sys_open+0x69/0x140
       [<ffffffff8112d9f0>] sys_open+0x20/0x30
       [<ffffffff81013132>] system_call_fastpath+0x16/0x1b
       [<ffffffffffffffff>] 0xffffffffffffffff

-> #0 (&mount_crypt_stat->global_auth_tok_list_mutex){+.+.+.}:
       [<ffffffff8108c675>] check_prev_add+0x85/0x370
       [<ffffffff8108cfc1>] validate_chain+0x661/0x750
       [<ffffffff8108d2e7>] __lock_acquire+0x237/0x430
       [<ffffffff8108d585>] lock_acquire+0xa5/0x150
       [<ffffffff815526cd>] __mutex_lock_common+0x4d/0x3d0
       [<ffffffff81552b56>] mutex_lock_nested+0x46/0x60
       [<ffffffff8121591e>] ecryptfs_find_global_auth_tok_for_sig+0x2e/0x90
       [<ffffffff812177d5>] ecryptfs_generate_key_packet_set+0x105/0x2b0
       [<ffffffff81212f49>] ecryptfs_write_headers_virt+0xc9/0x120
       [<ffffffff8121306d>] ecryptfs_write_metadata+0xcd/0x200
       [<ffffffff8120e44b>] ecryptfs_initialize_file+0x6b/0x140
       [<ffffffff8120e54d>] ecryptfs_create+0x2d/0x60
       [<ffffffff8113a7d4>] vfs_create+0xb4/0xe0
       [<ffffffff8113a8c4>] __open_namei_create+0xc4/0x110
       [<ffffffff8113d1c1>] do_filp_open+0xa01/0xae0
       [<ffffffff8112d8d9>] do_sys_open+0x69/0x140
       [<ffffffff8112d9f0>] sys_open+0x20/0x30
       [<ffffffff81013132>] system_call_fastpath+0x16/0x1b
       [<ffffffffffffffff>] 0xffffffffffffffff

other info that might help us debug this:

2 locks held by gdm/2640:
 #0:  (&sb->s_type->i_mutex_key#11){+.+.+.}, at: [<ffffffff8113cb8b>] do_filp_open+0x3cb/0xae0
 #1:  (&crypt_stat->keysig_list_mutex){+.+.+.}, at: [<ffffffff81217728>] ecryptfs_generate_key_packet_set+0x58/0x2b0

stack backtrace:
Pid: 2640, comm: gdm Tainted: G         C 2.6.31-2-generic #14~rbd2
Call Trace:
 [<ffffffff8108b988>] print_circular_bug_tail+0xa8/0xf0
 [<ffffffff8108c675>] check_prev_add+0x85/0x370
 [<ffffffff81094912>] ? __module_text_address+0x12/0x60
 [<ffffffff8108cfc1>] validate_chain+0x661/0x750
 [<ffffffff81017275>] ? print_context_stack+0x85/0x140
 [<ffffffff81089c68>] ? find_usage_backwards+0x38/0x160
 [<ffffffff8108d2e7>] __lock_acquire+0x237/0x430
 [<ffffffff8108d585>] lock_acquire+0xa5/0x150
 [<ffffffff8121591e>] ? ecryptfs_find_global_auth_tok_for_sig+0x2e/0x90
 [<ffffffff8108b0b0>] ? check_usage_backwards+0x0/0xb0
 [<ffffffff815526cd>] __mutex_lock_common+0x4d/0x3d0
 [<ffffffff8121591e>] ? ecryptfs_find_global_auth_tok_for_sig+0x2e/0x90
 [<ffffffff8121591e>] ? ecryptfs_find_global_auth_tok_for_sig+0x2e/0x90
 [<ffffffff8108c02c>] ? mark_held_locks+0x6c/0xa0
 [<ffffffff81125b0d>] ? kmem_cache_alloc+0xfd/0x1a0
 [<ffffffff8108c34d>] ? trace_hardirqs_on_caller+0x14d/0x190
 [<ffffffff81552b56>] mutex_lock_nested+0x46/0x60
 [<ffffffff8121591e>] ecryptfs_find_global_auth_tok_for_sig+0x2e/0x90
 [<ffffffff812177d5>] ecryptfs_generate_key_packet_set+0x105/0x2b0
 [<ffffffff81212f49>] ecryptfs_write_headers_virt+0xc9/0x120
 [<ffffffff8121306d>] ecryptfs_write_metadata+0xcd/0x200
 [<ffffffff81210240>] ? ecryptfs_init_persistent_file+0x60/0xe0
 [<ffffffff8120e44b>] ecryptfs_initialize_file+0x6b/0x140
 [<ffffffff8120e54d>] ecryptfs_create+0x2d/0x60
 [<ffffffff8113a7d4>] vfs_create+0xb4/0xe0
 [<ffffffff8113a8c4>] __open_namei_create+0xc4/0x110
 [<ffffffff8113d1c1>] do_filp_open+0xa01/0xae0
 [<ffffffff8129a93e>] ? _raw_spin_unlock+0x5e/0xb0
 [<ffffffff8155410b>] ? _spin_unlock+0x2b/0x40
 [<ffffffff81139e9b>] ? getname+0x3b/0x240
 [<ffffffff81148a5a>] ? alloc_fd+0xfa/0x140
 [<ffffffff8112d8d9>] do_sys_open+0x69/0x140
 [<ffffffff81553b8f>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff8112d9f0>] sys_open+0x20/0x30
 [<ffffffff81013132>] system_call_fastpath+0x16/0x1b

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
fs/ecryptfs/crypto.c
fs/ecryptfs/keystore.c