[PATCH] kernel32: Add a special case to EnumProcessModules for the local process

[Martin Storsjo]

LLVM's libunwind uses EnumProcessModules for locating .eh_frame
sections when unwinding dwarf exceptions (used on i686).

When running wine in docker (without adding custom additional
permissions), EnumProcessModules currently fails as it uses
ReadProcessMemory, which requires the SYS_PTRACE capability.

The current implementation also is slower than necessary (by a
couple orders of magnituide), for accessing the current process.
Currently, unwinding 10000 exception throws with libunwind on i686
takes 24 seconds when run in wine, while it runs in less than 0.1
second with this patch.

Signed-off-by: Martin Storsjo <martin(a)martin.st>
---
 dlls/kernel32/module.c | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

diff --git a/dlls/kernel32/module.c b/dlls/kernel32/module.c
index b7f19ffe36..45814dbff3 100644
--- a/dlls/kernel32/module.c
+++ b/dlls/kernel32/module.c
@@ -555,6 +555,40 @@ BOOL WINAPI K32EnumProcessModules(HANDLE process, HMODULE *lphModule,
     DWORD size = 0;
     INT ret;
 
+    if (process == GetCurrentProcess())
+    {
+        PPEB_LDR_DATA ldr_data = NtCurrentTeb()->Peb->LdrData;
+        PLIST_ENTRY head = &ldr_data->InLoadOrderModuleList;
+        PLIST_ENTRY entry = head->Flink;
+
+        if (cb && !lphModule)
+        {
+            SetLastError(ERROR_NOACCESS);
+            return FALSE;
+        }
+        while (entry != head)
+        {
+            PLDR_MODULE table_entry = (PLDR_MODULE)
+                ((PBYTE)entry - offsetof(LDR_MODULE, InLoadOrderModuleList));
+            if (cb >= sizeof(HMODULE))
+            {
+                *lphModule++ = table_entry->BaseAddress;
+                cb -= sizeof(HMODULE);
+            }
+            size += sizeof(HMODULE);
+            entry = entry->Flink;
+        }
+
+        if (!needed)
+        {
+            SetLastError(ERROR_NOACCESS);
+            return FALSE;
+        }
+        *needed = size;
+
+        return TRUE;
+    }
+
     if (!init_module_iterator(&iter, process))
         return FALSE;
 
-- 
2.17.1

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=58713

Your paranoid android.


=== debian10 (32 bit report) ===

kernel32:
debugger.c:320: Test failed: GetThreadContext failed: 5
debugger.c:320: Test failed: GetThreadContext failed: 5

=== debian10 (32 bit Chinese:China report) ===

kernel32:
debugger: Timeout

=== debian10 (64 bit WoW report) ===

kernel32:
debugger.c:320: Test failed: GetThreadContext failed: 5
debugger.c:320: Test failed: GetThreadContext failed: 5
debugger.c:320: Test failed: GetThreadContext failed: 5