On Mon, 18 Dec 2017, Masahiro Yamada wrote:
2017-12-15 17:23 GMT+09:00 Nicholas Mc Guire <der.herr@xxxxxxx>:
On Thu, Dec 14, 2017 at 08:54:10PM +0100, Lukas Bulwahn wrote:
Commit dee81e988674 ("fixdep: faster CONFIG_ search") introduces the memory
leak when `map = mmap(...)` was replaced with `map = malloc(...)` and
`read(fd, map, ...)`. It introduces a new second error path, which does not
free the allocated memory for `map`. We now correct that behavior and free
`map` before the do_config_file() function returns.
Facebook's static analysis tool Infer (http://fbinfer.com) found this
memory leak:
scripts/basic/fixdep.c:297: error: MEMORY_LEAK
memory dynamically allocated by call to `malloc()` at line 290, \
column 8 is not reachable after line 297, column 3.
Fixes: dee81e988674 ("fixdep: faster CONFIG_ search")
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@xxxxxxxxx>
Reviewed-by: Nicholas Mc Guire <der.herr@xxxxxxx>
---
scripts/basic/fixdep.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/scripts/basic/fixdep.c b/scripts/basic/fixdep.c
index bbf62cb..131c450 100644
--- a/scripts/basic/fixdep.c
+++ b/scripts/basic/fixdep.c
@@ -296,6 +296,7 @@ static void do_config_file(const char *filename)
if (read(fd, map, st.st_size) != st.st_size) {
perror("fixdep: read");
close(fd);
+ free(map);
This looks reasonable but actually it is not clear why do_config_file()
should return at all if read fails as the read error would go unnoticed
in the current code and allow the build to continue. so this probably
should be an exit(2) here and not a return which would then take care
of the free() anyway.
I agree.
This should be exit(2).
You can also remove close() as well since
the operation system will close it anyway.
Atleast I do not see the rational to allow continuation if the read
failed as the file should not be empty nor a mismatch with st.st_size
expected. If it were due to a EINTR then it still should terminate as
EINTR was not handled and we thus could miss a valid dependency.
Note: this probably also should be applied to the if (!map) condition
before as well, as at that point it is known that map > 0 and a malloc()
failure would allow skipping parse_config_file() for a valid config.
I agree. We should change this too.
I created a new patch with your requested changes, but I did not consider
it a second version of this patch by the time of writing. You can find the
new patch at:
v1: https://patchwork.kernel.org/patch/10126547/
v2: https://patchwork.kernel.org/patch/10128297/
This patch here is now superseded by the new patch. So, this patch here
can be abandoned.
Lukas
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html