| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Thu, 27 Jun 2019 19:38:10 -0700
From: Stanislav Fomichev <sdf@...ichev.me>
To: Andrii Nakryiko <andrii.nakryiko@...il.com>
Cc: kernel test robot <rong.a.chen@...el.com>,
Stanislav Fomichev <sdf@...gle.com>,
Daniel Borkmann <daniel@...earbox.net>,
Martin Lau <kafai@...com>, LKML <linux-kernel@...r.kernel.org>,
Stephen Rothwell <sfr@...b.auug.org.au>,
bpf <bpf@...r.kernel.org>, lkp@...org
Subject: Re: [bpf/tools] cd17d77705:
kernel_selftests.bpf.test_sock_addr.sh.fail
On 06/27, Andrii Nakryiko wrote:
> On Thu, Jun 27, 2019 at 10:29 AM Stanislav Fomichev <sdf@...ichev.me> wrote:
> >
> > On 06/27, Stanislav Fomichev wrote:
> > > On 06/27, kernel test robot wrote:
> > > > FYI, we noticed the following commit (built with gcc-7):
> > > >
> > > > commit: cd17d77705780e2270937fb3cbd2b985adab3edc ("bpf/tools: sync bpf.h")
> > > > https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
> > > >
> > > > in testcase: kernel_selftests
> > > > with following parameters:
> > > >
> > > > group: kselftests-00
> > > >
> > > > test-description: The kernel contains a set of "self tests" under the tools/testing/selftests/ directory. These are intended to be small unit tests to exercise individual code paths in the kernel.
> > > > test-url: https://www.kernel.org/doc/Documentation/kselftest.txt
> > > >
> > > >
> > > > on test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 8G
> > > >
> > > > caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
> > > >
> > > > # 55: (18) r1 = 0x100000000000000
> > > > # ; ctx->user_ip6[2] = bpf_htonl(DST_REWRITE_IP6_2);
> > > > # 57: (7b) *(u64 *)(r6 +16) = r1
> > > > # invalid bpf_context access off=16 size=8
> > > This looks like clang doing single u64 write for user_ip6[2] and
> > > user_ip6[3] instead of two u32. I don't think we allow that.
> > >
> > > I've seen this a couple of times myself while playing with some
> > > progs, but not sure what's the right way to 'fix' it.
> > >
> > Any thoughts about the patch below? Another way to "fix" it
>
> I'll give it a more thorough look a bit later, but see my comments below.
>
> > would be to mark context accesses 'volatile' in bpf progs, but that sounds
> > a bit gross.
> >
> > diff --git a/include/linux/filter.h b/include/linux/filter.h
> > index 43b45d6db36d..34a14c950e60 100644
> > --- a/include/linux/filter.h
> > +++ b/include/linux/filter.h
> > @@ -746,6 +746,20 @@ bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
> > return size <= size_default && (size & (size - 1)) == 0;
> > }
> >
> > +static inline bool __bpf_ctx_wide_store_ok(u32 off, u32 size)
>
> It seems like bpf_ctx_wide_store_ok and __bpf_ctx_wide_store_ok are
> used only inside net/core/filter.c, why declaring them in header file?
I wanted it to be next to bpf_ctx_narrow_access_ok which does the
reverse operation for reads.
> > +{
> > + /* u64 access is aligned and fits into the field size */
> > + return off % sizeof(__u64) == 0 && off + sizeof(__u64) <= size;
> > +}
> > +
> > +#define bpf_ctx_wide_store_ok(off, size, type, field) \
> > + (size == sizeof(__u64) && \
> > + off >= offsetof(type, field) && \
> > + off < offsetofend(type, field) ? \
> > + __bpf_ctx_wide_store_ok(off - offsetof(type, field), \
> > + FIELD_SIZEOF(type, field)) : 0)
>
> Why do you need ternary operator instead of just a chain of &&s?
Good point. I didn't spend too much time on the patch tbh :-)
If it looks good in general, I can add proper tests and do a
proper submition, this patch is just to get the discussion started.
> It also seems like you can avoid macro and use plain function if
> instead of providing (type, field) you provide values of offsetof and
> offsetofend (offsetofend - offsetof should equal FIELD_SIZEOF(type,
> field), shouldn't it?).
But then I'd have to copy-paste the args of offsetof/offsetofend at
the caller, right? I wanted the caller to be clean and simple.
> > #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
> >
> > static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
> > diff --git a/net/core/filter.c b/net/core/filter.c
> > index 2014d76e0d2a..2d3787a439ae 100644
> > --- a/net/core/filter.c
> > +++ b/net/core/filter.c
> > @@ -6849,6 +6849,16 @@ static bool sock_addr_is_valid_access(int off, int size,
> > if (!bpf_ctx_narrow_access_ok(off, size, size_default))
> > return false;
> > } else {
> > + if (bpf_ctx_wide_store_ok(off, size,
> > + struct bpf_sock_addr,
> > + user_ip6))
> > + return true;
> > +
> > + if (bpf_ctx_wide_store_ok(off, size,
> > + struct bpf_sock_addr,
> > + msg_src_ip6))
> > + return true;
> > +
> > if (size != size_default)
> > return false;
> > }
> > @@ -7689,9 +7699,6 @@ static u32 xdp_convert_ctx_access(enum bpf_access_type type,
> > /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
> > * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
> > *
> > - * It doesn't support SIZE argument though since narrow stores are not
> > - * supported for now.
> > - *
> > * In addition it uses Temporary Field TF (member of struct S) as the 3rd
> > * "register" since two registers available in convert_ctx_access are not
> > * enough: we can't override neither SRC, since it contains value to store, nor
> > @@ -7699,7 +7706,7 @@ static u32 xdp_convert_ctx_access(enum bpf_access_type type,
> > * instructions. But we need a temporary place to save pointer to nested
> > * structure whose field we want to store to.
> > */
> > -#define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
> > +#define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
> > do { \
> > int tmp_reg = BPF_REG_9; \
> > if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
> > @@ -7710,8 +7717,7 @@ static u32 xdp_convert_ctx_access(enum bpf_access_type type,
> > offsetof(S, TF)); \
> > *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
> > si->dst_reg, offsetof(S, F)); \
> > - *insn++ = BPF_STX_MEM( \
> > - BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
> > + *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
> > bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
> > target_size) \
> > + OFF); \
> > @@ -7723,8 +7729,8 @@ static u32 xdp_convert_ctx_access(enum bpf_access_type type,
> > TF) \
> > do { \
> > if (type == BPF_WRITE) { \
> > - SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
> > - TF); \
> > + SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
> > + OFF, TF); \
> > } else { \
> > SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
> > S, NS, F, NF, SIZE, OFF); \
Powered by blists - more mailing lists