[PATCH v3 bpf-next 0/2] libbpf: BTF dumper support for typed data

[PATCH v3 bpf-next 0/2] libbpf: BTF dumper support for typed data

[Alan Maguire]

Add a libbpf dumper function that supports dumping a representation
of data passed in using the BTF id associated with the data in a
manner similar to the bpf_snprintf_btf helper.

Default output format is identical to that dumped by bpf_snprintf_btf()
(bar using tabs instead of spaces for indentation); for example,
a "struct sk_buff" representation would look like this:

(struct sk_buff){
	(union){
		(struct){
			.next = (struct sk_buff *)0xffffffffffffffff,
			.prev = (struct sk_buff *)0xffffffffffffffff,
			(union){
				.dev = (struct net_device *)0xffffffffffffffff,
				.dev_scratch = (long unsigned int)18446744073709551615,
			},
	},
...

Patch 1 implements the dump functionality in a manner similar
to that in kernel/bpf/btf.c, but with a view to fitting into
libbpf more naturally.  For example, rather than using flags,
boolean dump options are used to control output.  In addition,
rather than combining checks for display (such as is this
field zero?) and actual display - as is done for the kernel
code - the code is organized to separate zero and overflow
checks from type display.

Patch 2 consists of selftests that utilize a dump printf function
to snprintf the dump output to a string for comparison with
expected output.  Tests deliberately mirror those in
snprintf_btf helper test to keep output consistent, but
also cover overflow handling, var/section display.

Apologies for the long time lag between v2 and this revision.

Changes since v2 [1]

- Renamed function to btf_dump__dump_type_data, reorganized
  arguments such that opts are last (Andrii)
- Modified code to separate questions about display such
  as have we overflowed?/is this field zero? from actual
  display of typed data, such that we ask those questions
  separately from the code that actually displays typed data
  (Andrii)
- Reworked code to handle overflow - where we do not provide
  enough data for the type we wish to display - by returning
  -E2BIG and attempting to present as much data as possible.
  Such a mode of operation allows for tracers which retrieve
  partial data (such as first 1024 bytes of a
  "struct task_struct" say), and want to display that partial
  data, while also knowing that it is not the full type.
  Such tracers can then denote this (perhaps via "..." or
  similar).
- Explored reusing existing type emit functions, such as
  passing in a type id stack with a single type id to
  btf_dump_emit_type_chain() to support the display of
  typed data where a "cast" is prepended to the data to
  denote its type; "(int)1", "(struct foo){", etc.
  However the task of emitting a
  ".field_name = (typecast)" did not match well with model
  of walking the stack to display innermost types first
  and made the resultant code harder to read.  Added a
  dedicated btf_dump_emit_type_name() function instead which
  is only ~70 lines (Andrii)
- Various cleanups around bitfield macros, unneeded member
  iteration macros, avoiding compiler complaints when
  displaying int da ta by casting to long long, etc (Andrii)
- Use DECLARE_LIBBPF_OPTS() in defining opts for tests (Andrii)
- Added more type tests, overflow tests, var tests and
  section tests.

Changes since RFC [2]

- The initial approach explored was to share the kernel code
  with libbpf using #defines to paper over the different needs;
  however it makes more sense to try and fit in with libbpf
  code style for maintenance.  A comment in the code points at
  the implementation in kernel/bpf/btf.c and notes that any
  issues found in it should be fixed there or vice versa;
  mirroring the tests should help with this also
  (Andrii)

[1] https://lore.kernel.org/bpf/1610921764-7526-1-git-send-email-alan.maguire@oracle.com/
[2] https://lore.kernel.org/bpf/1610386373-24162-1-git-send-email-alan.maguire@oracle.com/T/#t


Alan Maguire (2):
  libbpf: BTF dumper support for typed data
  selftests/bpf: add dump type data tests to btf dump tests

 tools/lib/bpf/btf.h                               |  17 +
 tools/lib/bpf/btf_dump.c                          | 901 ++++++++++++++++++++++
 tools/lib/bpf/libbpf.map                          |   5 +
 tools/testing/selftests/bpf/prog_tests/btf_dump.c | 524 +++++++++++++
 4 files changed, 1447 insertions(+)

-- 
1.8.3.1

[PATCH v3 bpf-next 1/2] libbpf: BTF dumper support for typed data

[Alan Maguire]

Add a BTF dumper for typed data, so that the user can dump a typed
version of the data provided.

The API is

int btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
                             void *data, size_t byte_sz,
                             const struct btf_dump_type_data_opts *opts);

...where the id is the BTF id of the data pointed to by the "void *"
argument; for example the BTF id of "struct sk_buff" for a
"struct skb *" data pointer.  Options supported are

 - a starting indent level (indent_lvl)
 - a set of boolean options to control dump display, similar to those
   used for BPF helper bpf_snprintf_btf().  Options are
        - compact : omit newlines and other indentation
        - skip_names: omit member names
        - emit_zeroes: show zero-value members

Default output format is identical to that dumped by bpf_snprintf_btf(),
for example a "struct sk_buff" representation would look like this:

struct sk_buff){
	(union){
		(struct){
			.next = (struct sk_buff *)0xffffffffffffffff,
			.prev = (struct sk_buff *)0xffffffffffffffff,
		(union){
			.dev = (struct net_device *)0xffffffffffffffff,
			.dev_scratch = (long unsigned int)18446744073709551615,
		},
	},
...

If the data structure is larger than the *byte_sz*
number of bytes that are available in *data*, as much
of the data as possible will be dumped and -E2BIG will
be returned.  This is useful as tracers will sometimes
not be able to capture all of the data associated with
a type; for example a "struct task_struct" is ~16k.
Being able to specify that only a subset is available is
important for such cases.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
---
 tools/lib/bpf/btf.h      |  17 +
 tools/lib/bpf/btf_dump.c | 901 +++++++++++++++++++++++++++++++++++++++++++++++
 tools/lib/bpf/libbpf.map |   5 +
 3 files changed, 923 insertions(+)

diff --git a/tools/lib/bpf/btf.h b/tools/lib/bpf/btf.h
index b54f1c3..234aa97 100644
--- a/tools/lib/bpf/btf.h
+++ b/tools/lib/bpf/btf.h
@@ -184,6 +184,23 @@ struct btf_dump_emit_type_decl_opts {
 btf_dump__emit_type_decl(struct btf_dump *d, __u32 id,
 			 const struct btf_dump_emit_type_decl_opts *opts);
 
+
+struct btf_dump_type_data_opts {
+	/* size of this struct, for forward/backward compatibility */
+	size_t sz;
+	int indent_level;
+	/* below match "show" flags for bpf_show_snprintf() */
+	bool compact;		/* no newlines/tabs */
+	bool skip_names;	/* skip member/type names */
+	bool emit_zeroes;	/* show 0-valued fields */
+};
+#define btf_dump_type_data_opts__last_field emit_zeroes
+
+LIBBPF_API int
+btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
+			 void *data, size_t byte_sz,
+			 const struct btf_dump_type_data_opts *opts);
+
 /*
  * A set of helpers for easier BTF types handling
  */
diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
index 5e2809d..27baa6a 100644
--- a/tools/lib/bpf/btf_dump.c
+++ b/tools/lib/bpf/btf_dump.c
@@ -10,6 +10,8 @@
 #include <stddef.h>
 #include <stdlib.h>
 #include <string.h>
+#include <ctype.h>
+#include <endian.h>
 #include <errno.h>
 #include <linux/err.h>
 #include <linux/btf.h>
@@ -19,6 +21,13 @@
 #include "libbpf.h"
 #include "libbpf_internal.h"
 
+#define BITS_PER_BYTE			8
+#define BITS_PER_U128			128
+#define BITS_PER_BYTE_MASK		(BITS_PER_BYTE - 1)
+#define BITS_PER_BYTE_MASKED(bits)	((bits) & BITS_PER_BYTE_MASK)
+#define BITS_ROUNDDOWN_BYTES(bits)	((bits) / 8)
+#define BITS_ROUNDUP_BYTES(bits)	(roundup(bits, 8))
+
 static const char PREFIXES[] = "\t\t\t\t\t\t\t\t\t\t\t\t\t";
 static const size_t PREFIX_CNT = sizeof(PREFIXES) - 1;
 
@@ -53,6 +62,25 @@ struct btf_dump_type_aux_state {
 	__u8 referenced: 1;
 };
 
+/*
+ * Common internal data for BTF type data dump operations.
+ */
+struct btf_dump_data {
+	void *data_end;		/* end of valid data to show */
+	bool compact;
+	bool skip_names;
+	bool emit_zeroes;
+	__u8 indent_lvl;	/* base indent level */
+	/* below are used during iteration */
+	struct {
+		__u8 depth;
+		__u8 array_member:1,
+		     array_terminated:1,
+		     array_ischar:1;
+		__u32 bitfield_size;
+	} state;
+};
+
 struct btf_dump {
 	const struct btf *btf;
 	const struct btf_ext *btf_ext;
@@ -89,6 +117,10 @@ struct btf_dump {
 	 * name occurrences
 	 */
 	struct hashmap *ident_names;
+	/*
+	 * data for typed display.
+	 */
+	struct btf_dump_data data;
 };
 
 static size_t str_hash_fn(const void *key, void *ctx)
@@ -1392,6 +1424,91 @@ static void btf_dump_emit_type_chain(struct btf_dump *d,
 	btf_dump_emit_name(d, fname, last_was_ptr);
 }
 
