Cygwin: Cygwin test failures analysis

[Blackhex]

Description

As a follow up of #257 investigate the root causes of Cygwin tests failures. Update definition of done of the following issues or create new ones if another unimplemented APIs will be identified based on your findings:

• Cygwin: Port signal APIs handling to AArch64 #249
• Cygwin: Port fork/exec APIs to AArch64 #283
• Cygwin: Port filesystem APIs to AArch64 #284
• Cygwin: Port pipes APIs to AArch64 #285

Acceptance Criteria

• We know answers for the follwing questions:
  • What are the root causes of Cygwin tests failures?
  • What Cygwin tests are failing because of definicencies in what Cygwin POSIX APIs?
• The definition of done of the issues listed above were updated or new issues requesting implementation of additional APIs were created.

Analysis

• Cause: Tests are Hanging due to WaitForMultipleObjects (num, wait_objects, FALSE, INFINITE) function call in cygwait.cc line no:79. Until any one of the object is signalled the function will wait. Signals to be ported for deeper analysis of these tests.

  • winsup.api/ltp/pause01
  • winsup.api/checksignal
  • winsup.api/ltp/signal03

• Cause: In fcntl07, calling do_exec(subprog_path, **tcp, *tcd) returns 1, leading to a switch-case that breaks due to waitpid returning status 2. Internally, waitpid() calls wait4(), which invokes cygwait(waitfor, cw_infinite, cw_cancel | cw_cancel_self), modifying w->status.

  • winsup.api/ltp/fcntl07
  • winsup.api/ltp/fcntl07B

• Cause: The waitpid call in spawn.cc at line 894 is returning 11, which is stored in the return variable. This value (11) indicates that the process was terminated by a SIGSEGV (segmentation fault), rather than exiting normally.

  • winsup.api/systemcall

• Cause:The tests fails due to cygserver being unavailable, causing the CreateFileW function in transport_pipes.cc to return an invalid handle value (-1).

  • Winsup.api/semtest
  • Winsup.api/shmtest
  • Winsup.api/msgtest

• Cause: Siganl is not handled properly in test file.Siganls to be ported for deeper analysis of these tests.

  • Winsup.api/sigchld

• cause: The condition for permission fails at line no:102 in test file.

  • Winsup.api/ltp/umask03

• cause:The posix_spawnp call fails with "No such file or directory" because the provided environment is empty and lacks a PATH, preventing it from locating the "sh" executable at test file line no:20.

  • Winsup.api/posix_spawn/spawnp

• Cause: Missing pthread_wrapper implemenentation at https://github.com/Windows-on-ARM-Experiments/newlib-cygwin/blob/woarm64/winsup/cygwin/create_posix_thread.cc#L103:

  • winsup.api/pthread/cancel11

  • Other pthread tests which were failing are fixed by Add inline assembly pthread wrapper for aarch64 newlib-cygwin#28

• Cause: The failures occur due to insufficient pointer validation in the Cygwin implementation. When invalid pointers are passed as arguments (e.g., -1, NULL), the functions crash with access violations instead of returning -1 and setting errno as expected for negative test cases.

  • winsup.api/ltp/link04
  • winsup.api/ltp/stat03
  • winsup.api/ltp/symlink01
  • winsup.api/ltp/symlink03
  • winsup.api/ltp/unlink07
  • winsup.api/ltp/stat06
  • winsup.api/ltp/rmdir05
  • winsup.api/ltp/mkdir01
  • winsup.api/ltp/munmap01
  • winsup.api/ltp/munmap02
  • winsup.api/ltp/mmap05
  • winsup.api/ltp/times02
  • winsup.api/ltp/mmaptest02
  • winsup.api/ltp/mmaptest03
  • winsup.api/ltp/rename08
  • winsup.api/ltp/chmod01
  • winsup.api/ltp/chdir04
  • winsup.api/ltp/access01
  • winsup.api/ltp/access03
  • winsup.api/ltp/access05

• The following tests not working in the Larger pipeline at mingw-woarm64-build repo even after introducing cygwin environment variable.

