[ Upstream commit e60b0d12a9 ]
If we ever get to a point again where we convert a bogus looking <ptr>_or_null
typed register containing a non-zero fixed or variable offset, then lets not
reset these bounds to zero since they are not and also don't promote the register
to a <ptr> type, but instead leave it as <ptr>_or_null. Converting to a unknown
register could be an avenue as well, but then if we run into this case it would
allow to leak a kernel pointer this way.
Fixes: f1174f77b5 ("bpf/verifier: rework value tracking")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ no upstream commit given implicitly fixed through the larger refactoring
in c25b2ae136 ]
While auditing some other code, I noticed missing checks inside the pointer
arithmetic simulation, more specifically, adjust_ptr_min_max_vals(). Several
*_OR_NULL types are not rejected whereas they are _required_ to be rejected
given the expectation is that they get promoted into a 'real' pointer type
for the success case, that is, after an explicit != NULL check.
One case which stands out and is accessible from unprivileged (iff enabled
given disabled by default) is BPF ring buffer. From crafting a PoC, the NULL
check can be bypassed through an offset, and its id marking will then lead
to promotion of mem_or_null to a mem type.
bpf_ringbuf_reserve() helper can trigger this case through passing of reserved
flags, for example.
func#0 @0
0: R1=ctx(id=0,off=0,imm=0) R10=fp0
0: (7a) *(u64 *)(r10 -8) = 0
1: R1=ctx(id=0,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm
1: (18) r1 = 0x0
3: R1_w=map_ptr(id=0,off=0,ks=0,vs=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm
3: (b7) r2 = 8
4: R1_w=map_ptr(id=0,off=0,ks=0,vs=0,imm=0) R2_w=invP8 R10=fp0 fp-8_w=mmmmmmmm
4: (b7) r3 = 0
5: R1_w=map_ptr(id=0,off=0,ks=0,vs=0,imm=0) R2_w=invP8 R3_w=invP0 R10=fp0 fp-8_w=mmmmmmmm
5: (85) call bpf_ringbuf_reserve#131
6: R0_w=mem_or_null(id=2,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
6: (bf) r6 = r0
7: R0_w=mem_or_null(id=2,ref_obj_id=2,off=0,imm=0) R6_w=mem_or_null(id=2,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
7: (07) r0 += 1
8: R0_w=mem_or_null(id=2,ref_obj_id=2,off=1,imm=0) R6_w=mem_or_null(id=2,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
8: (15) if r0 == 0x0 goto pc+4
R0_w=mem(id=0,ref_obj_id=0,off=0,imm=0) R6_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
9: R0_w=mem(id=0,ref_obj_id=0,off=0,imm=0) R6_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
9: (62) *(u32 *)(r6 +0) = 0
R0_w=mem(id=0,ref_obj_id=0,off=0,imm=0) R6_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
10: R0_w=mem(id=0,ref_obj_id=0,off=0,imm=0) R6_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
10: (bf) r1 = r6
11: R0_w=mem(id=0,ref_obj_id=0,off=0,imm=0) R1_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R6_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
11: (b7) r2 = 0
12: R0_w=mem(id=0,ref_obj_id=0,off=0,imm=0) R1_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R2_w=invP0 R6_w=mem(id=0,ref_obj_id=2,off=0,imm=0) R10=fp0 fp-8_w=mmmmmmmm refs=2
12: (85) call bpf_ringbuf_submit#132
13: R6=invP(id=0) R10=fp0 fp-8=mmmmmmmm
13: (b7) r0 = 0
14: R0_w=invP0 R6=invP(id=0) R10=fp0 fp-8=mmmmmmmm
14: (95) exit
from 8 to 13: safe
processed 15 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 0
OK
All three commits, that is b121b341e5 ("bpf: Add PTR_TO_BTF_ID_OR_NULL support"),
457f44363a ("bpf: Implement BPF ring buffer and verifier support for it"), and the
afbf21dce6 ("bpf: Support readonly/readwrite buffers in verifier") suffer the same
cause and their *_OR_NULL type pendants must be rejected in adjust_ptr_min_max_vals().
Make the test more robust by reusing reg_type_may_be_null() helper such that we catch
all *_OR_NULL types we have today and in future.
Note that pointer arithmetic on PTR_TO_BTF_ID, PTR_TO_RDONLY_BUF, and PTR_TO_RDWR_BUF
is generally allowed.
Fixes: b121b341e5 ("bpf: Add PTR_TO_BTF_ID_OR_NULL support")
Fixes: 457f44363a ("bpf: Implement BPF ring buffer and verifier support for it")
Fixes: afbf21dce6 ("bpf: Support readonly/readwrite buffers in verifier")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit e572ff80f0 upstream.
Make the bounds propagation in __reg_assign_32_into_64() slightly more
robust and readable by aligning it similarly as we did back in the
__reg_combine_64_into_32() counterpart. Meaning, only propagate or
pessimize them as a smin/smax pair.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3cf2b61eb0 upstream.
For the case where both s32_{min,max}_value bounds are positive, the
__reg_assign_32_into_64() directly propagates them to their 64 bit
counterparts, otherwise it pessimises them into [0,u32_max] universe and
tries to refine them later on by learning through the tnum as per comment
in mentioned function. However, that does not always happen, for example,
in mov32 operation we call zext_32_to_64(dst_reg) which invokes the
__reg_assign_32_into_64() as is without subsequent bounds update as
elsewhere thus no refinement based on tnum takes place.
Thus, not calling into the __update_reg_bounds() / __reg_deduce_bounds() /
__reg_bound_offset() triplet as we do, for example, in case of ALU ops via
adjust_scalar_min_max_vals(), will lead to more pessimistic bounds when
dumping the full register state:
Before fix:
0: (b4) w0 = -1
1: R0_w=invP4294967295
(id=0,imm=ffffffff,
smin_value=4294967295,smax_value=4294967295,
umin_value=4294967295,umax_value=4294967295,
var_off=(0xffffffff; 0x0),
s32_min_value=-1,s32_max_value=-1,
u32_min_value=-1,u32_max_value=-1)
1: (bc) w0 = w0
2: R0_w=invP4294967295
(id=0,imm=ffffffff,
smin_value=0,smax_value=4294967295,
umin_value=4294967295,umax_value=4294967295,
var_off=(0xffffffff; 0x0),
s32_min_value=-1,s32_max_value=-1,
u32_min_value=-1,u32_max_value=-1)
Technically, the smin_value=0 and smax_value=4294967295 bounds are not
incorrect, but given the register is still a constant, they break assumptions
about const scalars that smin_value == smax_value and umin_value == umax_value.
After fix:
0: (b4) w0 = -1
1: R0_w=invP4294967295
(id=0,imm=ffffffff,
smin_value=4294967295,smax_value=4294967295,
umin_value=4294967295,umax_value=4294967295,
var_off=(0xffffffff; 0x0),
s32_min_value=-1,s32_max_value=-1,
u32_min_value=-1,u32_max_value=-1)
1: (bc) w0 = w0
2: R0_w=invP4294967295
(id=0,imm=ffffffff,
smin_value=4294967295,smax_value=4294967295,
umin_value=4294967295,umax_value=4294967295,
var_off=(0xffffffff; 0x0),
s32_min_value=-1,s32_max_value=-1,
u32_min_value=-1,u32_max_value=-1)
Without the smin_value == smax_value and umin_value == umax_value invariant
being intact for const scalars, it is possible to leak out kernel pointers
from unprivileged user space if the latter is enabled. For example, when such
registers are involved in pointer arithmtics, then adjust_ptr_min_max_vals()
will taint the destination register into an unknown scalar, and the latter
can be exported and stored e.g. into a BPF map value.
Fixes: 3f50f132d8 ("bpf: Verifier, do explicit ALU32 bounds tracking")
Reported-by: Kuee K1r0a <liulin063@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 2fa7d94afc upstream.
The first commit cited below attempts to fix the off-by-one error that
appeared in some comparisons with an open range. Due to this error,
arithmetically equivalent pieces of code could get different verdicts
from the verifier, for example (pseudocode):
// 1. Passes the verifier:
if (data + 8 > data_end)
return early
read *(u64 *)data, i.e. [data; data+7]
// 2. Rejected by the verifier (should still pass):
if (data + 7 >= data_end)
return early
read *(u64 *)data, i.e. [data; data+7]
The attempted fix, however, shifts the range by one in a wrong
direction, so the bug not only remains, but also such piece of code
starts failing in the verifier:
// 3. Rejected by the verifier, but the check is stricter than in #1.
if (data + 8 >= data_end)
return early
read *(u64 *)data, i.e. [data; data+7]
The change performed by that fix converted an off-by-one bug into
off-by-two. The second commit cited below added the BPF selftests
written to ensure than code chunks like #3 are rejected, however,
they should be accepted.
This commit fixes the off-by-two error by adjusting new_range in the
right direction and fixes the tests by changing the range into the
one that should actually fail.
Fixes: fb2a311a31 ("bpf: fix off by one for range markings with L{T, E} patterns")
Fixes: b37242c773 ("bpf: add test cases to bpf selftests to cover all access tests")
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211130181607.593149-1-maximmi@nvidia.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 353050be4c upstream.
