Message-ID: <20180702220659.6baa77ba@cakuba.netronome.com>
Date:   Mon, 2 Jul 2018 22:06:59 -0700
From:   Jakub Kicinski <jakub.kicinski@...ronome.com>
To:     Okash Khawaja <osk@...com>
Cc:     Daniel Borkmann <daniel@...earbox.net>,
        Martin KaFai Lau <kafai@...com>,
        Alexei Starovoitov <ast@...nel.org>,
        Yonghong Song <yhs@...com>,
        Quentin Monnet <quentin.monnet@...ronome.com>,
        "David S. Miller" <davem@...emloft.net>, <netdev@...r.kernel.org>,
        <kernel-team@...com>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH bpf-next v2 2/3] bpf: btf: add btf print functionality

On Mon, 2 Jul 2018 11:39:15 -0700, Okash Khawaja wrote:
> This consumes functionality exported in the previous patch. It does the
> main job of printing with BTF data. This is used in the following patch
> to provide a more readable output of a map's dump. It relies on
> json_writer to do json printing. Below is sample output where map keys
> are ints and values are of type struct A:
> 
> typedef int int_type;
> enum E {
>         E0,
>         E1,
> };
> 
> struct B {
>         int x;
>         int y;
> };
> 
> struct A {
>         int m;
>         unsigned long long n;
>         char o;
>         int p[8];
>         int q[4][8];
>         enum E r;
>         void *s;
>         struct B t;
>         const int u;
>         int_type v;
>         unsigned int w1: 3;
>         unsigned int w2: 3;
> };
> 
> $ sudo bpftool map dump id 14
> [{
>         "key": 0,
>         "value": {
>             "m": 1,
>             "n": 2,
>             "o": "c",
>             "p": [15,16,17,18,15,16,17,18
>             ],
>             "q": [[25,26,27,28,25,26,27,28
>                 ],[35,36,37,38,35,36,37,38
>                 ],[45,46,47,48,45,46,47,48
>                 ],[55,56,57,58,55,56,57,58
>                 ]
>             ],
>             "r": 1,
>             "s": 0x7ffd80531cf8,
>             "t": {
>                 "x": 5,
>                 "y": 10
>             },
>             "u": 100,
>             "v": 20,
>             "w1": 0x7,
>             "w2": 0x3
>         }
>     }
> ]
> 
> This patch uses json's {} and [] to imply struct/union and array. More
> explicit information can be added later. For example, a command line
> option can be introduced to print whether a key or value is struct
> or union, name of a struct etc. This will however come at the expense
> of duplicating info when, for example, printing an array of structs.
> enums are printed as ints without their names.
> 
> Signed-off-by: Okash Khawaja <osk@...com>
> 
> ---
>  tools/bpf/bpftool/btf_dumper.c |  263 +++++++++++++++++++++++++++++++++++++++++
>  tools/bpf/bpftool/btf_dumper.h |   23 +++
>  2 files changed, 286 insertions(+)
> 
> --- /dev/null
> +++ b/tools/bpf/bpftool/btf_dumper.c
> @@ -0,0 +1,263 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Copyright (c) 2018 Facebook */
> +
> +#include <linux/btf.h>
> +#include <linux/err.h>
> +#include <stdio.h> /* for (FILE *) used by json_writer */
> +#include <linux/bitops.h>
> +#include <string.h>
> +#include <ctype.h>
> +
> +#include "btf.h"
> +#include "json_writer.h"
> +#include "btf_dumper.h"



> +#define BITS_PER_BYTE_MASK (BITS_PER_BYTE - 1)
> +#define BITS_PER_BYTE_MASKED(bits) ((bits) & BITS_PER_BYTE_MASK)

Perhaps it's just me but BIT_OFFSET or BIT_COUNT as a name of this macro
would make it more obvious to parse in the code below.

