diff mbox series

[v2,3/3] posix: Use posix_spawn on system

Message ID 20181019232609.25531-3-adhemerval.zanella@linaro.org
State New
Headers show
Series [v2,1/3] posix: Add internal symbols for posix_spawn interface | expand

Commit Message

Adhemerval Zanella Netto Oct. 19, 2018, 11:26 p.m. UTC
Changed from previous version:

  - Call __libc_signal_block_app for SPAWN_XFLAGS_ENABLE_CANCEL.

---

This patch uses posix_spawn on system implementation.  On Linux this has
the advantage of much lower memory consumption (usually 32 Kb minimum for
the mmap stack area).

Although POSIX does not require, glibc system implementation aims to be
thread and cancellation safe.  While reentracy handling does not require
any direct change of current strategy, cancellation requires a posix_spawn
to be cancellable.  This is done by adding an internal
__posix_spawn_cancellable which does not disable cancellation neither
change process signal mask.

The cancellation code is also moved to generic implementation and enabled
only if SIGCANCEL is defined (similar on how the cancellation handler is
enabled on nptl-init.c).

Checked on x86_64-linux-gnu and i686-linux-gnu.

	* include/spawn.h (__posix_spawn_cancellable): New prototype.
	* posix/spawn.c (__posix_spawn_cancellable): New function.
	* posix/spawn_int.h (SPAWN_XFLAGS_ENABLE_CANCEL): New define.
	* sysdeps/generic/not-errno.h (__kill_noerrno): New prototype.
	* sysdeps/unix/sysv/linux/not-errno.h (__kill_noerrno): New
	implementation.
	* sysdeps/posix/spawni.c (__spawnix): Disable cancellation for
	!SPAWN_XFLAGS_ENABLE_CANCEL.
	* sysdeps/unix/sysv/linux/spawni.c (__spawnix): Likewise.
	* sysdeps/posix/system.c [_LIBC_REENTRANT && SIGCANCEL]
	(cancel_handler): New function.
	(CLEANUP_HANDLER, CLEANUP_RESET): Likewise.
	(do_system): Use posix_spawn instead of fork and execl.
	* sysdeps/unix/sysv/linux/ia64/system.c: Remove file.
	* sysdeps/unix/sysv/linux/s390/system.c: Likewise.
	* sysdeps/unix/sysv/linux/sparc/system.c: Likewise.
	* sysdeps/unix/sysv/linux/system.c: Likewise.
---
 ChangeLog                              |  18 ++++
 include/spawn.h                        |   2 +
 posix/spawn.c                          |  14 +++
 posix/spawn_int.h                      |   1 +
 sysdeps/generic/not-errno.h            |   2 +
 sysdeps/posix/spawni.c                 |   8 +-
 sysdeps/posix/system.c                 | 132 ++++++++++++++-----------
 sysdeps/unix/sysv/linux/ia64/system.c  |  30 ------
 sysdeps/unix/sysv/linux/not-errno.h    |  13 +++
 sysdeps/unix/sysv/linux/s390/system.c  |  29 ------
 sysdeps/unix/sysv/linux/sparc/system.c |  29 ------
 sysdeps/unix/sysv/linux/spawni.c       |  22 +++--
 sysdeps/unix/sysv/linux/system.c       |  76 --------------
 13 files changed, 145 insertions(+), 231 deletions(-)
 delete mode 100644 sysdeps/unix/sysv/linux/ia64/system.c
 delete mode 100644 sysdeps/unix/sysv/linux/s390/system.c
 delete mode 100644 sysdeps/unix/sysv/linux/sparc/system.c
 delete mode 100644 sysdeps/unix/sysv/linux/system.c

-- 
2.17.1

Comments

Florian Weimer Oct. 24, 2018, 10:34 a.m. UTC | #1
* Adhemerval Zanella:

>  #ifdef _LIBC_REENTRANT


Should we remove this as a patch prior to this change?  I think removing
these indirections would make the patch easier to review.