Commit a23740ec43 ("bpf: Track contents of read-only maps as scalars") is
checking whether maps are read-only both from BPF program side and user space
side, and then, given their content is constant, reading out their data via
map->ops->map_direct_value_addr() which is then subsequently used as known
scalar value for the register, that is, it is marked as __mark_reg_known()
with the read value at verification time. Before a23740ec43, the register
content was marked as an unknown scalar so the verifier could not make any
assumptions about the map content.
The current implementation however is prone to a TOCTOU race, meaning, the
value read as known scalar for the register is not guaranteed to be exactly
the same at a later point when the program is executed, and as such, the
prior made assumptions of the verifier with regards to the program will be
invalid which can cause issues such as OOB access, etc.
While the BPF_F_RDONLY_PROG map flag is always fixed and required to be
specified at map creation time, the map->frozen property is initially set to
false for the map given the map value needs to be populated, e.g. for global
data sections. Once complete, the loader "freezes" the map from user space
such that no subsequent updates/deletes are possible anymore. For the rest
of the lifetime of the map, this freeze one-time trigger cannot be undone
anymore after a successful BPF_MAP_FREEZE cmd return. Meaning, any new BPF_*
cmd calls which would update/delete map entries will be rejected with -EPERM
since map_get_sys_perms() removes the FMODE_CAN_WRITE permission. This also
means that pending update/delete map entries must still complete before this
guarantee is given. This corner case is not an issue for loaders since they
create and prepare such program private map in successive steps.
However, a malicious user is able to trigger this TOCTOU race in two different
ways: i) via userfaultfd, and ii) via batched updates. For i) userfaultfd is
used to expand the competition interval, so that map_update_elem() can modify
the contents of the map after map_freeze() and bpf_prog_load() were executed.
This works, because userfaultfd halts the parallel thread which triggered a
map_update_elem() at the time where we copy key/value from the user buffer and
this already passed the FMODE_CAN_WRITE capability test given at that time the
map was not "frozen". Then, the main thread performs the map_freeze() and
bpf_prog_load(), and once that had completed successfully, the other thread
is woken up to complete the pending map_update_elem() which then changes the
map content. For ii) the idea of the batched update is similar, meaning, when
there are a large number of updates to be processed, it can increase the
competition interval between the two. It is therefore possible in practice to
modify the contents of the map after executing map_freeze() and bpf_prog_load().
One way to fix both i) and ii) at the same time is to expand the use of the
map's map->writecnt. The latter was introduced in fc9702273e ("bpf: Add mmap()
support for BPF_MAP_TYPE_ARRAY") and further refined in 1f6cb19be2 ("bpf:
Prevent re-mmap()'ing BPF map as writable for initially r/o mapping") with
the rationale to make a writable mmap()'ing of a map mutually exclusive with
read-only freezing. The counter indicates writable mmap() mappings and then
prevents/fails the freeze operation. Its semantics can be expanded beyond
just mmap() by generally indicating ongoing write phases. This would essentially
span any parallel regular and batched flavor of update/delete operation and
then also have map_freeze() fail with -EBUSY. For the check_mem_access() in
the verifier we expand upon the bpf_map_is_rdonly() check ensuring that all
last pending writes have completed via bpf_map_write_active() test. Once the
map->frozen is set and bpf_map_write_active() indicates a map->writecnt of 0
only then we are really guaranteed to use the map's data as known constants.
For map->frozen being set and pending writes in process of still being completed
we fall back to marking that register as unknown scalar so we don't end up
making assumptions about it. With this, both TOCTOU reproducers from i) and
ii) are fixed.
Note that the map->writecnt has been converted into a atomic64 in the fix in
order to avoid a double freeze_mutex mutex_{un,}lock() pair when updating
map->writecnt in the various map update/delete BPF_* cmd flavors. Spanning
the freeze_mutex over entire map update/delete operations in syscall side
would not be possible due to then causing everything to be serialized.
Similarly, something like synchronize_rcu() after setting map->frozen to wait
for update/deletes to complete is not possible either since it would also
have to span the user copy which can sleep. On the libbpf side, this won't
break d66562fba1 ("libbpf: Add BPF object skeleton support") as the
anonymous mmap()-ed "map initialization image" is remapped as a BPF map-backed
mmap()-ed memory where for .rodata it's non-writable.
Fixes: a23740ec43 ("bpf: Track contents of read-only maps as scalars")
Reported-by: w1tcher.bupt@gmail.com
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
[fix conflict to call bpf_map_write_active_dec() in err_put block.
fix conflict to insert new functions after find_and_alloc_map().]
Reference: CVE-2021-4001
Signed-off-by: Masami Ichikawa(CIP) <masami.ichikawa@cybertrust.co.jp>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit b9979db834 ]
Before this fix:
166: (b5) if r2 <= 0x1 goto pc+22
from 166 to 189: R2=invP(id=1,umax_value=1,var_off=(0x0; 0xffffffff))
After this fix:
166: (b5) if r2 <= 0x1 goto pc+22
from 166 to 189: R2=invP(id=1,umax_value=1,var_off=(0x0; 0x1))
While processing BPF_JLE the reg_set_min_max() would set true_reg->umax_value = 1
and call __reg_combine_64_into_32(true_reg).
Without the fix it would not pass the condition:
if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value))
since umin_value == 0 at this point.
Before commit 10bf4e8316 the umin was incorrectly ingored.
The commit 10bf4e8316 fixed the correctness issue, but pessimized
propagation of 64-bit min max into 32-bit min max and corresponding var_off.
Fixes: 10bf4e8316 ("bpf: Fix propagation of 32 bit unsigned bounds from 64 bit bounds")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211101222153.78759-1-alexei.starovoitov@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit d7af7e497f ]
Fix a verifier bug found by smatch static checker in [0].
This problem has never been seen in prod to my best knowledge. Fixing it
still seems to be a good idea since it's hard to say for sure whether
it's possible or not to have a scenario where a combination of
convert_ctx_access() and a narrow load would lead to an out of bound
write.
When narrow load is handled, one or two new instructions are added to
insn_buf array, but before it was only checked that
cnt >= ARRAY_SIZE(insn_buf)
And it's safe to add a new instruction to insn_buf[cnt++] only once. The
second try will lead to out of bound write. And this is what can happen
if `shift` is set.
Fix it by making sure that if the BPF_RSH instruction has to be added in
addition to BPF_AND then there is enough space for two more instructions
in insn_buf.
The full report [0] is below:
kernel/bpf/verifier.c:12304 convert_ctx_accesses() warn: offset 'cnt' incremented past end of array
kernel/bpf/verifier.c:12311 convert_ctx_accesses() warn: offset 'cnt' incremented past end of array
kernel/bpf/verifier.c
12282
12283 insn->off = off & ~(size_default - 1);
12284 insn->code = BPF_LDX | BPF_MEM | size_code;
12285 }
12286
12287 target_size = 0;
12288 cnt = convert_ctx_access(type, insn, insn_buf, env->prog,
12289 &target_size);
12290 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) ||
^^^^^^^^^^^^^^^^^^^^^^^^^^^
Bounds check.
12291 (ctx_field_size && !target_size)) {
12292 verbose(env, "bpf verifier is misconfigured\n");
12293 return -EINVAL;
12294 }
12295
12296 if (is_narrower_load && size < target_size) {
12297 u8 shift = bpf_ctx_narrow_access_offset(
12298 off, size, size_default) * 8;
12299 if (ctx_field_size <= 4) {
12300 if (shift)
12301 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH,
^^^^^
increment beyond end of array
12302 insn->dst_reg,
12303 shift);
--> 12304 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg,
^^^^^
out of bounds write
12305 (1 << size * 8) - 1);
12306 } else {
12307 if (shift)
12308 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH,
12309 insn->dst_reg,
12310 shift);
12311 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg,
^^^^^^^^^^^^^^^
Same.
12312 (1ULL << size * 8) - 1);
12313 }
12314 }
12315
12316 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
12317 if (!new_prog)
12318 return -ENOMEM;
12319
12320 delta += cnt - 1;
12321
12322 /* keep walking new program and skip insns we just inserted */
12323 env->prog = new_prog;
12324 insn = new_prog->insnsi + i + delta;
12325 }
12326
12327 return 0;
12328 }
[0] https://lore.kernel.org/bpf/20210817050843.GA21456@kili/
v1->v2:
- clarify that problem was only seen by static checker but not in prod;
Fixes: 46f53a65d2 ("bpf: Allow narrow loads with offset > 0")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrey Ignatov <rdna@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210820163935.1902398-1-rdna@fb.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 75f0fc7b48 ]
In bpf_patch_insn_data(), we first use the bpf_patch_insn_single() to
insert new instructions, then use adjust_insn_aux_data() to adjust
insn_aux_data. If the old env->prog have no enough room for new inserted
instructions, we use bpf_prog_realloc to construct new_prog and free the
old env->prog.
There have two errors here. First, if adjust_insn_aux_data() return
ENOMEM, we should free the new_prog. Second, if adjust_insn_aux_data()
return ENOMEM, bpf_patch_insn_data() will return NULL, and env->prog has
been freed in bpf_prog_realloc, but we will use it in bpf_check().
So in this patch, we make the adjust_insn_aux_data() never fails. In
bpf_patch_insn_data(), we first pre-malloc memory for the new
insn_aux_data, then call bpf_patch_insn_single() to insert new
instructions, at last call adjust_insn_aux_data() to adjust
insn_aux_data.
