[LU-13582] open/rmdir/close vs xattr set race may cause a transaction exec abort

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Severity: 3
• Assignee: Alexander Zarochentsev
• Reporter: Alexander Zarochentsev

Description

An attempt to close a file handle to an ophan striped dir may race with a delayed xattr set or just an chrown. Due to lacking of DLM locks for mdt_close(), the race is not prevented by LDLM and only gets resolved at transaction exec phase. For a striped dir , the xattr set itransaction is a distributed operation, its rollback is complex, potentially buggy, and should be avoided if it is possible.

The following errors indicate an tx exec error:

/tmp/test_logs/1589821660/sanityn.test_107.debug_log.devvm1.1589821723.log:00000020:00000040:1.0:1589821723.399086:0:14322:0:(out_lib.c:752:out_tx_xattr_set_exec()) lustre-MDT0002-osd: set xattr buf ffff8d9e52e05548 name user.17 flag 0
/tmp/test_logs/1589821660/sanityn.test_107.debug_log.devvm1.1589821723.log:00000020:02000000:1.0:1589821723.399089:0:14322:0:(out_lib.c:756:out_tx_xattr_set_exec()) No object found [0x280000402:0x2:0x0]
/tmp/test_logs/1589821660/sanityn.test_107.debug_log.devvm1.1589821723.log:00000020:00000001:1.0:1589821723.399098:0:14322:0:(out_lib.c:843:out_tx_xattr_set_exec()) Process leaving via out (rc=18446744073709551614 : -2 : 0xfffffffffffffffe)
/tmp/test_logs/1589821660/sanityn.test_107.debug_log.devvm1.1589821723.log:00000020:00000040:1.0:1589821723.399100:0:14322:0:(out_lib.c:847:out_tx_xattr_set_exec()) lustre-MDT0002-osd: insert xattr set reply ffff8d9de08361e0 index 0: rc = -2

I got them by a reproducer I will post later.