> +#define BITS_ROUNDDOWN_BYTES(bits) ((bits) >> 3)
> +#define BITS_ROUNDUP_BYTES(bits) \
> +	(BITS_ROUNDDOWN_BYTES(bits) + !!BITS_PER_BYTE_MASKED(bits))
> +
> +static int btf_dumper_do_type(const struct btf_dumper *d, uint32_t type_id,
> +			      uint8_t bit_offset, const void *data);
> +
> +static void btf_dumper_ptr(const void *data, json_writer_t *jw,
> +			   bool is_plain_text)
> +{
> +	if (is_plain_text)
> +		jsonw_printf(jw, "%p", *((uintptr_t *)data));
> +	else
> +		jsonw_printf(jw, "%u", *((uintptr_t *)data));
> +}
> +
> +static int btf_dumper_modifier(const struct btf_dumper *d, uint32_t type_id,
> +			       const void *data)
> +{
> +	int32_t actual_type_id = btf__resolve_type(d->btf, type_id);

Please prefer kernel types like __u32 wherever possible.

> +	int ret;
> +
> +	if (actual_type_id < 0)
> +		return actual_type_id;
> +
> +	ret = btf_dumper_do_type(d, actual_type_id, 0, data);
> +
> +	return ret;

ret is unnecessary.

> +}
> +
> +static void btf_dumper_enum(const void *data, json_writer_t *jw)
> +{
> +	jsonw_printf(jw, "%d", *((int32_t *)data));

Unnecessary parenthesis.  There is a lot of those, please remove them
all.

> +}
> +
> +static int btf_dumper_array(const struct btf_dumper *d, uint32_t type_id,
> +			    const void *data)
> +{
> +	const struct btf_type *t = btf__type_by_id(d->btf, type_id);
> +	struct btf_array *arr = (struct btf_array *)(t + 1);
> +	int64_t elem_size;
> +	int ret = 0;
> +	uint32_t i;
> +
> +	elem_size = btf__resolve_size(d->btf, arr->type);
> +	if (elem_size < 0)
> +		return elem_size;
> +
> +	jsonw_start_array(d->jw);
> +	for (i = 0; i < arr->nelems; i++) {
> +		ret = btf_dumper_do_type(d, arr->type, 0,
> +					 data + (i * elem_size));

Unnecessary parenthesis.

> +		if (ret)
> +			break;
> +	}
> +
> +	jsonw_end_array(d->jw);
> +	return ret;
> +}
> +
> +static void btf_dumper_int_bits(uint32_t int_type, uint8_t bit_offset,
> +				const void *data, json_writer_t *jw,
> +				bool is_plain_text)
> +{
> +	uint32_t bits = BTF_INT_BITS(int_type);
> +	uint16_t total_bits_offset;
> +	uint16_t bytes_to_copy;
> +	uint16_t bits_to_copy;

Please use normal int types for things which don't have to be
explicitly sized.  Using explicitly sized variables is bad style, 
and ALU operations other than on word or byte quantities are usually
slower on modern CPUs.

> +	uint8_t upper_bits;
> +	union {
> +		uint64_t u64_num;
> +		uint8_t u8_nums[8];

Are the int types in BTF constrained to 64bit at most?

> +	} print_num;
> +
> +	total_bits_offset = bit_offset + BTF_INT_OFFSET(int_type);
> +	data += BITS_ROUNDDOWN_BYTES(total_bits_offset);
> +	bit_offset = BITS_PER_BYTE_MASKED(total_bits_offset);
> +	bits_to_copy = bits + bit_offset;
> +	bytes_to_copy = BITS_ROUNDUP_BYTES(bits_to_copy);
> +
> +	print_num.u64_num = 0;
> +	memcpy(&print_num.u64_num, data, bytes_to_copy);

This scheme is unlikely to work on big endian machines...