+/* show type name as [.fname =] (type_name) */
+static void btf_dump_emit_type_name(struct btf_dump *d, __u32 id,
+				    const char *fname, int lvl, bool toplevel)
+{
+
+	const struct btf_type *t, *child;
+	const char *name;
+	__u16 kind;
+
+	/* for array members, we don't bother emitting type name for each
+	 * member to avoid the redundancy of
+	 * .name = (char[])[(char)'f',(char)'o',(char)'o',]
+	 */
+	if (d->data.state.array_member)
+		return;
+
+	t = btf__type_by_id(d->btf, id);
+	kind = btf_kind(t);
+
+	/* avoid type name specification for variable/section; it will be done
+	 * for the associated variable value(s).  Also skip for function
+	 * prototypes.
+	 */
+	switch (kind) {
+	case BTF_KIND_VAR:
+	case BTF_KIND_DATASEC:
+	case BTF_KIND_FUNC_PROTO:
+		return;
+	default:
+		break;
+	}
+
+	if (toplevel) {
+		if (fname && strlen(fname) > 0)
+			btf_dump_printf(d, ".%s = ", fname);
+		btf_dump_printf(d, "(");
+	}
+
+	switch (kind) {
+	case BTF_KIND_INT:
+		name = btf_name_of(d, t->name_off);
+		btf_dump_printf(d, "%s", name);
+		break;
+	case BTF_KIND_STRUCT:
+	case BTF_KIND_UNION:
+		name = btf_dump_type_name(d, id);
+		btf_dump_printf(d, "%s%s%s",
+				btf_is_struct(t) ? "struct" : "union",
+				strlen(name) > 0 ? " " : "",
+				name);
+		break;
+	case BTF_KIND_ENUM:
+		btf_dump_emit_enum_fwd(d, id, t);
+		break;
+	case BTF_KIND_TYPEDEF:
+		btf_dump_printf(d, "%s", btf_dump_ident_name(d, id));
+		break;
+	case BTF_KIND_VOLATILE:
+	case BTF_KIND_CONST:
+	case BTF_KIND_RESTRICT:
+		/* modifiers are omitted from the cast to save space */
+		btf_dump_emit_type_name(d, t->type, NULL, lvl, false);
+		break;
+	case BTF_KIND_PTR:
+		btf_dump_emit_type_name(d, t->type, NULL, lvl, false);
+		child = btf__type_by_id(d->btf, t->type);
+		btf_dump_printf(d,
+				btf_kind(child) == BTF_KIND_PTR ? "*" : " *");
+		break;
+	case BTF_KIND_ARRAY: {
+		const struct btf_array *a = btf_array(t);
+
+		btf_dump_emit_type_name(d, a->type, NULL, lvl, false);
+		btf_dump_printf(d, "[]");
+		break;
+	}
+	default:
+		pr_warn("unexpected type when emitting type name, kind %u, id:[%u]\n",
+			kind, id);
+		break;
+	}
+	if (toplevel)
+		btf_dump_printf(d, ")");
+}
+
 /* return number of duplicates (occurrences) of a given name */
 static size_t btf_dump_name_dups(struct btf_dump *d, struct hashmap *name_map,
 				 const char *orig_name)
@@ -1442,3 +1559,787 @@ static const char *btf_dump_ident_name(struct btf_dump *d, __u32 id)
 {
 	return btf_dump_resolve_name(d, id, d->ident_names);
 }
+
+static int btf_dump_dump_type_data(struct btf_dump *d,
+				   const char *fname,
+				   const struct btf_type *t,
+				   __u32 id,
+				   void *data,
+				   __u8 bits_offset);
+
+static const char *btf_dump_data_newline(struct btf_dump *d)
+{
+	return d->data.compact ? "" : "\n";
+}
+
+static const char *btf_dump_data_delim(struct btf_dump *d)
+{
+	if (d->data.state.depth == 0)
+		return "";
+
+	return ",";
+}
+
+static const char *btf_dump_data_pfx(struct btf_dump *d)
+{
+	int lvl = d->data.indent_lvl + d->data.state.depth;
+
+	if (d->data.compact)
+		lvl = 0;
+	return pfx(lvl);
+}
+
+/* A macro is used here as btf_type_value[s]() appends format specifiers
+ * to the format specifier passed in; these do the work of appending
+ * delimiters etc while the caller simply has to specify the type values
+ * in the format specifier + value(s).
+ */
+#define btf_dump_type_values(d, fmt, ...)				\
+	btf_dump_printf(d, fmt "%s%s",					\
+			__VA_ARGS__,					\
+			btf_dump_data_delim(d),				\
+			btf_dump_data_newline(d))
+
+static int btf_dump_df_data(struct btf_dump *d,
+			    const struct btf_type *t,
+			    __u32 id,
+			    void *data)
+{
+	btf_dump_printf(d, "<unsupported kind:%u>",
+			BTF_INFO_KIND(t->info));
+	return -ENOTSUP;
+}
+
+static void btf_dump_int128(struct btf_dump *d,
+			    const struct btf_type *t,
+			    void *data)
+{
+	/* data points to a __int128 number.
+	 * Suppose
+	 *	int128_num = *(__int128 *)data;
+	 * The below formulas shows what upper_num and lower_num represents:
+	 *     upper_num = int128_num >> 64;
+	 *     lower_num = int128_num & 0xffffffffFFFFFFFFULL;
+	 */
+	__u64 upper_num, lower_num;
+
+#ifdef __BIG_ENDIAN_BITFIELD
+	upper_num = *(__u64 *)data;
+	lower_num = *(__u64 *)(data + 8);
+#else
+	upper_num = *(__u64 *)(data + 8);
+	lower_num = *(__u64 *)data;
+#endif
+	if (upper_num == 0)
+		btf_dump_type_values(d, "0x%llx", (long long)lower_num);
+	else
+		btf_dump_type_values(d, "0x%llx%016llx", (long long)upper_num,
+				     (long long)lower_num);
+}
+
+static void btf_int128_shift(__u64 *print_num, __u16 left_shift_bits,
+			     __u16 right_shift_bits)
+{
+	__u64 upper_num, lower_num;
+
+#ifdef __BIG_ENDIAN_BITFIELD
+	upper_num = print_num[0];
+	lower_num = print_num[1];
+#else
+	upper_num = print_num[1];
+	lower_num = print_num[0];
+#endif
+
+	/* shake out un-needed bits by shift/or operations */
+	if (left_shift_bits >= 64) {
+		upper_num = lower_num << (left_shift_bits - 64);
+		lower_num = 0;
+	} else {
+		upper_num = (upper_num << left_shift_bits) |
+			    (lower_num >> (64 - left_shift_bits));
+		lower_num = lower_num << left_shift_bits;
+	}
+
+	if (right_shift_bits >= 64) {
+		lower_num = upper_num >> (right_shift_bits - 64);
+		upper_num = 0;
+	} else {
+		lower_num = (lower_num >> right_shift_bits) |
+			    (upper_num << (64 - right_shift_bits));
+		upper_num = upper_num >> right_shift_bits;
+	}
+
+#ifdef __BIG_ENDIAN_BITFIELD
+	print_num[0] = upper_num;
+	print_num[1] = lower_num;
+#else
+	print_num[0] = lower_num;
+	print_num[1] = upper_num;
+#endif
+}
+
+static int btf_dump_bitfield_get_data(struct btf_dump *d,
+				      void *data,
+				      __u8 bits_offset,
+				      __u8 nr_bits,
+				      __u64 *print_num)
+{
+	__u16 left_shift_bits, right_shift_bits;
+	__u8 nr_copy_bytes;
+	__u8 nr_copy_bits;
+
+	nr_copy_bits = nr_bits + bits_offset;
+	nr_copy_bytes = BITS_ROUNDUP_BYTES(nr_copy_bits);
+
+	memcpy(print_num, data, nr_copy_bytes);
+
+#ifdef __BIG_ENDIAN_BITFIELD
+	left_shift_bits = bits_offset;
+#else
+	left_shift_bits = BITS_PER_U128 - nr_copy_bits;
+#endif
+	right_shift_bits = BITS_PER_U128 - nr_bits;
+
+	btf_int128_shift(print_num, left_shift_bits, right_shift_bits);
+
+	return 0;
+}
+
+static int btf_dump_bitfield_data(struct btf_dump *d,
+				  const struct btf_type *t,
+				  void *data,
+				  __u8 bits_offset,
+				  __u8 nr_bits)
+{
+	__u64 print_num[2] = {};
+
+	btf_dump_bitfield_get_data(d, data, bits_offset, nr_bits, print_num);
+	btf_dump_int128(d, t, print_num);
+
+	return 0;
+}
+
+static int btf_dump_int_bits(struct btf_dump *d,
+			     const struct btf_type *t,
+			     void *data,
+			     __u8 bits_offset)
+{
+	__u8 nr_bits = d->data.state.bitfield_size ?: btf_int_bits(t);
+	__u8 total_bits_offset;
+
+	/*
+	 * bits_offset is at most 7.
+	 * BTF_INT_OFFSET() cannot exceed 128 bits.
+	 */
+	total_bits_offset = bits_offset + btf_int_offset(t);
+	data += BITS_ROUNDDOWN_BYTES(total_bits_offset);
+	bits_offset = BITS_PER_BYTE_MASKED(total_bits_offset);
+	return btf_dump_bitfield_data(d, t, data, bits_offset, nr_bits);
+}
+
+static int btf_dump_int_bits_check_zero(struct btf_dump *d,
+					const struct btf_type *t,
+					void *data,
+					__u8 bits_offset)
+{
+	__u64 print_num[2], zero[2] = { };
+	__u8 nr_bits = d->data.state.bitfield_size ?: btf_int_bits(t);
+	__u8 total_bits_offset;
+
+	total_bits_offset = bits_offset + btf_int_offset(t);
+	data += BITS_ROUNDDOWN_BYTES(total_bits_offset);
+	bits_offset = BITS_PER_BYTE_MASKED(total_bits_offset);
+	btf_dump_bitfield_get_data(d, data, bits_offset, nr_bits,
+				   (__u64 *)&print_num);
+	if (memcmp(print_num, zero, sizeof(zero)) == 0)
+		return -ENODATA;
+	return 0;
+}
+
+static int btf_dump_int_check_zero(struct btf_dump *d,
+				const struct btf_type *t,
+				void *data,
+				__u8 bits_offset)
+{
+	__u8 encoding = btf_int_encoding(t);
+	bool sign = encoding & BTF_INT_SIGNED;
+	__u8 nr_bits = btf_int_bits(t);
+	bool zero = false;
+
+	if (bits_offset || btf_int_offset(t) ||
+	    BITS_PER_BYTE_MASKED(nr_bits))
+		return btf_dump_int_bits_check_zero(d, t, data, bits_offset);
+
+	switch (nr_bits) {
+	case 128:
+		zero = sign ? (*(__int128 *)data) == 0 :
+			      (*(unsigned __int128 *)data) == 0;
+		break;
+	case 64:
+		zero = sign ? (*(__s64 *)data) == 0 :
+			      (*(__u64 *)data) == 0;
+		break;
+	case 32:
+		zero = sign ? (*(__s32 *)data) == 0 :
+			      (*(__u32 *)data) == 0;
+		break;
+	case 16:
+		zero = sign ? (*(__s16 *)data) == 0 :
+			      (*(__u16 *)data) == 0;
+		break;
+	case 8:
+		zero = sign ? (*(__s8 *)data) == 0 :
+			      (*(__u8 *)data) == 0;
+		break;
+	default:
+		break;
+	}
+	if (zero)
+		return -ENODATA;
+	return 0;
+}
+
+static int btf_dump_int_data(struct btf_dump *d,
+			     const struct btf_type *t,
+			     __u32 type_id,
+			     void *data,
+			     __u8 bits_offset)
+{
+	__u8 encoding = btf_int_encoding(t);
+	bool sign = encoding & BTF_INT_SIGNED;
+	__u8 nr_bits = btf_int_bits(t);
+
+	if (bits_offset || btf_int_offset(t) ||
+	    BITS_PER_BYTE_MASKED(nr_bits))
+		return btf_dump_int_bits(d, t, data, bits_offset);
+
+	switch (nr_bits) {
+	case 128:
+		btf_dump_int128(d, t, data);
+		break;
+	case 64:
+		if (sign)
+			btf_dump_type_values(d, "%lld", *(long long *)data);
+		else
+			btf_dump_type_values(d, "%llu",
+					     *(unsigned long long *)data);
+		break;
+	case 32:
+		if (sign)
+			btf_dump_type_values(d, "%d", *(__s32 *)data);
+		else
+			btf_dump_type_values(d, "%u", *(__u32 *)data);
+		break;
+	case 16:
+		if (sign)
+			btf_dump_type_values(d, "%d", *(__s16 *)data);
+		else
+			btf_dump_type_values(d, "%u", *(__u16 *)data);
+		break;
+	case 8:
+		if (d->data.state.array_ischar) {
+			/* check for null terminator */
+			if (d->data.state.array_terminated)
+				break;
+			if (*(char *)data == '\0') {
+				d->data.state.array_terminated = 1;
+				break;
+			}
+			if (isprint(*(char *)data)) {
+				btf_dump_type_values(d, "'%c'",
+						     *(char *)data);
+				break;
+			}
+		}
+		if (sign)
+			btf_dump_type_values(d, "%d", *(__s8 *)data);
+		else
+			btf_dump_type_values(d, "%u", *(__u8 *)data);
+		break;
+	default:
+		return btf_dump_int_bits(d, t, data, bits_offset);
+	}
+	return 0;
+}
+
+static int btf_dump_var_data(struct btf_dump *d,
+			     const struct btf_type *v,
+			     __u32 id,
+			     void *data)
+{
+	enum btf_func_linkage linkage = btf_var(v)->linkage;
+	const struct btf_type *t;
+	const char *l = "";
+	__u32 type_id;
+
+	switch (linkage) {
+	case BTF_FUNC_STATIC:
+		l = "static ";
+		break;
+	case BTF_FUNC_EXTERN:
+		l = "extern ";
+		break;
+	case BTF_FUNC_GLOBAL:
+	default:
+		break;
+	}
+
+	/* format of output here is [linkage] [type] [varname] = (type)value,
+	 * for example "static int cpu_profile_flip = (int)1"
+	 */
+	btf_dump_printf(d, "%s", l);
+	type_id = v->type;
+	t = btf__type_by_id(d->btf, type_id);
+	btf_dump_emit_type_name(d, type_id, NULL, 0, false);
+	btf_dump_printf(d, " %s = ", btf_name_of(d, v->name_off));
+	return btf_dump_dump_type_data(d, NULL,
+				       t, type_id, data, 0);
+}
+
+static int btf_dump_array_data(struct btf_dump *d,
+			       const struct btf_type *t,
+			       __u32 id,
+			       void *data)
+{
+	const struct btf_array *array = btf_array(t);
+	const struct btf_type *elem_type;
+	__u32 i, elem_size = 0, elem_type_id;
+	int array_member;
+
+	elem_type_id = array->type;
+	elem_type = skip_mods_and_typedefs(d->btf, elem_type_id, NULL);
+	if (!elem_type) {
+		pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
+			elem_type_id);
+		return -EINVAL;
+	}
+	elem_size = btf__resolve_size(d->btf, elem_type_id);
+
+	if (elem_type && btf_is_int(elem_type)) {
+		/*
+		 * BTF_INT_CHAR encoding never seems to be set for
+		 * char arrays, so if size is 1 and element is
