Re: [PATCH bpf-next v5 6/6] selftests/bpf: Add a series of tests for bpf_snprintf

From: Rasmus Villemoes
Date: Tue Apr 27 2021 - 02:35:57 EST


On 26/04/2021 23.08, Florent Revest wrote:
> On Mon, Apr 26, 2021 at 6:19 PM Andrii Nakryiko
> <andrii.nakryiko@xxxxxxxxx> wrote:
>>
>> On Mon, Apr 26, 2021 at 3:10 AM Florent Revest <revest@xxxxxxxxxxxx> wrote:
>>>
>>> On Sat, Apr 24, 2021 at 12:38 AM Andrii Nakryiko
>>> <andrii.nakryiko@xxxxxxxxx> wrote:
>>>>
>>>> On Mon, Apr 19, 2021 at 8:52 AM Florent Revest <revest@xxxxxxxxxxxx> wrote:
>>>>>
>>>>> The "positive" part tests all format specifiers when things go well.
>>>>>
>>>>> The "negative" part makes sure that incorrect format strings fail at
>>>>> load time.
>>>>>
>>>>> Signed-off-by: Florent Revest <revest@xxxxxxxxxxxx>
>>>>> ---
>>>>> .../selftests/bpf/prog_tests/snprintf.c | 125 ++++++++++++++++++
>>>>> .../selftests/bpf/progs/test_snprintf.c | 73 ++++++++++
>>>>> .../bpf/progs/test_snprintf_single.c | 20 +++
>>>>> 3 files changed, 218 insertions(+)
>>>>> create mode 100644 tools/testing/selftests/bpf/prog_tests/snprintf.c
>>>>> create mode 100644 tools/testing/selftests/bpf/progs/test_snprintf.c
>>>>> create mode 100644 tools/testing/selftests/bpf/progs/test_snprintf_single.c
>>>>>
>>>>> diff --git a/tools/testing/selftests/bpf/prog_tests/snprintf.c b/tools/testing/selftests/bpf/prog_tests/snprintf.c
>>>>> new file mode 100644
>>>>> index 000000000000..a958c22aec75
>>>>> --- /dev/null
>>>>> +++ b/tools/testing/selftests/bpf/prog_tests/snprintf.c
>>>>> @@ -0,0 +1,125 @@
>>>>> +// SPDX-License-Identifier: GPL-2.0
>>>>> +/* Copyright (c) 2021 Google LLC. */
>>>>> +
>>>>> +#include <test_progs.h>
>>>>> +#include "test_snprintf.skel.h"
>>>>> +#include "test_snprintf_single.skel.h"
>>>>> +
>>>>> +#define EXP_NUM_OUT "-8 9 96 -424242 1337 DABBAD00"
>>>>> +#define EXP_NUM_RET sizeof(EXP_NUM_OUT)
>>>>> +
>>>>> +#define EXP_IP_OUT "127.000.000.001 0000:0000:0000:0000:0000:0000:0000:0001"
>>>>> +#define EXP_IP_RET sizeof(EXP_IP_OUT)
>>>>> +
>>>>> +/* The third specifier, %pB, depends on compiler inlining so don't check it */
>>>>> +#define EXP_SYM_OUT "schedule schedule+0x0/"
>>>>> +#define MIN_SYM_RET sizeof(EXP_SYM_OUT)
>>>>> +
>>>>> +/* The third specifier, %p, is a hashed pointer which changes on every reboot */
>>>>> +#define EXP_ADDR_OUT "0000000000000000 ffff00000add4e55 "
>>>>> +#define EXP_ADDR_RET sizeof(EXP_ADDR_OUT "unknownhashedptr")
>>>>> +
>>>>> +#define EXP_STR_OUT "str1 longstr"
>>>>> +#define EXP_STR_RET sizeof(EXP_STR_OUT)
>>>>> +
>>>>> +#define EXP_OVER_OUT "%over"
>>>>> +#define EXP_OVER_RET 10
>>>>> +
>>>>> +#define EXP_PAD_OUT " 4 000"
>>>>
>>>> Roughly 50% of the time I get failure for this test case:
>>>>
>>>> test_snprintf_positive:FAIL:pad_out unexpected pad_out: actual ' 4
>>>> 0000' != expected ' 4 000'
>>>>
>>>> Re-running this test case immediately passes. Running again most
>>>> probably fails. Please take a look.
>>>
>>> Do you have more information on how to reproduce this ?
>>> I spinned up a VM at 87bd9e602 with ./vmtest -s and then run this script:
>>>
>>> #!/bin/sh
>>> for i in `seq 1000`
>>> do
>>> ./test_progs -t snprintf
>>> if [ $? -ne 0 ];
>>> then
>>> echo FAILURE
>>> exit 1
>>> fi
>>> done
>>>
>>> The thousand executions passed.
>>>
>>> This is a bit concerning because your unexpected_pad_out seems to have
>>> an extra '0' so it ends up with strlen(pad_out)=11 but
>>> sizeof(pad_out)=10. The actual string writing is not really done by
>>> our helper code but by the snprintf implementation (str and str_size
>>> are only given to snprintf()) so I'd expect the truncation to work
>>> well there. I'm a bit puzzled
>>
>> I'm puzzled too, have no idea. I also can't repro this with vmtest.sh.
>> But I can quite reliably reproduce with my local ArchLinux-based qemu
>> image with different config (see [0] for config itself). So please try
>> with my config and see if that helps to repro. If not, I'll have to
>> debug it on my own later.
>>
>> [0] https://gist.github.com/anakryiko/4b6ae21680842bdeacca8fa99d378048
>
> I tried that config on the same commit 87bd9e602 (bpf-next/master)
> with my debian-based qemu image and I still can't reproduce the issue
> :| If I can be of any help let me know, I'd be happy to help
>

It's not really clear to me if this is before or after the rewrite to
use bprintf, but regardless, in those two patches this caught my attention:

u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
- enum bpf_printf_mod_type mod[MAX_TRACE_PRINTK_VARARGS];
+ u32 *bin_args;
static char buf[BPF_TRACE_PRINTK_SIZE];
unsigned long flags;
int ret;

- ret = bpf_printf_prepare(fmt, fmt_size, args, args, mod,
- MAX_TRACE_PRINTK_VARARGS);
+ ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
+ MAX_TRACE_PRINTK_VARARGS);
if (ret < 0)
return ret;

- ret = snprintf(buf, sizeof(buf), fmt, BPF_CAST_FMT_ARG(0, args, mod),
- BPF_CAST_FMT_ARG(1, args, mod), BPF_CAST_FMT_ARG(2, args, mod));
- /* snprintf() will not append null for zero-length strings */
- if (ret == 0)
- buf[0] = '\0';
+ ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);

raw_spin_lock_irqsave(&trace_printk_lock, flags);
trace_bpf_trace_printk(buf);
raw_spin_unlock_irqrestore(&trace_printk_lock, flags);

Why isn't the write to buf[] protected by that spinlock? Or put another
way, what protects buf[] from concurrent writes?

Probably the test cases are not run in parallel, but this is the kind of
thing that would give those symptoms.

Rasmus