This is *the* fix we have all been waiting for, folks.
These 2 fixes were created by Anonymous author, who also developed unofficial U891711, Q918547, GDI*.*, KERNEL32.DLL + older SHELL98 fixes.
SHELL98:
http://www.mdgx.com/web.htm#9SU
Restored proper SHELL32.DLL 4.72.3812.634 English.
This SHELL32.DLL version corrects *almost all* EXPLORER.EXE lockups independent of free USER resources percentage:
* Unofficial Windows 98/98 SP1/98 SE Explorer Lockups With Internet Explorer 5.xx/6.xx SHELL32.DLL 4.72.3812.634 Fix:
- SHELL32.DLL Fix [475 KB, English]:
http://www.mdgx.com/files/SHELL98.EXE
- SHELL32.DLL Fix [475 KB, French]:
http://www.mdgx.com/files/SHELL98F.EXE
- SHELL32.DLL Fix [475 KB, Italian]:
http://www.mdgx.com/files/SHELL98I.EXE
This SHELL32.DLL Fix corrects Windows Explorer (and similar tools that use SHELL32.DLL APIs) lockups while trying to copy/create/move/delete/rename large number of files/folders with Internet Explorer 5.xx/6.xx already installed.
BUG: This SHELL32.DLL fix breaks the ability of removing annoying arrows from Desktop shortcut icons!
FIX: See "PATCHED SHELL32.DLL BUG + FIX" for complete details:
http://www.mdgx.com/98-5.htm#PSBF
SHELLME:
http://www.mdgx.com/web.htm#MEU
Same SHELL32.DLL version "rearranged" for easier localization (translation into languages other than English):
* Unofficial Windows ME Explorer Lockups With Internet Explorer 5.xx/6.xx SHELL32.DLL 5.50.4134.120 Fix:
Direct download [843 KB, English]:
http://www.mdgx.com/files/SHELLME.EXE
Direct download [847 KB, Italian]:
http://www.mdgx.com/files/ITSHELME.EXE
This SHELL32.DLL Fix corrects Windows Explorer (and similar tools that use SHELL32.DLL APIs) lockups while trying to copy/create/move/delete/rename large number of files/folders with Internet Explorer 5.xx/6.xx already installed.
How to uninstall this fix [valid for all versions]:
Start button -> Settings -> Control Panel -> Add/Remove Programs -> select "Remove Unofficial Explorer Lockups SHELL32.DLL Fix" -> click Add/Remove button -> reboot.
Author's comments...
Quote
It does not fix the underlying bug in USER.EXE and its WM processing, but is designed to prevent the series of EXPLORER.EXE hangs 100% even if User Resources are low etc.
Although some MSFN members seem to think otherwise, it is not a bug in SHELL32.DLL - it just happens to be the best place to take care of the problem.
As always, this patch comes with the notice "Use at your own risk!"
I am very sorry but I have no time to localize SHELL32.DLL and I also have no plans at this point to port the code from Win98/98 SE SHELL32.DLL 4.72.3812.634 or 635 to WinME 5.50.4134.120 ( --> .134) - the patch is more difficult to implement under WinME.
However, I am sending a "rearranged" SHELL32.DLL 5.50.4134.120 [for WinME], the same "workaround", which should be much easier to localize = translate localizable strings into languages other than English.
Rick Chauvin wrote on Jan 3 2007, 05:46 PM:
Quote
(SHELL98.EXE) fix for Explorer lockups appears? to be more effective if
the Kernel32.dll fix (Copy2gb.exe) is also installed. To explain more of
what I've noticed is that if only the Shell32.dll is updated the hang
problem is yes much improved (70% successful) over just having the
original one in place - but I still can make it hang although not nearly
as readily as without it that's for sure; however if I have both the
shell32 and Kernel32 dlls updated then in my observances as of yet
anyway, it appears to be even more improved (85% successful) and more
difficult to trigger the hang problem. (In any of my tests when it hangs
the User, System, and GDI resources still show high %) Even though not a
perfect fix, the Shell32.dll workaround is a good one and well worth it,
and many thanks again to anonymous for taking the time to look at this
problem.
Many thanks for testing the patch and for your comments.
Is there a sure-fire method that you know and use which triggers the bug successfully in most cases?
It would be very helpful to know.
It may help pinpoint the location of the actual bug in USER.EXE and whatever code in KRNL386.EXE/KERNEL32.DLL USER.EXE calls.
Which versions of KERNEL32.DLL have you been using for these tests?
4.10.2225 (official version) from 320798usa8.exe or an earlier version?
The updated version from COPY2GB.EXE on your machine is 4.10.2226 (unofficial version), correct?