+		 * printable as a char, we'll do that.
+		 */
+		if (elem_size == 1)
+			d->data.state.array_ischar = true;
+	}
+
+	if (!elem_type)
+		return 0;
+
+	btf_dump_printf(d, "[%s",
+			btf_dump_data_newline(d));
+	d->data.state.depth++;
+
+	/* may be a multidimemsional array, so store current "is array member"
+	 * status so we can restore it correctly later.
+	 */
+	array_member = d->data.state.array_member;
+	d->data.state.array_member = 1;
+	for (i = 0; i < array->nelems && !d->data.state.array_terminated; i++) {
+
+		btf_dump_dump_type_data(d, NULL, elem_type, elem_type_id,
+					  data, 0);
+		data += elem_size;
+	}
+	d->data.state.array_member = array_member;
+	d->data.state.depth--;
+	btf_dump_printf(d, "%s]%s%s",
+			btf_dump_data_pfx(d),
+			btf_dump_data_delim(d),
+			btf_dump_data_newline(d));
+
+	return 0;
+}
+
+static int btf_dump_struct_data(struct btf_dump *d,
+				const struct btf_type *t,
+				__u32 id,
+				void *data)
+{
+	const struct btf_member *member;
+	__u32 i;
+	int err;
+
+	btf_dump_printf(d, "{%s",
+			btf_dump_data_newline(d));
+	d->data.state.depth++;
+	for (i = 0, member = btf_members(t);
+	     i < btf_vlen(t);
+	     i++, member++) {
+		const struct btf_type *member_type;
+		__u32 bytes_offset, member_offset;
+		const char *member_name;
+		__u8 bits8_offset;
+
+		member_type = btf__type_by_id(d->btf, member->type);
+		member_name = btf_name_of(d, member->name_off);
+		member_offset = btf_member_bit_offset(t, i);
+		bytes_offset = BITS_ROUNDDOWN_BYTES(member_offset);
+		bits8_offset = BITS_PER_BYTE_MASKED(member_offset);
+
+		/* btf_int_bits() does not store member bitfield size;
+		 * bitfield size needs to be stored here so int display
+		 * of member can retrieve it.
+		 */
+		d->data.state.bitfield_size =
+			btf_member_bitfield_size(t, i);
+		err = btf_dump_dump_type_data(d,
+					      member_name,
+					      member_type,
+					      member->type,
+					      data + bytes_offset,
+					      bits8_offset);
+		d->data.state.bitfield_size = 0;
+		if (err)
+			return err;
+	}
+	d->data.state.depth--;
+	btf_dump_printf(d, "%s}%s%s",
+			btf_dump_data_pfx(d),
+			btf_dump_data_delim(d),
+			btf_dump_data_newline(d));
+	return err;
+}
+
+static int btf_dump_ptr_data(struct btf_dump *d,
+			      const struct btf_type *t,
+			      __u32 id,
+			      void *data)
+{
+	btf_dump_type_values(d, "%p", *(void **)data);
+	return 0;
+}
+
+static int btf_dump_get_enum_value(const struct btf_type *t,
+				   void *data,
+				   __u32 id,
+				   __s64 *value)
+{
+	switch (t->size) {
+	case 8:
+		*value = *(__s64 *)data;
+		return 0;
+	case 4:
+		*value = *(__s32 *)data;
+		return 0;
+	case 2:
+		*value = *(__s16 *)data;
+		return 0;
+	case 1:
+		*value = *(__s8 *)data;
+	default:
+		pr_warn("unexpected size %d for enum, id:[%u]\n",
+			t->size, id);
+		return -EINVAL;
+	}
+}
+
+static int btf_dump_enum_data(struct btf_dump *d,
+			      const struct btf_type *t,
+			      __u32 id,
+			      void *data)
+{
+	const struct btf_enum *e;
+	__s64 value;
+	int i, err;
+
+	err = btf_dump_get_enum_value(t, data, id, &value);
+	if (err)
+		return err;
+
+	for (i = 0, e = btf_enum(t); i < btf_vlen(t); i++, e++) {
+		if (value != e->val)
+			continue;
+		btf_dump_type_values(d, "%s",
+				     btf_name_of(d, e->name_off));
+		return 0;
+	}
+
+	btf_dump_type_values(d, "%d", value);
+	return 0;
+}
+
+static int btf_dump_datasec_data(struct btf_dump *d,
+				 const struct btf_type *t,
+				 __u32 id,
+				 void *data)
+{
+	const struct btf_var_secinfo *vsi;
+	const struct btf_type *var;
+	__u32 i;
+	int err;
+
+	btf_dump_type_values(d, "SEC(\"%s\") ",
+			     btf_name_of(d, t->name_off));
+	for (i = 0, vsi = btf_var_secinfos(t);
+	     i < btf_vlen(t);
+	     i++, vsi++) {
+		var = btf__type_by_id(d->btf, vsi->type);
+		err = btf_dump_dump_type_data(d, NULL, var,
+					      vsi->type,
+					      data + vsi->offset,
+					      0);
+		if (err)
+			return err;
+		btf_dump_printf(d, ";");
+	}
+	return 0;
+}
+
+static int btf_dump_type_data_check_overflow(struct btf_dump *d,
+					     const struct btf_type *t,
+					     __u32 id,
+					     void *data,
+					     __u8 bits_offset)
+{
+	__s64 size;
+
+	/* Only do overflow checking for base types; we do not want to
+	 * avoid showing part of a struct, union or array, even if we
+	 * do not have enough data to show the full object.  By
+	 * restricting overflow checking to base types we can ensure
+	 * that partial display succeeds, while avoiding overflowing
+	 * and using bogus data for display.
+	 */
+	t = skip_mods_and_typedefs(d->btf, id, NULL);
+	if (!t) {
+		pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
+			id);
+		return -EINVAL;
+	}
+
+	switch (BTF_INFO_KIND(t->info)) {
+	case BTF_KIND_INT:
+	case BTF_KIND_PTR:
+	case BTF_KIND_ENUM:
+		size = btf__resolve_size(d->btf, id);
+		if (size < 0) {
+			pr_warn("unexpected size [%llu] for id [%u]\n",
+				size, id);
+			return -EINVAL;
+		}
+		if (data + (bits_offset >> 3) + size > d->data.data_end)
+			return -E2BIG;
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static int btf_dump_type_data_check_zero(struct btf_dump *d,
+					 const struct btf_type *t,
+					 __u32 id,
+					 void *data,
+					 __u8 bits_offset)
+{
+	__s64 value;
+	int i, err;
+
+	/* toplevel exceptions; we show zero values if
+	 * - we ask for them (emit_zeros)
+	 * - if we are at top-level so we see "struct empty { }"
+	 * - or if we are an array member and the array is non-empty and
+	 *   not a char array; we don't want to be in a situation where we
+	 *   have an integer array 0, 1, 0, 1 and only show non-zero values.
+	 *   If the array contains zeroes only, or is a char array starting
+	 *   with a '\0', the array-level check_zero() will prevent showing it;
+	 *   we are concerned with determining zero value at the array member
+	 *   level here.
+	 */
+	if (d->data.emit_zeroes || d->data.state.depth == 0 ||
+	    (d->data.state.array_member && !d->data.state.array_ischar))
+		return 0;
+
+	t = skip_mods_and_typedefs(d->btf, id, NULL);
+	if (!t) {
+		pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
+			id);
+		return -EINVAL;
+	}
+
+
+	switch (BTF_INFO_KIND(t->info)) {
+	case BTF_KIND_INT:
+		if (d->data.state.bitfield_size)
+			return btf_dump_int_bits_check_zero(d, t, data,
+							    bits_offset);
+		return btf_dump_int_check_zero(d, t, data, bits_offset);
+	case BTF_KIND_PTR:
+		if (*((void **)data) == NULL)
+			return -ENODATA;
+		return 0;
+	case BTF_KIND_ARRAY: {
+		const struct btf_array *array = btf_array(t);
+		const struct btf_type *elem_type;
+		__u32 elem_type_id, elem_size;
+		bool ischar;
+
+		elem_type_id = array->type;
+		elem_size = btf__resolve_size(d->btf, elem_type_id);
+		elem_type =  btf__type_by_id(d->btf, elem_type_id);
+
+		ischar = btf_is_int(elem_type) && elem_size == 1;
+
+		/* check all elements; if _any_ element is nonzero, all
+		 * of array is displayed.  We make an exception however
+		 * for char arrays where the first element is 0; these
+		 * are considered zeroed also, even if later elements are
+		 * non-zero because the string is terminated.
+		 */
+		for (i = 0; i < array->nelems; i++) {
+			if (i == 0 && ischar && *(char *)data == 0)
+				return -ENODATA;
+			err = btf_dump_type_data_check_zero(d, elem_type,
+							    elem_type_id,
+							    data +
+							    (i * elem_size),
+							    bits_offset);
+			if (err != -ENODATA)
+				return err;
+		}
+		return -ENODATA;
+	}
+	case BTF_KIND_STRUCT:
+	case BTF_KIND_UNION: {
+		const struct btf_member *member;
+
+		/* if any struct/union member is non-zero, the struct/union
+		 * is considered non-zero and dumped.
+		 */
+		for (i = 0, member = btf_members(t);
+		     i < btf_vlen(t);
+		     i++, member++) {
+			const struct btf_type *member_type;
+			__u32 bytes_offset, member_offset;
+			__u8 bits8_offset;
+
+			member_type = btf__type_by_id(d->btf, member->type);
+			member_offset = btf_member_bit_offset(t, i);
+			bytes_offset = BITS_ROUNDDOWN_BYTES(member_offset);
+			bits8_offset = BITS_PER_BYTE_MASKED(member_offset);
+
+			/* btf_int_bits() does not store member bitfield size;
+			 * bitfield size needs to be stored here so int display
+			 * of member can retrieve it.
+			 */
+			d->data.state.bitfield_size =
+				btf_member_bitfield_size(t, i);
+
+			err = btf_dump_type_data_check_zero(d, member_type,
+							    member->type,
+							    data + bytes_offset,
+							    bits8_offset);
+			d->data.state.bitfield_size = 0;
+			if (err != ENODATA)
+				return err;
+		}
+		return -ENODATA;
+	}
+	case BTF_KIND_ENUM:
+		if (btf_dump_get_enum_value(t, data, id, &value))
+			return 0;
+		if (value == 0)
+			return -ENODATA;
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static int btf_dump_dump_type_data(struct btf_dump *d,
+				   const char *fname,
+				   const struct btf_type *t,
+				   __u32 id,
+				   void *data,
+				   __u8 bits_offset)
+{
+	int err;
+
+	err = btf_dump_type_data_check_overflow(d, t, id, data, bits_offset);
+	if (err)
+		return err;
+	err = btf_dump_type_data_check_zero(d, t, id, data, bits_offset);
+	if (err) {
+		/* zeroed data is expected and not an error, so simply skip
+		 * dumping such data.  Record other errors however.
+		 */
+		if (err == -ENODATA)
+			return 0;
+		return err;
+	}
+	btf_dump_printf(d, "%s", btf_dump_data_pfx(d));
+	if (!d->data.skip_names)
+		btf_dump_emit_type_name(d, id, fname, 0, true);
+
+	t = skip_mods_and_typedefs(d->btf, id, NULL);
+	if (!t) {
+		pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
+			id);
+		return -EINVAL;
+	}
+
+	switch (BTF_INFO_KIND(t->info)) {
+	case BTF_KIND_UNKN:
+	case BTF_KIND_FWD:
+	case BTF_KIND_FUNC:
+	case BTF_KIND_FUNC_PROTO:
+		return btf_dump_df_data(d, t, id, data);
+	case BTF_KIND_INT:
+		if (d->data.state.bitfield_size)
+			return btf_dump_bitfield_data(d, t, data,
+						      bits_offset,
+						      d->data.state.bitfield_size);
+		return btf_dump_int_data(d, t, id, data, bits_offset);
+	case BTF_KIND_PTR:
+		return btf_dump_ptr_data(d, t, id, data);
+	case BTF_KIND_ARRAY:
+		return btf_dump_array_data(d, t, id, data);
+	case BTF_KIND_STRUCT:
+	case BTF_KIND_UNION:
+		return btf_dump_struct_data(d, t, id, data);
+	case BTF_KIND_ENUM:
+		return btf_dump_enum_data(d, t, id, data);
+	case BTF_KIND_VAR:
+		return btf_dump_var_data(d, t, id, data);
+	case BTF_KIND_DATASEC:
+		return btf_dump_datasec_data(d, t, id, data);
+	default:
+		pr_warn("unexpected kind [%u] for id [%u]\n",
+			BTF_INFO_KIND(t->info), id);
+		return -EINVAL;
+	}
+}
+
+int btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
+			     void *data, size_t byte_sz,
+			     const struct btf_dump_type_data_opts *opts)
+{
+	const struct btf_type *t;
+	int err;
+
+	if (!OPTS_VALID(opts, btf_dump_type_data_opts))
+		return -EINVAL;
+
+	t = btf__type_by_id(d->btf, id);
+	if (!t)
+		return -ENOENT;
+
+	d->data.data_end = data + byte_sz;
+	d->data.indent_lvl = OPTS_GET(opts, indent_level, 0);
+	d->data.compact = OPTS_GET(opts, compact, false);
+	d->data.skip_names = OPTS_GET(opts, skip_names, false);
+	d->data.emit_zeroes = OPTS_GET(opts, emit_zeroes, false);
+	memset(&d->data.state, 0, sizeof(d->data.state));
+
+	err = btf_dump_dump_type_data(d, NULL, t, id, data, 0);
+
+	/* We reported all the data; return size of data we reported. */
+	if (err == 0)
+		err = btf__resolve_size(d->btf, id);
+
+	memset(&d->data, 0, sizeof(d->data));
+
+	return err;
+}
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
index 0229e01..76cfac6 100644
--- a/tools/lib/bpf/libbpf.map
+++ b/tools/lib/bpf/libbpf.map
@@ -370,3 +370,8 @@ LIBBPF_0.4.0 {
 		bpf_tc_hook_destroy;
 		bpf_tc_query;
 } LIBBPF_0.3.0;
+
+LIBBPF_0.5.0 {
+	global:
+		btf_dump__dump_type_data;
+} LIBBPF_0.4.0;
-- 
1.8.3.1

