[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 12 Dec 2018 04:38:10 +0000
From: Martin Lau <kafai@...com>
To: Daniel Borkmann <daniel@...earbox.net>
CC: Song Liu <songliubraving@...com>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"ast@...nel.org" <ast@...nel.org>, Kernel Team <Kernel-team@...com>
Subject: Re: [PATCH v2 bpf-next 1/2] bpf: include sub program tags in
bpf_prog_info
On Wed, Dec 12, 2018 at 03:22:38AM +0100, Daniel Borkmann wrote:
> On 12/11/2018 09:18 AM, Song Liu wrote:
> > Changes from v1:
> > 1. Fix error path as Martin suggested.
> >
> > This patch adds nr_prog_tags and prog_tags to bpf_prog_info. This is a
> > reliable way for user space to get tags of all sub programs. Before this
> > patch, user space need to find sub program tags via kallsyms.
> >
> > This feature will be used in BPF introspection, where user space queries
> > information about BPF programs via sys_bpf.
> >
> > Signed-off-by: Song Liu <songliubraving@...com>
> > ---
> > include/uapi/linux/bpf.h | 2 ++
> > kernel/bpf/syscall.c | 27 +++++++++++++++++++++++++++
> > 2 files changed, 29 insertions(+)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index 1bee1135866a..368d185aa32f 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -2703,6 +2703,8 @@ struct bpf_prog_info {
> > __u32 jited_line_info_cnt;
> > __u32 line_info_rec_size;
> > __u32 jited_line_info_rec_size;
> > + __u32 nr_prog_tags;
> > + __aligned_u64 prog_tags;
> > } __attribute__((aligned(8)));
> >
> > struct bpf_map_info {
> > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > index 19c88cff7880..49cb59177db9 100644
> > --- a/kernel/bpf/syscall.c
> > +++ b/kernel/bpf/syscall.c
> > @@ -2322,6 +2322,33 @@ static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
> > }
> > }
> >
> > + ulen = info.nr_prog_tags;
> > + info.nr_prog_tags = prog->aux->func_cnt ? : 1;
> > + if (ulen) {
> > + if (bpf_dump_raw_ok()) {
>
> Hm, why is bpf_dump_raw_ok() needed here? tag is not exposing a kernel
> address. prog tag is in general also visible from fdinfo from unpriv.
>
> Just looking at the recently merged func_info / line_info I'm not sure
> this is needed there either ... Martin, is there a specific reason this
> was added there as well?
It is mostly to follow info.jited_func_lens which also tests
bpf_dump_raw_ok(). We thought the check iss there even it is
not exposing the kernel address because jited_func_lens is not
useful in general without jited_ksyms.
Same go for func_info when dumping jited insn. func_info used to only
support jited insn dump.
Considering func_info was later made to support xlated insn also,
I think it makes sense to remove the bpf_dump_raw_ok() check
for func_info, line_info and prog_tags and ask the
userspace to decide if it has all needed details before dumping
info for the xlated insn and jited insn?
The bpf_dump_raw_ok() check for jited_line_info will stay
because jited_line_info contains kernel address.
> > + __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
> > + u32 i;
> > +
> > + user_prog_tags = u64_to_user_ptr(info.prog_tags);
> > + ulen = min_t(u32, info.nr_prog_tags, ulen);
> > + if (prog->aux->func_cnt) {
> > + for (i = 0; i < ulen; i++) {
> > + if (copy_to_user(
> > + user_prog_tags[i],
> > + prog->aux->func[i]->tag,
> > + BPF_TAG_SIZE))
> > + return -EFAULT;
> > + }
> > + } else {
> > + if (copy_to_user(user_prog_tags[0],
> > + prog->tag, BPF_TAG_SIZE))
> > + return -EFAULT;
> > + }
> > + } else {
> > + info.prog_tags = 0;
> > + }
> > + }
> > +
> > done:
> > if (copy_to_user(uinfo, &info, info_len) ||
> > put_user(info_len, &uattr->info.info_len))
> > --
> > 2.17.1
> >
>
Powered by blists - more mailing lists