> +	upper_bits = BITS_PER_BYTE_MASKED(bits_to_copy);
> +	if (upper_bits) {
> +		uint8_t mask = (1 << upper_bits) - 1;
> +
> +		print_num.u8_nums[bytes_to_copy - 1] &= mask;
> +	}
> +
> +	print_num.u64_num >>= bit_offset;
> +
> +	if (is_plain_text)
> +		jsonw_printf(jw, "0x%llx", print_num.u64_num);
> +	else
> +		jsonw_printf(jw, "%llu", print_num.u64_num);
> +}
> +
> +static int btf_dumper_int(const struct btf_type *t, uint8_t bit_offset,
> +			  const void *data, json_writer_t *jw,
> +			  bool is_plain_text)
> +{
> +	uint32_t *int_type = (uint32_t *)(t + 1);
> +	uint32_t bits = BTF_INT_BITS(*int_type);
> +	int ret = 0;
> +
> +	/* if this is bit field */
> +	if (bit_offset || BTF_INT_OFFSET(*int_type) ||
> +	    BITS_PER_BYTE_MASKED(bits)) {
> +		btf_dumper_int_bits(*int_type, bit_offset, data, jw,
> +				    is_plain_text);
> +		return ret;
> +	}
> +
> +	switch (BTF_INT_ENCODING(*int_type)) {
> +	case 0:
> +		if (BTF_INT_BITS(*int_type) == 64)
> +			jsonw_printf(jw, "%lu", *((uint64_t *)data));
> +		else if (BTF_INT_BITS(*int_type) == 32)
> +			jsonw_printf(jw, "%u", *((uint32_t *)data));
> +		else if (BTF_INT_BITS(*int_type) == 16)
> +			jsonw_printf(jw, "%hu", *((uint16_t *)data));
> +		else if (BTF_INT_BITS(*int_type) == 8)
> +			jsonw_printf(jw, "%hhu", *((uint8_t *)data));
> +		else
> +			btf_dumper_int_bits(*int_type, bit_offset, data, jw,
> +					    is_plain_text);
> +		break;
> +	case BTF_INT_SIGNED:
> +		if (BTF_INT_BITS(*int_type) == 64)
> +			jsonw_printf(jw, "%ld", *((int64_t *)data));
> +		else if (BTF_INT_BITS(*int_type) == 32)
> +			jsonw_printf(jw, "%d", *((int32_t *)data));
> +		else if (BTF_INT_BITS(*int_type) ==  16)

Please drop the double space.  Both for 16 where it makes no sense and
for 8 where it's marginally useful but not really.

> +			jsonw_printf(jw, "%hd", *((int16_t *)data));
> +		else if (BTF_INT_BITS(*int_type) ==  8)
> +			jsonw_printf(jw, "%hhd", *((int8_t *)data));
> +		else
> +			btf_dumper_int_bits(*int_type, bit_offset, data, jw,
> +					    is_plain_text);
> +		break;
> +	case BTF_INT_CHAR:
> +		if (*((char *)data) == '\0')
> +			jsonw_null(jw);

Mm.. I don't think 0 char is equivalent to null.

> +		else if (isprint(*((char *)data)))
> +			jsonw_printf(jw, "\"%c\"", *((char *)data));

This looks very suspicious.  So if I see a "6" for a char field it's
either a 6 ('\u0006') or a 54 ('6')...

> +		else
> +			if (is_plain_text)
> +				jsonw_printf(jw, "%hhx", *((char *)data));
> +			else
> +				jsonw_printf(jw, "%hhd", *((char *)data));

... I think you need to always print a string, and express it as
\u00%02hhx for non-printable.

> +		break;
> +	case BTF_INT_BOOL:
> +		jsonw_bool(jw, *((int *)data));
> +		break;
> +	default:
> +		/* shouldn't happen */
> +		ret = -EINVAL;

You only set ret to something else than 0 here just to break and
immediately return.  Please remove the ret variable.

> +		break;
> +	}
> +
> +	return ret;
> +}
> +
> +static int btf_dumper_struct(const struct btf_dumper *d, uint32_t type_id,
> +			     const void *data)
> +{
> +	const struct btf_type *t = btf__type_by_id(d->btf, type_id);

Please don't call functions which need error checking as initialized
the if below looks very awkward..