LLXX wrote on Jan 6 2007, 07:39 PM:
Quote
64-bit file pointers; the problem is in _llseek which interprets its
32-bit file pointer as signed irregardless of the desired seek (from the
beginning, from the current position, or from the end), and thus
attempts to move the file pointer past the beginning of the file even
when the desired seek is from the beginning (in which case a negative
offset would be nonsensical). In the unfixed version of _llseek,
SetFilePointer gets passed the sign-extended file pointer, causing *it*
to attempt to seek past the beginning of the file. I have fixed this by
forcing it to zero-extend the file pointer to 64 bits (thus keeping its
sign positive) *only* when a seek from the beginning is specified.
Otherwise the file pointer is signed and can be moved within the file as
usual.
Anonymous seems to have fixed shell32.dll to use SetFilePointer instead,
although this still leaves _llseek flawed. SetFilePointer doesn't need
any changes.
Quote
was fixed. Edit: This problem doesn't affect explorer.exe in WinME so
maybe that file is just a dummy.
KERNEL32.SetFilePointer has two modes: If called with a 32-bit pointer to the high-order DWORD of a 64-bit file pointer, it uses a 64-bit distance to move and uses it correctly.
However, if the pointer to the high-order DWORD is NULL, KERNEL32.SetFilePointer uses a 32-bit distance to move.
This is how KERNEL32._llseek calls KERNEL32.SetFilePointer (if unpatched) - KERNEL32._llseek itself is just a stub to call KERNEL32.SetFilePointer!
This is very different from a (real = non-zero) pointer pointing to a high-order DWORD that is zero.
AFAIK, _llseek and SetFilePointer are flawed in all versions of Windows (Win9x, WinNT, Win2k and WinXP) and probably in Vista, too.
Interestingly, the official description on msdn2.microsoft.com of the SetFilePointer API (and its 32-bit call mechanism!) has changed several times over the years. One may speculate if this was done to disguise the flaw.
P.S.:
The code below is an example of a bug in USER.EXE.
Unfortunately, it is *not* the bug that causes the EXPLORER.EXE hang, but the type of bug I was alluding to in an earlier message.
This bug can cause mayhem when Resources are too low and KERNEL.LOCALALLOC fails.
This type of bug occurs more than once in USER.EXE, but also exists in GDI.EXE, which highlights that coding practices apparently were not at their best when Windows 95 was developed.
push 0x40 push 0x14 call KERNEL.LOCALALLOC mov cx, ds <-- ds<>0 mov [bp-4], ax mov [bp-2], ds or cx, ax <-- always <>0 jnz ... xor ax, ax jmp ...
Quote
- SHELL32.DLL 5.50.4134.120 for Windows ME
- SHELL32.DLL 4.72.3812.648 for Windows 98/98 SP1/98 SE
that prevents the EXPLORER hang from occuring when a large number of files is deleted, moved etc.
As I wrote earlier, the problem arises because USER.EXE runs completely out of resources (=0%) right after such a file operation is completed.
However, this is not a bug in USER.EXE. The patch I am providing is not a fix in the true sense, it is more of a workaround and, as always, is provided 'as is' without warranties expressed or implied and with the note 'Use it at your own risk.'
The problem may still occur if a large number of files is deleted, moved etc, while USER resources are low, say below 30%.
I also believe this problem has nothing to do with what versions of SHLWAPI.DLL and/or BROWSEUI.DLL are installed.
It may occur much more frequently with the IE6 SP1 versions because they have much more code, which also takes longer to execute (and may no longer be "optimized" for a Win9x environment).
May I use this opportunity to address a common misconception about GDI and
USER resources in Win9x/ME: The OS and all applications, whether they are
16-bit or 32-bit versions, share the same GDI and USER resources. These
resources are managed by GDI.EXE and USER.EXE, respectively. One of the
two data segments of GDI.EXE is for GDI resources and one of the two data
segments of USER.EXE (plus two extra heaps) is for USER resources. GDI and
USER objects (in these data segments) are accessed through *16-bit-wide*
handles. 32-bit applications (and DLLs) just use zero-extended 32-bit
versions of the same handles. The data segments I am referring to are
32-bit segments with an *initial* size of approx. 2.1 MiByte each. The
16-bit resources are just the lowest 65536 bytes each in these two 32-bit
data segments. 16-bit-wide handles and the 16-bit resource limit cause the
real bottle neck in resources of Win9x. Increasing the default size of the
two 32-bit data segments (which would be possible using hidden system
settings for them) makes very, very little, if any difference.
Please post here your test results.
Thanks.
Enjoy.



Help

Back to top










