]> git.kernelconcepts.de Git - karo-tx-linux.git/commit
target: Fix a deadlock between the XCOPY code and iSCSI session shutdown
authorBart Van Assche <bart.vanassche@sandisk.com>
Tue, 23 May 2017 23:48:29 +0000 (16:48 -0700)
committerNicholas Bellinger <nab@linux-iscsi.org>
Fri, 7 Jul 2017 05:58:00 +0000 (22:58 -0700)
commitd877d7275be34ad70ce92bcbb4bb36cec77ed004
tree0d41926243642d076f085e4cb1ebb01a49c26f5a
parenta85d667e58bddf73be84d1981b41eaac985ed216
target: Fix a deadlock between the XCOPY code and iSCSI session shutdown

Move the code for parsing an XCOPY command from the context of
the iSCSI receiver thread to the context of the XCOPY workqueue.
Keep the simple XCOPY checks in the context of the iSCSI receiver
thread. Move the code for allocating and freeing struct xcopy_op
from the code that parses an XCOPY command to its caller.

This patch fixes the following deadlock:

======================================================
[ INFO: possible circular locking dependency detected ]
4.10.0-rc7-dbg+ #1 Not tainted
-------------------------------------------------------
rmdir/13321 is trying to acquire lock:
 (&sess->cmdsn_mutex){+.+.+.}, at: [<ffffffffa02cb47d>] iscsit_free_all_ooo_cmdsns+0x2d/0xb0 [iscsi_target_mod]

but task is already holding lock:
 (&sb->s_type->i_mutex_key#14){++++++}, at: [<ffffffff811c6e20>] vfs_rmdir+0x50/0x140

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:
-> #1 (&sb->s_type->i_mutex_key#14){++++++}:
 lock_acquire+0x71/0x90
 down_write+0x3f/0x70
 configfs_depend_item+0x3a/0xb0 [configfs]
 target_depend_item+0x13/0x20 [target_core_mod]
 target_xcopy_locate_se_dev_e4+0xdd/0x1a0 [target_core_mod]
 target_do_xcopy+0x34b/0x970 [target_core_mod]
 __target_execute_cmd+0x22/0xa0 [target_core_mod]
 target_execute_cmd+0x233/0x2c0 [target_core_mod]
 iscsit_execute_cmd+0x208/0x270 [iscsi_target_mod]
 iscsit_sequence_cmd+0x10b/0x190 [iscsi_target_mod]
 iscsit_get_rx_pdu+0x37d/0xcd0 [iscsi_target_mod]
 iscsi_target_rx_thread+0x6e/0xa0 [iscsi_target_mod]
 kthread+0x102/0x140
 ret_from_fork+0x31/0x40

-> #0 (&sess->cmdsn_mutex){+.+.+.}:
 __lock_acquire+0x10e6/0x1260
 lock_acquire+0x71/0x90
 mutex_lock_nested+0x5f/0x670
 iscsit_free_all_ooo_cmdsns+0x2d/0xb0 [iscsi_target_mod]
 iscsit_close_session+0xac/0x200 [iscsi_target_mod]
 lio_tpg_close_session+0x9f/0xb0 [iscsi_target_mod]
 target_shutdown_sessions+0xc3/0xd0 [target_core_mod]
 core_tpg_del_initiator_node_acl+0x91/0x140 [target_core_mod]
 target_fabric_nacl_base_release+0x20/0x30 [target_core_mod]
 config_item_release+0x5a/0xc0 [configfs]
 config_item_put+0x1d/0x1f [configfs]
 configfs_rmdir+0x1a6/0x300 [configfs]
 vfs_rmdir+0xb7/0x140
 do_rmdir+0x1f4/0x200
 SyS_rmdir+0x11/0x20
 entry_SYSCALL_64_fastpath+0x23/0xc6

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(&sb->s_type->i_mutex_key#14);
                               lock(&sess->cmdsn_mutex);
                               lock(&sb->s_type->i_mutex_key#14);
  lock(&sess->cmdsn_mutex);

 *** DEADLOCK ***

3 locks held by rmdir/13321:
 #0:  (sb_writers#10){.+.+.+}, at: [<ffffffff811e1aff>] mnt_want_write+0x1f/0x50
 #1:  (&default_group_class[depth - 1]#2/1){+.+.+.}, at: [<ffffffff811cc8ce>] do_rmdir+0x15e/0x200
 #2:  (&sb->s_type->i_mutex_key#14){++++++}, at: [<ffffffff811c6e20>] vfs_rmdir+0x50/0x140

stack backtrace:
CPU: 2 PID: 13321 Comm: rmdir Not tainted 4.10.0-rc7-dbg+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Call Trace:
 dump_stack+0x86/0xc3
 print_circular_bug+0x1c7/0x220
 __lock_acquire+0x10e6/0x1260
 lock_acquire+0x71/0x90
 mutex_lock_nested+0x5f/0x670
 iscsit_free_all_ooo_cmdsns+0x2d/0xb0 [iscsi_target_mod]
 iscsit_close_session+0xac/0x200 [iscsi_target_mod]
 lio_tpg_close_session+0x9f/0xb0 [iscsi_target_mod]
 target_shutdown_sessions+0xc3/0xd0 [target_core_mod]
 core_tpg_del_initiator_node_acl+0x91/0x140 [target_core_mod]
 target_fabric_nacl_base_release+0x20/0x30 [target_core_mod]
 config_item_release+0x5a/0xc0 [configfs]
 config_item_put+0x1d/0x1f [configfs]
 configfs_rmdir+0x1a6/0x300 [configfs]
 vfs_rmdir+0xb7/0x140
 do_rmdir+0x1f4/0x200
 SyS_rmdir+0x11/0x20
 entry_SYSCALL_64_fastpath+0x23/0xc6

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Andy Grover <agrover@redhat.com>
Cc: David Disseldorp <ddiss@suse.de>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
drivers/target/target_core_xcopy.c