Re: [PATCH v3 bpf-next 1/2] libbpf: BTF dumper support for typed data

[Andrii Nakryiko]

On Thu, May 27, 2021 at 9:00 AM Alan Maguire <alan.maguire@oracle.com> wrote:
>
> Add a BTF dumper for typed data, so that the user can dump a typed
> version of the data provided.
>
> The API is
>
> int btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
>                              void *data, size_t byte_sz,
>                              const struct btf_dump_type_data_opts *opts);
>
> ...where the id is the BTF id of the data pointed to by the "void *"
> argument; for example the BTF id of "struct sk_buff" for a
> "struct skb *" data pointer.  Options supported are
>
>  - a starting indent level (indent_lvl)
>  - a set of boolean options to control dump display, similar to those
>    used for BPF helper bpf_snprintf_btf().  Options are
>         - compact : omit newlines and other indentation
>         - skip_names: omit member names
>         - emit_zeroes: show zero-value members
>
> Default output format is identical to that dumped by bpf_snprintf_btf(),
> for example a "struct sk_buff" representation would look like this:
>
> struct sk_buff){
>         (union){
>                 (struct){
>                         .next = (struct sk_buff *)0xffffffffffffffff,
>                         .prev = (struct sk_buff *)0xffffffffffffffff,
>                 (union){
>                         .dev = (struct net_device *)0xffffffffffffffff,
>                         .dev_scratch = (long unsigned int)18446744073709551615,
>                 },
>         },
> ...

Is the '\t' as a default indentation filler a requirement? With 8
spaces per tab this gets wide really fast. 2 or 4 spaces might be a
better default? And it probably would be good to let users specify
this as a string (e.g., "indent_str"), which could be anything the
user wants, from empty string to multiple tabs to "..." to whatever
else. If I needed to dump something with this API, I'd probably use 2
spaces, for instance. But I can see how other users might want
something else, so making this customizable seems like a good thing?

>
> If the data structure is larger than the *byte_sz*
> number of bytes that are available in *data*, as much
> of the data as possible will be dumped and -E2BIG will
> be returned.  This is useful as tracers will sometimes
> not be able to capture all of the data associated with
> a type; for example a "struct task_struct" is ~16k.
> Being able to specify that only a subset is available is
> important for such cases.
>
> Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
> ---

I'll do another pass next week, I ran out of time, it's a bit of a big
patch, sorry. But left few comments, please take a look.

>  tools/lib/bpf/btf.h      |  17 +
>  tools/lib/bpf/btf_dump.c | 901 +++++++++++++++++++++++++++++++++++++++++++++++
>  tools/lib/bpf/libbpf.map |   5 +
>  3 files changed, 923 insertions(+)
>
> diff --git a/tools/lib/bpf/btf.h b/tools/lib/bpf/btf.h
> index b54f1c3..234aa97 100644
> --- a/tools/lib/bpf/btf.h
> +++ b/tools/lib/bpf/btf.h
> @@ -184,6 +184,23 @@ struct btf_dump_emit_type_decl_opts {
>  btf_dump__emit_type_decl(struct btf_dump *d, __u32 id,
>                          const struct btf_dump_emit_type_decl_opts *opts);
>
> +
> +struct btf_dump_type_data_opts {
> +       /* size of this struct, for forward/backward compatibility */
> +       size_t sz;
> +       int indent_level;
> +       /* below match "show" flags for bpf_show_snprintf() */
> +       bool compact;           /* no newlines/tabs */
> +       bool skip_names;        /* skip member/type names */
> +       bool emit_zeroes;       /* show 0-valued fields */

please add size_t: 0; as the last member to avoid various issues with
non-zero padding. We've been doing it for a few not perfectly aligned
opts recently and that seems to work good in practice.

> +};
> +#define btf_dump_type_data_opts__last_field emit_zeroes
> +
> +LIBBPF_API int
> +btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
> +                        void *data, size_t byte_sz,

can't get over data vs byte_sz naming inconsistency.... why not data
and data_sz?

also, const void *data?

