debug: make __read_chk a cancellation point (bug 29274)

  The __read_chk function, as the implementation behind the fortified read
function, must be a cancellation point, thus it cannot use INLINE_SYSCALL.
---
 debug/Makefile              |  7 ++++++
 debug/read_chk.c            | 10 --------
 debug/tst-read-chk-cancel.c | 50 +++++++++++++++++++++++++++++++++++++
 3 files changed, 57 insertions(+), 10 deletions(-)
 create mode 100644 debug/tst-read-chk-cancel.c
  

On 22/06/2022 20:14, Andreas Schwab via Libc-alpha wrote:
> The __read_chk function, as the implementation behind the fortified read
> function, must be a cancellation point, thus it cannot use INLINE_SYSCALL.
> ---
>   debug/Makefile              |  7 ++++++
>   debug/read_chk.c            | 10 --------
>   debug/tst-read-chk-cancel.c | 50 +++++++++++++++++++++++++++++++++++++
>   3 files changed, 57 insertions(+), 10 deletions(-)
>   create mode 100644 debug/tst-read-chk-cancel.c

LGTM.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>

> 
> diff --git a/debug/Makefile b/debug/Makefile
> index 96029f32ee..456b349c4d 100644
> --- a/debug/Makefile
> +++ b/debug/Makefile
> @@ -110,6 +110,7 @@ CPPFLAGS-tst-longjmp_chk2.c += -D_FORTIFY_SOURCE=1
>   CFLAGS-tst-longjmp_chk3.c += -fexceptions -fasynchronous-unwind-tables
>   CPPFLAGS-tst-longjmp_chk3.c += -D_FORTIFY_SOURCE=1
>   CPPFLAGS-tst-realpath-chk.c += -D_FORTIFY_SOURCE=2
> +CPPFLAGS-tst-read-chk-cancel.c += -D_FORTIFY_SOURCE=2
>   
>   # _FORTIFY_SOURCE tests.
>   # Auto-generate tests for _FORTIFY_SOURCE for different levels, compilers and
> @@ -204,6 +205,10 @@ ifeq ($(have-ssp),yes)
>   tests += tst-ssp-1
>   endif
>   
> +ifeq ($(have-thread-library), yes)
> +tests += tst-read-chk-cancel
> +endif
> +
>   ifeq (,$(CXX))
>   tests-unsupported = $(tests-cc-chk)
>   endif
> @@ -242,3 +247,5 @@ $(objpfx)xtrace: xtrace.sh
>   	    -e 's|@BINDIR@|$(bindir)|' -e 's|@PKGVERSION@|$(PKGVERSION)|' \
>   	    -e 's|@REPORT_BUGS_TO@|$(REPORT_BUGS_TO)|' $^ > $@.new \
>   	&& rm -f $@ && mv $@.new $@ && chmod +x $@
> +
> +$(objpfx)tst-read-chk-cancel: $(shared-thread-library)
> diff --git a/debug/read_chk.c b/debug/read_chk.c
> index 0cd58db8cb..274b4f93e9 100644
> --- a/debug/read_chk.c
> +++ b/debug/read_chk.c
> @@ -16,12 +16,6 @@
>      <https://www.gnu.org/licenses/>.  */
>   
>   #include <unistd.h>
> -#include <sys/param.h>
> -#ifdef HAVE_INLINED_SYSCALLS
> -# include <errno.h>
> -# include <sysdep.h>
> -#endif
> -
>   
>   ssize_t
>   __read_chk (int fd, void *buf, size_t nbytes, size_t buflen)
> @@ -29,9 +23,5 @@ __read_chk (int fd, void *buf, size_t nbytes, size_t buflen)
>     if (nbytes > buflen)
>       __chk_fail ();
>   
> -#ifdef HAVE_INLINED_SYSCALLS
> -  return INLINE_SYSCALL (read, 3, fd, buf, nbytes);
> -#else
>     return __read (fd, buf, nbytes);
> -#endif
>   }
> diff --git a/debug/tst-read-chk-cancel.c b/debug/tst-read-chk-cancel.c
> new file mode 100644
> index 0000000000..7e06afb596
> --- /dev/null
> +++ b/debug/tst-read-chk-cancel.c
> @@ -0,0 +1,50 @@
> +/* Test that __read_chk is a cancellation point (BZ #29274)
> +   Copyright (C) 2022 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
> +   <https://www.gnu.org/licenses/>.  */
> +
> +#include <stdint.h>
> +#include <support/xunistd.h>
> +#include <support/xthread.h>
> +
> +static int pipe_fds[2];
> +static pthread_barrier_t barrier;
> +
> +static void *
> +read_thread (void *n)
> +{
> +  xpthread_barrier_wait (&barrier);
> +  char c;
> +  /* This call should be forwarded to __read_chk because the buffer size
> +     is known, but the read length is non-constant.  */
> +  if (read (pipe_fds[0], &c, (uintptr_t) n) != 1)
> +    return (void *) -1L;
> +  return 0;
> +}
> +
> +static int
> +do_test (void)
> +{
> +  xpthread_barrier_init (&barrier, 0, 2);
> +  xpipe (pipe_fds);
> +  pthread_t thr = xpthread_create (0, read_thread, (void *) 1L);
> +  xpthread_barrier_wait (&barrier);
> +  xpthread_cancel (thr);
> +  xpthread_join (thr);
> +  return 0;
> +}
> +
> +#include <support/test-driver.c>
  

> On 22 Jun 2022, at 11:44, Andreas Schwab via Libc-alpha <libc-alpha@sourceware.org> wrote:
> 
> The __read_chk function, as the implementation behind the fortified read
> function, must be a cancellation point, thus it cannot use INLINE_SYSCALL.
> ---
> debug/Makefile              |  7 ++++++
> debug/read_chk.c            | 10 --------
> debug/tst-read-chk-cancel.c | 50 +++++++++++++++++++++++++++++++++++++

Maybe add a tst-cancel4-chk that builds with _FORTIFY_SOURCE=2 so we can tests
more cancellation entrypoints?