https://jira.whamcloud.com This file is an XML representation of an issue en-us 9.4.14 940014 05-12-2023 https://jira.whamcloud.com/browse/LU-5285 Lustre <p>Running racer on 2 nodes with MDSCOUNT=1 and <a href="http://review.whamcloud.com/#/c/5936/" class="external-link" target="_blank" rel="nofollow noopener">http://review.whamcloud.com/#/c/5936/</a> (add more operations to racer) I see this frequently.</p> <div class="preformatted panel" style="border-width: 1px;"><div class="preformattedContent panelContent"> <pre>[ 803.128945] Lustre: lustre-MDT0000: Client 045a894d-d2a7-e744-df1f-a695740583d8 (at 19\ 2.168.122.108@tcp) reconnecting [ 803.134578] LustreError: 5113:0:(lu_object.h:852:lu_object_attr()) ASSERTION( ((o)-&gt;lo\ _header-&gt;loh_attr &amp; LOHA_EXISTS) != 0 ) failed: [ 803.136583] LustreError: 5113:0:(lu_object.h:852:lu_object_attr()) LBUG [ 803.137789] Pid: 5113, comm: mdt00_005 [ 803.138439] [ 803.138440] Call Trace: [ 803.139129] [&lt;ffffffffa02be8c5&gt;] libcfs_debug_dumpstack+0x55/0x80 [libcfs] [ 803.140296] [&lt;ffffffffa02beec7&gt;] lbug_with_loc+0x47/0xb0 [libcfs] [ 803.141337] [&lt;ffffffffa0c034ac&gt;] mdt_attr_get_complex+0xaec/0xb80 [mdt] [ 803.142457] [&lt;ffffffffa0c20425&gt;] mdt_reconstruct_setattr+0xd5/0x410 [mdt] [ 803.143631] [&lt;ffffffffa04737e0&gt;] ? lu_ucred+0x20/0x30 [obdclass] [ 803.144677] [&lt;ffffffffa0c1fde5&gt;] mdt_reconstruct+0x45/0x120 [mdt] [ 803.145722] [&lt;ffffffffa0bfdbfa&gt;] mdt_reint_internal+0x6fa/0x7c0 [mdt] [ 803.146820] [&lt;ffffffffa0bfe24b&gt;] mdt_reint+0x6b/0x120 [mdt] [ 803.147818] [&lt;ffffffffa06e9c45&gt;] tgt_request_handle+0x245/0xad0 [ptlrpc] [ 803.148977] [&lt;ffffffffa069ce31&gt;] ptlrpc_main+0xcf1/0x1880 [ptlrpc] [ 803.150053] [&lt;ffffffffa069c140&gt;] ? ptlrpc_main+0x0/0x1880 [ptlrpc] [ 803.151093] [&lt;ffffffff8109eab6&gt;] kthread+0x96/0xa0 [ 803.151886] [&lt;ffffffff8100c30a&gt;] child_rip+0xa/0x20 [ 803.152714] [&lt;ffffffff81554710&gt;] ? _spin_unlock_irq+0x30/0x40 [ 803.153677] [&lt;ffffffff8100bb10&gt;] ? restore_args+0x0/0x30 [ 803.154572] [&lt;ffffffff8109ea20&gt;] ? kthread+0x0/0xa0 [ 803.155403] [&lt;ffffffff8100c300&gt;] ? child_rip+0x0/0x20 [ 803.156261] </pre> </div></div> <p>When I hit this I see the following sequence of events:</p> <ol> <li>The MDT crashes.</li> <li>The client node stays up.</li> <li>I restart the MDT.</li> <li>The client resends the setattr.</li> <li>GOTO 1.</li> </ol> LU-5285

mdt_reconstruct_setattr() calls mdt_attr_get_complex() without checking that object exists

Bug Critical Resolved Fixed Alex Zhuravlev John Hammond mdt mn4 Wed, 2 Jul 2014 17:34:56 +0000 Mon, 12 Jan 2015 07:37:41 +0000 Fri, 11 Jul 2014 11:19:53 +0000 Lustre 2.6.0 Lustre 2.6.0 Lustre 2.5.3 0 5 <p>Looks like this was added in <a href="http://review.whamcloud.com/3782" class="external-link" target="_blank" rel="nofollow noopener">http://review.whamcloud.com/3782</a></p> <p>From the source mdt_reconstruct_create() has exactly the same issue. I haven't gotten a crash there but it does pass an object to mdt_attr_get_complex() without checking that it exists. mdt_reconstruct_open() looks OK. Probably all uses of mdt_attr_get_complex() should be audited...</p> <p>Andreas, 2.1 doesn't check that either.</p> <p>why do we need to return the attributes in the setattr path? we're in the reconstruct path and given the original transaction was committed, rep-ack lock was cancelled and another client was able to destroy the object. so we can't return valid attribues. with no LDLM lock the attributes can't be used either?</p> <p>I'm OK with not returning the attributes for setattr if the object is missing. The setattr itself must have been processed, so it shouldn't return an error to the caller (i.e. no <tt>-ENOENT</tt>), but it is OK if the attributes on the client inode are not updated, and it could even be purged from cache if there is a mechanism to do that (e.g. return only <tt>mbo_nlink=0</tt> and <tt>mbo_flags=OBD_MD_FLNLINK</tt> to the client).</p> <p><a href="http://review.whamcloud.com/11025" class="external-link" target="_blank" rel="nofollow noopener">http://review.whamcloud.com/11025</a></p> <p>Landed for 2.6</p> Related LU-5125 LU-6085 LU-5388 LU-6088 LU-5340 Development Rank 1|hzwqfr: Rank (Obsolete) 14748 Severity