June 2012 - Wine-Devel - list.winehq.org

Re: Incompatibility in Kernel32

by Ben Peddell

On 21/06/2012 7:11 PM, Dmitry Timoshkov wrote: > Hans Leidekker <hans(a)codeweavers.com> wrote: > >>> Is there any reason that you call DeleteFile() on a still being opened file? >> >> That's the whole point of this test. The observed behavior is that you can pass a >> filename to MsiRecordSetStream, successfully delete the file, and keep on using the >> stream. >> >> We currently implement MsiRecordSetStream by creating an in-memory copy of the >> stream. As Robert noticed, this will fail if the stream is too large. >> >> The idea is to move to an implementation on top of a file handle, but it should >> still allow DeleteFile to succeed while the file is in use. > > Thanks, I see. Removing GENERIC_WRITE from flags passed to NtCreateFile by > DeleteFile makes the test I just sent for that behaviour pass (and that's > most like a correct fix), but that leads to some other test failures. > This needs further investigation. server/fd.c:open_fd() doesn't check that the read-only bit is unset (i.e. the file is writable) when the DELETE access flag is set. Thus DeleteFile() succeeding when the tests expect it to fail. -- Ben Peddell IT Support Bowen, Collinsville and Proserpine Catholic schools http://klightspeed.killerwolves.net/

