[
https://issues.apache.org/jira/browse/WW-5677?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18107156#comment-18107156
]
Lukasz Lenart commented on WW-5677:
-----------------------------------
PR [#1859|https://github.com/apache/struts/pull/1859] opened against main.
Both call sites done, no behaviour change:
# {{checkDefaultPackageAccess}} now tests {{toPackageName(clazz).isEmpty()}}
instead of {{getPackage() == null || getPackage().getName().isEmpty()}}, which
also collapses the double {{getPackage()}} evaluation per class.
# {{isExcludedPackageNamePatterns}} resolves the package name once per call
instead of once per configured pattern.
Per the constraint on this ticket, the equivalence is asserted rather than
argued. {{defaultPackageConditionMatchesLegacyAcrossClassShapes}} runs the
replaced condition — frozen verbatim as an oracle that calls {{getPackage()}}
directly and never delegates to production code — against the new one over the
existing {{classShapes()}} matrix (arrays, primitives, {{void}}, a
default-package class, a lambda, a JDK proxy). Two behavioural tests cover the
gate itself: a named-package class still passes, and an array target stays
blocked.
All three were mutation-checked. Resolving arrays to {{java.lang}} fails the
equivalence test and the array-target test; inverting the member-class
condition fails the named-package test.
Two things deliberately left out of scope:
* The {{getPackage()}} calls in {{checkExclusionList}}'s {{LOG.warn}}
arguments. Deny path only, so no hot-path value, and changing them would alter
the log text from {{package java.io}} to {{java.io}}.
* Threading one package name through {{isPackageExcluded}} into both helpers.
It would remove the last redundant call, but
{{isExcludedPackageNames}}/{{isExcludedPackageNamePatterns}} are {{protected}}
and may be overridden, so the signature change is source-breaking — WW-5678
owns that for 8.0.0. {{toPackageName}} is now a cached field read plus a
branch, so calling it twice is negligible.
Neither path runs by default, so this is a consistency fix for deployments that
enable {{struts.disallowDefaultPackageAccess}} or configure
{{struts.excludedPackageNamePatterns}}, not where the WW-5667 9% lived — that
was WW-5675.
{{mvn test -DskipAssembly -pl core}}: 3190 tests, 0 failures, 0 errors.
> Remove the remaining redundant getPackage() lookups on the OGNL member-access
> path
> ----------------------------------------------------------------------------------
>
> Key: WW-5677
> URL: https://issues.apache.org/jira/browse/WW-5677
> Project: Struts 2
> Issue Type: Sub-task
> Reporter: Lukasz Lenart
> Assignee: Lukasz Lenart
> Priority: Major
> Fix For: 7.4.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Sub-task of WW-5667. Residual items found while reviewing WW-5674, which
> optimised {{isClassBelongsToPackages}} and {{toPackageName}} but deliberately
> left these two call sites alone as out of scope.
> Both sit on the same per-access hot path WW-5674 just optimised, within
> twenty lines of it, so leaving them is a visible inconsistency.
> h2. 1. {{checkDefaultPackageAccess}} makes four {{getPackage()}} calls
> {{SecurityMemberAccess.checkDefaultPackageAccess}} still calls
> {{clazz.getPackage()}} directly, twice per class and for up to two classes:
> {code:java}
> if (memberClass.getPackage() == null ||
> memberClass.getPackage().getName().isEmpty()) {
> {code}
> {{Class.getPackage()}} resolves through the defining classloader's package
> map on every call. WW-5674 removed exactly this lookup from {{toPackageName}}
> by switching to the cached {{Class.getPackageName()}}, so this method is now
> the only remaining user of the slow form.
> The condition is equivalent to {{toPackageName(memberClass).isEmpty()}} —
> including for arrays and primitives, where both forms treat the class as
> being in the default package and therefore block. Replacing it would also
> collapse the double {{getPackage()}} evaluation per class.
> Note that this method only runs when {{struts.disallowDefaultPackageAccess}}
> is enabled, so the impact is limited to deployments that turn it on.
> h2. 2. {{isExcludedPackageNamePatterns}} recomputes the package name per
> pattern
> {code:java}
> protected boolean isExcludedPackageNamePatterns(Class clazz) {
> return excludedPackageNamePatterns.stream().anyMatch(pattern ->
> pattern.matcher(toPackageName(clazz)).matches());
> }
> {code}
> {{toPackageName(clazz)}} is evaluated inside the lambda, so it runs once per
> pattern rather than once per call. Hoisting it out of the stream is a
> one-line change.
> The call is cheaper after WW-5674, but it is still N redundant calls.
> {{struts.excludedPackageNamePatterns}} is empty by default (both pattern
> constants are commented out in {{struts-excluded-classes.xml}}), so this only
> affects deployments that configure it.
> h2. Constraints
> This is the OGNL security gate. Both changes must be behaviour-preserving,
> and the equivalence of {{getPackage() == null ||
> getPackage().getName().isEmpty()}} with {{toPackageName(...).isEmpty()}} must
> be asserted by test rather than argued — in particular for arrays,
> primitives, {{void}} and default-package classes.
> {{SecurityMemberAccessPackageMatchingTest}} already provides the class-shape
> matrix and the frozen {{legacyToPackageName}} reference oracle to build on.
> h2. Related
> {{ConfigParseUtil.validatePackageNames}} evaluating
> {{Pattern.compile("\\s")}} once per package name rather than once overall is
> a per-instantiation cost, not a per-access one, and is tracked on WW-5675
> instead.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
