Hi Aaron,
On Fri, Jan 24, 2025 at 08:32:58PM -0500, Aaron Merey wrote:
> skel_fd is passed to create_dwfl, which calls dup() on skel_fd.
> create_dwfl handles closing the dup'ed fd but not the original.
>
> Ensure the original skel_fd is closed after it's passed to create_dwfl.
Nice find.
We should add --track-fds=yes to our valgrind testing. It would have
found this earlier. Two of our tests currently fail with:
-==75005== Open file descriptor 5: testfile-splitdwarf-4
-==75005== at 0x4A1E023: open (in /usr/lib64/libc.so.6)
-==75005== by 0x41392B: open (fcntl2.h:55)
-==75005== by 0x41392B: print_debug (readelf.c:11919)
-==75005== by 0x416451: process_elf_file (readelf.c:1084)
-==75005== by 0x4170CC: process_dwflmod (readelf.c:840)
-==75005== by 0x489CAE0: dwfl_getmodules (dwfl_getmodules.c:86)
-==75005== by 0x406964: process_file (readelf.c:948)
-==75005== by 0x401D11: main (readelf.c:417)
This file descriptor leak disappears with your patch.
Thanks,
Mark
> Signed-off-by: Aaron Merey <ame...@redhat.com>
> ---
> src/readelf.c | 8 +++++++-
> 1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/src/readelf.c b/src/readelf.c
> index 3e97b64c..6526db07 100644
> --- a/src/readelf.c
> +++ b/src/readelf.c
> @@ -11921,7 +11921,13 @@ print_debug (Dwfl_Module *dwflmod, Ebl *ebl,
> GElf_Ehdr *ehdr)
> fprintf (stderr, "Warning: Couldn't open DWARF skeleton file"
> " '%s'\n", skel_name);
> else
> - skel_dwfl = create_dwfl (skel_fd, skel_name);
> + {
> + skel_dwfl = create_dwfl (skel_fd, skel_name);
> +
> + /* skel_fd was dup'ed by create_dwfl. We can close the
> + original now. */
> + close (skel_fd);
> + }
>
> if (skel_dwfl != NULL)
> {
> --
> 2.47.1
>