June 21, 2012

Re: Wine automated testing update

by Francois Gouget

On Wed, 1 Feb 2012, Jeremy White wrote: [...] > However, Alexandre has persuaded me that we should first explore > integrating the dotest.sh functionality into the existing WineTestBot, > as that would allow us to have a very simple clean web page, and would > also allow us to integrate most cleanly with Alexandre's back end tools. > > Of course, WineTestBot is Perl code, and my Perl fu is low. So I have > recently asked Francois to help with that effort as well. > > So the basic plan is as follows: [...] > 2. If so, modify WineTestBot to use qemu instead of vmware. > (This is 'easy' because the WineTestBot code nicely isolates > the VM functions, but 'hard', because VMWare provides some > APIs that may have to be replicated). So I have finally found the time to get started on this. Yay! It's not done yet but I figure it's time for me to give a status report and ask some questions. --- LibVirt I updated the Perl scripts to remove the dependency on VMware. Now I use libvirt which is a virtualization API that supports multiple virtualization technologies such as KVM/QEMU, but also VirtualBox, Xen and even VMware. libvirt also makes it possible to control VMs on remote boxes so we can distribute the load over multiple machines. For more details see the project's web site: http://libvirt.org/ --- Host / VM communication One piece that's missing is the ability to copy files from/to a VM and to run specific commands in a VM. This was provided by VMware's VIX API combined with a guest agent. See VMCopyFileFromHostToGuest() and VM::RunScriptInGuest() on the following page for instance: http://www.vmware.com/support/developer/vix-api/vix111_reference/index2.html All replacements involve installing some application in the Windows VM. So given that we want to run Winetest too I have the following requirements: * The application should run on everything from Windows 8 down to NT4. * It should not bring in dependencies that would modify the behavior of Windows. So anything that depends on installing .Net or upgrading standard dlls located in c:\windows\system32 is out. * Ideally it should also have a 64bit version so we can have 64bit-only Windows VMs. So I looked for a replacement but came away not entirely satisfied: * There is some talk of a QEmu Guest Agent but as far as I can tell it's just vaporware. If someone knows more I'd be interested (even if it would tie us to QEmu). http://wiki.qemu.org/Features/QAPI/GuestAgent * There is the BuildBot slave which probably provides the functionality we need. But it is implemented in Python which only works on Windows XP and later, and comes as an MSI installer. I did not check exactly how to interface with it. http://trac.buildbot.net/wiki/RunningBuildbotOnWindows * There's Cygwin's OpenSSH server which would give us what we need through scp and ssh. Cygwin does not install on NT4 and I'm not convinced it's really free of side-effects. In particular copying the cygwin directory to another computer does not seem to work. http://cygwin.com/install.html * There's more limited OpenSSH installations derived from Cygwin but I'm not sure they're much better on the side-effects side. http://sshwindows.sourceforge.net/download/ * For my own Windows VMs (see the fg-xxx VMs on test.winehq.org) what I use is a Windows implementation of netcat. http://joncraton.org/blog/46/netcat-for-windows I essentially run it in a batch that does: REM Receive winetest and how to invoke it nc95 -l -p%port% >app.exe nc95 -l -p%port% >cmdline.bat REM Run winetest, sending its output to the client nc95 -l -p%port% -e .\cmdline.bat I'm not entirely satisfied with it (and not because netcat is totally insecure, the VMs are behind firewalls anyway). Rather it's because it's not very flexible and because it has race conditions (which I work around with pauses on the client side). I could make it more flexible with something like: :start nc95 -l -p%port1% >cmd.bat nc95 -l -p%port1% -e .\cmd.bat goto start And put more nc95 commands running on a second port in cmd.bat. * Modified Windows netcat. I'd like to be able to run a netcat server so that on the client side I could do: ( echo "write c:/winetest.exe" cat winetest.exe ) | netcat $host $port echo "run c:/winetest.exe my options here" | netcat $host $port >winetest.log The server would see the command on the first line and do the relevant action. --- Network trouble I'm having trouble getting the network to work fine in the QEmu VMs. Eventually I discovered that if I bridged to eth0 then it's just the communication between the host and the VM that does not work. Communication with other machines on the LAN or with the Internet works. However that's going to be a big problem to send the jobs to the VMs (see previous section, plus the dhcp and dns is also running on the host in my case). Now, rather than bridge to eth0 what I'd want on my machine is to bridge to tap0, a local network where I have all my VMware VMs. What I typically see if I ping the host from the VM is that the echo requests go out and the echo replies are sent. But ping on the VM sees nothing. It could be a firewalling issue though my VMware VMs on the same network work fine. --- Patches My code is far from ready for commits but I stumbled upon a few fixes. Where should I send those? --- Build VM The installation instructions mention setting up a 'Wine build VM' but are short on details. Is it actually supposed to be a VM. What name? Identified through the 'Type' field in the VMs table? --- Miscellaneous * In Engine.pl, HandleBuildNotification() calls a ${ProjectName}RetrieveBuild.pl script which concerns me because there is no such script. There is a WineRunBuild.pl script that could match though. Still that indicates the Git code may not be exactly what's running on the current WineTestBot. Does anyone know more? * The virtio disk and network drivers appear to be a must. Replacing the default vga emulation with the cirrus one makes it possible to suspend / hibernate the VM (which is probably useless but still instills more confidence to me). Switching to vmware seems to give the best performance but Libvirt does not recognize it :-( It looks like the XML schema is simply being out of date. Does anyone know more? Still on the subject of optimizing Windows to work with QEmu, does anyone know if it's possible to expose the full capabilities of the graphics card to the VM? By declaring the graphics card as 'passthrough' maybe? That would be nice for Direct3D but I'm not getting my hopes up :-( -- Francois Gouget <fgouget(a)codeweavers.com>

June 21, 2012

Re: kernel32: Add a test for deleting an opened for reading file. Take 2.

