RFR(XS) 8190809 JVM crashes while generating appcds for classpath with empty directory entry
Ioi Lam
ioi.lam at oracle.com
Fri Dec 1 21:57:36 UTC 2017
Hi,
I had to change the test because of JDK-8192927 "os::dir_is_empty is
incorrect on Windows"
import jdk.test.lib.Platform;
import jdk.test.lib.process.OutputAnalyzer;
import java.io.File;
public class DirClasspathTest {
public static void main(String[] args) throws Exception {
File dir = new File(System.getProperty("user.dir"));
File emptydir = new File(dir, "emptydir");
emptydir.mkdir();
// Empty dir in -cp: should be OK
OutputAnalyzer output;
if (!Platform.isWindows()) {
// This block fails on Windows because of JDK-8192927
output = TestCommon.dump(emptydir.getPath(),
TestCommon.list("DoesntMatter"), "-Xlog:class+path=info");
TestCommon.checkDump(output);
}
// Non-empty dir in -cp: should fail
// <dir> is not empty because it has at least one subdirectory,
i.e., <emptydir>
output = TestCommon.dump(dir.getPath(),
TestCommon.list("DoesntMatter"), "-Xlog:class+path=info");
output.shouldNotHaveExitValue(0);
output.shouldContain("CDS allows only empty directories in
archived classpaths");
}
}
So the test for "-cp emptydir" is now disabled on Windows.
Thanks
- Ioi
On 12/1/17 11:44 AM, Leonid Mesnik wrote:
> Thank you for update. Looks good.
>
> Leonid
>
>> On Dec 1, 2017, at 11:42 AM, Ioi Lam <ioi.lam at oracle.com
>> <mailto:ioi.lam at oracle.com>> wrote:
>>
>> Hi Leonid,
>>
>> Thanks for the review. I've changed that line to
>>
>> File dir = new File(System.getProperty("user.dir"));
>>
>> (which is similar to what TestCommon.java does with other temp files).
>>
>> I'll rerun the tests now.
>>
>> Thanks
>>
>> - Ioi
>>
>>
>> On 12/1/17 11:32 AM, Leonid Mesnik wrote:
>>> Fix looks good.
>>> The only one question. Why don’t just always use current dir in 39?
>>> 39 File dir = new File(System.getProperty("test.classes", "."));
>>> I think that jtreg always run test in new scratch directory however
>>> “test.classes” might be re-used between runs and people could have
>>> fail to create new directory if they re-run test locally.
>>>
>>> Leonid
>>>
>>>> On Dec 1, 2017, at 10:53 AM, Ioi Lam <ioi.lam at oracle.com
>>>> <mailto:ioi.lam at oracle.com>> wrote:
>>>>
>>>> Hi
>>>>
>>>> Please review this very small change. I hope to invoke the trivial
>>>> rule so that I can push before today's 8PM PDT deadline for JDK 10
>>>> integration.
>>>>
>>>> https://bugs.openjdk.java.net/browse/JDK-8190809
>>>> http://cr.openjdk.java.net/~iklam/jdk10/8190809-cds-crash-empty-cp.v01/
>>>>
>>>> The crash happened because CDS was trying to use a classpath as a
>>>> JAR file, before checking it's actually a JAR file.
>>>>
>>>> I added test case for the use of directories in -cp during CDS
>>>> dumping, both for empty and non-empty cases.
>>>>
>>>> I will test with tier1/tier2 testing before pushing.
>>>>
>>>> Thanks
>>>> - Ioi
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> PS: the check for non-emptiness is done in here:
>>>>
>>>> void ClassLoader::check_shared_classpath(const char *path) {
>>>> if (strcmp(path, "") == 0) {
>>>> exit_with_path_failure("Cannot have empty path in archived
>>>> classpaths", NULL);
>>>> }
>>>>
>>>> struct stat st;
>>>> if (os::stat(path, &st) == 0) {
>>>> if ((st.st_mode & S_IFMT) != S_IFREG) { // is not a regular file
>>>> if (!os::dir_is_empty(path)) {
>>>> tty->print_cr("Error: non-empty directory '%s'", path);
>>>> exit_with_path_failure("CDS allows only empty directories
>>>> in archived classpaths", NULL);
>>>> }
>>>> }
>>>> }
>>>> }
>>>>
>>>>
>>>
>>
>
More information about the hotspot-runtime-dev
mailing list