[prev in list] [next in list] [prev in thread] [next in thread]
List: xdp-newbies
Subject: Re: fentry/fexit trace to BPF_PROG_TYPE_EXT BPF program not working
From: "Eelco Chaudron" <echaudro () redhat ! com>
Date: 2020-05-28 9:10:27
Message-ID: 20F0986B-1EFD-42B8-AF84-07F41E4F8B54 () redhat ! com
[Download RAW message or body]
On 27 May 2020, at 18:54, Eelco Chaudron wrote:
> On 14 May 2020, at 13:56, Eelco Chaudron wrote:
>
>> On 13 May 2020, at 16:58, Eelco Chaudron wrote:
>>
>>> On 6 May 2020, at 3:45, Alexei Starovoitov wrote:
>>>
>>>> On Wed, Apr 29, 2020 at 4:51 AM Eelco Chaudron
>>>> <echaudro@redhat.com> wrote:
>>>>>
>>>>> Hi Alexie at al.
>>>>>
>>>>> I was trying to attach a fentry/fexit trace to BPF_PROG_TYPE_EXT
>>>>> BPF
>>>>> program but I'm getting a verifier error, and not sure why. Is
>>>>> this
>>>>> supported?
>>>>
>
> <SNIP>
>
>> xdp-fexit]$ bpftool btf dump file xdpdump_bpf.o
>> [1] PTR '(anon)' type_id=2
>> [2] INT 'long long unsigned int' size=8 bits_offset=0 nr_bitsd
>> encoding=(none)
>> [3] FUNC_PROTO '(anon)' ret_type_id=4 vlen=1
>> 'ctx' type_id=1
>> [4] INT 'int' size=4 bits_offset=0 nr_bits2 encoding=SIGNED
>> [5] FUNC 'trace_on_exit' type_id=3
>> [6] STRUCT 'xdp_buff' sizeH vlen=6
>> 'data' type_id=7 bits_offset=0
>> 'data_end' type_id=7 bits_offsetd
>> 'data_meta' type_id=7 bits_offset8
>> 'data_hard_start' type_id=7 bits_offset2
>> 'handle' type_id=8 bits_offset%6
>> 'rxq' type_id=9 bits_offset20
>> [7] PTR '(anon)' type_id=0
>> [8] INT 'long unsigned int' size=8 bits_offset=0 nr_bitsd
>> encoding=(none)
>> [9] PTR '(anon)' type_id
>> [10] STRUCT 'xdp_rxq_info' size vlen=2
>> 'dev' type_id bits_offset=0
>> 'queue_index' type_id bits_offsetd
>> [11] PTR '(anon)' type_id
>> [12] TYPEDEF '__u32' type_id
>> [13] INT 'unsigned int' size=4 bits_offset=0 nr_bits2
>> encoding=(none)
>> [14] STRUCT 'net_device' size=4 vlen=1
>> 'ifindex' type_id=4 bits_offset=0
>> [15] INT 'char' size=1 bits_offset=0 nr_bits=8 encoding=SIGNED
>> [16] ARRAY '(anon)' type_id index_type_id nr_elems=4
>> [17] INT '__ARRAY_SIZE_TYPE__' size=4 bits_offset=0 nr_bits2
>> encoding=(none)
>> [18] VAR '_license' type_id, linkage=global-alloc
>> [19] DATASEC 'license' size=0 vlen=1
>> type_id offset=0 size=4
>>
>> Maybe you can take a quick peek why it's failing as I have no real
>> experience with the verifier. I created some sample code that you can
>> easily use with the following instructions:
>>
>> Checkout the xdp tools repo and build it:
>>
>> # git clone https://github.com/chaudron/xdp-tools.git
>> # cd xdp-tools
>> # git checkout dev/xdpdump_multi
>> # make
>>
>> # ulimit -l unlimited
>> # cd xdp-test
>> # cp ../lib/libxdp/xdp-dispatcher.o .
