[Bug 55138] New: msys2: bash.exe is not able to fork/start another process.
https://bugs.winehq.org/show_bug.cgi?id=55138 Bug ID: 55138 Summary: msys2: bash.exe is not able to fork/start another process. Product: Wine Version: 8.10 Hardware: x86-64 OS: Linux Status: UNCONFIRMED Severity: normal Priority: P2 Component: ntdll Assignee: wine-bugs(a)winehq.org Reporter: bernhardu(a)mailbox.org Distribution: --- Created attachment 74697 --> https://bugs.winehq.org/attachment.cgi?id=74697 allocation stacks in parent and child bash.exe This is a followup to bug 47808. If I understand it right, bug 47808 was primarily about native libraries getting called at a win32 stack which was not yet committed. Those native libraries then jumped over the guard page, therefore the stack got not grown, instead a crash reported. This aspect was fixed by using the syscall interface and therefore bug 47808 was closed. But there is another issue that causes a msys2-forked process crashing, because after the stack was grown it still misses the PAGE_READWRITE flags. For this Zeb Figura included in the first attachement to bug 47808 "always expand the stack if within the stack reserve" a line adding the flags "VPROT_COMMITTED | VPROT_READ | VPROT_WRITE" to the page that gets the VPROT_GUARD removed in grow_thread_stack. This modification still enables a forked process to run. The issue can be reproduced with an unmodified wine by just - running msys2 bash.exe interactively - then trying to start e.g. notepad from it. Attached file shows the stack when the stack gets allocated - in a working msys2-parent process and - in a failing msys2-forked child process. Therefore it looks like the forked stack get really setup differently, but it is still not clear if or when this memory gets the READ/WRITE added in native windows. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 Joel Holdsworth <joel(a)airwebreathe.org.uk> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |joel(a)airwebreathe.org.uk -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu(a)mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|msys2: bash.exe is not able |msys2/bash.exe: is not able |to fork/start another |to fork/start another |process. |process (cygwin/mintty: | |silently finishes, fails | |top open window) See Also| |https://bugs.winehq.org/sho | |w_bug.cgi?id=47808 -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 --- Comment #1 from Bernhard Übelacker <bernhardu(a)mailbox.org> --- Created attachment 75738 --> https://bugs.winehq.org/attachment.cgi?id=75738 ntdll: Add read and write to new guard pages. This makes also cygwin64 mintty silently fail to open the window. It looks like cygwins child_info_fork::alloc_stack setups the stack with three guard pages. This pages end up with vprot bits 0x33 set (read,write,guard,committed). Below pages stay with having no vprot bits set. The function grow_thread_stack seems to be able to grow the stack for these top three pages without problem. It is able to set vprot bit 0x30 to the next guard page (guard,committed). Unfortunately on the next grow_thread_stack it changes the vprot bits to 0x20, and therefore get_unix_prot(0x20) returns prot=0, therefore the page looses its guard page status, but gets not committed. (mprotect is called with unix_prot=0) Attached patch, adds always read,write to guard pages. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 Fabian Maurer <dark.shadow4(a)web.de> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |dark.shadow4(a)web.de -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu(a)mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Attachment #75738|0 |1 is obsolete| | --- Comment #2 from Bernhard Übelacker <bernhardu(a)mailbox.org> --- Created attachment 76702 --> https://bugs.winehq.org/attachment.cgi?id=76702 ntdll: Add read and write to new guard pages. (rebased wine-9.12) Still present with wine-9.11. Test was done with a wine tree including patches related to bug 52105. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu(a)mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Keywords| |download, source -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=55138 manschwetus(a)gmail.com changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |manschwetus(a)gmail.com -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu(a)mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Attachment #76702|0 |1 is obsolete| | --- Comment #3 from Bernhard Übelacker <bernhardu(a)mailbox.org> --- Created attachment 78506 --> http://bugs.winehq.org/attachment.cgi?id=78506 ntdll: Add read and write to new guard pages. (rebased wine-10.7) Still present with wine-10.7. This is currently causes msys2-x86_64-20250221.exe to show this error message at 50%: ##### Installer Error Error during installation process (com.msys2.root): Execution failed (Unexpected exit code: 1): "C:/msys64\usr\bin\bash.exe --login -c exit" ##### Starting above command directly shows following error: ##### /etc/profile: Zeile 43: /usr/bin/cygpath: Bad address /etc/profile: Zeile 43: /usr/bin/cygpath: No error ##### Applying attached patch makes the installer making more progress, but then hangs because of bug #53574. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 pipcet(a)protonmail.com changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |pipcet(a)protonmail.com --- Comment #4 from pipcet(a)protonmail.com --- Potentially related to https://bugs.winehq.org/show_bug.cgi?id=58420 ? I definitely saw the "/usr/bin/cygpath: Bad address" error message, but I don't recall seeing one that said "No error". -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 --- Comment #5 from Bernhard Übelacker <bernhardu(a)mailbox.org> --- (In reply to pipcet from comment #4)
Potentially related to https://bugs.winehq.org/show_bug.cgi?id=58420 ?