Fixes: 8041902dae ("bpf: adjust insn_aux_data when patching insns")
Signed-off-by: He Fengqing <hefengqing@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20210714101815.164322-1-hefengqing@huawei.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 5b029a32cf upstream.
Commit 457f44363a ("bpf: Implement BPF ring buffer and verifier support
for it") extended check_map_func_compatibility() by enforcing map -> helper
function match, but not helper -> map type match.
Due to this all of the bpf_ringbuf_*() helper functions could be used with
a wrong map type such as array or hash map, leading to invalid access due
to type confusion.
Also, both BPF_FUNC_ringbuf_{submit,discard} have ARG_PTR_TO_ALLOC_MEM as
argument and not a BPF map. Therefore, their check_map_func_compatibility()
presence is incorrect since it's only for map type checking.
Fixes: 457f44363a ("bpf: Implement BPF ring buffer and verifier support for it")
Reported-by: Ryota Shiga (Flatt Security)
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 45c709f8c7 ]
"access skb fields ok" verifier test fails on s390 with the "verifier
bug. zext_dst is set, but no reg is defined" message. The first insns
of the test prog are ...
0: 61 01 00 00 00 00 00 00 ldxw %r0,[%r1+0]
8: 35 00 00 01 00 00 00 00 jge %r0,0,1
10: 61 01 00 08 00 00 00 00 ldxw %r0,[%r1+8]
... and the 3rd one is dead (this does not look intentional to me, but
this is a separate topic).
sanitize_dead_code() converts dead insns into "ja -1", but keeps
zext_dst. When opt_subreg_zext_lo32_rnd_hi32() tries to parse such
an insn, it sees this discrepancy and bails. This problem can be seen
only with JITs whose bpf_jit_needs_zext() returns true.
Fix by clearning dead insns' zext_dst.
The commits that contributed to this problem are:
1. 5aa5bd14c5 ("bpf: add initial suite for selftests"), which
introduced the test with the dead code.
2. 5327ed3d44 ("bpf: verifier: mark verified-insn with
sub-register zext flag"), which introduced the zext_dst flag.
3. 83a2881903 ("bpf: Account for BPF_FETCH in
insn_has_def32()"), which introduced the sanity check.
4. 9183671af6 ("bpf: Fix leakage under speculation on
mispredicted branches"), which bisect points to.
It's best to fix this on stable branches that contain the second one,
since that's the point where the inconsistency was introduced.
Fixes: 5327ed3d44 ("bpf: verifier: mark verified-insn with sub-register zext flag")
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210812151811.184086-2-iii@linux.ibm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit e042aa532c upstream.
In 7fedb63a83 ("bpf: Tighten speculative pointer arithmetic mask") we
narrowed the offset mask for unprivileged pointer arithmetic in order to
mitigate a corner case where in the speculative domain it is possible to
advance, for example, the map value pointer by up to value_size-1 out-of-
bounds in order to leak kernel memory via side-channel to user space.
The verifier's state pruning for scalars leaves one corner case open
where in the first verification path R_x holds an unknown scalar with an
aux->alu_limit of e.g. 7, and in a second verification path that same
register R_x, here denoted as R_x', holds an unknown scalar which has
tighter bounds and would thus satisfy range_within(R_x, R_x') as well as
tnum_in(R_x, R_x') for state pruning, yielding an aux->alu_limit of 3:
Given the second path fits the register constraints for pruning, the final
generated mask from aux->alu_limit will remain at 7. While technically
not wrong for the non-speculative domain, it would however be possible
to craft similar cases where the mask would be too wide as in 7fedb63a83.
One way to fix it is to detect the presence of unknown scalar map pointer
arithmetic and force a deeper search on unknown scalars to ensure that
we do not run into a masking mismatch.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 59089a189e upstream.
Follow-up to fe9a5ca7e3 ("bpf: Do not mark insn as seen under speculative
path verification"). The sanitize_insn_aux_data() helper does not serve a
particular purpose in today's code. The original intention for the helper
was that if function-by-function verification fails, a given program would
be cleared from temporary insn_aux_data[], and then its verification would
be re-attempted in the context of the main program a second time.
However, a failure in do_check_subprogs() will skip do_check_main() and
propagate the error to the user instead, thus such situation can never occur.
Given its interaction is not compatible to the Spectre v1 mitigation (due to
comparing aux->seen with env->pass_cnt), just remove sanitize_insn_aux_data()
to avoid future bugs in this area.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 2039f26f3a ]
Spectre v4 gadgets make use of memory disambiguation, which is a set of
techniques that execute memory access instructions, that is, loads and
stores, out of program order; Intel's optimization manual, section 2.4.4.5:
A load instruction micro-op may depend on a preceding store. Many
microarchitectures block loads until all preceding store addresses are
known. The memory disambiguator predicts which loads will not depend on
any previous stores. When the disambiguator predicts that a load does
not have such a dependency, the load takes its data from the L1 data
cache. Eventually, the prediction is verified. If an actual conflict is
detected, the load and all succeeding instructions are re-executed.
af86ca4e30 ("bpf: Prevent memory disambiguation attack") tried to mitigate
this attack by sanitizing the memory locations through preemptive "fast"
(low latency) stores of zero prior to the actual "slow" (high latency) store
of a pointer value such that upon dependency misprediction the CPU then
speculatively executes the load of the pointer value and retrieves the zero
value instead of the attacker controlled scalar value previously stored at
that location, meaning, subsequent access in the speculative domain is then
redirected to the "zero page".
The sanitized preemptive store of zero prior to the actual "slow" store is
done through a simple ST instruction based on r10 (frame pointer) with
relative offset to the stack location that the verifier has been tracking
on the original used register for STX, which does not have to be r10. Thus,
there are no memory dependencies for this store, since it's only using r10
and immediate constant of zero; hence af86ca4e30 /assumed/ a low latency
operation.
However, a recent attack demonstrated that this mitigation is not sufficient
since the preemptive store of zero could also be turned into a "slow" store
and is thus bypassed as well:
[...]
// r2 = oob address (e.g. scalar)
// r7 = pointer to map value
31: (7b) *(u64 *)(r10 -16) = r2
// r9 will remain "fast" register, r10 will become "slow" register below
32: (bf) r9 = r10
// JIT maps BPF reg to x86 reg:
// r9 -> r15 (callee saved)
// r10 -> rbp
// train store forward prediction to break dependency link between both r9
// and r10 by evicting them from the predictor's LRU table.
33: (61) r0 = *(u32 *)(r7 +24576)
34: (63) *(u32 *)(r7 +29696) = r0
35: (61) r0 = *(u32 *)(r7 +24580)
36: (63) *(u32 *)(r7 +29700) = r0
37: (61) r0 = *(u32 *)(r7 +24584)
38: (63) *(u32 *)(r7 +29704) = r0
39: (61) r0 = *(u32 *)(r7 +24588)
40: (63) *(u32 *)(r7 +29708) = r0
[...]
543: (61) r0 = *(u32 *)(r7 +25596)
544: (63) *(u32 *)(r7 +30716) = r0
// prepare call to bpf_ringbuf_output() helper. the latter will cause rbp
// to spill to stack memory while r13/r14/r15 (all callee saved regs) remain
// in hardware registers. rbp becomes slow due to push/pop latency. below is
// disasm of bpf_ringbuf_output() helper for better visual context:
//
// ffffffff8117ee20: 41 54 push r12
// ffffffff8117ee22: 55 push rbp
// ffffffff8117ee23: 53 push rbx
// ffffffff8117ee24: 48 f7 c1 fc ff ff ff test rcx,0xfffffffffffffffc
// ffffffff8117ee2b: 0f 85 af 00 00 00 jne ffffffff8117eee0 <-- jump taken
// [...]
// ffffffff8117eee0: 49 c7 c4 ea ff ff ff mov r12,0xffffffffffffffea
// ffffffff8117eee7: 5b pop rbx
// ffffffff8117eee8: 5d pop rbp
// ffffffff8117eee9: 4c 89 e0 mov rax,r12
// ffffffff8117eeec: 41 5c pop r12
// ffffffff8117eeee: c3 ret
545: (18) r1 = map[id:4]
547: (bf) r2 = r7
548: (b7) r3 = 0
549: (b7) r4 = 4
550: (85) call bpf_ringbuf_output#194288
// instruction 551 inserted by verifier \
551: (7a) *(u64 *)(r10 -16) = 0 | /both/ are now slow stores here
// storing map value pointer r7 at fp-16 | since value of r10 is "slow".
552: (7b) *(u64 *)(r10 -16) = r7 /
// following "fast" read to the same memory location, but due to dependency
// misprediction it will speculatively execute before insn 551/552 completes.
553: (79) r2 = *(u64 *)(r9 -16)
// in speculative domain contains attacker controlled r2. in non-speculative
// domain this contains r7, and thus accesses r7 +0 below.
554: (71) r3 = *(u8 *)(r2 +0)
// leak r3
As can be seen, the current speculative store bypass mitigation which the
verifier inserts at line 551 is insufficient since /both/, the write of
the zero sanitation as well as the map value pointer are a high latency
instruction due to prior memory access via push/pop of r10 (rbp) in contrast
to the low latency read in line 553 as r9 (r15) which stays in hardware
registers. Thus, architecturally, fp-16 is r7, however, microarchitecturally,
fp-16 can still be r2.