  • winsup.api/ltp/access01
  • winsup.api/ltp/chmod01
  • winsup.api/ltp/link02
  • winsup.api/ltp/link03
  • winsup.api/ltp/link05
  • winsup.api/ltp/lseek04
  • winsup.api/ltp/lseek10
  • winsup.api/ltp/lstat02
  • winsup.api/ltp/readlink01
  • winsup.api/ltp/readlink02
  • winsup.api/ltp/rename01
  • winsup.api/ltp/select03
  • winsup.api/ltp/stat02
  • winsup.api/ltp/symlink04
  • winsup.api/ltp/symlink05
  • winsup.api/ltp/unlink06
  • winsup.api/ltp/mknod01
  • winsup.api/ltp/dup05

• Cause: The fork-based child process failed to copy Cygwin internal memory (cygheap) due to an invalid handle (Win32 error 6), leading to a segmentation fault (signal 11) and termination of the child process.

  • winsup.api/posix_spawn/fds
  • winsup.api/posix_spawn/signals
  • winsup.api/posix_spawn/chdir
  • winsup.api/signal-into-win32

n at path.cc line no:715.

[thiru-mcw]

In addition to the APIs identified through test failures, a visual inspection revealed // TODO comments specifically referencing AArch64 in the following locations:

• cpuid.h:21
• cpuid.h:47
• autoload.cc:87
• autoload.cc:131
• autoload.cc:214

While these may not be directly responsible for the current test failures, they are noted here to track pending AArch64-related work across the codebase.

[thiru-mcw]

• The following tests are passing in windows environment but failing in WSL environment. All the sub-test cases in these tests are passing in windows environment when executed from /mnt/c/test folder with wsl env or C:/test with command prompt.
  • winsup.api/ltp/access01.exe
    • The second test case fails at line no:188 access function call. The access is getting denied in base.cc line no:408 ,so the test is getting failed.
  • winsup.api/ltp/chmod01.exe
    • The test fails at line no:155 due to the mismatch in modes. So the test returns the fail status.
  • winsup.api/ltp/link02.exe
    • The link function fails in line no:171 so the test is getting failed . The NtsetInformationFile function call in disk_file line no:1494 fails , so the link function returns -1. NtsetInformationFile function
  • winsup.api/ltp/link03.exe
    • The link function fails in line no:213.Similar to link02 NtsetInformationFile function in disk_file fails here too.
  • winsup.api/ltp/link05.exe
    • The test fails t line no:278 , where the unlink function fails similar to mknod01 test. So the test is throwing warning at line no:279.
  • winsup.api/ltp/lseek04.exe
    • The open function call in line no:225 fails , so the test case is broken. Cygheap_fdnew object creation fails in syscalls.cc line no:1576 so it returns -1.
  • winsup.api/ltp/lseek10.exe
    • The close function call fails in line no:328, so the test case is broken. Fd1 value that is getting passed to close function is -1, so in cygheap_fdget function the if conditions fails so it returns -1.
  • winsup.api/ltp/lstat02.exe
    • The condition fails in line no:224 and throws warning.The readlink function fails and returns -1.In readlink function , the failure occurs at line no:3708 in path.cc .
  • winsup.api/ltp/readlink01.exe
    • The readlink function fails in line no:115 ,so the test returns -1 and it fails. In readlink function fails occurs in line no:3708 in path.cc and retruns -1.This is similar to Lstat02 test failure reason.
  • winsup.api/ltp/readlink02.exe
    • The readlink02 test also fails due to to same reason as readlink01 and lstat02 .The readlink function call occurs at line no:164.
  • winsup.api/ltp/rename01.exe
    • The if condition in line no:152 fails and the test is getting failed. In If condition the second condition is true , the inode number after rename function is changed.
  • winsup.api/ltp/select03.exe
    • Same reason as lseek04.The open function fails an returns -1.
  • winsup.api/ltp/stat02.exe
    • The test fails in line no:148 , where the results doesn't match the expected data for the stat function.
  • winsup.api/ltp/symlink04.exe & winsup.api/ltp/symlink05.exe
    • The tests fails at IS_LINK function call. The file pointed is not a symbolic so the tests are failing.
  • winsup.api/ltp/unlink06.exe
    • The test fails at unlink function call in line no:172.This is similar to mknod01 where the unlink function there too.
  • winsup.api/ltp/mknod01.exe
    • The test fails in line no:211 unlink function and in line no:212 rmdir function. The unlink function fails at syscalls.cc line no:1136 , file seems to nonexistent file.so it returns -1.The rmdir function return -1 as the fh doesn't exist in line no:487 in dir.cc.
  • winsup.api/ltp/dup05.exe
    • The test fails at line no:229 , where open functions fails so the test is broken.