Thanks,
Florian
Florian Weimer Oct. 24, 2018, 10:40 a.m. UTC | #2
* Adhemerval Zanella:

> Although POSIX does not require, glibc system implementation aims to be

> thread and cancellation safe.  While reentracy handling does not require

> any direct change of current strategy, cancellation requires a posix_spawn

> to be cancellable.


I don't think this is actually true.  My understanding is that
posix_spawn does not block indefinitely in the parent, beyond things
that are not cancellation points anyway (resource allocation).  I/O
happens in the child process.  A posix_spawn that is a cancellation
point will lead to marginally faster reaction to a cancellation request.
But I think in practice, making the subsequent waitpid cancellable is
sufficient, and it should help to reduce the size of the patch.

Thanks,
Florian
Adhemerval Zanella Netto Oct. 24, 2018, 6:57 p.m. UTC | #3
On 24/10/2018 07:40, Florian Weimer wrote:
> * Adhemerval Zanella:

> 

>> Although POSIX does not require, glibc system implementation aims to be

>> thread and cancellation safe.  While reentracy handling does not require

>> any direct change of current strategy, cancellation requires a posix_spawn

>> to be cancellable.

> 

> I don't think this is actually true.  My understanding is that

> posix_spawn does not block indefinitely in the parent, beyond things

> that are not cancellation points anyway (resource allocation).  I/O

> happens in the child process.  A posix_spawn that is a cancellation

> point will lead to marginally faster reaction to a cancellation request.

> But I think in practice, making the subsequent waitpid cancellable is

> sufficient, and it should help to reduce the size of the patch.


Indeed I think I overthought the cancellation handling requirements for
posix_spawn and I agree waitpid is suffice to handle system cancellation
(which is already a cancellable entrypoint). I will send an update version.
Adhemerval Zanella Netto Oct. 24, 2018, 7:11 p.m. UTC | #4
On 24/10/2018 07:34, Florian Weimer wrote:
> * Adhemerval Zanella:

> 

>>  #ifdef _LIBC_REENTRANT

> 

> Should we remove this as a patch prior to this change?  I think removing

> these indirections would make the patch easier to review.

> 

> Thanks,

> Florian

> 


Alright, I think it is worth to clean this up.
diff mbox series

Patch

diff --git a/include/spawn.h b/include/spawn.h
index 7fdd965bd7..995ab9a8f4 100644
--- a/include/spawn.h
+++ b/include/spawn.h
@@ -5,6 +5,8 @@ 
 __typeof (posix_spawn) __posix_spawn;
 libc_hidden_proto (__posix_spawn)
 
+__typeof (posix_spawn) __posix_spawn_cancellable attribute_hidden;
+
 __typeof (posix_spawn_file_actions_addclose)
   __posix_spawn_file_actions_addclose attribute_hidden;
 
diff --git a/posix/spawn.c b/posix/spawn.c
index a82f1c84e2..24a416f53c 100644
--- a/posix/spawn.c
+++ b/posix/spawn.c
@@ -33,6 +33,20 @@  versioned_symbol (libc, __posix_spawn, posix_spawn, GLIBC_2_15);
 libc_hidden_def (__posix_spawn)
 
 
+/* Same as __posix_spawn but allows asynchronous cancellation.  It is used
+   internally on SYSTEM implementation to avoid leak a process ID through
+   a cancellation handler.  */
+int
+__posix_spawn_cancellable (pid_t *pid, const char *path,
+			   const posix_spawn_file_actions_t *file_actions,
+			   const posix_spawnattr_t *attrp, char *const argv[],
+			   char *const envp[])
+{
+  return __spawni (pid, path, file_actions, attrp, argv, envp,
+		   SPAWN_XFLAGS_ENABLE_CANCEL);
+}
+
+
 #if SHLIB_COMPAT (libc, GLIBC_2_2, GLIBC_2_15)
 int
 attribute_compat_text_section
diff --git a/posix/spawn_int.h b/posix/spawn_int.h
index 171f67c649..5cb8ce49b3 100644
--- a/posix/spawn_int.h
+++ b/posix/spawn_int.h
@@ -55,6 +55,7 @@  struct __spawn_action
 
