Lockdep complains about potential recursive locking during mount
because the client configuration log is holding a lock on the MGC
obd_device to prevent it from being torn down, while also getting
locks on the MDC and OSC devices as they are instantiated:
Lustre: Mounted myth-client
=============================================
[ INFO: possible recursive locking detected ]
---------------------------------------------
2 locks held by ll_cfg_requeue/1460:
#0: (&cli->cl_sem){.+.+..}, at mgc_requeue_thread+0x27d/0x5d8 [mgc]
#1: (&cld->cld_lock){+.+...}, at mgc_process_log+0x88/0xc01 [mgc]
CPU: 3 PID: 1460 Comm: ll_cfg_requeue
Call Trace:
[<
ffffffff814cb360>] dump_stack+0x4f/0x6f
[<
ffffffff81081d73>] __lock_acquire+0xc06/0xc74
[<
ffffffff81081f4f>] lock_acquire+0x16e/0x194
[<
ffffffff814ce5d7>] down_read+0x4c/0x61
[<
ffffffffa029f370>] sptlrpc_conf_client_adapt+0x95/0x142 [ptlrpc]
[<
ffffffffa03a7a69>] mdc_set_info_async+0xeb/0x420 [mdc]
[<
ffffffffa01ad69e>] class_notify_sptlrpc_conf+0x2f4/0x347 [obdclass]
[<
ffffffffa04a805c>] mgc_process_log+0xa31/0xc01 [mgc]
[<
ffffffffa04a94ef>] mgc_requeue_thread+0x31f/0x5d8 [mgc]
[<
ffffffff81061f77>] kthread+0xfa/0x102
Add a separate lock class for the MGC callpath, since it will always
be held first, and none of the other obd_device locks should ever
be held concurrently.
Signed-off-by: Andreas Dilger <andreas.dilger@intel.com>
Change-Id: Ic36e64608d12dcd4f5b12056e3e2072ca6500c1e
Reviewed-on: http://review.whamcloud.com/17582
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Dmitry Eremin <dmitry.eremin@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
*/
#define OBD_MAX_DEFAULT_EA_SIZE 4096
+enum obd_cl_sem_lock_class {
+ OBD_CLI_SEM_NORMAL,
+ OBD_CLI_SEM_MGC,
+ OBD_CLI_SEM_MDCOSC,
+};
+
struct mdc_rpc_lock;
struct obd_import;
struct client_obd {
LASSERT(atomic_read(&cld->cld_refcount) > 0);
- /* Do not run mgc_process_log on a disconnected export or an
+ /*
+ * Do not run mgc_process_log on a disconnected export or an
* export which is being disconnected. Take the client
- * semaphore to make the check non-racy. */
- down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
+ * semaphore to make the check non-racy.
+ */
+ down_read_nested(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem,
+ OBD_CLI_SEM_MGC);
if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) {
CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname);
rc = mgc_process_log(cld->cld_mgcexp->exp_obd, cld);
strcmp(obd->obd_type->typ_name, LUSTRE_LWP_NAME) == 0);
CDEBUG(D_SEC, "obd %s\n", obd->u.cli.cl_target_uuid.uuid);
- /* serialize with connect/disconnect import */
- down_read(&obd->u.cli.cl_sem);
+ /* serialize with connect/disconnect import */
+ down_read_nested(&obd->u.cli.cl_sem, OBD_CLI_SEM_MDCOSC);
imp = obd->u.cli.cl_import;
if (imp) {