Fix #1796 and skip probe rewriter for bpf_probe_read #1812
Conversation
| // return TCP_SKB_CB(skb)->tcp_gso_size; | ||
| u16 val = 0; | ||
| bpf_probe_read(&val, sizeof(val), &(TCP_SKB_CB(skb)->tcp_gso_size)); | ||
| return val + skb->protocol; |
There was a problem hiding this comment.
Without this pull request, this second test fails because the AST traversal is stopped after meeting the bpf_probe_read call, even though a rewrite is needed for the later expression skb->protocol.
|
I don't understand the Jenkins' failures. The tests passed locally (Ubuntu 16.04 and 17.10), and they're passing for the Fedora builds... |
|
The function Currently, we already track LLVM major numbers in the cmake file. We can also track minor number, and guard reference of What do you think? |
I agree, though here, we can actually implement the same behavior without |
pchaigno
force-pushed
the
skip-bpf_probe_read
branch
2 times, most recently
from
0f86ff3 to
6048c00
Compare
20fb64c stops the whole AST traversal if it meets a bpf_probe_read call. I think the original intent was to simply not rewrite the third argument, so this commit fixes it by remembering the third argument on bpf_probe_read call traversals and overriding TraverseStmt to skip the traversal of that argument when we meet it later.
pchaigno
force-pushed
the
skip-bpf_probe_read
branch
from
6048c00 to
c289d92
Compare
|
LGTM. Thanks! |
20fb64c stops the whole AST traversal if it meets a bpf_probe_read call. I think the original intent was to simply not rewrite the third argument, so this commit fixes it by remembering the third argument on bpf_probe_read call traversals and overriding TraverseStmt to skip the traversal of that argument when we meet it later.
20fb64c stops the whole AST traversal if it meets a bpf_probe_read call. I think the original intent was to simply not rewrite the third argument, so this commit fixes it by remembering the third argument on bpf_probe_read call traversals and overriding TraverseStmt to skip the traversal of that argument when we meet it later.
#1796 stops the whole AST traversal if it meets a
bpf_probe_readcall. I think the original intent was to simply not rewrite the third argument, so this commit fixes it by remembering the third argument onbpf_probe_readcall traversals and overridingdataTraverseStmtPreto skip the traversal of thatargument when we meet it later.
/cc @yonghong-song