In the Windows getopt_internal function the condition freeing the memory allocated by _dupenv_s is correct only for the first call to the function. the next callers will try to free the buffer even though the _dupenv_s call is skipped if the POSIXLY_CORRECT env isn't found (undefined behavior). Fixed by releasing the buffer in the scope of the same if statement calling _dupenv_s Fixes: 5e373e456e6acdc ("eal/windows: add getopt implementation") Cc: stable@dpdk.org Signed-off-by: Tal Shnaiderman <talshn@nvidia.com> --- lib/librte_eal/windows/getopt.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/lib/librte_eal/windows/getopt.c b/lib/librte_eal/windows/getopt.c index a08f7c109b..20da225a68 100644 --- a/lib/librte_eal/windows/getopt.c +++ b/lib/librte_eal/windows/getopt.c @@ -253,16 +253,17 @@ getopt_internal(int nargc, char **nargv, const char *options, * Disable GNU extensions if POSIXLY_CORRECT is set or options * string begins with a '+'. */ - if (posixly_correct == -1) + if (posixly_correct == -1) { posixly_correct = _dupenv_s(&buf, &len, "POSIXLY_CORRECT"); + if (!posixly_correct) + free(buf); + } if (!posixly_correct || *options == '+') flags &= ~FLAG_PERMUTE; else if (*options == '-') flags |= FLAG_ALLARGS; if (*options == '+' || *options == '-') options++; - if (!posixly_correct) - free(buf); /* * reset if requested */ -- 2.16.1.windows.4
On Wed, 23 Sep 2020 17:15:38 +0300, Tal Shnaiderman wrote: > In the Windows getopt_internal function the condition freeing > the memory allocated by _dupenv_s is correct only for the first > call to the function. Hi Tal, a few days back Khoa To and me privately discussed a patch that makes MinGW and Clang bith use getopt.c from librte_eal (now only Clang does). That patch includes adjustments that just remove _dupenv_s() along with the bug. Just discovered you were not Cc'd, sorry. I put your name in Reported-by and mentioned the bug in commit message: http://patchwork.dpdk.org/patch/78765/
> From: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>
> On Wed, 23 Sep 2020 17:15:38 +0300, Tal Shnaiderman wrote:
> > In the Windows getopt_internal function the condition freeing the
> > memory allocated by _dupenv_s is correct only for the first call to
> > the function.
>
> Hi Tal,
>
> a few days back Khoa To and me privately discussed a patch that makes
> MinGW and Clang bith use getopt.c from librte_eal (now only Clang does).
> That patch includes adjustments that just remove _dupenv_s() along with
> the bug. Just discovered you were not Cc'd, sorry. I put your name in
> Reported-by and mentioned the bug in commit message:
> https://nam11.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpatch
> work.dpdk.org%2Fpatch%2F78765%2F&data=02%7C01%7Ctalshn%40nvi
> dia.com%7Cab3a10f2d2ec458a7ef908d860e00e69%7C43083d15727340c1b7db
> 39efd9ccc17a%7C0%7C0%7C637365862717871619&sdata=1xiI9JZ%2BUw
> sv2mj8rkAcff6%2B3KFNYZUiM72ZbnkyokY%3D&reserved=0
Thanks for the update Dmitry, I'll mark this commit as superseded.