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-10579 Lustre <p>There seems to be an issue with XATTR changelogs for some specific prefixes such as <tt>trusted</tt> or <tt>security</tt>, but not <tt>user</tt>.</p> <div class="preformatted panel" style="border-width: 1px;"><div class="preformattedContent panelContent"> <pre># bash lustre/tests/llmount.sh # touch /mnt/lustre/blob # lctl --device lustre-MDT0000 changelog_register # setfattr -n trusted.test -v test /mnt/lustre/blob # lfs changelog lustre-MDT0000 1 -1(null) 08:13:35.501636480 2018.01.30 0x0 t=[0x200000401:0x1:0x0] j=setfattr.0 ef=0x1 u=0:0 # setfattr -n user.test -v test /mnt/lustre/blob # lfs changelog lustre-MDT0000 1 -1(null) 08:13:35.501636480 2018.01.30 0x0 t=[0x200000401:0x1:0x0] j=setfattr.0 ef=0x1 u=0:0 2 15XATTR 08:13:21.191376815 2018.01.30 0x0 t=[0x200000401:0x1:0x0] j=setfattr.0 ef=0x1 u=0:0 </pre> </div></div> LU-10579

chlg: -1(null) changelog for setfattr on trusted.*

Bug Minor Resolved Fixed John Hammond CEA Tue, 30 Jan 2018 08:24:30 +0000 Wed, 5 Aug 2020 13:50:25 +0000 Tue, 9 Oct 2018 23:11:29 +0000 Lustre 2.10.6 0 5 <p>Bruno</p> <p>Could you please investigate?</p> <p>Thanks</p> <p>Peter</p> <p>It seems that we're not handling these prefixes in <tt>mdd_xattr_changelog_type()</tt>:</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java"><span class="code-keyword">static</span> <span class="code-keyword">enum</span> changelog_rec_type mdd_xattr_changelog_type(<span class="code-keyword">const</span> struct lu_env *env, struct mdd_device *mdd, <span class="code-keyword">const</span> <span class="code-object">char</span> *xattr_name) { <span class="code-comment">/* Layout changes systematically recorded */</span> <span class="code-keyword">if</span> (strcmp(XATTR_NAME_LOV, xattr_name) == 0 || strncmp(XATTR_LUSTRE_LOV, xattr_name, strlen(XATTR_LUSTRE_LOV)) == 0) <span class="code-keyword">return</span> CL_LAYOUT; <span class="code-comment">/* HSM information changes systematically recorded */</span> <span class="code-keyword">if</span> (strcmp(XATTR_NAME_HSM, xattr_name) == 0) <span class="code-keyword">return</span> CL_HSM; <span class="code-keyword">if</span> (has_prefix(xattr_name, XATTR_USER_PREFIX) || has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_ACCESS) || has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_DEFAULT)) <span class="code-keyword">return</span> CL_XATTR; <span class="code-keyword">return</span> -1; } </pre> </div></div> <p>Mike</p> <p>Can you please investigate?</p> <p>Peter</p> <p>I ran into this working on changelog processing for HPE DMF 7.</p> <p><tt>mdd_xattr_changelog_type()</tt> returns -1 to indicate that no record should be emitted. This is itself a rather dubious practice: I'd prefer that CL_XATTR records be generated for those cases as well. At present I got half my wish, because they're generated as -1 records.</p> <p>The reason is that the return type is not <tt>int</tt> but <tt>enum changelog_rec_type</tt>, and this is an unsigned type. So checks like the one in  <tt>mdd_declare_xattr_set()</tt> always fail, and the record is generated with type <tt>((enum changelog_rec_type)-1)</tt>.</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java"> <span class="code-keyword">static</span> <span class="code-object">int</span> mdd_declare_xattr_set(<span class="code-keyword">const</span> struct lu_env *env, struct mdd_device *mdd, struct mdd_object *obj, <span class="code-keyword">const</span> struct lu_buf *buf, <span class="code-keyword">const</span> <span class="code-object">char</span> *name, <span class="code-object">int</span> fl, struct thandle *handle) { <span class="code-keyword">enum</span> changelog_rec_type type; <span class="code-object">int</span> rc; rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle); <span class="code-keyword">if</span> (rc) <span class="code-keyword">return</span> rc; type = mdd_xattr_changelog_type(env, mdd, name); &gt;&gt;&gt; <span class="code-keyword">if</span> (type &lt; 0) &gt;&gt;&gt; <span class="code-keyword">return</span> 0; <span class="code-comment">/* no changelog to store */</span> <span class="code-keyword">return</span> mdd_declare_changelog_store(env, mdd, type, NULL, NULL, handle); } </pre> </div></div> <p>My preferred solution is that