> +                        const struct btf_dump_type_data_opts *opts);
> +
>  /*
>   * A set of helpers for easier BTF types handling
>   */
> diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
> index 5e2809d..27baa6a 100644
> --- a/tools/lib/bpf/btf_dump.c
> +++ b/tools/lib/bpf/btf_dump.c
> @@ -10,6 +10,8 @@
>  #include <stddef.h>
>  #include <stdlib.h>
>  #include <string.h>
> +#include <ctype.h>
> +#include <endian.h>
>  #include <errno.h>
>  #include <linux/err.h>
>  #include <linux/btf.h>
> @@ -19,6 +21,13 @@
>  #include "libbpf.h"
>  #include "libbpf_internal.h"
>
> +#define BITS_PER_BYTE                  8
> +#define BITS_PER_U128                  128
> +#define BITS_PER_BYTE_MASK             (BITS_PER_BYTE - 1)
> +#define BITS_PER_BYTE_MASKED(bits)     ((bits) & BITS_PER_BYTE_MASK)
> +#define BITS_ROUNDDOWN_BYTES(bits)     ((bits) / 8)
> +#define BITS_ROUNDUP_BYTES(bits)       (roundup(bits, 8))
> +
>  static const char PREFIXES[] = "\t\t\t\t\t\t\t\t\t\t\t\t\t";
>  static const size_t PREFIX_CNT = sizeof(PREFIXES) - 1;
>
> @@ -53,6 +62,25 @@ struct btf_dump_type_aux_state {
>         __u8 referenced: 1;
>  };
>
> +/*
> + * Common internal data for BTF type data dump operations.
> + */
> +struct btf_dump_data {
> +       void *data_end;         /* end of valid data to show */
> +       bool compact;
> +       bool skip_names;
> +       bool emit_zeroes;
> +       __u8 indent_lvl;        /* base indent level */
> +       /* below are used during iteration */
> +       struct {
> +               __u8 depth;
> +               __u8 array_member:1,
> +                    array_terminated:1,
> +                    array_ischar:1;
> +               __u32 bitfield_size;
> +       } state;
> +};
> +
>  struct btf_dump {
>         const struct btf *btf;
>         const struct btf_ext *btf_ext;
> @@ -89,6 +117,10 @@ struct btf_dump {
>          * name occurrences
>          */
>         struct hashmap *ident_names;
> +       /*
> +        * data for typed display.
> +        */
> +       struct btf_dump_data data;

given it's not always used, should this be allocated lazily only if necessary?

>  };
>
>  static size_t str_hash_fn(const void *key, void *ctx)
> @@ -1392,6 +1424,91 @@ static void btf_dump_emit_type_chain(struct btf_dump *d,
>         btf_dump_emit_name(d, fname, last_was_ptr);
>  }
>
> +/* show type name as [.fname =] (type_name) */
> +static void btf_dump_emit_type_name(struct btf_dump *d, __u32 id,
> +                                   const char *fname, int lvl, bool toplevel)

Sorry, I didn't get from your cover letter why
btf_dump__emit_type_decl() doesn't do what you need it to do. Can you
please elaborate with a simple example?

> +{
> +
> +       const struct btf_type *t, *child;
> +       const char *name;
> +       __u16 kind;
> +
> +       /* for array members, we don't bother emitting type name for each
> +        * member to avoid the redundancy of
> +        * .name = (char[])[(char)'f',(char)'o',(char)'o',]
> +        */
> +       if (d->data.state.array_member)
> +               return;
> +
> +       t = btf__type_by_id(d->btf, id);
> +       kind = btf_kind(t);
> +
> +       /* avoid type name specification for variable/section; it will be done
> +        * for the associated variable value(s).  Also skip for function
> +        * prototypes.
> +        */
> +       switch (kind) {
> +       case BTF_KIND_VAR:
> +       case BTF_KIND_DATASEC:
> +       case BTF_KIND_FUNC_PROTO:

hm... why skipping func_proto? seems useful when printing all kinds of
opts struct?

> +               return;
> +       default:
> +               break;
> +       }
> +
> +       if (toplevel) {
> +               if (fname && strlen(fname) > 0)
> +                       btf_dump_printf(d, ".%s = ", fname);
> +               btf_dump_printf(d, "(");
> +       }

".field = " part is not really a type name, can't it be done outside?

> +
> +       switch (kind) {
> +       case BTF_KIND_INT:
> +               name = btf_name_of(d, t->name_off);
> +               btf_dump_printf(d, "%s", name);
> +               break;
> +       case BTF_KIND_STRUCT:
> +       case BTF_KIND_UNION:
> +               name = btf_dump_type_name(d, id);
> +               btf_dump_printf(d, "%s%s%s",
> +                               btf_is_struct(t) ? "struct" : "union",
> +                               strlen(name) > 0 ? " " : "",
> +                               name);
> +               break;
> +       case BTF_KIND_ENUM:
> +               btf_dump_emit_enum_fwd(d, id, t);
> +               break;
> +       case BTF_KIND_TYPEDEF:
> +               btf_dump_printf(d, "%s", btf_dump_ident_name(d, id));
> +               break;
> +       case BTF_KIND_VOLATILE:
> +       case BTF_KIND_CONST:
> +       case BTF_KIND_RESTRICT:
> +               /* modifiers are omitted from the cast to save space */
> +               btf_dump_emit_type_name(d, t->type, NULL, lvl, false);
> +               break;
> +       case BTF_KIND_PTR:
> +               btf_dump_emit_type_name(d, t->type, NULL, lvl, false);
> +               child = btf__type_by_id(d->btf, t->type);
> +               btf_dump_printf(d,
> +                               btf_kind(child) == BTF_KIND_PTR ? "*" : " *");
> +               break;
> +       case BTF_KIND_ARRAY: {
> +               const struct btf_array *a = btf_array(t);
> +
> +               btf_dump_emit_type_name(d, a->type, NULL, lvl, false);
> +               btf_dump_printf(d, "[]");
> +               break;
> +       }
> +       default:
> +               pr_warn("unexpected type when emitting type name, kind %u, id:[%u]\n",
> +                       kind, id);
> +               break;
> +       }
> +       if (toplevel)
> +               btf_dump_printf(d, ")");
> +}
> +
>  /* return number of duplicates (occurrences) of a given name */
>  static size_t btf_dump_name_dups(struct btf_dump *d, struct hashmap *name_map,
>                                  const char *orig_name)
> @@ -1442,3 +1559,787 @@ static const char *btf_dump_ident_name(struct btf_dump *d, __u32 id)
>  {
>         return btf_dump_resolve_name(d, id, d->ident_names);
>  }
> +
> +static int btf_dump_dump_type_data(struct btf_dump *d,
> +                                  const char *fname,
> +                                  const struct btf_type *t,
> +                                  __u32 id,
> +                                  void *data,
> +                                  __u8 bits_offset);
> +
> +static const char *btf_dump_data_newline(struct btf_dump *d)
> +{
> +       return d->data.compact ? "" : "\n";
> +}
> +
> +static const char *btf_dump_data_delim(struct btf_dump *d)
> +{
> +       if (d->data.state.depth == 0)
> +               return "";
> +
> +       return ",";

nit: use ternary like for newline?

> +}
> +
> +static const char *btf_dump_data_pfx(struct btf_dump *d)
> +{
> +       int lvl = d->data.indent_lvl + d->data.state.depth;
> +
> +       if (d->data.compact)
> +               lvl = 0;
> +       return pfx(lvl);
> +}
> +
> +/* A macro is used here as btf_type_value[s]() appends format specifiers
> + * to the format specifier passed in; these do the work of appending
> + * delimiters etc while the caller simply has to specify the type values
> + * in the format specifier + value(s).
> + */
> +#define btf_dump_type_values(d, fmt, ...)                              \
> +       btf_dump_printf(d, fmt "%s%s",                                  \
> +                       __VA_ARGS__,                                    \
> +                       btf_dump_data_delim(d),                         \
> +                       btf_dump_data_newline(d))
> +
> +static int btf_dump_df_data(struct btf_dump *d,

what does "df" stand for?


> +                           const struct btf_type *t,
> +                           __u32 id,
> +                           void *data)
> +{
> +       btf_dump_printf(d, "<unsupported kind:%u>",
> +                       BTF_INFO_KIND(t->info));
> +       return -ENOTSUP;
> +}
> +
> +static void btf_dump_int128(struct btf_dump *d,
> +                           const struct btf_type *t,
> +                           void *data)
> +{
> +       /* data points to a __int128 number.
> +        * Suppose
> +        *      int128_num = *(__int128 *)data;
> +        * The below formulas shows what upper_num and lower_num represents:
> +        *     upper_num = int128_num >> 64;
> +        *     lower_num = int128_num & 0xffffffffFFFFFFFFULL;
> +        */
> +       __u64 upper_num, lower_num;
> +
> +#ifdef __BIG_ENDIAN_BITFIELD
> +       upper_num = *(__u64 *)data;
> +       lower_num = *(__u64 *)(data + 8);
> +#else
> +       upper_num = *(__u64 *)(data + 8);
> +       lower_num = *(__u64 *)data;
> +#endif
> +       if (upper_num == 0)
> +               btf_dump_type_values(d, "0x%llx", (long long)lower_num);
> +       else
> +               btf_dump_type_values(d, "0x%llx%016llx", (long long)upper_num,
> +                                    (long long)lower_num);
> +}
> +
> +static void btf_int128_shift(__u64 *print_num, __u16 left_shift_bits,
> +                            __u16 right_shift_bits)
> +{
> +       __u64 upper_num, lower_num;
> +
> +#ifdef __BIG_ENDIAN_BITFIELD
> +       upper_num = print_num[0];
> +       lower_num = print_num[1];
> +#else
> +       upper_num = print_num[1];
> +       lower_num = print_num[0];
> +#endif
> +
> +       /* shake out un-needed bits by shift/or operations */
> +       if (left_shift_bits >= 64) {
> +               upper_num = lower_num << (left_shift_bits - 64);
> +               lower_num = 0;
> +       } else {
> +               upper_num = (upper_num << left_shift_bits) |
> +                           (lower_num >> (64 - left_shift_bits));
> +               lower_num = lower_num << left_shift_bits;
> +       }
> +
> +       if (right_shift_bits >= 64) {
> +               lower_num = upper_num >> (right_shift_bits - 64);
> +               upper_num = 0;
> +       } else {
> +               lower_num = (lower_num >> right_shift_bits) |
> +                           (upper_num << (64 - right_shift_bits));
> +               upper_num = upper_num >> right_shift_bits;
> +       }
> +
> +#ifdef __BIG_ENDIAN_BITFIELD
> +       print_num[0] = upper_num;
> +       print_num[1] = lower_num;
> +#else
> +       print_num[0] = lower_num;
> +       print_num[1] = upper_num;
> +#endif
> +}
> +
> +static int btf_dump_bitfield_get_data(struct btf_dump *d,
> +                                     void *data,
> +                                     __u8 bits_offset,
> +                                     __u8 nr_bits,
> +                                     __u64 *print_num)
> +{
> +       __u16 left_shift_bits, right_shift_bits;
> +       __u8 nr_copy_bytes;
> +       __u8 nr_copy_bits;
> +
> +       nr_copy_bits = nr_bits + bits_offset;
> +       nr_copy_bytes = BITS_ROUNDUP_BYTES(nr_copy_bits);
> +
> +       memcpy(print_num, data, nr_copy_bytes);
> +
> +#ifdef __BIG_ENDIAN_BITFIELD
> +       left_shift_bits = bits_offset;
> +#else
> +       left_shift_bits = BITS_PER_U128 - nr_copy_bits;
> +#endif
> +       right_shift_bits = BITS_PER_U128 - nr_bits;
> +
> +       btf_int128_shift(print_num, left_shift_bits, right_shift_bits);
> +
> +       return 0;
> +}
> +
> +static int btf_dump_bitfield_data(struct btf_dump *d,
> +                                 const struct btf_type *t,
> +                                 void *data,
> +                                 __u8 bits_offset,
> +                                 __u8 nr_bits)
> +{
> +       __u64 print_num[2] = {};