Initial thoughts to address this issue was to track spilled pointer loads
from stack and enforce their load via LDX through r10 as well so that /both/
the preemptive store of zero /as well as/ the load use the /same/ register
such that a dependency is created between the store and load. However, this
option is not sufficient either since it can be bypassed as well under
speculation. An updated attack with pointer spill/fills now _all_ based on
r10 would look as follows:
[...]
// r2 = oob address (e.g. scalar)
// r7 = pointer to map value
[...]
// longer store forward prediction training sequence than before.
2062: (61) r0 = *(u32 *)(r7 +25588)
2063: (63) *(u32 *)(r7 +30708) = r0
2064: (61) r0 = *(u32 *)(r7 +25592)
2065: (63) *(u32 *)(r7 +30712) = r0
2066: (61) r0 = *(u32 *)(r7 +25596)
2067: (63) *(u32 *)(r7 +30716) = r0
// store the speculative load address (scalar) this time after the store
// forward prediction training.
2068: (7b) *(u64 *)(r10 -16) = r2
// preoccupy the CPU store port by running sequence of dummy stores.
2069: (63) *(u32 *)(r7 +29696) = r0
2070: (63) *(u32 *)(r7 +29700) = r0
2071: (63) *(u32 *)(r7 +29704) = r0
2072: (63) *(u32 *)(r7 +29708) = r0
2073: (63) *(u32 *)(r7 +29712) = r0
2074: (63) *(u32 *)(r7 +29716) = r0
2075: (63) *(u32 *)(r7 +29720) = r0
2076: (63) *(u32 *)(r7 +29724) = r0
2077: (63) *(u32 *)(r7 +29728) = r0
2078: (63) *(u32 *)(r7 +29732) = r0
2079: (63) *(u32 *)(r7 +29736) = r0
2080: (63) *(u32 *)(r7 +29740) = r0
2081: (63) *(u32 *)(r7 +29744) = r0
2082: (63) *(u32 *)(r7 +29748) = r0
2083: (63) *(u32 *)(r7 +29752) = r0
2084: (63) *(u32 *)(r7 +29756) = r0
2085: (63) *(u32 *)(r7 +29760) = r0
2086: (63) *(u32 *)(r7 +29764) = r0
2087: (63) *(u32 *)(r7 +29768) = r0
2088: (63) *(u32 *)(r7 +29772) = r0
2089: (63) *(u32 *)(r7 +29776) = r0
2090: (63) *(u32 *)(r7 +29780) = r0
2091: (63) *(u32 *)(r7 +29784) = r0
2092: (63) *(u32 *)(r7 +29788) = r0
2093: (63) *(u32 *)(r7 +29792) = r0
2094: (63) *(u32 *)(r7 +29796) = r0
2095: (63) *(u32 *)(r7 +29800) = r0
2096: (63) *(u32 *)(r7 +29804) = r0
2097: (63) *(u32 *)(r7 +29808) = r0
2098: (63) *(u32 *)(r7 +29812) = r0
// overwrite scalar with dummy pointer; same as before, also including the
// sanitation store with 0 from the current mitigation by the verifier.
2099: (7a) *(u64 *)(r10 -16) = 0 | /both/ are now slow stores here
2100: (7b) *(u64 *)(r10 -16) = r7 | since store unit is still busy.
// load from stack intended to bypass stores.
2101: (79) r2 = *(u64 *)(r10 -16)
2102: (71) r3 = *(u8 *)(r2 +0)
// leak r3
[...]
Looking at the CPU microarchitecture, the scheduler might issue loads (such
as seen in line 2101) before stores (line 2099,2100) because the load execution
units become available while the store execution unit is still busy with the
sequence of dummy stores (line 2069-2098). And so the load may use the prior
stored scalar from r2 at address r10 -16 for speculation. The updated attack
may work less reliable on CPU microarchitectures where loads and stores share
execution resources.
This concludes that the sanitizing with zero stores from af86ca4e30 ("bpf:
Prevent memory disambiguation attack") is insufficient. Moreover, the detection
of stack reuse from af86ca4e30 where previously data (STACK_MISC) has been
written to a given stack slot where a pointer value is now to be stored does
not have sufficient coverage as precondition for the mitigation either; for
several reasons outlined as follows:
1) Stack content from prior program runs could still be preserved and is
therefore not "random", best example is to split a speculative store
bypass attack between tail calls, program A would prepare and store the
oob address at a given stack slot and then tail call into program B which
does the "slow" store of a pointer to the stack with subsequent "fast"
read. From program B PoV such stack slot type is STACK_INVALID, and
therefore also must be subject to mitigation.
2) The STACK_SPILL must not be coupled to register_is_const(&stack->spilled_ptr)
condition, for example, the previous content of that memory location could
also be a pointer to map or map value. Without the fix, a speculative
store bypass is not mitigated in such precondition and can then lead to
a type confusion in the speculative domain leaking kernel memory near
these pointer types.
While brainstorming on various alternative mitigation possibilities, we also
stumbled upon a retrospective from Chrome developers [0]:
[...] For variant 4, we implemented a mitigation to zero the unused memory
of the heap prior to allocation, which cost about 1% when done concurrently
and 4% for scavenging. Variant 4 defeats everything we could think of. We
explored more mitigations for variant 4 but the threat proved to be more
pervasive and dangerous than we anticipated. For example, stack slots used
by the register allocator in the optimizing compiler could be subject to
type confusion, leading to pointer crafting. Mitigating type confusion for
stack slots alone would have required a complete redesign of the backend of
the optimizing compiler, perhaps man years of work, without a guarantee of
completeness. [...]
From BPF side, the problem space is reduced, however, options are rather
limited. One idea that has been explored was to xor-obfuscate pointer spills
to the BPF stack:
[...]
// preoccupy the CPU store port by running sequence of dummy stores.
[...]
2106: (63) *(u32 *)(r7 +29796) = r0
2107: (63) *(u32 *)(r7 +29800) = r0
2108: (63) *(u32 *)(r7 +29804) = r0
2109: (63) *(u32 *)(r7 +29808) = r0
2110: (63) *(u32 *)(r7 +29812) = r0
// overwrite scalar with dummy pointer; xored with random 'secret' value
// of 943576462 before store ...
2111: (b4) w11 = 943576462
2112: (af) r11 ^= r7
2113: (7b) *(u64 *)(r10 -16) = r11
2114: (79) r11 = *(u64 *)(r10 -16)
2115: (b4) w2 = 943576462
2116: (af) r2 ^= r11
// ... and restored with the same 'secret' value with the help of AX reg.
2117: (71) r3 = *(u8 *)(r2 +0)
[...]
While the above would not prevent speculation, it would make data leakage
infeasible by directing it to random locations. In order to be effective
and prevent type confusion under speculation, such random secret would have
to be regenerated for each store. The additional complexity involved for a
tracking mechanism that prevents jumps such that restoring spilled pointers
would not get corrupted is not worth the gain for unprivileged. Hence, the
fix in here eventually opted for emitting a non-public BPF_ST | BPF_NOSPEC
instruction which the x86 JIT translates into a lfence opcode. Inserting the
latter in between the store and load instruction is one of the mitigations
options [1]. The x86 instruction manual notes:
[...] An LFENCE that follows an instruction that stores to memory might
complete before the data being stored have become globally visible. [...]
The latter meaning that the preceding store instruction finished execution
and the store is at minimum guaranteed to be in the CPU's store queue, but
it's not guaranteed to be in that CPU's L1 cache at that point (globally
visible). The latter would only be guaranteed via sfence. So the load which
is guaranteed to execute after the lfence for that local CPU would have to
rely on store-to-load forwarding. [2], in section 2.3 on store buffers says:
[...] For every store operation that is added to the ROB, an entry is
allocated in the store buffer. This entry requires both the virtual and
physical address of the target. Only if there is no free entry in the store
buffer, the frontend stalls until there is an empty slot available in the
store buffer again. Otherwise, the CPU can immediately continue adding
subsequent instructions to the ROB and execute them out of order. On Intel
CPUs, the store buffer has up to 56 entries. [...]
One small upside on the fix is that it lifts constraints from af86ca4e30
where the sanitize_stack_off relative to r10 must be the same when coming
from different paths. The BPF_ST | BPF_NOSPEC gets emitted after a BPF_STX
or BPF_ST instruction. This happens either when we store a pointer or data
value to the BPF stack for the first time, or upon later pointer spills.
The former needs to be enforced since otherwise stale stack data could be
leaked under speculation as outlined earlier. For non-x86 JITs the BPF_ST |
BPF_NOSPEC mapping is currently optimized away, but others could emit a
speculation barrier as well if necessary. For real-world unprivileged
programs e.g. generated by LLVM, pointer spill/fill is only generated upon
register pressure and LLVM only tries to do that for pointers which are not
used often. The program main impact will be the initial BPF_ST | BPF_NOSPEC
sanitation for the STACK_INVALID case when the first write to a stack slot
occurs e.g. upon map lookup. In future we might refine ways to mitigate
the latter cost.
