glibc/sysdeps/unix/sysv/linux/ia64/clone2.S
Adhemerval Zanella 0cb313f7cb Fix clone (CLONE_VM) pid/tid reset (BZ#19957)
As discussed in libc-alpha [1] current clone with CLONE_VM (without
CLONE_THREAD set) will reset the pthread pid/tid fields to -1.  The
issue is since memory is shared between the parent and child it will
clobber parent's cached pid/tid leading to internal inconsistencies
if the value is not restored.

And even it is restored it may lead to racy conditions when between
set/restore a thread might invoke pthread function that validate the
pthread with INVALID_TD_P/INVALID_NOT_TERMINATED_TD_P and thus get
wrong results.

As stated in BZ19957, previously reports of this behaviour was close
with EWONTFIX due the fact usage of clone outside glibc is tricky
since glibc requires consistent internal pthread, while using clone
directly may not provide it. However since now posix_spawn uses
clone (CLONE_VM) to fixes various issues related to previous vfork
usage this issue requires fixing.

The vfork implementation also does something similar, but instead
it negates and restores only the *pid* field and functions that
might access its value know to handle such case (getpid, raise
and pthread ones that uses INVALID_TD_P/INVALID_NOT_TERMINATED_TD_P
macros that check only *tid* field).  Also vfork does not call
__clone directly, instead calling either __NR_vfork or __NR_clone
directly.

So this patch removes this clone behavior by avoiding setting
the pthread pid/tid field for CLONE_VM. There is no need to
check for CLONE_THREAD, since the minimum supported kernel in all
architecture implies that CLONE_VM must be used with CLONE_THREAD,
otherwise clone returns EINVAL.

Instead of current approach of:

   int clone(int (*fn)(void *), void *child_stack, int flags, ...)
      [...]
      if (flags & CLONE_THREAD)
        goto do_syscall;
      pid_t new_value;
      if (flags & CLONE_VM)
        new_value = -1;
      else
        new_value = getpid ();
      THREAD_SETMEM (THREAD_SELF, pid, new_value);
      THREAD_SETMEM (THREAD_SELF, tid, new_value);

    do_syscall:
      [...]

The new approach uses:

   int clone(int (*fn)(void *), void *child_stack, int flags, ...)
      [...]
      if (flags & CLONE_VM)
        goto do_syscall;
      pid_t new_value = getpid ();
      THREAD_SETMEM (THREAD_SELF, pid, new_value);
      THREAD_SETMEM (THREAD_SELF, tid, new_value);

    do_syscall:
      [...]

It also removes the linux tst-getpid2.c test which expects the previous
behavior and instead add another clone test.

Tested on x86_64, i686, x32, powerpc64le, aarch64, armhf, s390, and
s390x. I also did limited check on mips32 and sparc64 (using the new
added test).

I also got reviews from both m68k, hppa, and tile.  So I presume for
these architecture the patch works.

The fixes for alpha, microblaze, sh, ia64, and nio2 have not been
tested.

[1] https://sourceware.org/ml/libc-alpha/2016-04/msg00307.html

	* sysdeps/unix/sysv/linux/Makefile [$(subdir) == nptl] (test): Remove
	tst-getpid2.
	(test): Add tst-clone2.
	* sysdeps/unix/sysv/linux/tst-clone2.c: New file.
	* sysdeps/unix/sysv/linux/aarch64/clone.S (__clone): Do not change
	pid/tid fields for CLONE_VM.
	* sysdeps/unix/sysv/linux/arm/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/i386/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/mips/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/s390/s390-32/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/s390/s390-64/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/sparc/sparc32/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/sparc/sparc64/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/clone.S: Likewise.
	* sysdeps/unix/sysv/linux/tst-getpid2.c: Remove file.
2016-04-29 18:19:30 -03:00

104 lines
3.3 KiB
ArmAsm

/* Copyright (C) 2000-2016 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <asm/errno.h>
/* int __clone2(int (*fn) (void *arg), void *child_stack_base, */
/* size_t child_stack_size, int flags, void *arg, */
/* pid_t *parent_tid, void *tls, pid_t *child_tid) */
#define CHILD p8
#define PARENT p9
ENTRY(__clone2)
.prologue
alloc r2=ar.pfs,8,1,6,0
cmp.eq p6,p0=0,in0
cmp.eq p7,p0=0,in1
mov r8=EINVAL
mov out0=in3 /* Flags are first syscall argument. */
mov out1=in1 /* Stack address. */
(p6) br.cond.spnt.many __syscall_error /* no NULL function pointers */
(p7) br.cond.spnt.many __syscall_error /* no NULL stack pointers */
;;
mov out2=in2 /* Stack size. */
mov out3=in5 /* Parent TID Pointer */
mov out4=in7 /* Child TID Pointer */
mov out5=in6 /* TLS pointer */
/*
* clone2() is special: the child cannot execute br.ret right
* after the system call returns, because it starts out
* executing on an empty stack. Because of this, we can't use
* the new (lightweight) syscall convention here. Instead, we
* just fall back on always using "break".
*
* Furthermore, since the child starts with an empty stack, we
* need to avoid unwinding past invalid memory. To that end,
* we'll pretend now that __clone2() is the end of the
* call-chain. This is wrong for the parent, but only until
* it returns from clone2() but it's better than the
* alternative.
*/
mov r15=SYS_ify (clone2)
.save rp, r0
break __BREAK_SYSCALL
.body
cmp.eq p6,p0=-1,r10
cmp.eq CHILD,PARENT=0,r8 /* Are we the child? */
(p6) br.cond.spnt.many __syscall_error
;;
(CHILD) mov loc0=gp
(PARENT) ret
;;
tbit.nz p6,p0=in3,8 /* CLONE_VM */
(p6) br.cond.dptk 1f
;;
mov r15=SYS_ify (getpid)
(p7) break __BREAK_SYSCALL
;;
add r9=PID,r13
add r10=TID,r13
;;
st4 [r9]=r8
st4 [r10]=r8
;;
1: ld8 out1=[in0],8 /* Retrieve code pointer. */
mov out0=in4 /* Pass proper argument to fn */
;;
ld8 gp=[in0] /* Load function gp. */
mov b6=out1
br.call.dptk.many rp=b6 /* Call fn(arg) in the child */
;;
mov out0=r8 /* Argument to _exit */
mov gp=loc0
.globl HIDDEN_JUMPTARGET(_exit)
br.call.dpnt.many rp=HIDDEN_JUMPTARGET(_exit)
/* call _exit with result from fn. */
ret /* Not reached. */
PSEUDO_END(__clone2)
libc_hidden_def (__clone2)
/* For now we leave __clone undefined. This is unlikely to be a */
/* problem, since at least the i386 __clone in glibc always failed */
/* with a 0 sp (eventhough the kernel explicitly handled it). */
/* Thus all such calls needed to pass an explicit sp, and as a result, */
/* would be unlikely to work on ia64. */