Created attachment 36943 [details] Corrupted XLSX file with malformed _rels/.rels XML When a user creates XSSFWorkbook instance from some file or stream resource, but exception happens in the process, ZipFile can't be closed by the developer (neither explicitly catching an exception nor using `try-with-resource` statement), but closes later invoking ZipFile#finalize() during garbage collecting. The message "Cleaning up unclosed ZipFile for archive ..." is printed to stderr. Because the behavior is depended on encapsulated GC behavior it's hard to supply a simple working code snippet to reproduce. The bug is observable in a web application when multiple XLSX files are uploaded and parsed with some time pauses. But from the source code perspective it could be also understood. Imagine we want to print sheets number in an XLSX workbook: 1) try (final var workbook = new XSSFWorkbook(inputFile)) { System.out.println(workbook.getNumberOfSheets()); } catch (Exception e) { // constructor throws POIXMLException, or XMLException, or InvalidFormatException // workbook is null } 2) The steps in the source code starting from XSSFWorkbook(File) XSSFWorkbook(File) -> OPCPackage#open(File) -> ZipPackage(File) -> ZipFile(...) https://github.com/apache/poi/blob/a0770034fcc878a697b715d0b6a6e53406b2d02d/src/ooxml/java/org/apache/poi/xssf/usermodel/XSSFWorkbook.java#L324 https://github.com/apache/commons-compress/blob/master/src/main/java/org/apache/commons/compress/archivers/zip/ZipFile.java#L363 This creates a private ZipFile closable instance inside OPCPackage and continues the workbook constructor XSSFWorkbook(OPCPackage pkg): POIXMLDocument(OPCPackage pkg) -> POIXMLDocumentPart(OPCPackage pkg, String coreDocumentRel) -> POIXMLDocumentPart#getPartFromOPCPackage(OPCPackage pkg, String coreDocumentRel) -> throw new POIXMLException("OOXML file structure broken/invalid - no core document found!"); https://github.com/apache/poi/blob/a0770034fcc878a697b715d0b6a6e53406b2d02d/src/ooxml/java/org/apache/poi/ooxml/POIXMLDocumentPart.java#L753 Even if we catch the exception we can't close/release the ZipFile instance because it's private! In the same time we don't have a XSSFWorkbook to close/autoclose it in case of an exception. The object is detached from the thread and becomes available for GC, but the file stream is still not closed! Note: 1) other ZIP/OOXML parsers likely affected (like DOCX, PPTX) 2) other exceptions (like "POIXMLException: org.apache.xmlbeans.XmlException: error: DOCTYPE is disallowed when the feature "http://apache.org/xml/features/disallow-doctype-decl" set to true.") also create such detached closable source instances. 2) Some methods have proper exception handlers, like OPCPackage#open(File, PackageAccess): https://github.com/apache/poi/blob/a0770034fcc878a697b715d0b6a6e53406b2d02d/src/ooxml/java/org/apache/poi/openxml4j/opc/OPCPackage.java#L228 catch (InvalidFormatException | RuntimeException e) { IOUtils.closeQuietly(pack); throw e; }
I added a few changes: https://github.com/apache/poi/commit/22266e209a8f13a1ee88bafecb80e005303b165d Hopefully this build will succeed and new poi jars should be available to test with. Please suggest any other changes you think are useful.
Reviewing the POI sources i found some places which could be considered in the fix: 1) `PackageHelper#open(InputStream)` (and maybe also `clone()`) is used directly in `super` and `this` constructor calls for XmlVisioDocument, XMLSlideShow, XWPFDocument, XSSFWorkbook 2) same for: - POIXMLDocument#openPackage() - XSSFWorkbook#newPackage() - OPCPackage: - open(XXX) - openOrCreate(XXX) - create(XXXX) It seems there are a lot of places where `OPCPackage` (or other Closeable local instances) are wrapped/created and passed as arguments to init methods and constructor, but not properly handled in case of exceptions.
I fixed a few additional cases where file-handles could leak via r1872287, mostly in test-code itself, but also one case in EMF handling. We use file-leak-detector to find any leak that is triggered when running unit-tests. So a good way to help poi-devs would be to provide patches for unit-tests which actually trigger the remaining cases. This way we ensure these are caught automatically in the future.
As for the code-locations that you list as still causing file-handle leaks: I could not find any actual possible way to cause a leak in a quick check for any of these, can you point out in more detail how you think those can leak file-handles?