[0] https://arxiv.org/pdf/1902.05178.pdf
[1] https://msrc-blog.microsoft.com/2018/05/21/analysis-and-mitigation-of-speculative-store-bypass-cve-2018-3639/
[2] https://arxiv.org/pdf/1905.05725.pdf
Fixes: af86ca4e30 ("bpf: Prevent memory disambiguation attack")
Fixes: f7cf25b202 ("bpf: track spill/fill of constants")
Co-developed-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Benedict Schlueter <benedict.schlueter@rub.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Benedict Schlueter <benedict.schlueter@rub.de>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit f263a81451 upstream.
Subprograms are calling map_poke_track(), but on program release there is no
hook to call map_poke_untrack(). However, on program release, the aux memory
(and poke descriptor table) is freed even though we still have a reference to
it in the element list of the map aux data. When we run map_poke_run(), we then
end up accessing free'd memory, triggering KASAN in prog_array_map_poke_run():
[...]
[ 402.824689] BUG: KASAN: use-after-free in prog_array_map_poke_run+0xc2/0x34e
[ 402.824698] Read of size 4 at addr ffff8881905a7940 by task hubble-fgs/4337
[ 402.824705] CPU: 1 PID: 4337 Comm: hubble-fgs Tainted: G I 5.12.0+ #399
[ 402.824715] Call Trace:
[ 402.824719] dump_stack+0x93/0xc2
[ 402.824727] print_address_description.constprop.0+0x1a/0x140
[ 402.824736] ? prog_array_map_poke_run+0xc2/0x34e
[ 402.824740] ? prog_array_map_poke_run+0xc2/0x34e
[ 402.824744] kasan_report.cold+0x7c/0xd8
[ 402.824752] ? prog_array_map_poke_run+0xc2/0x34e
[ 402.824757] prog_array_map_poke_run+0xc2/0x34e
[ 402.824765] bpf_fd_array_map_update_elem+0x124/0x1a0
[...]
The elements concerned are walked as follows:
for (i = 0; i < elem->aux->size_poke_tab; i++) {
poke = &elem->aux->poke_tab[i];
[...]
The access to size_poke_tab is a 4 byte read, verified by checking offsets
in the KASAN dump:
[ 402.825004] The buggy address belongs to the object at ffff8881905a7800
which belongs to the cache kmalloc-1k of size 1024
[ 402.825008] The buggy address is located 320 bytes inside of
1024-byte region [ffff8881905a7800, ffff8881905a7c00)
The pahole output of bpf_prog_aux:
struct bpf_prog_aux {
[...]
/* --- cacheline 5 boundary (320 bytes) --- */
u32 size_poke_tab; /* 320 4 */
[...]
In general, subprograms do not necessarily manage their own data structures.
For example, BTF func_info and linfo are just pointers to the main program
structure. This allows reference counting and cleanup to be done on the latter
which simplifies their management a bit. The aux->poke_tab struct, however,
did not follow this logic. The initial proposed fix for this use-after-free
bug further embedded poke data tracking into the subprogram with proper
reference counting. However, Daniel and Alexei questioned why we were treating
these objects special; I agree, its unnecessary. The fix here removes the per
subprogram poke table allocation and map tracking and instead simply points
the aux->poke_tab pointer at the main programs poke table. This way, map
tracking is simplified to the main program and we do not need to manage them
per subprogram.
This also means, bpf_prog_free_deferred(), which unwinds the program reference
counting and kfrees objects, needs to ensure that we don't try to double free
the poke_tab when free'ing the subprog structures. This is easily solved by
NULL'ing the poke_tab pointer. The second detail is to ensure that per
subprogram JIT logic only does fixups on poke_tab[] entries it owns. To do
this, we add a pointer in the poke structure to point at the subprogram value
so JITs can easily check while walking the poke_tab structure if the current
entry belongs to the current program. The aux pointer is stable and therefore
suitable for such comparison. On the jit_subprogs() error path, we omit
cleaning up the poke->aux field because these are only ever referenced from
the JIT side, but on error we will never make it to the JIT, so its fine to
leave them dangling. Removing these pointers would complicate the error path
for no reason. However, we do need to untrack all poke descriptors from the
main program as otherwise they could race with the freeing of JIT memory from
the subprograms. Lastly, a748c6975d ("bpf: propagate poke descriptors to
subprograms") had an off-by-one on the subprogram instruction index range
check as it was testing 'insn_idx >= subprog_start && insn_idx <= subprog_end'.
However, subprog_end is the next subprogram's start instruction.
Fixes: a748c6975d ("bpf: propagate poke descriptors to subprograms")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210707223848.14580-2-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 7506d211b9 ]
The sub-programs prog->aux->poke_tab[] is populated in jit_subprogs() and
then used when emitting 'BPF_JMP|BPF_TAIL_CALL' insn->code from the
individual JITs. The poke_tab[] to use is stored in the insn->imm by
the code adding it to that array slot. The JIT then uses imm to find the
right entry for an individual instruction. In the x86 bpf_jit_comp.c
this is done by calling emit_bpf_tail_call_direct with the poke_tab[]
of the imm value.
However, we observed the below null-ptr-deref when mixing tail call
programs with subprog programs. For this to happen we just need to
mix bpf-2-bpf calls and tailcalls with some extra calls or instructions
that would be patched later by one of the fixup routines. So whats
happening?
Before the fixup_call_args() -- where the jit op is done -- various
code patching is done by do_misc_fixups(). This may increase the
insn count, for example when we patch map_lookup_up using map_gen_lookup
hook. This does two things. First, it means the instruction index,
insn_idx field, of a tail call instruction will move by a 'delta'.
In verifier code,
struct bpf_jit_poke_descriptor desc = {
.reason = BPF_POKE_REASON_TAIL_CALL,
.tail_call.map = BPF_MAP_PTR(aux->map_ptr_state),
.tail_call.key = bpf_map_key_immediate(aux),
.insn_idx = i + delta,
};
Then subprog start values subprog_info[i].start will be updated
with the delta and any poke descriptor index will also be updated
with the delta in adjust_poke_desc(). If we look at the adjust
subprog starts though we see its only adjusted when the delta
occurs before the new instructions,
/* NOTE: fake 'exit' subprog should be updated as well. */
for (i = 0; i <= env->subprog_cnt; i++) {
if (env->subprog_info[i].start <= off)
continue;
Earlier subprograms are not changed because their start values
are not moved. But, adjust_poke_desc() does the offset + delta
indiscriminately. The result is poke descriptors are potentially
corrupted.
Then in jit_subprogs() we only populate the poke_tab[]
when the above insn_idx is less than the next subprogram start. From
above we corrupted our insn_idx so we might incorrectly assume a
poke descriptor is not used in a subprogram omitting it from the
subprogram. And finally when the jit runs it does the deref of poke_tab
when emitting the instruction and crashes with below. Because earlier
step omitted the poke descriptor.
The fix is straight forward with above context. Simply move same logic
from adjust_subprog_starts() into adjust_poke_descs() and only adjust
insn_idx when needed.
[ 82.396354] bpf_testmod: version magic '5.12.0-rc2alu+ SMP preempt mod_unload ' should be '5.12.0+ SMP preempt mod_unload '
[ 82.623001] loop10: detected capacity change from 0 to 8
[ 88.487424] ==================================================================
[ 88.487438] BUG: KASAN: null-ptr-deref in do_jit+0x184a/0x3290
[ 88.487455] Write of size 8 at addr 0000000000000008 by task test_progs/5295
[ 88.487471] CPU: 7 PID: 5295 Comm: test_progs Tainted: G I 5.12.0+ #386
[ 88.487483] Hardware name: Dell Inc. Precision 5820 Tower/002KVM, BIOS 1.9.2 01/24/2019
[ 88.487490] Call Trace:
[ 88.487498] dump_stack+0x93/0xc2
[ 88.487515] kasan_report.cold+0x5f/0xd8
[ 88.487530] ? do_jit+0x184a/0x3290
[ 88.487542] do_jit+0x184a/0x3290
...
[ 88.487709] bpf_int_jit_compile+0x248/0x810
...
[ 88.487765] bpf_check+0x3718/0x5140
...
[ 88.487920] bpf_prog_load+0xa22/0xf10
Fixes: a748c6975d ("bpf: propagate poke descriptors to subprograms")
Reported-by: Jussi Maki <joamaki@gmail.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit fe9a5ca7e3 ]
... in such circumstances, we do not want to mark the instruction as seen given
the goal is still to jmp-1 rewrite/sanitize dead code, if it is not reachable
from the non-speculative path verification. We do however want to verify it for
safety regardless.
With the patch as-is all the insns that have been marked as seen before the
patch will also be marked as seen after the patch (just with a potentially
different non-zero count). An upcoming patch will also verify paths that are
unreachable in the non-speculative domain, hence this extension is needed.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Reviewed-by: Benedict Schlueter <benedict.schlueter@rub.de>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit d203b0fd86 ]
Instead of relying on current env->pass_cnt, use the seen count from the
old aux data in adjust_insn_aux_data(), and expand it to the new range of
patched instructions. This change is valid given we always expand 1:n
with n>=1, so what applies to the old/original instruction needs to apply
for the replacement as well.