by Eric Pouech

one tricky question... could it be that on windows this test behaves differently on ntfs vs FAT filesystem ? or what would give re-opening the file after the deletefile call ? A+ 2012/6/21, Dmitry Timoshkov <dmitry(a)baikal.ru>: > Previous version had a typo in file name. > --- > dlls/kernel32/tests/file.c | 26 ++++++++++++++++++++++++++ > 1 file changed, 26 insertions(+) > > diff --git a/dlls/kernel32/tests/file.c b/dlls/kernel32/tests/file.c > index 8768b99..0a662d9 100644 > --- a/dlls/kernel32/tests/file.c > +++ b/dlls/kernel32/tests/file.c > @@ -3236,6 +3236,31 @@ static void test_CreatFile(void) > > DeleteFile(file_name); > } > +static void test_delete_file(void) > +{ > + char temp_path[MAX_PATH], temp_file[MAX_PATH]; > + HANDLE hfile; > + int ret; > + > + GetTempPathA(MAX_PATH, temp_path); > + GetTempFileName(temp_path, "tst", 0, temp_file); > + > + SetLastError(0xdeadbeef); > + hfile = CreateFile(temp_file, GENERIC_READ, FILE_SHARE_DELETE | > FILE_SHARE_READ, NULL, OPEN_EXISTING, 0, 0); > + ok(hfile != INVALID_HANDLE_VALUE, "CreateFile error %d\n", > GetLastError()); > + > + SetLastError(0xdeadbeef); > + ret = DeleteFile(temp_file); > +todo_wine > + ok(ret, "DeleteFile error %d\n", GetLastError()); > + > + SetLastError(0xdeadbeef); > + ret = CloseHandle(hfile); > + ok(ret, "CloseHandle error %d\n", GetLastError()); > + ret = DeleteFile(temp_file); > +todo_wine > + ok(!ret, "DeleteFile should fail\n"); > +} > > START_TEST(file) > { > @@ -3257,6 +3282,7 @@ START_TEST(file) > test_CreateFileW(); > test_DeleteFileA(); > test_DeleteFileW(); > + test_delete_file(); > test_MoveFileA(); > test_MoveFileW(); > test_FindFirstFileA(); > -- > 1.7.11 > > > > -- -- Eric Pouech

June 21, 2012

Re: wintrust/tests: Fix build with MSVC.

by Thomas Faber

On 2012-06-21 12:34, Dmitry Timoshkov wrote: > Thomas Faber <thfabba(a)gmx.de> wrote: > >> FIELD_OFFSET isn't constant "enough". > > The problem is not with FIELD_OFFSET, but with applying shift and mask > operations to a constant. You're right, it's actually a combination of multiple factors. I'm not sure what exactly its problem is - the gist is, FIELD_OFFSET cannot reliably be assumed constant. E.g. the following code only complains about c4: struct a { char a; int b[5]; int c; }; int c1 = (int)&(((struct a *)0)->a) << 8; int c2 = (int)&(((struct a *)0)->b[0]); int c3 = (int)&(((struct a *)0)->b[0]) + 1; int c4 = (int)&(((struct a *)0)->b[0]) << 8; int c5 = (int)&(((struct a *)0)->b) << 8; int c6 = (int)&(((struct a *)0)->c) << 8; On 2012-06-21 12:11, Dmitry Timoshkov wrote: > Thomas Faber <thfabba(a)gmx.de> wrote: > >> Empty braces are no valid initializer. >> FIELD_OFFSET isn't constant "enough". > >> + /* add 'jmp ExitProcess' instruction */ >> + count = 2; >> + bin.text_section[count++] = 0xFF; >> + bin.text_section[count++] = 0x25; >> + bin.text_section[count++] = EXIT_PROCESS&0xFF; >> + bin.text_section[count++] = (EXIT_PROCESS>>8)&0xFF; >> + bin.text_section[count++] = (EXIT_PROCESS>>16)&0xFF; >> + bin.text_section[count++] = (EXIT_PROCESS>>24)&0xFF; Which in turn means it doesn't consider these constant, which is why I filled them in at runtime instead. I thought this cleaner than hardcoding the offset or otherwise forcing it to a constant. I hope that makes sense. Thanks for your comments. Best regards, Tom

June 21, 2012

Re: wintrust/tests: Fix build with MSVC.

by Dmitry Timoshkov

Thomas Faber <thfabba(a)gmx.de> wrote: > FIELD_OFFSET isn't constant "enough". The problem is not with FIELD_OFFSET, but with applying shift and mask operations to a constant. -- Dmitry.