these changelog records be generated for these cases, rather than dropped on the floor. This can be done by changing <tt>mdd_xattr_changelog_type()</tt> to look like this:</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java"> <span class="code-keyword">static</span> <span class="code-keyword">enum</span> changelog_rec_type mdd_xattr_changelog_type(<span class="code-keyword">const</span> struct lu_env *env, struct mdd_device *mdd, <span class="code-keyword">const</span> <span class="code-object">char</span> *xattr_name) { <span class="code-comment">/* Layout changes systematically recorded */</span> <span class="code-keyword">if</span> (strcmp(XATTR_NAME_LOV, xattr_name) == 0 || strncmp(XATTR_LUSTRE_LOV, xattr_name, strlen(XATTR_LUSTRE_LOV)) == 0) <span class="code-keyword">return</span> CL_LAYOUT; <span class="code-comment">/* HSM information changes systematically recorded */</span> <span class="code-keyword">if</span> (strcmp(XATTR_NAME_HSM, xattr_name) == 0) <span class="code-keyword">return</span> CL_HSM; <span class="code-comment">/* Everything <span class="code-keyword">else</span> systematically recorded */</span> <span class="code-keyword">return</span> CL_XATTR; } </pre> </div></div> <p> If this is done then the checks of the return value can also be removed of course.</p> <p>This looks like fallout from <a href="https://jira.whamcloud.com/browse/LU-7264" title="store ChangeLog record for security.* xattr changes" class="issue-link" data-issue-key="LU-7264"><del>LU-7264</del></a> so would be present in 2.10 and later.</p> <p>Fixed in 2.11 by <a href="https://review.whamcloud.com/28251" class="external-link" target="_blank" rel="nofollow noopener">https://review.whamcloud.com/28251</a> <a href="https://jira.whamcloud.com/browse/LU-9727" title="Lustre Audit with Changelogs" class="issue-link" data-issue-key="LU-9727"><del>LU-9727</del></a> lustre: add CL_GETXATTR for Changelogs.</p> <p>John L. Hammond (jhammond@whamcloud.com) uploaded a new patch: <a href="https://review.whamcloud.com/33048" class="external-link" target="_blank" rel="nofollow noopener">https://review.whamcloud.com/33048</a><br/> Subject: <a href="https://jira.whamcloud.com/browse/LU-10579" title="chlg: -1(null) changelog for setfattr on trusted.*" class="issue-link" data-issue-key="LU-10579"><del>LU-10579</del></a> mdd: emit changelogs for securit* and trusted xattrs<br/> Project: fs/lustre-release<br/> Branch: b2_10<br/> Current Patch Set: 1<br/> Commit: 9958d6f33212d818a439801b88c480207e449908</p> <p>Note that 33048 doesn't address the issue that returning -1 from <tt>mdd_xattr_changelog_type()</tt> does not, in fact, suppress the generation of a changelog entry.</p> <p>Hi Olaf,</p> <p>Sorry, I didn't read your comments as carefully as I should have.</p> <p>I think we can add a negative sentinel value to the enum that forces the compiler to use a signed type for it. Or return <tt>CL_SETXATTR</tt> as a catch all type.</p> <p>Hi John,</p> <p>As noted, my preference would be to just emit the changelog event in all cases. It is simpler to code, and I do not see the value of suppressing the events for certain xattrs. So returning CL_SETXATTR for all the "other" cases has my vote.</p> <p>Expanding a bit on my previous comment.</p> <p>What I'm looking for is the completion of this sentence: "No changelog processor should be notified of changes to an extended attribute named like <em>this</em> because ..."</p> <p>Maybe there is one. But until there is one, I continue to think we should emit changelog entries for changes to every extended attribute.</p> <p>John L. Hammond (jhammond@whamcloud.com) merged in patch <a href="https://review.whamcloud.com/33048/" class="external-link" target="_blank" rel="nofollow noopener">https://review.whamcloud.com/33048/</a><br/> Subject: <a href="https://jira.whamcloud.com/browse/LU-10579" title="chlg: -1(null) changelog for setfattr on trusted.*" class="issue-link" data-issue-key="LU-10579"><del>LU-10579</del></a> mdd: emit changelogs for security and trusted xattrs<br/> Project: fs/lustre-release<br/> Branch: b2_10<br/> Current Patch Set: <br/> Commit: ab246d23566d7da1a41479b688833f47976b6428</p> <p>Landed for 2.10.6. Not needed for master</p> Development Rank 1|hzzrvj: Rank (Obsolete) 9223372036854775807 Severity