[Bug 2007796] Re: glibc __read_chk not a cancellation point

[Brian Murray]

** Also affects: glibc (Ubuntu Kinetic)
   Importance: Undecided
       Status: New

** Changed in: glibc (Ubuntu Kinetic)
       Status: New => Fix Released

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to glibc in Ubuntu.
https://bugs.launchpad.net/bugs/2007796

Title:
  glibc __read_chk not a cancellation point

Status in glibc package in Ubuntu:
  Fix Released
Status in glibc source package in Jammy:
  In Progress
Status in glibc source package in Kinetic:
  Fix Released

Bug description:
  [Impact]
  I'm working with Xen and libxenstore.  libxenstore, when using a "watch", spawns a pthread (read_thread).  When libxenstore shuts down, it pthread_cancel()s and pthread_join()s the "watch" thread.

  That thread never exits and the process shutdown hangs.

  read_threads is sitting in __read_chk().  In glibc 2.35, __read_chk is
  not a cancellation point, so the thread never reacts to the
  cancellation.

  Upstream glibc fixed it in 2.36 in
  https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=dc30acf20bd635d71cd4c84100e842fdf0429e48

  Here's the 2.35 disassembly - the lack of __pthread_enable_asynccancel() indicates the missing cancellation support:
  (gdb) disassemble
  Dump of assembler code for function __read_chk:
     0x00007ffff7ea04d0 <+0>:	endbr64
     0x00007ffff7ea04d4 <+4>:	cmp    %rcx,%rdx
     0x00007ffff7ea04d7 <+7>:	ja     0x7ffff7ea0504 <__read_chk+52>
     0x00007ffff7ea04d9 <+9>:	xor    %eax,%eax
     0x00007ffff7ea04db <+11>:	syscall
  => 0x00007ffff7ea04dd <+13>:	cmp    $0xfffffffffffff000,%rax
     0x00007ffff7ea04e3 <+19>:	ja     0x7ffff7ea04f0 <__read_chk+32>
     0x00007ffff7ea04e5 <+21>:	ret
     0x00007ffff7ea04e6 <+22>:	cs nopw 0x0(%rax,%rax,1)
     0x00007ffff7ea04f0 <+32>:	mov    0xe3919(%rip),%rdx        # 0x7ffff7f83e10
     0x00007ffff7ea04f7 <+39>:	neg    %eax
     0x00007ffff7ea04f9 <+41>:	mov    %eax,%fs:(%rdx)
     0x00007ffff7ea04fc <+44>:	mov    $0xffffffffffffffff,%rax
     0x00007ffff7ea0503 <+51>:	ret
     0x00007ffff7ea0504 <+52>:	push   %rax
     0x00007ffff7ea0505 <+53>:	call   0x7ffff7ea00b0 <__GI___chk_fail>
  End of assembler dump.

  [Test procedure]

  The patch includes a test for this that is run at build time.

  [Regression potential]

  Besides the usual risks with any glibc update, this could potentially
  surface some race conditions at thread shutdown in user applications
  that were thus far hidden by the lack of cancellation point.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/glibc/+bug/2007796/+subscriptions