June 21, 2012

Re: [PATCH 1/7] ntdll: Add support for junction point creation.

by Hans Leidekker

On Wed, 2012-06-20 at 17:26 -0600, Erich E. Hoover wrote: > + TRACE("Linking %s to %s\n", unix_src.Buffer, unix_dest.Buffer); > + if (rmdir( unix_src.Buffer ) < 0) > + { > + status = FILE_GetNtStatus(); > + goto cleanup; > + } > + if (symlink( unix_dest.Buffer, unix_src.Buffer ) < 0) > + { > + status = FILE_GetNtStatus(); > + goto cleanup; > + } > + status = STATUS_SUCCESS; This is not an atomic operation since you need two Unix calls. So you would need locking or rollback to deal with possible races. Should permissions and ownership be preserved on the directory when a junction is removed?

June 21, 2012

Re: wintrust/tests: Fix build with MSVC.

by Dmitry Timoshkov

Thomas Faber <thfabba(a)gmx.de> wrote: > Empty braces are no valid initializer. > FIELD_OFFSET isn't constant "enough". > + /* add 'jmp ExitProcess' instruction */ > + count = 2; > + bin.text_section[count++] = 0xFF; > + bin.text_section[count++] = 0x25; > + bin.text_section[count++] = EXIT_PROCESS&0xFF; > + bin.text_section[count++] = (EXIT_PROCESS>>8)&0xFF; > + bin.text_section[count++] = (EXIT_PROCESS>>16)&0xFF; > + bin.text_section[count++] = (EXIT_PROCESS>>24)&0xFF; This part of the patch obviously is not related to the patch description. -- Dmitry.

June 21, 2012

Re: Incompatibility in Kernel32

by Dmitry Timoshkov

Hans Leidekker <hans(a)codeweavers.com> wrote: > > Is there any reason that you call DeleteFile() on a still being opened file? > > That's the whole point of this test. The observed behavior is that you can pass a > filename to MsiRecordSetStream, successfully delete the file, and keep on using the > stream. > > We currently implement MsiRecordSetStream by creating an in-memory copy of the > stream. As Robert noticed, this will fail if the stream is too large. > > The idea is to move to an implementation on top of a file handle, but it should > still allow DeleteFile to succeed while the file is in use. Thanks, I see. Removing GENERIC_WRITE from flags passed to NtCreateFile by DeleteFile makes the test I just sent for that behaviour pass (and that's most like a correct fix), but that leads to some other test failures. This needs further investigation. -- Dmitry.

June 21, 2012

Re: Incompatibility in Kernel32

by Hans Leidekker

On Thu, 2012-06-21 at 16:03 +0900, Dmitry Timoshkov wrote: > > Shall I file a bug report? > > Try to add a test case to dlls/kenel32/tests/file.c,test_file_sharing(). > > > HANDLE h2 = CreateFileA(filename, GENERIC_READ, FILE_SHARE_DELETE | FILE_SHARE_READ, NULL, OPEN_EXISTING, 0, 0); > > assert (h2 != 0); > > > > if (!DeleteFile(filename)) { > > r = GetLastError(); > > fprintf(stderr, "This only happens in Wine. I got error: %d\n", r); > > assert(0); > > } > > Is there any reason that you call DeleteFile() on a still being opened file? That's the whole point of this test. The observed behavior is that you can pass a filename to MsiRecordSetStream, successfully delete the file, and keep on using the stream. We currently implement MsiRecordSetStream by creating an in-memory copy of the stream. As Robert noticed, this will fail if the stream is too large. The idea is to move to an implementation on top of a file handle, but it should still allow DeleteFile to succeed while the file is in use.

June 21, 2012

Re: kernel32: Add a test for deleting an opened for reading file.

by Marvin

Hi, While running your changed tests on Windows, 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 http://testbot.winehq.org/JobDetails.pl?Key=19407 Your paranoid android. === WNT4WSSP6 (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === W2KPROSP4 (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === WXPPROSP3 (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === W2K3R2SESP2 (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === WVISTAADM (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === W2K8SE (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === W7PRO (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === W7PROX64 (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === TEST64_W7SP1 (32 bit file) === file.c:3255: Test failed: DeleteFile error 2 === W7PROX64 (64 bit file) === file.c:3255: Test failed: DeleteFile error 2 === TEST64_W7SP1 (64 bit file) === file.c:3255: Test failed: DeleteFile error 2

June 21, 2012