Not relying on env->pass_cnt is a prerequisite for a later change where we
want to avoid marking an instruction seen when verified under speculative
execution path.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Reviewed-by: Benedict Schlueter <benedict.schlueter@rub.de>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 9183671af6 ]
The verifier only enumerates valid control-flow paths and skips paths that
are unreachable in the non-speculative domain. And so it can miss issues
under speculative execution on mispredicted branches.
For example, a type confusion has been demonstrated with the following
crafted program:
// r0 = pointer to a map array entry
// r6 = pointer to readable stack slot
// r9 = scalar controlled by attacker
1: r0 = *(u64 *)(r0) // cache miss
2: if r0 != 0x0 goto line 4
3: r6 = r9
4: if r0 != 0x1 goto line 6
5: r9 = *(u8 *)(r6)
6: // leak r9
Since line 3 runs iff r0 == 0 and line 5 runs iff r0 == 1, the verifier
concludes that the pointer dereference on line 5 is safe. But: if the
attacker trains both the branches to fall-through, such that the following
is speculatively executed ...
r6 = r9
r9 = *(u8 *)(r6)
// leak r9
... then the program will dereference an attacker-controlled value and could
leak its content under speculative execution via side-channel. This requires
to mistrain the branch predictor, which can be rather tricky, because the
branches are mutually exclusive. However such training can be done at
congruent addresses in user space using different branches that are not
mutually exclusive. That is, by training branches in user space ...
A: if r0 != 0x0 goto line C
B: ...
C: if r0 != 0x0 goto line D
D: ...
... such that addresses A and C collide to the same CPU branch prediction
entries in the PHT (pattern history table) as those of the BPF program's
lines 2 and 4, respectively. A non-privileged attacker could simply brute
force such collisions in the PHT until observing the attack succeeding.
Alternative methods to mistrain the branch predictor are also possible that
avoid brute forcing the collisions in the PHT. A reliable attack has been
demonstrated, for example, using the following crafted program:
// r0 = pointer to a [control] map array entry
// r7 = *(u64 *)(r0 + 0), training/attack phase
// r8 = *(u64 *)(r0 + 8), oob address
// [...]
// r0 = pointer to a [data] map array entry
1: if r7 == 0x3 goto line 3
2: r8 = r0
// crafted sequence of conditional jumps to separate the conditional
// branch in line 193 from the current execution flow
3: if r0 != 0x0 goto line 5
4: if r0 == 0x0 goto exit
5: if r0 != 0x0 goto line 7
6: if r0 == 0x0 goto exit
[...]
187: if r0 != 0x0 goto line 189
188: if r0 == 0x0 goto exit
// load any slowly-loaded value (due to cache miss in phase 3) ...
189: r3 = *(u64 *)(r0 + 0x1200)
// ... and turn it into known zero for verifier, while preserving slowly-
// loaded dependency when executing:
190: r3 &= 1
191: r3 &= 2
// speculatively bypassed phase dependency
192: r7 += r3
193: if r7 == 0x3 goto exit
194: r4 = *(u8 *)(r8 + 0)
// leak r4
As can be seen, in training phase (phase != 0x3), the condition in line 1
turns into false and therefore r8 with the oob address is overridden with
the valid map value address, which in line 194 we can read out without
issues. However, in attack phase, line 2 is skipped, and due to the cache
miss in line 189 where the map value is (zeroed and later) added to the
phase register, the condition in line 193 takes the fall-through path due
to prior branch predictor training, where under speculation, it'll load the
byte at oob address r8 (unknown scalar type at that point) which could then
be leaked via side-channel.
One way to mitigate these is to 'branch off' an unreachable path, meaning,
the current verification path keeps following the is_branch_taken() path
and we push the other branch to the verification stack. Given this is
unreachable from the non-speculative domain, this branch's vstate is
explicitly marked as speculative. This is needed for two reasons: i) if
this path is solely seen from speculative execution, then we later on still
want the dead code elimination to kick in in order to sanitize these
instructions with jmp-1s, and ii) to ensure that paths walked in the
non-speculative domain are not pruned from earlier walks of paths walked in
the speculative domain. Additionally, for robustness, we mark the registers
which have been part of the conditional as unknown in the speculative path
given there should be no assumptions made on their content.
The fix in here mitigates type confusion attacks described earlier due to
i) all code paths in the BPF program being explored and ii) existing
verifier logic already ensuring that given memory access instruction
references one specific data structure.
An alternative to this fix that has also been looked at in this scope was to
mark aux->alu_state at the jump instruction with a BPF_JMP_TAKEN state as
well as direction encoding (always-goto, always-fallthrough, unknown), such
that mixing of different always-* directions themselves as well as mixing of
always-* with unknown directions would cause a program rejection by the
verifier, e.g. programs with constructs like 'if ([...]) { x = 0; } else
{ x = 1; }' with subsequent 'if (x == 1) { [...] }'. For unprivileged, this
would result in only single direction always-* taken paths, and unknown taken
paths being allowed, such that the former could be patched from a conditional
jump to an unconditional jump (ja). Compared to this approach here, it would
have two downsides: i) valid programs that otherwise are not performing any
pointer arithmetic, etc, would potentially be rejected/broken, and ii) we are
required to turn off path pruning for unprivileged, where both can be avoided
in this work through pushing the invalid branch to the verification stack.
The issue was originally discovered by Adam and Ofek, and later independently
discovered and reported as a result of Benedict and Piotr's research work.
Fixes: b2157399cc ("bpf: prevent out-of-bounds speculation")
Reported-by: Adam Morrison <mad@cs.tau.ac.il>
Reported-by: Ofek Kirzner <ofekkir@gmail.com>
Reported-by: Benedict Schlueter <benedict.schlueter@rub.de>
Reported-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Reviewed-by: Benedict Schlueter <benedict.schlueter@rub.de>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit a703619127 upstream.
In 801c6058d1 ("bpf: Fix leakage of uninitialized bpf stack under
speculation") we replaced masking logic with direct loads of immediates
if the register is a known constant. Given in this case we do not apply
any masking, there is also no reason for the operation to be truncated
under the speculative domain.
Therefore, there is also zero reason for the verifier to branch-off and
simulate this case, it only needs to do it for unknown but bounded scalars.
As a side-effect, this also enables few test cases that were previously
rejected due to simulation under zero truncation.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit bb01a1bba5 upstream.
Masking direction as indicated via mask_to_left is considered to be
calculated once and then used to derive pointer limits. Thus, this
needs to be placed into bpf_sanitize_info instead so we can pass it
to sanitize_ptr_alu() call after the pointer move. Piotr noticed a
corner case where the off reg causes masking direction change which
then results in an incorrect final aux->alu_limit.
Fixes: 7fedb63a83 ("bpf: Tighten speculative pointer arithmetic mask")
Reported-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3d0220f686 upstream.
Add a container structure struct bpf_sanitize_info which holds
the current aux info, and update call-sites to sanitize_ptr_alu()
to pass it in. This is needed for passing in additional state
later on.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 049c4e1371 upstream.
Fix a bug in the verifier's scalar32_min_max_*() functions which leads to
incorrect tracking of 32 bit bounds for the simulation of and/or/xor bitops.
When both the src & dst subreg is a known constant, then the assumption is
that scalar_min_max_*() will take care to update bounds correctly. However,
this is not the case, for example, consider a register R2 which has a tnum
of 0xffffffff00000000, meaning, lower 32 bits are known constant and in this
case of value 0x00000001. R2 is then and'ed with a register R3 which is a
64 bit known constant, here, 0x100000002.
What can be seen in line '10:' is that 32 bit bounds reach an invalid state
where {u,s}32_min_value > {u,s}32_max_value. The reason is scalar32_min_max_*()
delegates 32 bit bounds updates to scalar_min_max_*(), however, that really
only takes place when both the 64 bit src & dst register is a known constant.
Given scalar32_min_max_*() is intended to be designed as closely as possible
to scalar_min_max_*(), update the 32 bit bounds in this situation through
__mark_reg32_known() which will set all {u,s}32_{min,max}_value to the correct
constant, which is 0x00000000 after the fix (given 0x00000001 & 0x00000002 in
32 bit space). This is possible given var32_off already holds the final value
as dst_reg->var_off is updated before calling scalar32_min_max_*().
Before fix, invalid tracking of R2:
[...]
9: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=-9223372036854775807 (0x8000000000000001),smax_value=9223372032559808513 (0x7fffffff00000001),umin_value=1,umax_value=0xffffffff00000001,var_off=(0x1; 0xffffffff00000000),s32_min_value=1,s32_max_value=1,u32_min_value=1,u32_max_value=1) R3_w=inv4294967298 R10=fp0
9: (5f) r2 &= r3
10: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=0,smax_value=4294967296 (0x100000000),umin_value=0,umax_value=0x100000000,var_off=(0x0; 0x100000000),s32_min_value=1,s32_max_value=0,u32_min_value=1,u32_max_value=0) R3_w=inv4294967298 R10=fp0
[...]
After fix, correct tracking of R2:
[...]
9: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=-9223372036854775807 (0x8000000000000001),smax_value=9223372032559808513 (0x7fffffff00000001),umin_value=1,umax_value=0xffffffff00000001,var_off=(0x1; 0xffffffff00000000),s32_min_value=1,s32_max_value=1,u32_min_value=1,u32_max_value=1) R3_w=inv4294967298 R10=fp0
9: (5f) r2 &= r3
10: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=0,smax_value=4294967296 (0x100000000),umin_value=0,umax_value=0x100000000,var_off=(0x0; 0x100000000),s32_min_value=0,s32_max_value=0,u32_min_value=0,u32_max_value=0) R3_w=inv4294967298 R10=fp0
[...]