 #define SPAWN_XFLAGS_USE_PATH	0x1
 #define SPAWN_XFLAGS_TRY_SHELL	0x2
+#define SPAWN_XFLAGS_ENABLE_CANCEL 0x3
 
 extern int __posix_spawn_file_actions_realloc (posix_spawn_file_actions_t *
 					       file_actions)
diff --git a/sysdeps/generic/not-errno.h b/sysdeps/generic/not-errno.h
index 93617a3266..0fd66b5c5e 100644
--- a/sysdeps/generic/not-errno.h
+++ b/sysdeps/generic/not-errno.h
@@ -17,3 +17,5 @@ 
    <http://www.gnu.org/licenses/>.  */
 
 extern __typeof (__access) __access_noerrno attribute_hidden;
+
+extern __typeof (__kill) __kill_noerrno attribute_hidden;
diff --git a/sysdeps/posix/spawni.c b/sysdeps/posix/spawni.c
index b138ab4393..e1a391659b 100644
--- a/sysdeps/posix/spawni.c
+++ b/sysdeps/posix/spawni.c
@@ -256,8 +256,9 @@  __spawnix (pid_t *pid, const char *file,
 
   /* Disable asynchronous cancellation.  */
   int state;
-  __libc_ptf_call (__pthread_setcancelstate,
-                   (PTHREAD_CANCEL_DISABLE, &state), 0);
+  if ((xflags & SPAWN_XFLAGS_ENABLE_CANCEL) == 0)
+    __libc_ptf_call (__pthread_setcancelstate,
+                     (PTHREAD_CANCEL_DISABLE, &state), 0);
 
   ptrdiff_t argc = 0;
   ptrdiff_t limit = INT_MAX - 1;
@@ -299,7 +300,8 @@  __spawnix (pid_t *pid, const char *file,
   if ((ec == 0) && (pid != NULL))
     *pid = new_pid;
 
-  __libc_ptf_call (__pthread_setcancelstate, (state, NULL), 0);
+  if ((xflags & SPAWN_XFLAGS_ENABLE_CANCEL) == 0)
+    __libc_ptf_call (__pthread_setcancelstate, (state, NULL), 0);
 
   return ec;
 }
diff --git a/sysdeps/posix/system.c b/sysdeps/posix/system.c
index d7594436ed..38d5e28ca2 100644
--- a/sysdeps/posix/system.c
+++ b/sysdeps/posix/system.c
@@ -17,20 +17,22 @@ 
 
 #include <errno.h>
 #include <signal.h>
-#include <stddef.h>
 #include <stdlib.h>
 #include <unistd.h>
+#include <sigsetops.h>
+#include <spawn.h>
+#include <pthread.h>
 #include <sys/types.h>
 #include <sys/wait.h>
-#include <libc-lock.h>
-#include <sysdep-cancel.h>
-#include <sigsetops.h>
+#include <stdio.h>
 
+#include <libc-lock.h>
+#include <not-errno.h>
+#include <internal-signals.h>
 
 #define	SHELL_PATH	"/bin/sh"	/* Path of the shell.  */
 #define	SHELL_NAME	"sh"		/* Name to give it.  */
 
-
 #ifdef _LIBC_REENTRANT
 static struct sigaction intr, quit;
 static int sa_refcntr;
@@ -38,17 +40,53 @@  __libc_lock_define_initialized (static, lock);
 
 # define DO_LOCK() __libc_lock_lock (lock)
 # define DO_UNLOCK() __libc_lock_unlock (lock)
-# define INIT_LOCK() ({ __libc_lock_init (lock); sa_refcntr = 0; })
 # define ADD_REF() sa_refcntr++
 # define SUB_REF() --sa_refcntr
+
 #else
 # define DO_LOCK()
 # define DO_UNLOCK()
-# define INIT_LOCK()
 # define ADD_REF() 0
 # define SUB_REF() 0
 #endif
 