= {} is unnecessary as you always initialize it anyways?
> +
> +       btf_dump_bitfield_get_data(d, data, bits_offset, nr_bits, print_num);
> +       btf_dump_int128(d, t, print_num);
> +
> +       return 0;
> +}
> +
> +static int btf_dump_int_bits(struct btf_dump *d,
> +                            const struct btf_type *t,
> +                            void *data,
> +                            __u8 bits_offset)
> +{
> +       __u8 nr_bits = d->data.state.bitfield_size ?: btf_int_bits(t);
> +       __u8 total_bits_offset;
> +
> +       /*
> +        * bits_offset is at most 7.
> +        * BTF_INT_OFFSET() cannot exceed 128 bits.
> +        */
> +       total_bits_offset = bits_offset + btf_int_offset(t);

BTF_INT_OFFSET() is long deprecated, pahole doesn't generate such
INTs, many versions of Clang don't generate such int, so I don't think
we should bother supporting it.

> +       data += BITS_ROUNDDOWN_BYTES(total_bits_offset);
> +       bits_offset = BITS_PER_BYTE_MASKED(total_bits_offset);
> +       return btf_dump_bitfield_data(d, t, data, bits_offset, nr_bits);
> +}
> +
> +static int btf_dump_int_bits_check_zero(struct btf_dump *d,
> +                                       const struct btf_type *t,
> +                                       void *data,
> +                                       __u8 bits_offset)
> +{
> +       __u64 print_num[2], zero[2] = { };
> +       __u8 nr_bits = d->data.state.bitfield_size ?: btf_int_bits(t);
> +       __u8 total_bits_offset;
> +
> +       total_bits_offset = bits_offset + btf_int_offset(t);
> +       data += BITS_ROUNDDOWN_BYTES(total_bits_offset);
> +       bits_offset = BITS_PER_BYTE_MASKED(total_bits_offset);
> +       btf_dump_bitfield_get_data(d, data, bits_offset, nr_bits,
> +                                  (__u64 *)&print_num);
> +       if (memcmp(print_num, zero, sizeof(zero)) == 0)
> +               return -ENODATA;
> +       return 0;
> +}
> +
> +static int btf_dump_int_check_zero(struct btf_dump *d,
> +                               const struct btf_type *t,
> +                               void *data,
> +                               __u8 bits_offset)
> +{
> +       __u8 encoding = btf_int_encoding(t);
> +       bool sign = encoding & BTF_INT_SIGNED;
> +       __u8 nr_bits = btf_int_bits(t);
> +       bool zero = false;
> +
> +       if (bits_offset || btf_int_offset(t) ||
> +           BITS_PER_BYTE_MASKED(nr_bits))
> +               return btf_dump_int_bits_check_zero(d, t, data, bits_offset);
> +
> +       switch (nr_bits) {
> +       case 128:
> +               zero = sign ? (*(__int128 *)data) == 0 :
> +                             (*(unsigned __int128 *)data) == 0;
> +               break;
> +       case 64:
> +               zero = sign ? (*(__s64 *)data) == 0 :
> +                             (*(__u64 *)data) == 0;
> +               break;
> +       case 32:
> +               zero = sign ? (*(__s32 *)data) == 0 :
> +                             (*(__u32 *)data) == 0;
> +               break;
> +       case 16:
> +               zero = sign ? (*(__s16 *)data) == 0 :
> +                             (*(__u16 *)data) == 0;
> +               break;
> +       case 8:
> +               zero = sign ? (*(__s8 *)data) == 0 :
> +                             (*(__u8 *)data) == 0;

for zero checks sign doesn't matter

> +               break;
> +       default:
> +               break;
> +       }
> +       if (zero)
> +               return -ENODATA;
> +       return 0;
> +}
> +
> +static int btf_dump_int_data(struct btf_dump *d,
> +                            const struct btf_type *t,
> +                            __u32 type_id,
> +                            void *data,
> +                            __u8 bits_offset)
> +{
> +       __u8 encoding = btf_int_encoding(t);
> +       bool sign = encoding & BTF_INT_SIGNED;
> +       __u8 nr_bits = btf_int_bits(t);
> +
> +       if (bits_offset || btf_int_offset(t) ||
> +           BITS_PER_BYTE_MASKED(nr_bits))
> +               return btf_dump_int_bits(d, t, data, bits_offset);
> +
> +       switch (nr_bits) {
> +       case 128:
> +               btf_dump_int128(d, t, data);
> +               break;
> +       case 64:
> +               if (sign)
> +                       btf_dump_type_values(d, "%lld", *(long long *)data);
> +               else
> +                       btf_dump_type_values(d, "%llu",
> +                                            *(unsigned long long *)data);
> +               break;
> +       case 32:
> +               if (sign)
> +                       btf_dump_type_values(d, "%d", *(__s32 *)data);
> +               else
> +                       btf_dump_type_values(d, "%u", *(__u32 *)data);
> +               break;
> +       case 16:
> +               if (sign)
> +                       btf_dump_type_values(d, "%d", *(__s16 *)data);
> +               else
> +                       btf_dump_type_values(d, "%u", *(__u16 *)data);
> +               break;
> +       case 8:
> +               if (d->data.state.array_ischar) {
> +                       /* check for null terminator */
> +                       if (d->data.state.array_terminated)
> +                               break;
> +                       if (*(char *)data == '\0') {
> +                               d->data.state.array_terminated = 1;
> +                               break;
> +                       }
> +                       if (isprint(*(char *)data)) {
> +                               btf_dump_type_values(d, "'%c'",
> +                                                    *(char *)data);
> +                               break;
> +                       }
> +               }
> +               if (sign)
> +                       btf_dump_type_values(d, "%d", *(__s8 *)data);
> +               else
> +                       btf_dump_type_values(d, "%u", *(__u8 *)data);
> +               break;
> +       default:
> +               return btf_dump_int_bits(d, t, data, bits_offset);

error out in such case?

> +       }
> +       return 0;
> +}
> +
> +static int btf_dump_var_data(struct btf_dump *d,
> +                            const struct btf_type *v,
> +                            __u32 id,
> +                            void *data)
> +{
> +       enum btf_func_linkage linkage = btf_var(v)->linkage;
> +       const struct btf_type *t;
> +       const char *l = "";
> +       __u32 type_id;
> +
> +       switch (linkage) {
> +       case BTF_FUNC_STATIC:
> +               l = "static ";
> +               break;
> +       case BTF_FUNC_EXTERN:
> +               l = "extern ";
> +               break;
> +       case BTF_FUNC_GLOBAL:
> +       default:
> +               break;
> +       }
> +
> +       /* format of output here is [linkage] [type] [varname] = (type)value,
> +        * for example "static int cpu_profile_flip = (int)1"
> +        */
> +       btf_dump_printf(d, "%s", l);
> +       type_id = v->type;
> +       t = btf__type_by_id(d->btf, type_id);
> +       btf_dump_emit_type_name(d, type_id, NULL, 0, false);
> +       btf_dump_printf(d, " %s = ", btf_name_of(d, v->name_off));
> +       return btf_dump_dump_type_data(d, NULL,
> +                                      t, type_id, data, 0);
> +}
> +
> +static int btf_dump_array_data(struct btf_dump *d,
> +                              const struct btf_type *t,
> +                              __u32 id,
> +                              void *data)
> +{
> +       const struct btf_array *array = btf_array(t);
> +       const struct btf_type *elem_type;
> +       __u32 i, elem_size = 0, elem_type_id;
> +       int array_member;
> +
> +       elem_type_id = array->type;
> +       elem_type = skip_mods_and_typedefs(d->btf, elem_type_id, NULL);
> +       if (!elem_type) {

can't really happen in well-formed BTF, I don't think libbpf ever
checks this for NULL

> +               pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
> +                       elem_type_id);
> +               return -EINVAL;
> +       }
> +       elem_size = btf__resolve_size(d->btf, elem_type_id);

this can actually fail and return -Exxx, so I'd check <= 0 case here,
otherwise it might get confusing later

> +
> +       if (elem_type && btf_is_int(elem_type)) {
> +               /*
> +                * BTF_INT_CHAR encoding never seems to be set for
> +                * char arrays, so if size is 1 and element is
> +                * printable as a char, we'll do that.
> +                */
> +               if (elem_size == 1)
> +                       d->data.state.array_ischar = true;
> +       }
> +
> +       if (!elem_type)
> +               return 0;

already checked above

> +
> +       btf_dump_printf(d, "[%s",
> +                       btf_dump_data_newline(d));

nit: keep on single line, it's not that long

> +       d->data.state.depth++;
> +
> +       /* may be a multidimemsional array, so store current "is array member"