> +	struct btf_member *m;
> +	int ret = 0;
> +
> +	int i, vlen;
> +
> +	if (!t)
> +		return -EINVAL;
> +
> +	vlen = BTF_INFO_VLEN(t->info);
> +	jsonw_start_object(d->jw);
> +	m = (struct btf_member *)(t + 1);
> +
> +	for (i = 0; i < vlen; i++) {
> +		jsonw_name(d->jw, btf__name_by_offset(d->btf, m[i].name_off));
> +		ret = btf_dumper_do_type(d, m[i].type,
> +					 BITS_PER_BYTE_MASKED(m[i].offset), data
> +					 + BITS_ROUNDDOWN_BYTES(m[i].offset));

Please use a temp variable to avoid this awkward multi-line sum.

> +		if (ret)
> +			return ret;

You can't return without jsonw_end_object().

> +	}
> +
> +	jsonw_end_object(d->jw);
> +
> +	return 0;
> +}
> +
> +static int btf_dumper_do_type(const struct btf_dumper *d, uint32_t type_id,
> +			      uint8_t bit_offset, const void *data)
> +{
> +	const struct btf_type *t = btf__type_by_id(d->btf, type_id);
> +	int ret = 0;
> +
> +	switch (BTF_INFO_KIND(t->info)) {
> +	case BTF_KIND_INT:
> +		ret = btf_dumper_int(t, bit_offset, data, d->jw,
> +				     d->is_plain_text);
> +		break;
> +	case BTF_KIND_STRUCT:
> +	case BTF_KIND_UNION:
> +		ret = btf_dumper_struct(d, type_id, data);
> +		break;
> +	case BTF_KIND_ARRAY:
> +		ret = btf_dumper_array(d, type_id, data);
> +		break;
> +	case BTF_KIND_ENUM:
> +		btf_dumper_enum(data, d->jw);
> +		break;
> +	case BTF_KIND_PTR:
> +		btf_dumper_ptr(data, d->jw, d->is_plain_text);
> +		break;
> +	case BTF_KIND_UNKN:
> +		jsonw_printf(d->jw, "(unknown)");
> +		break;
> +	case BTF_KIND_FWD:
> +		/* map key or value can't be forward */

Right, but you have to print _something_, otherwise we would have a
name without a value, which would break JSON, no?

> +		ret = -EINVAL;
> +		break;
> +	case BTF_KIND_TYPEDEF:
> +	case BTF_KIND_VOLATILE:
> +	case BTF_KIND_CONST:
> +	case BTF_KIND_RESTRICT:
> +		ret = btf_dumper_modifier(d, type_id, data);
> +		break;
> +	default:
> +		jsonw_printf(d->jw, "(unsupported-kind");
> +		ret = -EINVAL;
> +		break;
> +	}
> +
> +	return ret;

Why return ret; at all, there is no code after the switch just return
directly from cases and save 9 LOC.

> +}
> +
> +int32_t btf_dumper_type(const struct btf_dumper *d, uint32_t type_id,
> +			const void *data)
> +{
> +	if (!d)
> +		return -EINVAL;

No need for defensive programming.

> +	return btf_dumper_do_type(d, type_id, 0, data);
> +}
> --- /dev/null
> +++ b/tools/bpf/bpftool/btf_dumper.h
> @@ -0,0 +1,23 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/* Copyright (c) 2018 Facebook */
> +
> +#ifndef BTF_DUMPER_H
> +#define BTF_DUMPER_H
> +
> +struct btf_dumper {
> +	const struct btf *btf;
> +	json_writer_t *jw;
> +	bool is_plain_text;
> +};
> +
> +/* btf_dumper_type - print data along with type information
> + * @d: an instance containing context for dumping types
> + * @type_id: index in btf->types array. this points to the type to be dumped
> + * @data: pointer the actual data, i.e. the values to be printed
> + *
> + * Returns zero on success and negative error code otherwise
> + */
> +int32_t btf_dumper_type(const struct btf_dumper *d, uint32_t type_id,
> +			const void *data);
> +
> +#endif

Please don't add header files for a single struct and single function.
Just put this in main.h.

Powered by blists - more mailing lists