Fixes: 3f50f132d8 ("bpf: Verifier, do explicit ALU32 bounds tracking")
Fixes: 2921c90d47 ("bpf: Fix a verifier failure with xor")
Reported-by: Manfred Paul (@_manfp)
Reported-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 10bf4e8316 ]
Similarly as b02709587e ("bpf: Fix propagation of 32-bit signed bounds
from 64-bit bounds."), we also need to fix the propagation of 32 bit
unsigned bounds from 64 bit counterparts. That is, really only set the
u32_{min,max}_value when /both/ {umin,umax}_value safely fit in 32 bit
space. For example, the register with a umin_value == 1 does /not/ imply
that u32_min_value is also equal to 1, since umax_value could be much
larger than 32 bit subregister can hold, and thus u32_min_value is in
the interval [0,1] instead.
Before fix, invalid tracking result of R2_w=inv1:
[...]
5: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0) R10=fp0
5: (35) if r2 >= 0x1 goto pc+1
[...] // goto path
7: R0=inv1337 R1=ctx(id=0,off=0,imm=0) R2=inv(id=0,umin_value=1) R10=fp0
7: (b6) if w2 <= 0x1 goto pc+1
[...] // goto path
9: R0=inv1337 R1=ctx(id=0,off=0,imm=0) R2=inv(id=0,smin_value=-9223372036854775807,smax_value=9223372032559808513,umin_value=1,umax_value=18446744069414584321,var_off=(0x1; 0xffffffff00000000),s32_min_value=1,s32_max_value=1,u32_max_value=1) R10=fp0
9: (bc) w2 = w2
10: R0=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv1 R10=fp0
[...]
After fix, correct tracking result of R2_w=inv(id=0,umax_value=1,var_off=(0x0; 0x1)):
[...]
5: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0) R10=fp0
5: (35) if r2 >= 0x1 goto pc+1
[...] // goto path
7: R0=inv1337 R1=ctx(id=0,off=0,imm=0) R2=inv(id=0,umin_value=1) R10=fp0
7: (b6) if w2 <= 0x1 goto pc+1
[...] // goto path
9: R0=inv1337 R1=ctx(id=0,off=0,imm=0) R2=inv(id=0,smax_value=9223372032559808513,umax_value=18446744069414584321,var_off=(0x0; 0xffffffff00000001),s32_min_value=0,s32_max_value=1,u32_max_value=1) R10=fp0
9: (bc) w2 = w2
10: R0=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,umax_value=1,var_off=(0x0; 0x1)) R10=fp0
[...]
Thus, same issue as in b02709587e holds for unsigned subregister tracking.
Also, align __reg64_bound_u32() similarly to __reg64_bound_s32() as done in
b02709587e to make them uniform again.
Fixes: 3f50f132d8 ("bpf: Verifier, do explicit ALU32 bounds tracking")
Reported-by: Manfred Paul (@_manfp)
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 801c6058d1 upstream.
The current implemented mechanisms to mitigate data disclosure under
speculation mainly address stack and map value oob access from the
speculative domain. However, Piotr discovered that uninitialized BPF
stack is not protected yet, and thus old data from the kernel stack,
potentially including addresses of kernel structures, could still be
extracted from that 512 bytes large window. The BPF stack is special
compared to map values since it's not zero initialized for every
program invocation, whereas map values /are/ zero initialized upon
their initial allocation and thus cannot leak any prior data in either
domain. In the non-speculative domain, the verifier ensures that every
stack slot read must have a prior stack slot write by the BPF program
to avoid such data leaking issue.
However, this is not enough: for example, when the pointer arithmetic
operation moves the stack pointer from the last valid stack offset to
the first valid offset, the sanitation logic allows for any intermediate
offsets during speculative execution, which could then be used to
extract any restricted stack content via side-channel.
Given for unprivileged stack pointer arithmetic the use of unknown
but bounded scalars is generally forbidden, we can simply turn the
register-based arithmetic operation into an immediate-based arithmetic
operation without the need for masking. This also gives the benefit
of reducing the needed instructions for the operation. Given after
the work in 7fedb63a83 ("bpf: Tighten speculative pointer arithmetic
mask"), the aux->alu_limit already holds the final immediate value for
the offset register with the known scalar. Thus, a simple mov of the
immediate to AX register with using AX as the source for the original
instruction is sufficient and possible now in this case.
Reported-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit b9b34ddbe2 upstream.
The negation logic for the case where the off_reg is sitting in the
dst register is not correct given then we cannot just invert the add
to a sub or vice versa. As a fix, perform the final bitwise and-op
unconditionally into AX from the off_reg, then move the pointer from
the src to dst and finally use AX as the source for the original
pointer arithmetic operation such that the inversion yields a correct
result. The single non-AX mov in between is possible given constant
blinding is retaining it as it's not an immediate based operation.
Fixes: 979d63d50c ("bpf: prevent out of bounds speculation on pointer arithmetic")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 7fedb63a83 ]
This work tightens the offset mask we use for unprivileged pointer arithmetic
in order to mitigate a corner case reported by Piotr and Benedict where in
the speculative domain it is possible to advance, for example, the map value
pointer by up to value_size-1 out-of-bounds in order to leak kernel memory
via side-channel to user space.
Before this change, the computed ptr_limit for retrieve_ptr_limit() helper
represents largest valid distance when moving pointer to the right or left
which is then fed as aux->alu_limit to generate masking instructions against
the offset register. After the change, the derived aux->alu_limit represents
the largest potential value of the offset register which we mask against which
is just a narrower subset of the former limit.
For minimal complexity, we call sanitize_ptr_alu() from 2 observation points
in adjust_ptr_min_max_vals(), that is, before and after the simulated alu
operation. In the first step, we retieve the alu_state and alu_limit before
the operation as well as we branch-off a verifier path and push it to the
verification stack as we did before which checks the dst_reg under truncation,
in other words, when the speculative domain would attempt to move the pointer
out-of-bounds.
In the second step, we retrieve the new alu_limit and calculate the absolute
distance between both. Moreover, we commit the alu_state and final alu_limit
via update_alu_sanitation_state() to the env's instruction aux data, and bail
out from there if there is a mismatch due to coming from different verification
paths with different states.
Reported-by: Piotr Krysiuk <piotras@gmail.com>
Reported-by: Benedict Schlueter <benedict.schlueter@rub.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Benedict Schlueter <benedict.schlueter@rub.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 073815b756 ]
Move the bounds check in adjust_ptr_min_max_vals() into a small helper named
sanitize_check_bounds() in order to simplify the former a bit.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 01f810ace9 ]
Before this patch, variable offset access to the stack was dissalowed
for regular instructions, but was allowed for "indirect" accesses (i.e.
helpers). This patch removes the restriction, allowing reading and
writing to the stack through stack pointers with variable offsets. This
makes stack-allocated buffers more usable in programs, and brings stack
pointers closer to other types of pointers.
The motivation is being able to use stack-allocated buffers for data
manipulation. When the stack size limit is sufficient, allocating
buffers on the stack is simpler than per-cpu arrays, or other
alternatives.
In unpriviledged programs, variable-offset reads and writes are
disallowed (they were already disallowed for the indirect access case)
because the speculative execution checking code doesn't support them.
Additionally, when writing through a variable-offset stack pointer, if
any pointers are in the accessible range, there's possilibities of later
leaking pointers because the write cannot be tracked precisely.
Writes with variable offset mark the whole range as initialized, even
though we don't know which stack slots are actually written. This is in
order to not reject future reads to these slots. Note that this doesn't
affect writes done through helpers; like before, helpers need the whole
stack range to be initialized to begin with.
All the stack slots are in range are considered scalars after the write;
variable-offset register spills are not tracked.
For reads, all the stack slots in the variable range needs to be
initialized (but see above about what writes do), otherwise the read is
rejected. All register spilled in stack slots that might be read are
marked as having been read, however reads through such pointers don't do
register filling; the target register will always be either a scalar or
a constant zero.
Signed-off-by: Andrei Matei <andreimatei1@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210207011027.676572-2-andreimatei1@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit f528819334 upstream.
Add a small sanitize_needed() helper function and move sanitize_val_alu()
out of the main opcode switch. In upcoming work, we'll move sanitize_ptr_alu()
as well out of its opcode switch so this helps to streamline both.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6f55b2f2a1 ]
Small refactor to drag off_reg into sanitize_ptr_alu(), so we later on can
use off_reg for generalizing some of the checks for all pointer types.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 24c109bb15 ]
The mixed signed bounds check really belongs into retrieve_ptr_limit()
instead of outside of it in adjust_ptr_min_max_vals(). The reason is
that this check is not tied to PTR_TO_MAP_VALUE only, but to all pointer
types that we handle in retrieve_ptr_limit() and given errors from the latter
propagate back to adjust_ptr_min_max_vals() and lead to rejection of the
program, it's a better place to reside to avoid anything slipping through
for future types. The reason why we must reject such off_reg is that we
otherwise would not be able to derive a mask, see details in 9d7eceede7
("bpf: restrict unknown scalars of mixed signed bounds for unprivileged").