>> # ../xdp-loader/xdp-loader load -m skb eth0 test_two_bpf.o
>> test_bpf.o test_long_func_name.o --force
>>
>>
>> Bpftool should now show you the programs loaded, look for xdp_test_I:
>>
>> # bpftool prog | grep "xdp_test_I "
>> 199: ext name xdp_test_I tag b5a46c6e9935298c gpl
>>
>>
>> Now use the id (199) in the xdp-fexit test program
>>
>> # cd ../xdp-fexit/
>> # git diff
>>
>> - prog_fd = bpf_prog_get_fd_by_id(56);
>> + prog_fd = bpf_prog_get_fd_by_id(199);
>>
>> # make
>>
>>
>> Run it:
>>
>> # ./fexit
>>
>> - Found prog_fd = 3
>> - Opening object file
>> - Opened object file: 0xa526b0
>> libbpf: load bpf program failed: Permission denied
>> libbpf: -- BEGIN DUMP LOG ---
>> libbpf:
>> func#0 @0
>> arg#0 type is not a struct
>> Unrecognized arg#0 type PTR
>> 0: R1=ctx(id=0,off=0,imm=0) R10=fp0
>> ; int BPF_PROG(trace_on_exit, struct xdp_buff *xdp, int ret)
>> 0: (79) r1 = *(u64 *)(r1 +0)
>> invalid bpf_context access off=0 size=8
>> verification time 145 usec
>> stack depth 0
>> processed 1 insns (limit 1000000) max_states_per_insn 0
>> total_states 0 peak_states 0 mark_read 0
>>
>> libbpf: -- END LOG --
>> libbpf: failed to load program 'fexit/xdp_test_I'
>> libbpf: failed to load object './xdpdump_bpf.o'
>> ERROR: Failed to load object file: Permission denied
>>
>> Any ideas are welcome…
>
> Hi Alexei,
>
> I'm still trying to get attach an fentry to a EXT program, however I
> did found the verifier issues…
> If I attach fentry to an XDP entry point, the BPF_PROG() macro casts
> the ctx to long int *, which works:
>
> SEC("fentry/func")
> int BPF_PROG(trace_on_entry, struct xdp_buff *xdp)
> {
> }
>
> Is there a difference between attaching to an XDP program vs EXT that
> it is requiring a struct?
>
> So if I change my function to the below the verifier stops complaining
> for the EXT program:
>
>
> #define bpf_debug(fmt, ...) \
> { \
> char __fmt[] = fmt; \
> bpf_trace_printk(__fmt, sizeof(__fmt), \
> ##__VA_ARGS__); \
> }
>
> SEC("fentry/func")
> int trace_on_entry(struct xdp_md *xdp)
> {
> bpf_debug("HALLO ID: %u", xdp->rx_queue_index);
> return 0;
> }
>
>
> However now the verifier complains about the following and have no
> clue why:
>
> libbpf: map:xdpdump_perf_map
> container_name:____btf_map_xdpdump_perf_map cannot be found in BTF.
> Missing BPF_ANNOTATE_KV_PAIR?
> libbpf: created map xdpdump_perf_map: fd=7
> libbpf: created map xdpdump_.data: fd=8
> libbpf: load_attr.prog_type 26,load_attr.attach_prog_fd 6,
> load_attr.attach_btf_id 7
> libbpf: load bpf program failed: Permission denied
> libbpf: -- BEGIN DUMP LOG ---
> libbpf:
> Unrecognized arg#0 type PTR
> ; int trace_on_entry(struct xdp_md *xdp)
> 0: (b7) r2 = 1965367354
> ; bpf_debug("HALLO ID: %u", xdp->rx_queue_index);
> 1: (63) *(u32 *)(r10 -8) = r2
> 2: (18) r2 = 0x4449204f4c4c4148
> 4: (7b) *(u64 *)(r10 -16) = r2
> 5: (b7) r2 = 0
> 6: (73) *(u8 *)(r10 -4) = r2
> last_idx 6 first_idx 0
> regs=4 stack=0 before 5: (b7) r2 = 0
> 7: (61) r3 = *(u32 *)(r1 +16)
> func 'xdp_test_I' doesn't have 3-th argument
> invalid bpf_context access off size=4
> processed 7 insns (limit 1000000) max_states_per_insn 0 total_states
> 0 peak_states 0 mark_read 0
>
> libbpf: -- END LOG --
> libbpf: failed to load program 'fentry/func'
> libbpf: failed to load object './xdpdump_multi_bpf.o'
> ERROR: Can't load eBPF object: Kernel verifier blocks program
> loading(-4007)
>
>
> This is the program details I try to attach too:
>
> 49: type 28 name xdp_test_I tag b5a46c6e9935298c gpl
> loaded_at 2020-05-27T11:40:55+0000 uid 0
> xlated 136B jited 108B memlock 4096B
> btf_id 12
>
> # bpftool btf dump id 12
> [1] PTR '(anon)' type_id=2
> [2] STRUCT 'xdp_md' size vlen=5
> 'data' type_id=3 bits_offset=0
> 'data_end' type_id=3 bits_offset2
> 'data_meta' type_id=3 bits_offsetd
> 'ingress_ifindex' type_id=3 bits_offset–
> 'rx_queue_index' type_id=3 bits_offset8
> [3] TYPEDEF '__u32' type_id=4
> [4] INT 'unsigned int' size=4 bits_offset=0 nr_bits2 encoding=(none)
> [5] FUNC_PROTO '(anon)' ret_type_id=6 vlen=1
> 'ctx' type_id=1
> [6] INT 'int' size=4 bits_offset=0 nr_bits2 encoding=SIGNED
> [7] FUNC 'xdp_test_I' type_id=5 linkage=global
> [8] INT 'char' size=1 bits_offset=0 nr_bits=8 encoding=SIGNED
> [9] ARRAY '(anon)' type_id=8 index_type_id nr_elems=4
> [10] INT '__ARRAY_SIZE_TYPE__' size=4 bits_offset=0 nr_bits2
> encoding=(none)
> [11] VAR '_license' type_id=9, linkage=global-alloc
> [12] DATASEC 'license' size=4 vlen=1
> type_id offset=0 size=4
>
As attaching to an XDP entry point with fentry required not to use the
xdp_md struct, but the xdp_buff struct, I tried this also.
