[llvm-bugs] [Bug 43994] New: BPF code generation emits "|" instead of "+", causing problems for BPF verifier
via llvm-bugs
llvm-bugs at lists.llvm.org
Wed Nov 13 12:38:49 PST 2019
https://bugs.llvm.org/show_bug.cgi?id=43994
Bug ID: 43994
Summary: BPF code generation emits "|" instead of "+", causing
problems for BPF verifier
Product: clang
Version: 9.0
Hardware: Other
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: -New Bugs
Assignee: unassignedclangbugs at nondot.org
Reporter: eugene.loh at oracle.com
CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org,
neeilans at live.com, richard-llvm at metafoo.co.uk
Created attachment 22808
--> https://bugs.llvm.org/attachment.cgi?id=22808&action=edit
C code, run script, generated BPF code
Here is an issue with clang/LLVM's BPF code generation.
Technically, the generated code is not wrong, but it causes
an unnecessary problem for the BPF verifier.
Consider the C source code BPF.c. The generated BPF code can be
seen in objdump.txt and verifier.txt. Here are the C and BPF code,
redacted for readability and shown side-by-side:
BPF.c objdump.txt/verifier.txt
==================================== =====================
u64 dest = -1; *(u64 *)(r10-24) = -1
u64 offset;
u64 garbage = 0x3141592654; *(u64 *)(r10- 8) = 0x3141592654
// set offset to some mystery value r1 = r10-16
bpf_probe_read(&offset, r2 = 8
sizeof (offset), &garbage); r3 = r10- 8
call bpf_probe_read#4
// clip offset to be offset within dest r1 = *(u64 *)(r10 - 16)
offset &= sizeof (dest) - 1; r1 &= 7
*(u64 *)(r10 - 16) = r1
// copy byte to mystery offset in dest r2 = r10-24
bpf_probe_read(((u64) &dest) + offset, r1 |= r2 ======> PROBLEM!
1, &garbage); r2 = 1
r3 = r10- 8
call bpf_probe_read#4
return 0; r0 = 0
exit
To compute ((u64)&dest)+offset, the generated BPF uses "offset |= &dest".
The replacement of "+" with "|" presumably gives the correct result
in this particular case, but then the BPF verifier complains:
R1 bitwise operator |= on pointer prohibited
That is, the replacement of "+" with "|" causes this correct program to be
rejected unnecessarily by the BPF verifier.
This problem occurred on a system with an upstream kernel with
% uname -r
5.3.0+
with clang/LLVM built from source:
% clang --version
clang version 9.0.0 (trunk 364493)
[...]
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20191113/8a289fdf/attachment.html>
More information about the llvm-bugs
mailing list