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 9601148392 ]
We forbid adding unknown scalars with mixed signed bounds due to the
spectre v1 masking mitigation. Hence this also needs bypass_spec_v1
flag instead of allow_ptr_leaks.
Fixes: 2c78ee898d ("bpf: Implement CAP_BPF")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 12aa8a9467 upstream.
With the introduction of the struct_ops program type, it became possible to
implement kernel functionality in BPF, making it viable to use BPF in place
of a regular kernel module for these particular operations.
Thus far, the only user of this mechanism is for implementing TCP
congestion control algorithms. These are clearly marked as GPL-only when
implemented as modules (as seen by the use of EXPORT_SYMBOL_GPL for
tcp_register_congestion_control()), so it seems like an oversight that this
was not carried over to BPF implementations. Since this is the only user
of the struct_ops mechanism, just enforcing GPL-only for the struct_ops
program type seems like the simplest way to fix this.
Fixes: 0baf26b0fc ("bpf: tcp: Support tcp_congestion_ops in bpf")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210326100314.121853-1-toke@redhat.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 1b1597e64e upstream.
Given we know the max possible value of ptr_limit at the time of retrieving
the latter, add basic assertions, so that the verifier can bail out if
anything looks odd and reject the program. Nothing triggered this so far,
but it also does not hurt to have these.
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit b5871dca25 upstream.
Instead of having the mov32 with aux->alu_limit - 1 immediate, move this
operation to retrieve_ptr_limit() instead to simplify the logic and to
allow for subsequent sanity boundary checks inside retrieve_ptr_limit().
This avoids in future that at the time of the verifier masking rewrite
we'd run into an underflow which would not sign extend due to the nature
of mov32 instruction.
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 10d2bb2e6b upstream.
retrieve_ptr_limit() computes the ptr_limit for registers with stack and
map_value type. ptr_limit is the size of the memory area that is still
valid / in-bounds from the point of the current position and direction
of the operation (add / sub). This size will later be used for masking
the operation such that attempting out-of-bounds access in the speculative
domain is redirected to remain within the bounds of the current map value.
When masking to the right the size is correct, however, when masking to
the left, the size is off-by-one which would lead to an incorrect mask
and thus incorrect arithmetic operation in the non-speculative domain.
Piotr found that if the resulting alu_limit value is zero, then the
BPF_MOV32_IMM() from the fixup_bpf_calls() rewrite will end up loading
0xffffffff into AX instead of sign-extending to the full 64 bit range,
and as a result, this allows abuse for executing speculatively out-of-
bounds loads against 4GB window of address space and thus extracting the
contents of kernel memory via side-channel.
Fixes: 979d63d50c ("bpf: prevent out of bounds speculation on pointer arithmetic")
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit f232326f69 upstream.
The purpose of this patch is to streamline error propagation and in particular
to propagate retrieve_ptr_limit() errors for pointer types that are not defining
a ptr_limit such that register-based alu ops against these types can be rejected.
The main rationale is that a gap has been identified by Piotr in the existing
protection against speculatively out-of-bounds loads, for example, in case of
ctx pointers, unprivileged programs can still perform pointer arithmetic. This
can be abused to execute speculatively out-of-bounds loads without restrictions
and thus extract contents of kernel memory.
Fix this by rejecting unprivileged programs that attempt any pointer arithmetic
on unprotected pointer types. The two affected ones are pointer to ctx as well
as pointer to map. Field access to a modified ctx' pointer is rejected at a
later point in time in the verifier, and 7c69673262 ("bpf: Permit map_ptr
arithmetic with opcode add and offset 0") only relevant for root-only use cases.
Risk of unprivileged program breakage is considered very low.
Fixes: 7c69673262 ("bpf: Permit map_ptr arithmetic with opcode add and offset 0")
Fixes: b2157399cc ("bpf: prevent out-of-bounds speculation")
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 45159b2763 ]
test_global_func4 fails on s390 as reported by Yauheni in [1].
The immediate problem is that the zext code includes the instruction,
whose result needs to be zero-extended, into the zero-extension
patchlet, and if this instruction happens to be a branch, then its
delta is not adjusted. As a result, the verifier rejects the program
later.
However, according to [2], as far as the verifier's algorithm is
concerned and as specified by the insn_no_def() function, branching
insns do not define anything. This includes call insns, even though
one might argue that they define %r0.
This means that the real problem is that zero extension kicks in at
all. This happens because clear_caller_saved_regs() sets BPF_REG_0's
subreg_def after global function calls. This can be fixed in many
ways; this patch mimics what helper function call handling already
does.
[1] https://lore.kernel.org/bpf/20200903140542.156624-1-yauheni.kaliuta@redhat.com/
[2] https://lore.kernel.org/bpf/CAADnVQ+2RPKcftZw8d+B1UwB35cpBhpF5u3OocNh90D9pETPwg@mail.gmail.com/
Fixes: 51c39bb1d5 ("bpf: Introduce function-by-function verification")
Reported-by: Yauheni Kaliuta <yauheni.kaliuta@redhat.com>
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210212040408.90109-1-iii@linux.ibm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 9b00f1b788 upstream.
Recently noticed that when mod32 with a known src reg of 0 is performed,
then the dst register is 32-bit truncated in verifier:
0: R1=ctx(id=0,off=0,imm=0) R10=fp0
0: (b7) r0 = 0
1: R0_w=inv0 R1=ctx(id=0,off=0,imm=0) R10=fp0
1: (b7) r1 = -1
2: R0_w=inv0 R1_w=inv-1 R10=fp0
2: (b4) w2 = -1
3: R0_w=inv0 R1_w=inv-1 R2_w=inv4294967295 R10=fp0
3: (9c) w1 %= w0
4: R0_w=inv0 R1_w=inv(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R2_w=inv4294967295 R10=fp0
4: (b7) r0 = 1
5: R0_w=inv1 R1_w=inv(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R2_w=inv4294967295 R10=fp0
5: (1d) if r1 == r2 goto pc+1
R0_w=inv1 R1_w=inv(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R2_w=inv4294967295 R10=fp0
6: R0_w=inv1 R1_w=inv(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R2_w=inv4294967295 R10=fp0
6: (b7) r0 = 2
7: R0_w=inv2 R1_w=inv(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R2_w=inv4294967295 R10=fp0
7: (95) exit
7: R0=inv1 R1=inv(id=0,umin_value=4294967295,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R2=inv4294967295 R10=fp0
7: (95) exit
However, as a runtime result, we get 2 instead of 1, meaning the dst
register does not contain (u32)-1 in this case. The reason is fairly
straight forward given the 0 test leaves the dst register as-is:
# ./bpftool p d x i 23
0: (b7) r0 = 0
1: (b7) r1 = -1
2: (b4) w2 = -1
3: (16) if w0 == 0x0 goto pc+1
4: (9c) w1 %= w0
5: (b7) r0 = 1
6: (1d) if r1 == r2 goto pc+1
7: (b7) r0 = 2
8: (95) exit
This was originally not an issue given the dst register was marked as
completely unknown (aka 64 bit unknown). However, after 468f6eafa6
("bpf: fix 32-bit ALU op verification") the verifier casts the register
output to 32 bit, and hence it becomes 32 bit unknown. Note that for
the case where the src register is unknown, the dst register is marked
64 bit unknown. After the fix, the register is truncated by the runtime
and the test passes:
# ./bpftool p d x i 23
0: (b7) r0 = 0
1: (b7) r1 = -1
2: (b4) w2 = -1
3: (16) if w0 == 0x0 goto pc+2
4: (9c) w1 %= w0
5: (05) goto pc+1
6: (bc) w1 = w1
7: (b7) r0 = 1
8: (1d) if r1 == r2 goto pc+1
9: (b7) r0 = 2
10: (95) exit
Semantics also match with {R,W}x mod{64,32} 0 -> {R,W}x. Invalid div
has always been {R,W}x div{64,32} 0 -> 0. Rewrites are as follows:
mod32: mod64:
(16) if w0 == 0x0 goto pc+2 (15) if r0 == 0x0 goto pc+1
(9c) w1 %= w0 (9f) r1 %= r0
(05) goto pc+1
(bc) w1 = w1
Fixes: 468f6eafa6 ("bpf: fix 32-bit ALU op verification")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit ee114dd64c upstream.
Fix incorrect is_branch{32,64}_taken() analysis for the jsgt case. The return
code for both will tell the caller whether a given conditional jump is taken
or not, e.g. 1 means branch will be taken [for the involved registers] and the
goto target will be executed, 0 means branch will not be taken and instead we
fall-through to the next insn, and last but not least a -1 denotes that it is
not known at verification time whether a branch will be taken or not. Now while
the jsgt has the branch-taken case correct with reg->s32_min_value > sval, the
branch-not-taken case is off-by-one when testing for reg->s32_max_value < sval
since the branch will also be taken for reg->s32_max_value == sval. The jgt
branch analysis, for example, gets this right.
Fixes: 3f50f132d8 ("bpf: Verifier, do explicit ALU32 bounds tracking")
Fixes: 4f7b3e8258 ("bpf: improve verifier branch analysis")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Daniel Borkmann