Unfortunately i'll be able to continue only next Tuesday. I'll post the update
@dominik.stadler and @fanningpj I tested current sources (https://github.com/apache/poi/commit/4aa8334e3b695cc34a2bfb035b9cc35dc9271e73) on my project - can't reproduce "Cleaning up unclosed ZipFile for archive" any more! Thank you. > So a good way to help poi-devs would be to provide patches for unit-tests which actually trigger the remaining cases. I'd like to do this, but the solution and the test cases are quite huge, so hard to do this without a good introduction to the sources from actual developers. > can you point out in more detail how you think those can leak file-handles? That weren't real cases, just my assumption based on the sources, but as i mentioned above - i don't see the entire picture. BUT! When i run the gradle tests locally for ooxml subroject (`gradle --info --console=verbose :ooxml:clean :ooxml:test`) I see 2 another stderr traces about unclosed ZipFile: 1) >org.apache.poi.xddf.usermodel.chart.TestXDDFChartRemoveSeries > testRemoveSeries0 STANDARD_OUT > testRemoveSeries0: test file written to /home/andrii/workspace/poi-report/poi/build/ooxml/build/custom-reports-test/testRemoveSeries0.xlsx >org.apache.poi.openxml4j.opc.internal.TestContentTypeManager > testContentTypeRemovalPackage SKIPPED >org.apache.poi.openxml4j.opc.TestPackage > removePartRecursive SKIPPED >org.apache.poi.util.TestTempFileThreaded > runTest STANDARD_ERROR > Cleaning up unclosed ZipFile for archive /home/andrii/workspace/poi-report/poi/test-data/openxml4j/TestPackageCommon.docx >org.apache.poi.util.TestTempFileThreaded > runTest STANDARD_OUT > thread: 7, iter: 0: /tmp/poifiles/TestTempFileThreaded/26/TestTempFile-7-0-5000898654042792707.xlsx > thread: 0, iter: 0: /tmp/poifiles/TestTempFileThreaded/19/TestTempFile-0-0-9041124478588812993.xlsx > thread: 5, iter: 0: /tmp/poifiles/TestTempFileThreaded/24/TestTempFile-5-0-4091795158149176089.xlsx 2) >org.apache.poi.xssf.streaming.TestSXSSFCell > setCellFormula_isExceptionSafe_onBlankCell SKIPPED >org.apache.poi.xssf.streaming.TestSXSSFCell > setBlank_throwsISE_ifCellIsPartOfAnArrayFormulaGroupContainingOtherCells SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > test54764WithSAXHelper STANDARD_ERROR > [Fatal Error] :1:1: JAXP00010001: The parser has encountered more than "1" entity expansions in this document; this is the limit imposed by the JDK. >org.apache.poi.xssf.usermodel.TestXSSFBugs > bug48877 STANDARD_ERROR > Cleaning up unclosed ZipFile for archive /home/andrii/workspace/poi-report/poi/test-data/spreadsheet/54764.xlsx >org.apache.poi.xssf.usermodel.TestXSSFBugs > test55273 SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > test57699 SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > test58043 SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > testSetRGBBackgroundColor SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > testBug53798XLSXStream SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > testBug56688_3 STANDARD_ERROR > Cleaning up unclosed ZipFile for archive /home/andrii/workspace/poi-report/poi/test-data/spreadsheet/54764.xlsx >org.apache.poi.xssf.usermodel.TestXSSFBugs > bug46670 SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > bug59393_commentsCanHaveSameAnchor SKIPPED >org.apache.poi.xssf.usermodel.TestXSSFBugs > test57929 SKIPPED **Do you think these could be some undiscovered cases?**
Another thing which confuses me a bit: Considering these changes in `XSSWorkbook#newPackage(XSSFWorkbookType)`: - https://github.com/apache/poi/compare/adb8424bc1a1c9a502d2cd07757615b711d32c50..4aa8334e3b695cc34a2bfb035b9cc35dc9271e73#diff-512a44831b35fadfdff4e2cd0bb33374 - https://github.com/apache/poi/blob/fb8f17190064f39f8f35700aa6ba1fbc2cfd88e1/src/ooxml/java/org/apache/poi/xssf/usermodel/XSSFWorkbook.java#L495-L495 now the implementation looks diverse from similar one in `XWPFDocument#newPackage()`: https://github.com/apache/poi/blob/fb8f17190064f39f8f35700aa6ba1fbc2cfd88e1/src/ooxml/java/org/apache/poi/xwpf/usermodel/XWPFDocument.java#L156-L156 I mean the `try-catch` block with following `IOUtils.closeQuietly(pkg);`. From the first view: - either it doesn't have sense in `XSSWorkbook` - or it also shall be used in `XWPFDocument` Do you understand what i mean?
XSSFWorkbook and XWPFWorkbook were made similar via r1872453. I tried to reproduce the warnings during running unit-tests that you posted, unfortunately I could not reproduce those for now. Unfortunately the feature in commons-compress which we use here is not very well-suited for debugging such findings: it does not print any stacktrace from where the file was allocated and it depends on finalizer/garbage-collection which is unpredictable and thus makes it impossible to properly reproduce things with simple test-cases. We usually use file-leak-detector from https://github.com/kohsuke/file-leak-detector/ to check for unclosed resources. It is much better in showing details and being reproducible but commons-compress actually prevents file-leak-detector from detecting the leaks here. Therefore I have reported https://issues.apache.org/jira/browse/COMPRESS-502 to allow to disable this feature in commons-compress in the future so our own testing for file-leaks can be used fully again. However unless that happens I don't see more work possible here for now, so I am closing this for now. Please report any additional leaking in new issues so we can adjust code accordingly.