On 3/19/21 4:33 PM, Alexei Starovoitov wrote:
On Fri, Mar 19, 2021 at 8:17 AM Yonghong Song <yhs@xxxxxx> wrote:
On 3/19/21 12:21 AM, Daniel Borkmann wrote:
On 3/19/21 3:11 AM, Piotr Krysiuk wrote:
Hi Daniel,
On Fri, Mar 19, 2021 at 12:16 AM Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
wrote:
diff --cc kernel/bpf/verifier.c
index 44e4ec1640f1,f9096b049cd6..000000000000
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@@ -5876,10 -6056,22 +6060,23 @@@ static int
retrieve_ptr_limit(const str
if (mask_to_left)
*ptr_limit = MAX_BPF_STACK + off;
else
- *ptr_limit = -off;
- return 0;
+ *ptr_limit = -off - 1;
+ return *ptr_limit >= max ? -ERANGE : 0;
+ case PTR_TO_MAP_KEY:
+ /* Currently, this code is not exercised as the only use
+ * is bpf_for_each_map_elem() helper which requires
+ * bpf_capble. The code has been tested manually for
+ * future use.
+ */
+ if (mask_to_left) {
+ *ptr_limit = ptr_reg->umax_value + ptr_reg->off;
+ } else {
+ off = ptr_reg->smin_value + ptr_reg->off;
+ *ptr_limit = ptr_reg->map_ptr->key_size - off;
+ }
+ return 0;
PTR_TO_MAP_VALUE logic above looks like copy-paste of old
PTR_TO_MAP_VALUE
code from before "bpf: Fix off-by-one for area size in creating mask to
left" and is apparently affected by the same off-by-one, except this time
on "key_size" area and not "value_size".
This needs to be fixed in the same way as we did with PTR_TO_MAP_VALUE.
What is the best way to proceed?
Hm, not sure why PTR_TO_MAP_KEY was added by 69c087ba6225 in the first
place, I
presume noone expects this to be used from unprivileged as the comment
says.
Resolution should be to remove the PTR_TO_MAP_KEY case entirely from
that switch
until we have an actual user.
Alexei suggested so that we don't forget it in the future if
bpf_capable() requirement is removed.
https://lore.kernel.org/bpf/c837ae55-2487-2f39-47f6-a18781dc6fcc@xxxxxx/
I am okay with either way, fix it or remove it.
I prefer to fix it.
If the bpf_capable() is removed, the verifier would bail out on PTR_TO_MAP_KEY
if not covered in the switch given the recent fixes we did. I can fix it up after
merge if we think bpf_for_each_map_elem() will be used by unpriv in future..
