[#LU-14621] Broken lock-transaction ordering in MDS code

there are few places where local lock is taken before transaction starts which breaks transaction-then-locks rule:

 lbug_with_loc.cold.6+0x18/0x18 [libcfs]
 ? osd_trans_start+0x2f1/0x5a0 [osd_ldiskfs]
 osd_trans_start+0x2fd/0x5a0 [osd_ldiskfs]
 top_trans_start+0x40c/0x940 [ptlrpc]
 ? mdd_orphan_declare_delete+0x176/0x5c0 [mdd]
 mdd_orphan_cleanup_thread+0xaa1/0x18f0 [mdd]
 ? mdd_orphan_declare_delete+0x5c0/0x5c0 [mdd]
 kthread+0x11a/0x130

[<0>] libcfs_call_trace+0x76/0xa0 [libcfs]
[<0>] lbug_with_loc+0x3e/0x80 [libcfs]
[<0>] osd_trans_start+0x2fd/0x5a0 [osd_ldiskfs]
[<0>] top_trans_start+0x40c/0x940 [ptlrpc]
[<0>] mdd_swap_layouts+0x12f5/0x2350 [mdd]
[<0>] mdt_swap_layouts+0x40e/0x9a0 [mdt]

[<0>] osd_trans_start+0x2fd/0x5a0 [osd_ldiskfs]
[<0>] top_trans_start+0x40c/0x940 [ptlrpc]
[<0>] mdd_xattr_set+0x18c7/0x2e50 [mdd]
[<0>] mdt_close_handle_layouts+0xe23/0x1160 [mdt]
[<0>] mdt_mfd_close+0x5af/0x3110 [mdt]
[<0>] mdt_close_internal+0xfd/0x230 [mdt]
[<0>] mdt_close+0x60a/0x840 [mdt]

Call Trace:
[<0>] libcfs_call_trace+0x76/0xa0 [libcfs]
[<0>] lbug_with_loc+0x3e/0x80 [libcfs]
[<0>] osd_trans_start+0x2fd/0x5a0 [osd_ldiskfs]
[<0>] __lfsck_layout_update_pfid+0x15f/0x580 [lfsck]
[<0>] lfsck_layout_slave_in_notify_local+0x4ed/0x710 [lfsck]
[<0>] lfsck_in_notify_local+0x81/0x3f0 [lfsck]
[<0>] ofd_inconsistency_verification_main+0x1f2/0xa70 [ofd]

it's trivial to reproduce with the following patch:

index 04984f5d9f..b91b7c73e5 100644
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -1913,6 +1913,9 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 
        ENTRY;
 
+       LASSERT(oti->oti_w_locks == 0);
+       LASSERT(oti->oti_r_locks == 0);
+
        LASSERT(current->journal_info == NULL);
 
        oh = container_of(th, struct osd_thandle, ot_super);

[LU-14621] Broken lock-transaction ordering in MDS code Created: 18/Apr/21 Updated: 29/Aug/23 Resolved: 31/Jul/21
Status:	Resolved
Project:	Lustre
Component/s:	None
Affects Version/s:	Upstream
Fix Version/s:	Lustre 2.15.0

[LU-14621] Broken lock-transaction ordering in MDS code Created: 18/Apr/21 Updated: 29/Aug/23 Resolved: 31/Jul/21