+/* We have to and actually can handle cancelable system().  The big
+   problem: we have to kill the child process if necessary.  To do
+   this a cleanup handler has to be registered and is has to be able
+   to find the PID of the child.  The main problem is to reliable have
+   the PID when needed.  It is not necessary for the parent thread to
+   return.  It might still be in the kernel when the cancellation
+   request comes.  Therefore we have to use the clone() calls ability
+   to have the kernel write the PID into the user-level variable.  */
+
+#if defined(_LIBC_REENTRANT) && defined(SIGCANCEL)
+static void cancel_handler (void *arg)
+{
+  pid_t child = *(pid_t *) arg;
+
+  __kill_noerrno (child, SIGKILL);
+
+  TEMP_FAILURE_RETRY (__waitpid (child, NULL, 0));
+
+  DO_LOCK ();
+
+  if (SUB_REF () == 0)
+    {
+      __sigaction (SIGQUIT, &quit, NULL);
+      __sigaction (SIGINT, &intr, NULL);
+    }
+
+  DO_UNLOCK ();
+}
+# define CLEANUP_HANDLER() \
+  __libc_cleanup_region_start (1, cancel_handler, &pid)
+# define CLEANUP_RESET() \
+  __libc_cleanup_region_end (0)
+#else
+# define CLEANUP_HANDLER()
+# define CLEANUP_RESET()
+#endif
+
 
 /* Execute LINE as a shell command, returning its status.  */
 static int
@@ -60,7 +98,9 @@  do_system (const char *line)
 #ifndef _LIBC_REENTRANT
   struct sigaction intr, quit;
 #endif
+  struct sigaction oldintr, oldquit;
   sigset_t omask;
+  sigset_t reset;
 
   sa.sa_handler = SIG_IGN;
   sa.sa_flags = 0;
@@ -98,9 +138,9 @@  do_system (const char *line)
 	  if (SUB_REF () == 0)
 	    {
 	      save = errno;
-	      (void) __sigaction (SIGQUIT, &quit, (struct sigaction *) NULL);
+	      __sigaction (SIGQUIT, &quit, &oldquit);
 	    out_restore_sigint:
-	      (void) __sigaction (SIGINT, &intr, (struct sigaction *) NULL);
+	      __sigaction (SIGINT, &intr, &oldintr);
 	      __set_errno (save);
 	    }
 	out:
@@ -109,39 +149,30 @@  do_system (const char *line)
 	}
     }
 