typo: multidimensional

> +        * status so we can restore it correctly later.
> +        */
> +       array_member = d->data.state.array_member;
> +       d->data.state.array_member = 1;
> +       for (i = 0; i < array->nelems && !d->data.state.array_terminated; i++) {
> +
> +               btf_dump_dump_type_data(d, NULL, elem_type, elem_type_id,
> +                                         data, 0);
> +               data += elem_size;
> +       }
> +       d->data.state.array_member = array_member;
> +       d->data.state.depth--;
> +       btf_dump_printf(d, "%s]%s%s",
> +                       btf_dump_data_pfx(d),
> +                       btf_dump_data_delim(d),
> +                       btf_dump_data_newline(d));
> +
> +       return 0;
> +}
> +
> +static int btf_dump_struct_data(struct btf_dump *d,
> +                               const struct btf_type *t,
> +                               __u32 id,
> +                               void *data)
> +{
> +       const struct btf_member *member;
> +       __u32 i;
> +       int err;
> +
> +       btf_dump_printf(d, "{%s",
> +                       btf_dump_data_newline(d));
> +       d->data.state.depth++;
> +       for (i = 0, member = btf_members(t);
> +            i < btf_vlen(t);
> +            i++, member++) {
> +               const struct btf_type *member_type;
> +               __u32 bytes_offset, member_offset;
> +               const char *member_name;
> +               __u8 bits8_offset;
> +
> +               member_type = btf__type_by_id(d->btf, member->type);
> +               member_name = btf_name_of(d, member->name_off);
> +               member_offset = btf_member_bit_offset(t, i);
> +               bytes_offset = BITS_ROUNDDOWN_BYTES(member_offset);
> +               bits8_offset = BITS_PER_BYTE_MASKED(member_offset);
> +
> +               /* btf_int_bits() does not store member bitfield size;
> +                * bitfield size needs to be stored here so int display
> +                * of member can retrieve it.
> +                */
> +               d->data.state.bitfield_size =
> +                       btf_member_bitfield_size(t, i);
> +               err = btf_dump_dump_type_data(d,
> +                                             member_name,
> +                                             member_type,
> +                                             member->type,
> +                                             data + bytes_offset,
> +                                             bits8_offset);
> +               d->data.state.bitfield_size = 0;
> +               if (err)
> +                       return err;
> +       }
> +       d->data.state.depth--;
> +       btf_dump_printf(d, "%s}%s%s",
> +                       btf_dump_data_pfx(d),
> +                       btf_dump_data_delim(d),
> +                       btf_dump_data_newline(d));
> +       return err;
> +}
> +
> +static int btf_dump_ptr_data(struct btf_dump *d,
> +                             const struct btf_type *t,
> +                             __u32 id,
> +                             void *data)
> +{
> +       btf_dump_type_values(d, "%p", *(void **)data);
> +       return 0;
> +}
> +
> +static int btf_dump_get_enum_value(const struct btf_type *t,
> +                                  void *data,
> +                                  __u32 id,
> +                                  __s64 *value)
> +{
> +       switch (t->size) {
> +       case 8:
> +               *value = *(__s64 *)data;
> +               return 0;
> +       case 4:
> +               *value = *(__s32 *)data;
> +               return 0;
> +       case 2:
> +               *value = *(__s16 *)data;
> +               return 0;
> +       case 1:
> +               *value = *(__s8 *)data;
> +       default:
> +               pr_warn("unexpected size %d for enum, id:[%u]\n",
> +                       t->size, id);
> +               return -EINVAL;
> +       }
> +}
> +
> +static int btf_dump_enum_data(struct btf_dump *d,
> +                             const struct btf_type *t,
> +                             __u32 id,
> +                             void *data)
> +{
> +       const struct btf_enum *e;
> +       __s64 value;
> +       int i, err;
> +
> +       err = btf_dump_get_enum_value(t, data, id, &value);
> +       if (err)
> +               return err;
> +
> +       for (i = 0, e = btf_enum(t); i < btf_vlen(t); i++, e++) {
> +               if (value != e->val)
> +                       continue;
> +               btf_dump_type_values(d, "%s",
> +                                    btf_name_of(d, e->name_off));
> +               return 0;
> +       }
> +
> +       btf_dump_type_values(d, "%d", value);
> +       return 0;
> +}
> +
> +static int btf_dump_datasec_data(struct btf_dump *d,
> +                                const struct btf_type *t,
> +                                __u32 id,
> +                                void *data)
> +{
> +       const struct btf_var_secinfo *vsi;
> +       const struct btf_type *var;
> +       __u32 i;
> +       int err;
> +
> +       btf_dump_type_values(d, "SEC(\"%s\") ",
> +                            btf_name_of(d, t->name_off));
> +       for (i = 0, vsi = btf_var_secinfos(t);
> +            i < btf_vlen(t);
> +            i++, vsi++) {
> +               var = btf__type_by_id(d->btf, vsi->type);
> +               err = btf_dump_dump_type_data(d, NULL, var,
> +                                             vsi->type,
> +                                             data + vsi->offset,
> +                                             0);
> +               if (err)
> +                       return err;
> +               btf_dump_printf(d, ";");
> +       }
> +       return 0;
> +}
> +
> +static int btf_dump_type_data_check_overflow(struct btf_dump *d,
> +                                            const struct btf_type *t,
> +                                            __u32 id,
> +                                            void *data,
> +                                            __u8 bits_offset)
> +{
> +       __s64 size;
> +
> +       /* Only do overflow checking for base types; we do not want to
> +        * avoid showing part of a struct, union or array, even if we
> +        * do not have enough data to show the full object.  By
> +        * restricting overflow checking to base types we can ensure
> +        * that partial display succeeds, while avoiding overflowing
> +        * and using bogus data for display.
> +        */
> +       t = skip_mods_and_typedefs(d->btf, id, NULL);
> +       if (!t) {
> +               pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
> +                       id);
> +               return -EINVAL;
> +       }
> +
> +       switch (BTF_INFO_KIND(t->info)) {
> +       case BTF_KIND_INT:
> +       case BTF_KIND_PTR:
> +       case BTF_KIND_ENUM:
> +               size = btf__resolve_size(d->btf, id);
> +               if (size < 0) {
> +                       pr_warn("unexpected size [%llu] for id [%u]\n",
> +                               size, id);
> +                       return -EINVAL;
> +               }
> +               if (data + (bits_offset >> 3) + size > d->data.data_end)

bits_offset / 8...

and considering you have a whole dedicated BITS_PER_BYTE (btw, why?..
are we expecting it to ever change?), it's surprising to see bit
shifts instead of division

> +                       return -E2BIG;
> +               return 0;
> +       default:
> +               return 0;
> +       }
> +}
> +
> +static int btf_dump_type_data_check_zero(struct btf_dump *d,
> +                                        const struct btf_type *t,
> +                                        __u32 id,
> +                                        void *data,
> +                                        __u8 bits_offset)
> +{
> +       __s64 value;
> +       int i, err;
> +
> +       /* toplevel exceptions; we show zero values if
> +        * - we ask for them (emit_zeros)
> +        * - if we are at top-level so we see "struct empty { }"
> +        * - or if we are an array member and the array is non-empty and
> +        *   not a char array; we don't want to be in a situation where we
> +        *   have an integer array 0, 1, 0, 1 and only show non-zero values.
> +        *   If the array contains zeroes only, or is a char array starting
> +        *   with a '\0', the array-level check_zero() will prevent showing it;
> +        *   we are concerned with determining zero value at the array member
> +        *   level here.
> +        */
> +       if (d->data.emit_zeroes || d->data.state.depth == 0 ||
> +           (d->data.state.array_member && !d->data.state.array_ischar))
> +               return 0;
> +
> +       t = skip_mods_and_typedefs(d->btf, id, NULL);
> +       if (!t) {

don't check, if anything btf_dump_type_data_check_overflow() already
validated that it's ok

> +               pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
> +                       id);
> +               return -EINVAL;
> +       }
> +
> +
> +       switch (BTF_INFO_KIND(t->info)) {
> +       case BTF_KIND_INT:
> +               if (d->data.state.bitfield_size)
> +                       return btf_dump_int_bits_check_zero(d, t, data,
> +                                                           bits_offset);
> +               return btf_dump_int_check_zero(d, t, data, bits_offset);
> +       case BTF_KIND_PTR:
> +               if (*((void **)data) == NULL)
> +                       return -ENODATA;
> +               return 0;
> +       case BTF_KIND_ARRAY: {
> +               const struct btf_array *array = btf_array(t);
> +               const struct btf_type *elem_type;
> +               __u32 elem_type_id, elem_size;
> +               bool ischar;
> +
> +               elem_type_id = array->type;
> +               elem_size = btf__resolve_size(d->btf, elem_type_id);
> +               elem_type =  btf__type_by_id(d->btf, elem_type_id);
> +
> +               ischar = btf_is_int(elem_type) && elem_size == 1;
> +
> +               /* check all elements; if _any_ element is nonzero, all
> +                * of array is displayed.  We make an exception however
> +                * for char arrays where the first element is 0; these
> +                * are considered zeroed also, even if later elements are
> +                * non-zero because the string is terminated.
> +                */
> +               for (i = 0; i < array->nelems; i++) {
> +                       if (i == 0 && ischar && *(char *)data == 0)
> +                               return -ENODATA;
> +                       err = btf_dump_type_data_check_zero(d, elem_type,
> +                                                           elem_type_id,
> +                                                           data +
> +                                                           (i * elem_size),
> +                                                           bits_offset);
> +                       if (err != -ENODATA)
> +                               return err;
> +               }
> +               return -ENODATA;
> +       }
> +       case BTF_KIND_STRUCT:
> +       case BTF_KIND_UNION: {
> +               const struct btf_member *member;
> +
> +               /* if any struct/union member is non-zero, the struct/union
> +                * is considered non-zero and dumped.
> +                */
> +               for (i = 0, member = btf_members(t);
> +                    i < btf_vlen(t);
> +                    i++, member++) {

in a lot of other places in libbpf we use more succinct names and this pattern

n = btf_vlen(t);
m = btf_members(t);
for (i = 0; i < n; i++, m++) {

it's nice, short, and familiar in libbpf code base

> +                       const struct btf_type *member_type;
> +                       __u32 bytes_offset, member_offset;
> +                       __u8 bits8_offset;
> +
> +                       member_type = btf__type_by_id(d->btf, member->type);
> +                       member_offset = btf_member_bit_offset(t, i);
> +                       bytes_offset = BITS_ROUNDDOWN_BYTES(member_offset);
> +                       bits8_offset = BITS_PER_BYTE_MASKED(member_offset);
> +
> +                       /* btf_int_bits() does not store member bitfield size;
> +                        * bitfield size needs to be stored here so int display
> +                        * of member can retrieve it.
> +                        */
> +                       d->data.state.bitfield_size =
> +                               btf_member_bitfield_size(t, i);
> +
> +                       err = btf_dump_type_data_check_zero(d, member_type,
> +                                                           member->type,
> +                                                           data + bytes_offset,
> +                                                           bits8_offset);
> +                       d->data.state.bitfield_size = 0;
> +                       if (err != ENODATA)
> +                               return err;
> +               }
> +               return -ENODATA;
> +       }
> +       case BTF_KIND_ENUM:
> +               if (btf_dump_get_enum_value(t, data, id, &value))
> +                       return 0;
> +               if (value == 0)
> +                       return -ENODATA;
> +               return 0;
> +       default:
> +               return 0;
> +       }
> +}
> +
> +static int btf_dump_dump_type_data(struct btf_dump *d,
> +                                  const char *fname,
> +                                  const struct btf_type *t,
> +                                  __u32 id,
> +                                  void *data,
> +                                  __u8 bits_offset)
> +{
> +       int err;
> +
> +       err = btf_dump_type_data_check_overflow(d, t, id, data, bits_offset);
> +       if (err)
> +               return err;
> +       err = btf_dump_type_data_check_zero(d, t, id, data, bits_offset);
> +       if (err) {
> +               /* zeroed data is expected and not an error, so simply skip
> +                * dumping such data.  Record other errors however.
> +                */
> +               if (err == -ENODATA)
> +                       return 0;
> +               return err;
> +       }
> +       btf_dump_printf(d, "%s", btf_dump_data_pfx(d));
> +       if (!d->data.skip_names)
> +               btf_dump_emit_type_name(d, id, fname, 0, true);
> +
> +       t = skip_mods_and_typedefs(d->btf, id, NULL);
> +       if (!t) {
> +               pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
> +                       id);
> +               return -EINVAL;
> +       }
> +
> +       switch (BTF_INFO_KIND(t->info)) {
> +       case BTF_KIND_UNKN:
> +       case BTF_KIND_FWD:
> +       case BTF_KIND_FUNC:
> +       case BTF_KIND_FUNC_PROTO:
> +               return btf_dump_df_data(d, t, id, data);
> +       case BTF_KIND_INT:
> +               if (d->data.state.bitfield_size)