[thiru-mcw]

The following are the test that are failing due to unable to handle invalid params.[Negative testing: sub-testcases]

• link04: calling link(fname1, fname2)) function with fname1 set to invalid pointer is to handle invalid addresses. The invalid pointer flows through the internal call chain: fn → name → src inside path_conv::check, but due to insufficient pointer validation the code crashes instead of returning -1 and setting error_value. This failure occurs at winsup/cygwin/path.cc#L715:

• stat03: calling stat(file_name, &stat_buf) function with file_name set to invalid pointer is to handle invalid addresses. The invalid pointer flows through the internal call chain: fn → name → src inside path_conv::check, the code crashes with access voilation instead of returning -1 and setting error_value. This failure occurs at winsup/cygwin/path.cc#L715:

• Symlink01 calling unlink(path2) function with path2 set to invalid pointer is to handle invalid addresses. The invalid pointer flows through the internal call chain: fn → name → src inside path_conv::check. the code crashes instead of returning -1 and setting error_value. This failure occurs at winsup/cygwin/path.cc#L715

• Symlink03 Calling symlink(test_file,sym_file) with sym_file set to invalid addresses. The flow goes to symlink in winsup/cygwin/path.cc file but due to insufficient pointer validation, the code crashes instead of returning -1 and setting error_value.

• Unlink07 calling unlink(fname) function with fname set to invalid address. The invalid pointer flows through the internal call chain: fn → name → src inside path_conv::check, the code crashes instead of returning -1 and setting error_value. This failure occurs at winsup/cygwin/path.cc#L715

• Stat06 calling stat(fname,stbuf) function with fname set to invalid address. The invalid pointer flows through the internal call chain: fn → name → src inside path_conv::check, the code crashes instead of returning -1 and setting error_value. This failure occurs at winsup/cygwin/path.cc#L715

• Rmdir05 calling rmdir((char *)-1) function to handle invalid pointer. The invalid pointer flows goes to winsup/cygwin/dir.cc file but due to insufficient pointer validation , the code crashes instead of returning -1 and setting error_value.

• Mkdir01 calling mkdir((char*)-1.0777) function to handle invalid pointer .The invalid pointer flows goes to winsup/cygwin/dir.cc file but due to insufficient pointer validation , the code crashes instead of returning -1 and setting error_value.

• Munmap01 calling munmap01(addr,map_len) function with addr is invalid pointer. When addr is used to store value it throws access violation.

• Munmap02 calling munmap02(addr,map_len) with addr is invalid pointer. When addr is used to store value it throws access violation.

• Mmap05 : file_content = addr[0]; causes access violation as mmap function is called with PROT_NONE in line no:129.So the test fails to handle this case and throws access violation.

• Times02 : times((void*)-1) is called in line no:82 in times02 file. So this causes access violation at times.cc line no:76.The times() fails to handle this input.

• mmaptest02 : The test fails at mmap.cc line no:1140.The len virtual address is invalid so the test is failing.

• mmaptest03 : The buf1 is unmapped at line no:120,so accessing buf1[0] is causing the access violation in line no:126.

• Shmtest :The test fails at line no:150 , The shm_buf passed to strcpy is invalid so the test fails to handle it.

• semtest : The test fails at line no:181 ,In sem.cc line no:88 where the request is not happening. So the test fails.

• Systemcall : The test fails at systemcall.c line no:31.The spawnvp function for "bin/sh" fails at syscalls.cc line no:2869.