I definitely saw the "/usr/bin/cygpath: Bad address" error message, but I don't recall seeing one that said "No error".
If I understand 58420 right, it should just happen on CPUs showing in e.g. lscpu in Flags an entry "pku"? Unfortunately my CPU is too old to show this, but I definitely get this fork issue here. Therefore I currently don't see a connection between both issues at the moment. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 Hamish McIntyre-Bhatty <winehq@regd.hamishmb.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |winehq@regd.hamishmb.com --- Comment #6 from Hamish McIntyre-Bhatty <winehq@regd.hamishmb.com> --- Still present in Wine 11.5. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 --- Comment #7 from Hamish McIntyre-Bhatty <winehq@regd.hamishmb.com> --- *** Bug 59566 has been marked as a duplicate of this bug. *** -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 --- Comment #8 from Hamish McIntyre-Bhatty <winehq@regd.hamishmb.com> --- Please note, if you resize the Cygwin setup program to make it a bit larger on the first screen, at least in Wine 11.5, it behaves much better later. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu@mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Staged patchset| |https://gitlab.winehq.org/w | |ine/wine-staging/-/tree/mas | |ter/patches/ntdll-gaurd_pag | |e Ever confirmed|0 |1 Status|UNCONFIRMED |STAGED -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu@mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Blocks| |60374 Referenced Bugs: http://bugs.winehq.org/show_bug.cgi?id=60374 [Bug 60374] msys2: forked process died unexpectedly, retry 0, exit code 0xE0000269, errno 11 -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 Bernhard Übelacker <bernhardu@mailbox.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Staged patchset|https://gitlab.winehq.org/w |https://gitlab.winehq.org/w |ine/wine-staging/-/tree/mas |ine/wine-staging/-/blob/mas |ter/patches/ntdll-gaurd_pag |ter/patches/ntdll-guard_pag |e |e/definition -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=55138 Wolfgang Fahl <info@bitplan.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |info@bitplan.com --- Comment #9 from Wolfgang Fahl <info@bitplan.com> --- Confirming on macOS with Wine 11.18, with the exact fault and a real-world application. Environment: macOS 14.8.9, x86_64 (Intel), Wine 11.18 built from source via MacPorts (wine-devel). Application: Git for Windows 2.56.0 (64-bit), which bundles msys2-runtime 3.6.10. Reproduce: wine "C:\Program Files\Git\usr\bin\bash.exe" --norc --noprofile -c "/usr/bin/uname -sr" /usr/bin/bash: line 1: /usr/bin/uname: Bad address Every program started from bash fails this way, git included; bash -c "exec /usr/bin/uname -sr" works, so only the forked child is affected. Same on 11.0. Trace (+seh,+virtual) of the forked child: the runtime reserves a 2 MB stack at 0x7ffe00000, commits 4 pages at the top and 3 PAGE_GUARD pages below (NtAllocateVirtualMemory 0x7ffff9000 0x3000 PAGE_READWRITE|PAGE_GUARD). Spawning allocates ~64 KB on that stack; the _alloca probe faults with EXCEPTION_ACCESS_VIOLATION (write) at 0x7ffff8fb8, the first page below the original guard pages, i.e. the page grow_thread_stack() made the new guard page without VPROT_READ|VPROT_WRITE. With attachment 75738 ("ntdll: Add read and write to new guard pages") applied to 11.18 (it still applies cleanly, together with the other patches the MSYS2 project carries at https://github.com/msys2/msys2-docker/tree/main/build-base/patches): the command above prints MSYS_NT-10.0-19045 3.6.10-5a1665c8.x86_64, bash --login works, and a bash+git script over 16 repositories runs to completion. So the one-line change is sufficient for this symptom on macOS as well. Note for the record: on macOS this bug only becomes visible from Wine 10.5 on, since before the GSBASE swap the runtime did not get past startup at all. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
participants (2)
-
WineHQ Bugzilla -
WineHQ Bugzilla