-#ifdef CLEANUP_HANDLER
-  CLEANUP_HANDLER;
-#endif
-
-#ifdef FORK
-  pid = FORK ();
-#else
-  pid = __fork ();
-#endif
-  if (pid == (pid_t) 0)
-    {
-      /* Child side.  */
-      const char *new_argv[4];
-      new_argv[0] = SHELL_NAME;
-      new_argv[1] = "-c";
-      new_argv[2] = line;
-      new_argv[3] = NULL;
-
-      /* Restore the signals.  */
-      (void) __sigaction (SIGINT, &intr, (struct sigaction *) NULL);
-      (void) __sigaction (SIGQUIT, &quit, (struct sigaction *) NULL);
-      (void) __sigprocmask (SIG_SETMASK, &omask, (sigset_t *) NULL);
-      INIT_LOCK ();
-
-      /* Exec the shell.  */
-      (void) __execve (SHELL_PATH, (char *const *) new_argv, __environ);
-      _exit (127);
-    }
-  else if (pid < (pid_t) 0)
-    /* The fork failed.  */
-    status = -1;
-  else
-    /* Parent side.  */
+  __sigemptyset (&reset);
+  if (oldintr.sa_handler != SIG_IGN)
+    __sigaddset(&reset, SIGINT);
+  if (oldquit.sa_handler != SIG_IGN)
+    __sigaddset(&reset, SIGQUIT);
+
+  posix_spawnattr_t spawn_attr;
+  /* None of the posix_spawnattr_* function returns, including
+     posix_spawnattr_setflags for the follow specific usage (using valid
+     flags).  */
+  __posix_spawnattr_init (&spawn_attr);
+  __posix_spawnattr_setsigmask (&spawn_attr, &omask);
+  __posix_spawnattr_setsigdefault (&spawn_attr, &reset);
+  __posix_spawnattr_setflags (&spawn_attr,
+			      POSIX_SPAWN_SETSIGDEF | POSIX_SPAWN_SETSIGMASK);
+
+  CLEANUP_HANDLER ();
+  status = __posix_spawn_cancellable (&pid, SHELL_PATH, 0, &spawn_attr,
+				      (char *const[]){ (char*) SHELL_NAME,
+						       (char*) "-c",
+						       (char *) line, NULL },
+				      __environ);
+  __posix_spawnattr_destroy (&spawn_attr);
+  if (status == 0)
     {
       /* Note the system() is a cancellation point.  But since we call
 	 waitpid() which itself is a cancellation point we do not
@@ -149,25 +180,16 @@  do_system (const char *line)
       if (TEMP_FAILURE_RETRY (__waitpid (pid, &status, 0)) != pid)
 	status = -1;
     }
-
-#ifdef CLEANUP_HANDLER
-  CLEANUP_RESET;
-#endif
+  CLEANUP_RESET ();
 
   save = errno;
   DO_LOCK ();
   if ((SUB_REF () == 0
-       && (__sigaction (SIGINT, &intr, (struct sigaction *) NULL)
-	   | __sigaction (SIGQUIT, &quit, (struct sigaction *) NULL)) != 0)
+       && (__sigaction (SIGINT, &intr, NULL)
+	   | __sigaction (SIGQUIT, &quit, NULL)) != 0)
       || __sigprocmask (SIG_SETMASK, &omask, (sigset_t *) NULL) != 0)
     {
-#ifndef _LIBC
-      /* glibc cannot be used on systems without waitpid.  */
-      if (errno == ENOSYS)
-	__set_errno (save);
-      else
-#endif
-	status = -1;
+      status = -1;
     }
   DO_UNLOCK ();
 