But as you can see below, also here it compliant about the xdp_test_I
function an argument (note the xdp_test_I function is the EXT function I
try to attach to).
This is the new code:
struct xdp_rxq_info {
/* Structure does not need to contain all entries,
* as "preserve_access_index" will use BTF to fix this...
*/
struct net_device *dev;
__u32 queue_index;
} __attribute__((preserve_access_index));
struct xdp_buff {
void *data;
void *data_end;
void *data_meta;
void *data_hard_start;
unsigned long handle;
struct xdp_rxq_info *rxq;
} __attribute__((preserve_access_index));
SEC("fentry/func")
int trace_on_entry(struct xdp_buff *xdp)
{
bpf_debug("HALLO ID: %u", xdp->rxq->queue_index);
return 0;
}
The result of the kernel:
libbpf: map:xdpdump_perf_map
container_name:____btf_map_xdpdump_perf_map cannot be found in BTF.
Missing BPF_ANNOTATE_KV_PAIR?
libbpf: created map xdpdump_perf_map: fd=7
libbpf: created map xdpdump_.data: fd=8
libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
libbpf: prog 'fentry/func': performing 2 CO-RE offset relocs
libbpf: prog 'fentry/func': relo #0: kind 0, spec is [2] xdp_buff +
0:5 => 40.0 @ &x[0].rxq
libbpf: [2] xdp_buff: found candidate [19906] xdp_buff
libbpf: prog 'fentry/func': relo #0: matching candidate #0 xdp_buff
against spec [19906] xdp_buff + 0:4 => 32.0 @ &x[0].rxq: 1
libbpf: prog 'fentry/func': relo #0: patched insn #7 (LDX/ST/STX) off
40 -> 32
libbpf: prog 'fentry/func': relo #1: kind 0, spec is [9] xdp_rxq_info
+ 0:1 => 8.0 @ &x[0].queue_index
libbpf: [9] xdp_rxq_info: found candidate [3409] xdp_rxq_info
libbpf: prog 'fentry/func': relo #1: matching candidate #0
xdp_rxq_info against spec [3409] xdp_rxq_info + 0:1 => 8.0 @
&x[0].queue_index: 1
libbpf: prog 'fentry/func': relo #1: patched insn #8 (LDX/ST/STX) off
8 -> 8
libbpf: load_attr.prog_type 26,load_attr.attach_prog_fd 6,
load_attr.attach_btf_id 7
libbpf: load bpf program failed: Permission denied
libbpf: -- BEGIN DUMP LOG ---
libbpf:
Unrecognized arg#0 type PTR
; int trace_on_entry(struct xdp_buff *xdp)
0: (b7) r2 = 1965367354
; bpf_debug("HALLO ID: %u", xdp->rxq->queue_index);
1: (63) *(u32 *)(r10 -8) = r2
2: (18) r2 = 0x4449204f4c4c4148
4: (7b) *(u64 *)(r10 -16) = r2
5: (b7) r2 = 0
6: (73) *(u8 *)(r10 -4) = r2
last_idx 6 first_idx 0
regs=4 stack=0 before 5: (b7) r2 = 0
7: (79) r1 = *(u64 *)(r1 +32)
func 'xdp_test_I' doesn't have 5-th argument
invalid bpf_context access off2 size=8
processed 7 insns (limit 1000000) max_states_per_insn 0 total_states
0 peak_states 0 mark_read 0
libbpf: -- END LOG --
libbpf: failed to load program 'fentry/func'
libbpf: failed to load object './xdpdump_multi_bpf.o'
ERROR: Can't load eBPF object: Kernel verifier blocks program
loading(-4007)
Any input is welcome. I included the bpf list to see if others might
have any input (thought it was already).
Cheers,
Eelco
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic