On Thu, Apr 21, 2022 at 7:55 PM cuigaosheng <[email protected]> wrote:
>
> This email adjusts the code format.
>
> I don't understand why we don't check path for NULL, bpf_link__pin is an
> external
> interface, It will be called by external functions and provide input
> parameters,
that external interface expects non-NULL string as input argument,
which is a default throughout libbpf's API. You will get SIGSEGV in
lots of cases if you pass NULL where you are not supposed to, e.g.,
bpf_object__open_file() and many others. It doesn't mean that libbpf
should check any pointer argument for NULL.
You can argue that strdup(NULL) shouldn't crash but it doesn't. It's
because strdup() has a contract that it shouldn't be passed NULL. So
is the case here.
> for example in samples/bpf/hbm.c:
>
> > 201 link = bpf_program__attach_cgroup(bpf_prog, cg1);
> > 202 if (libbpf_get_error(link)) {
> > 203 fprintf(stderr, "ERROR: bpf_program__attach_cgroup
> > failed\n");
> > 204 goto err;
> > 205 }
> > 206
> > 207 sprintf(cg_pin_path, "/sys/fs/bpf/hbm%d", cg_id);
> > 208 rc = bpf_link__pin(link, cg_pin_path);
> > 209 if (rc < 0) {
> > 210 printf("ERROR: bpf_link__pin failed: %d\n", rc);
> > 211 goto err;
> > 212 }
>
> if cg_pin_path is NULL, strdup(NULL) will trigger a segmentation fault in
> make_parent_dir, I think we should avoid this and add null-pointer checking
> for path, just like check_path:
> > 7673 static int check_path(const char *path)
> > 7674 {
> > 7675 char *cp, errmsg[STRERR_BUFSIZE];
> > 7676 struct statfs st_fs;
> > 7677 char *dname, *dir;
> > 7678 int err = 0;
> > 7679
> > 7680 if (path == NULL)
> > 7681 return -EINVAL;
> > 7682
> > 7683 dname = strdup(path);
> > 7684 if (dname == NULL)
> > 7685 return -ENOMEM;
> > 7686
> > 7687 dir = dirname(dname);
> > 7688 if (statfs(dir, &st_fs)) {
> > 7689 cp = libbpf_strerror_r(errno, errmsg,
> > sizeof(errmsg));
> > 7690 pr_warn("failed to statfs %s: %s\n", dir, cp);
> > 7691 err = -errno;
> > 7692 }
> > 7693 free(dname);
> > 7694
> > 7695 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
> > 7696 pr_warn("specified path %s is not on BPF FS\n",
> > path);
> > 7697 err = -EINVAL;
> > 7698 }
> > 7699
> > 7700 return err;
> > 7701 }
>
> Thanks.
>
>
> 在 2022/4/22 0:55, Andrii Nakryiko 写道:
> > On Thu, Apr 21, 2022 at 6:01 AM Gaosheng Cui <[email protected]> wrote:
> >> The make_parent_dir is called without null-pointer checking for path,
> >> such as bpf_link__pin. To ensure there is no null-pointer dereference
> >> in make_parent_dir, so make_parent_dir requires additional null-pointer
> >> checking for path.
> >>
> >> Signed-off-by: Gaosheng Cui <[email protected]>
> >> ---
> >> tools/lib/bpf/libbpf.c | 3 +++
> >> 1 file changed, 3 insertions(+)
> >>
> >> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> >> index b53e51884f9e..5786e6184bf5 100644
> >> --- a/tools/lib/bpf/libbpf.c
> >> +++ b/tools/lib/bpf/libbpf.c
> >> @@ -7634,6 +7634,9 @@ static int make_parent_dir(const char *path)
> >> char *dname, *dir;
> >> int err = 0;
> >>
> >> + if (path == NULL)
> >> + return -EINVAL;
> >> +
> > API contract is that path shouldn't be NULL. Just like we don't check
> > link, obj, prog for NULL in every single API, I don't think we need to
> > do it here, unless I'm missing something?
> >
> >> dname = strdup(path);
> >> if (dname == NULL)
> >> return -ENOMEM;
> >> --
> >> 2.25.1
> >>
> > .