diff --git a/sysdeps/unix/sysv/linux/ia64/system.c b/sysdeps/unix/sysv/linux/ia64/system.c
deleted file mode 100644
index d09fefefe6..0000000000
--- a/sysdeps/unix/sysv/linux/ia64/system.c
+++ /dev/null
@@ -1,30 +0,0 @@ 
-/* Copyright (C) 2002-2018 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/>.  */
-
-/* We have to and actually can handle cancelable system().  The big
-   problem: we have to kill the child process if necessary.  To do
-   this a cleanup handler has to be registered and is has to be able
-   to find the PID of the child.  The main problem is to reliable have
-   the PID when needed.  It is not necessary for the parent thread to
-   return.  It might still be in the kernel when the cancellation
-   request comes.  Therefore we have to use the clone() calls ability
-   to have the kernel write the PID into the user-level variable.  */
-#define FORK() \
-  INLINE_SYSCALL (clone2, 6, CLONE_PARENT_SETTID | SIGCHLD, NULL, 0, \
-		  &pid, NULL, NULL)
-
-#include <sysdeps/unix/sysv/linux/system.c>
diff --git a/sysdeps/unix/sysv/linux/not-errno.h b/sysdeps/unix/sysv/linux/not-errno.h
index 106ba5c72e..4c5ae1a4c2 100644
--- a/sysdeps/unix/sysv/linux/not-errno.h
+++ b/sysdeps/unix/sysv/linux/not-errno.h
@@ -16,6 +16,8 @@ 
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
+#include <sysdep.h>
+
 /* This function is used on maybe_enable_malloc_check (elf/dl-tunables.c)
    and to avoid having to build/use multiple versions if stack protection
    in enabled it is defined as inline.  */
@@ -33,3 +35,14 @@  __access_noerrno (const char *pathname, int mode)
     return INTERNAL_SYSCALL_ERRNO (res, err);
   return 0;
 }
+
+static inline int
+__kill_noerrno (pid_t pid, int sig)
+{
+  int res;
+  INTERNAL_SYSCALL_DECL (err);
+  res = INTERNAL_SYSCALL_CALL (kill, err, pid, sig);
+  if (INTERNAL_SYSCALL_ERROR_P (res, err))
+    return INTERNAL_SYSCALL_ERRNO (res, err);
+  return 0;
+}
diff --git a/sysdeps/unix/sysv/linux/s390/system.c b/sysdeps/unix/sysv/linux/s390/system.c
deleted file mode 100644
index d8ef461334..0000000000
--- a/sysdeps/unix/sysv/linux/s390/system.c
+++ /dev/null
@@ -1,29 +0,0 @@ 
-/* Copyright (C) 2003-2018 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/>.  */
-
-/* We have to and actually can handle cancelable system().  The big
-   problem: we have to kill the child process if necessary.  To do
-   this a cleanup handler has to be registered and is has to be able
-   to find the PID of the child.  The main problem is to reliable have
-   the PID when needed.  It is not necessary for the parent thread to
-   return.  It might still be in the kernel when the cancellation
-   request comes.  Therefore we have to use the clone() calls ability
-   to have the kernel write the PID into the user-level variable.  */
-#define FORK() \
-  INLINE_SYSCALL (clone, 3, 0, CLONE_PARENT_SETTID | SIGCHLD, &pid)
-
-#include "../system.c"
diff --git a/sysdeps/unix/sysv/linux/sparc/system.c b/sysdeps/unix/sysv/linux/sparc/system.c
deleted file mode 100644
index 1f65c83399..0000000000
--- a/sysdeps/unix/sysv/linux/sparc/system.c
+++ /dev/null
@@ -1,29 +0,0 @@ 
-/* Copyright (C) 2003-2018 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/>.  */
-
-/* We have to and actually can handle cancelable system().  The big
-   problem: we have to kill the child process if necessary.  To do
-   this a cleanup handler has to be registered and is has to be able
-   to find the PID of the child.  The main problem is to reliable have
-   the PID when needed.  It is not necessary for the parent thread to
-   return.  It might still be in the kernel when the cancellation
-   request comes.  Therefore we have to use the clone() calls ability
-   to have the kernel write the PID into the user-level variable.  */
-#define FORK() \
-  INLINE_CLONE_SYSCALL (CLONE_PARENT_SETTID | SIGCHLD, 0, &pid, NULL, NULL)
-
-#include "../system.c"
diff --git a/sysdeps/unix/sysv/linux/spawni.c b/sysdeps/unix/sysv/linux/spawni.c
index 85239cedbf..7f17fb2f89 100644
--- a/sysdeps/unix/sysv/linux/spawni.c
+++ b/sysdeps/unix/sysv/linux/spawni.c
@@ -138,11 +138,11 @@  __spawni_child (void *arguments)
   for (int sig = 1; sig < _NSIG; ++sig)
     {
       if ((attr->__flags & POSIX_SPAWN_SETSIGDEF)
-	  && sigismember (&attr->__sd, sig))
+	  && __sigismember (&attr->__sd, sig))
 	{
 	  sa.sa_handler = SIG_DFL;
 	}
-      else if (sigismember (&hset, sig))
+      else if (__sigismember (&hset, sig))
 	{
 	  if (__is_internal_signal (sig))
 	    sa.sa_handler = SIG_IGN;
@@ -330,10 +330,16 @@  __spawnix (pid_t * pid, const char *file,
   if (__glibc_unlikely (stack == MAP_FAILED))
     return errno;
 
-  /* Disable asynchronous cancellation.  */
   int state;
-  __libc_ptf_call (__pthread_setcancelstate,
-                   (PTHREAD_CANCEL_DISABLE, &state), 0);
+  if (xflags & SPAWN_XFLAGS_ENABLE_CANCEL)
+    __libc_signal_block_app (&args.oldmask);
+  else
+    {
+      /* Disable asynchronous cancellation.  */
+      __libc_ptf_call (__pthread_setcancelstate,
+		       (PTHREAD_CANCEL_DISABLE, &state), 0);
+      __libc_signal_block_all (&args.oldmask);
+    }
 
   /* Child must set args.err to something non-negative - we rely on
      the parent and child sharing VM.  */
@@ -347,8 +353,6 @@  __spawnix (pid_t * pid, const char *file,
   args.envp = envp;
   args.xflags = xflags;
 
-  __libc_signal_block_all (&args.oldmask);
-
   /* The clone flags used will create a new child that will run in the same
      memory space (CLONE_VM) and the execution of calling thread will be
      suspend until the child calls execve or _exit.
@@ -390,8 +394,8 @@  __spawnix (pid_t * pid, const char *file,
     *pid = new_pid;
 
   __libc_signal_restore_set (&args.oldmask);
-
-  __libc_ptf_call (__pthread_setcancelstate, (state, NULL), 0);
+  if ((xflags & SPAWN_XFLAGS_ENABLE_CANCEL) == 0)
+    __libc_ptf_call (__pthread_setcancelstate, (state, NULL), 0);
 
   return ec;
 }
diff --git a/sysdeps/unix/sysv/linux/system.c b/sysdeps/unix/sysv/linux/system.c
deleted file mode 100644
index 7cc68a1528..0000000000
--- a/sysdeps/unix/sysv/linux/system.c
+++ /dev/null
@@ -1,76 +0,0 @@ 
-/* Copyright (C) 2002-2018 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 <sched.h>
-#include <signal.h>
-#include <string.h>	/* For the real memset prototype.  */
-#include <sysdep.h>
-#include <unistd.h>
-#include <sys/wait.h>
-#include <libc-lock.h>
-
-/* We have to and actually can handle cancelable system().  The big
-   problem: we have to kill the child process if necessary.  To do
-   this a cleanup handler has to be registered and is has to be able
-   to find the PID of the child.  The main problem is to reliable have
-   the PID when needed.  It is not necessary for the parent thread to
-   return.  It might still be in the kernel when the cancellation
-   request comes.  Therefore we have to use the clone() calls ability
-   to have the kernel write the PID into the user-level variable.  */
-#ifndef FORK
-# define FORK() \
-  INLINE_SYSCALL (clone, 3, CLONE_PARENT_SETTID | SIGCHLD, 0, &pid)
-#endif
-
-#ifdef _LIBC_REENTRANT
-static void cancel_handler (void *arg);
-
-# define CLEANUP_HANDLER \
-  __libc_cleanup_region_start (1, cancel_handler, &pid)
-
-# define CLEANUP_RESET \
-  __libc_cleanup_region_end (0)
-#endif
-
-
-/* Linux has waitpid(), so override the generic unix version.  */
-#include <sysdeps/posix/system.c>
-
-
-#ifdef _LIBC_REENTRANT
-/* The cancellation handler.  */
-static void
-cancel_handler (void *arg)
-{
-  pid_t child = *(pid_t *) arg;
-
-  INTERNAL_SYSCALL_DECL (err);
-  INTERNAL_SYSCALL (kill, err, 2, child, SIGKILL);
-
-  TEMP_FAILURE_RETRY (__waitpid (child, NULL, 0));
-
-  DO_LOCK ();
-
-  if (SUB_REF () == 0)
-    {
-      (void) __sigaction (SIGQUIT, &quit, (struct sigaction *) NULL);
-      (void) __sigaction (SIGINT, &intr, (struct sigaction *) NULL);
-    }
-
-  DO_UNLOCK ();
-}
-#endif