public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
* [PATCH] gdbsupport: fix path_join crash with -std=c++17 and -D_GLIBCXX_DEBUG
@ 2022-05-04 14:18 Simon Marchi
  2022-05-19 12:08 ` Simon Marchi
  0 siblings, 1 reply; 2+ messages in thread
From: Simon Marchi @ 2022-05-04 14:18 UTC (permalink / raw)
  To: gdb-patches

When building GDB with -std=c++17 and -D_GLIBCXX_DEBUG=1, I get:

  $ ./gdb -nx --data-directory=data-directory -q -ex "maint selftest path_join"
  /usr/include/c++/11.2.0/string_view:233: constexpr const value_type& std::basic_string_view<_CharT, _Traits>::operator[](std::basic_string_view<_CharT, _Traits>::size_type) const [with _CharT = char; _Traits = std::char_traits<char>; std::basic_string_view<_CharT, _Traits>::const_reference = const char&; std::basic_string_view<_CharT, _Traits>::size_type = long unsigned int]: Assertion  '__pos < this->_M_len' failed.

The problem is that we're passing an empty string_view to
IS_ABSOLUTE_PATH.  IS_ABSOLUTE_PATH accesses [0] on that string_view,
which is out-of-bounds.

The reason this is not seen with -std less than c++17 is that our local
copy of string_view (used with C++ < 17) does not have the assert in
operator[], as that wouldn't work in a constexpr method:

  https://gitlab.com/gnutools/binutils-gdb/-/blob/5890af36e5112bcbb8d7555e63570f68466e6944/gdbsupport/gdb_string_view.h#L180

IS_ABSOLUTE_PATH is normally used with null-terminated string.  It's
fine to pass an empty null-terminated string to IS_ABSOLUTE_PATH,
because index 0 in such a string is valid.  But not with an empty
string_view.

Fix that by avoiding the "call" to IS_ABSOLUTE_PATH if the string_view
is empty.

Change-Id: Idf4df961b63f513b3389235e93814c02b89ea32e
---
 gdbsupport/pathstuff.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gdbsupport/pathstuff.cc b/gdbsupport/pathstuff.cc
index 5b5a8eea9047..af10c6ebd2e8 100644
--- a/gdbsupport/pathstuff.cc
+++ b/gdbsupport/pathstuff.cc
@@ -200,7 +200,7 @@ path_join (gdb::array_view<const gdb::string_view> paths)
       const gdb::string_view path = paths[i];
 
       if (i > 0)
-	gdb_assert (!IS_ABSOLUTE_PATH (path));
+	gdb_assert (path.empty () || !IS_ABSOLUTE_PATH (path));
 
       if (!ret.empty () && !IS_DIR_SEPARATOR (ret.back ()))
 	  ret += '/';

base-commit: 5890af36e5112bcbb8d7555e63570f68466e6944
-- 
2.36.0


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] gdbsupport: fix path_join crash with -std=c++17 and -D_GLIBCXX_DEBUG
  2022-05-04 14:18 [PATCH] gdbsupport: fix path_join crash with -std=c++17 and -D_GLIBCXX_DEBUG Simon Marchi
@ 2022-05-19 12:08 ` Simon Marchi
  0 siblings, 0 replies; 2+ messages in thread
From: Simon Marchi @ 2022-05-19 12:08 UTC (permalink / raw)
  To: gdb-patches



On 2022-05-04 10:18, Simon Marchi wrote:
> When building GDB with -std=c++17 and -D_GLIBCXX_DEBUG=1, I get:
> 
>   $ ./gdb -nx --data-directory=data-directory -q -ex "maint selftest path_join"
>   /usr/include/c++/11.2.0/string_view:233: constexpr const value_type& std::basic_string_view<_CharT, _Traits>::operator[](std::basic_string_view<_CharT, _Traits>::size_type) const [with _CharT = char; _Traits = std::char_traits<char>; std::basic_string_view<_CharT, _Traits>::const_reference = const char&; std::basic_string_view<_CharT, _Traits>::size_type = long unsigned int]: Assertion  '__pos < this->_M_len' failed.
> 
> The problem is that we're passing an empty string_view to
> IS_ABSOLUTE_PATH.  IS_ABSOLUTE_PATH accesses [0] on that string_view,
> which is out-of-bounds.
> 
> The reason this is not seen with -std less than c++17 is that our local
> copy of string_view (used with C++ < 17) does not have the assert in
> operator[], as that wouldn't work in a constexpr method:
> 
>   https://gitlab.com/gnutools/binutils-gdb/-/blob/5890af36e5112bcbb8d7555e63570f68466e6944/gdbsupport/gdb_string_view.h#L180
> 
> IS_ABSOLUTE_PATH is normally used with null-terminated string.  It's
> fine to pass an empty null-terminated string to IS_ABSOLUTE_PATH,
> because index 0 in such a string is valid.  But not with an empty
> string_view.
> 
> Fix that by avoiding the "call" to IS_ABSOLUTE_PATH if the string_view
> is empty.
> 
> Change-Id: Idf4df961b63f513b3389235e93814c02b89ea32e
> ---
>  gdbsupport/pathstuff.cc | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/gdbsupport/pathstuff.cc b/gdbsupport/pathstuff.cc
> index 5b5a8eea9047..af10c6ebd2e8 100644
> --- a/gdbsupport/pathstuff.cc
> +++ b/gdbsupport/pathstuff.cc
> @@ -200,7 +200,7 @@ path_join (gdb::array_view<const gdb::string_view> paths)
>        const gdb::string_view path = paths[i];
>  
>        if (i > 0)
> -	gdb_assert (!IS_ABSOLUTE_PATH (path));
> +	gdb_assert (path.empty () || !IS_ABSOLUTE_PATH (path));
>  
>        if (!ret.empty () && !IS_DIR_SEPARATOR (ret.back ()))
>  	  ret += '/';
> 
> base-commit: 5890af36e5112bcbb8d7555e63570f68466e6944

I pushed this.

Simon

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2022-05-19 12:09 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-05-04 14:18 [PATCH] gdbsupport: fix path_join crash with -std=c++17 and -D_GLIBCXX_DEBUG Simon Marchi
2022-05-19 12:08 ` Simon Marchi

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).