btw, enum can be part of bitfields as well

> +                       return btf_dump_bitfield_data(d, t, data,
> +                                                     bits_offset,
> +                                                     d->data.state.bitfield_size);
> +               return btf_dump_int_data(d, t, id, data, bits_offset);
> +       case BTF_KIND_PTR:
> +               return btf_dump_ptr_data(d, t, id, data);
> +       case BTF_KIND_ARRAY:
> +               return btf_dump_array_data(d, t, id, data);
> +       case BTF_KIND_STRUCT:
> +       case BTF_KIND_UNION:
> +               return btf_dump_struct_data(d, t, id, data);
> +       case BTF_KIND_ENUM:
> +               return btf_dump_enum_data(d, t, id, data);
> +       case BTF_KIND_VAR:
> +               return btf_dump_var_data(d, t, id, data);
> +       case BTF_KIND_DATASEC:
> +               return btf_dump_datasec_data(d, t, id, data);
> +       default:

since last time we also added BTF_KIND_FLOAT

> +               pr_warn("unexpected kind [%u] for id [%u]\n",
> +                       BTF_INFO_KIND(t->info), id);
> +               return -EINVAL;
> +       }
> +}
> +
> +int btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
> +                            void *data, size_t byte_sz,
> +                            const struct btf_dump_type_data_opts *opts)
> +{
> +       const struct btf_type *t;
> +       int err;
> +
> +       if (!OPTS_VALID(opts, btf_dump_type_data_opts))
> +               return -EINVAL;
> +
> +       t = btf__type_by_id(d->btf, id);
> +       if (!t)
> +               return -ENOENT;

public APIs should use libbpf_err() to ensure errno is set properly,
see my recent patch set. This should be

return libbpf_err(-ENOENT);

> +
> +       d->data.data_end = data + byte_sz;
> +       d->data.indent_lvl = OPTS_GET(opts, indent_level, 0);
> +       d->data.compact = OPTS_GET(opts, compact, false);
> +       d->data.skip_names = OPTS_GET(opts, skip_names, false);
> +       d->data.emit_zeroes = OPTS_GET(opts, emit_zeroes, false);
> +       memset(&d->data.state, 0, sizeof(d->data.state));
> +
> +       err = btf_dump_dump_type_data(d, NULL, t, id, data, 0);
> +
> +       /* We reported all the data; return size of data we reported. */
> +       if (err == 0)
> +               err = btf__resolve_size(d->btf, id);

hm.. why can't btf_dump_dump_type_data() return the consumed amount of bytes?

> +
> +       memset(&d->data, 0, sizeof(d->data));
> +
> +       return err;
> +}
> diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> index 0229e01..76cfac6 100644
> --- a/tools/lib/bpf/libbpf.map
> +++ b/tools/lib/bpf/libbpf.map
> @@ -370,3 +370,8 @@ LIBBPF_0.4.0 {
>                 bpf_tc_hook_destroy;
>                 bpf_tc_query;
>  } LIBBPF_0.3.0;
> +
> +LIBBPF_0.5.0 {
> +       global:
> +               btf_dump__dump_type_data;
> +} LIBBPF_0.4.0;
> --
> 1.8.3.1
>

[PATCH v3 bpf-next 2/2] selftests/bpf: add dump type data tests to btf dump tests

[Alan Maguire]

Test various type data dumping operations by comparing expected
format with the dumped string; an snprintf-style printf function
is used to record the string dumped.  Also verify overflow handling
where the data passed does not cover the full size of a type,
such as would occur if a tracer has a portion of the 8k
"struct task_struct".

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
---
 tools/testing/selftests/bpf/prog_tests/btf_dump.c | 524 ++++++++++++++++++++++
 1 file changed, 524 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/btf_dump.c b/tools/testing/selftests/bpf/prog_tests/btf_dump.c
index 5e129dc..50a1c1d 100644
--- a/tools/testing/selftests/bpf/prog_tests/btf_dump.c
+++ b/tools/testing/selftests/bpf/prog_tests/btf_dump.c
@@ -232,6 +232,528 @@ void test_btf_dump_incremental(void)
 	btf__free(btf);
 }
 
+#define STRSIZE				4096
+
+void btf_dump_snprintf(void *ctx, const char *fmt, va_list args)
+{
+	char *s = ctx, new[STRSIZE];
+
+	vsnprintf(new, STRSIZE, fmt, args);
+	strncat(s, new, STRSIZE);
+}
+
+/* skip "enum "/"struct " prefixes */
+#define SKIP_PREFIX(_typestr, _prefix)					\
+	do {								\
+		if (strncmp(_typestr, _prefix, strlen(_prefix)) == 0)	\
+			_typestr += strlen(_prefix) + 1;		\
+	} while (0)
+
+int btf_dump_data(struct btf *btf, struct btf_dump *d,
+		  char *name, __u64 flags, void *ptr,
+		  size_t ptrsize, char *str, const char *expectedval)
+{
+	DECLARE_LIBBPF_OPTS(btf_dump_type_data_opts, opts);
+	int ret = 0, cmp;
+	size_t typesize;
+	__s32 type_id;
+
+	if (flags & BTF_F_COMPACT)
+		opts.compact = true;
+	if (flags & BTF_F_NONAME)
+		opts.skip_names = true;
+	if (flags & BTF_F_ZERO)
+		opts.emit_zeroes = true;
+	SKIP_PREFIX(name, "enum");
+	SKIP_PREFIX(name, "struct");
+	SKIP_PREFIX(name, "union");
+	type_id = btf__find_by_name(btf, name);
+	if (CHECK(type_id <= 0, "find type id",
+		  "no '%s' in BTF: %d\n", name, type_id)) {
+		ret = -ENOENT;
+		goto err;
+	}
+	typesize = btf__resolve_size(btf, type_id);
+	str[0] = '\0';
+	ret = btf_dump__dump_type_data(d, type_id, ptr, ptrsize, &opts);
+	if (typesize <= ptrsize) {
+		if (CHECK(ret != typesize, "btf_dump__dump_type_data",
+			  "failed/unexpected typesize: %d\n", ret))
+			goto err;
+	} else {
+		if (CHECK(ret != -E2BIG, "btf_dump__dump_type_data -E2BIG",
+			  "failed to return -E2BIG: %d\n", ret))
+			goto err;
+		ret = 0;
+	}
+
+	cmp = strcmp(str, expectedval);
+	if (CHECK(cmp, "ensure expected/actual match",
+		  "'%s' does not match expected '%s': %d\n",
+		  str, expectedval, cmp))
+		ret = -EFAULT;
+err:
+	if (ret < 0)
+		btf_dump__free(d);
+	return ret;
+}
+
+#define TEST_BTF_DUMP_DATA(_b, _d, _str, _type, _flags, _expected, ...)	\
+	do {								\
+		char __ptrtype[64] = #_type;				\
+		char *_ptrtype = (char *)__ptrtype;			\
+		_type _ptrdata = __VA_ARGS__;				\
+		void *_ptr = &_ptrdata;					\
+		int _err;						\
+									\
+		_err = btf_dump_data(_b, _d, _ptrtype, _flags, _ptr,	\
+				     sizeof(_type), _str, _expected);	\
+		if (_err < 0)						\
+			return _err;					\
+	} while (0)
+
+/* Use where expected data string matches its stringified declaration */
+#define TEST_BTF_DUMP_DATA_C(_b, _d, _str, _type, _flags, ...)		\
+	TEST_BTF_DUMP_DATA(_b, _d, _str, _type, _flags,			\
+			   "(" #_type ")" #__VA_ARGS__,	__VA_ARGS__)
+
+/* overflow test; pass typesize < expected type size, ensure E2BIG returned */
+#define TEST_BTF_DUMP_DATA_OVER(_b, _d, _str, _type, _typesize, _expected, ...)\
+	do {								\
+		char __ptrtype[64] = #_type;				\
+		char *_ptrtype = (char *)__ptrtype;			\
+		_type _ptrdata = __VA_ARGS__;				\
+		void *_ptr = &_ptrdata;					\
+		int _err;						\
+									\
+		_err = btf_dump_data(_b, _d, _ptrtype, 0, _ptr,		\
+				     _typesize, _str, _expected);	\
+		if (_err < 0)						\
+			return _err;					\
+	} while (0)
+
+#define TEST_BTF_DUMP_VAR(_b, _d, _str, _var, _type, _flags, _expected, ...) \
+	do {								\
+		_type _ptrdata = __VA_ARGS__;				\
+		void *_ptr = &_ptrdata;					\
+		int _err;						\
+									\
+		_err = btf_dump_data(_b, _d, _var, _flags, _ptr,	\
+				     sizeof(_type), _str, _expected);	\
+		if (_err < 0)						\
+			return _err;					\
+	} while (0)
+
+int test_btf_dump_int_data(struct btf *btf, struct btf_dump *d, char *str)
+{
+	/* simple int */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, int, BTF_F_COMPACT, 1234);
+	TEST_BTF_DUMP_DATA(btf, d, str, int, BTF_F_COMPACT | BTF_F_NONAME,
+			   "1234", 1234);
+	TEST_BTF_DUMP_DATA(btf, d, str, int, 0, "(int)1234\n", 1234);
+
+	/* zero value should be printed at toplevel */
+	TEST_BTF_DUMP_DATA(btf, d, str, int, BTF_F_COMPACT, "(int)0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, int, BTF_F_COMPACT | BTF_F_NONAME,
+			   "0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, int, BTF_F_COMPACT | BTF_F_ZERO,
+			   "(int)0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, int,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "0", 0);
+	TEST_BTF_DUMP_DATA_C(btf, d, str, int, BTF_F_COMPACT, -4567);
+	TEST_BTF_DUMP_DATA(btf, d, str, int, BTF_F_COMPACT | BTF_F_NONAME,
+			   "-4567", -4567);
+	TEST_BTF_DUMP_DATA(btf, d, str, int, 0, "(int)-4567\n", -4567);
+
+	TEST_BTF_DUMP_DATA_OVER(btf, d, str, int, sizeof(int)-1, "", 1);
+
+	return 0;
+}
+
+int test_btf_dump_char_data(struct btf *btf, struct btf_dump *d, char *str)
+{
+	/* simple char */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, char, BTF_F_COMPACT, 100);
+	TEST_BTF_DUMP_DATA(btf, d, str, char, BTF_F_COMPACT | BTF_F_NONAME,
+			   "100", 100);
+	TEST_BTF_DUMP_DATA(btf, d, str, char, 0, "(char)100\n", 100);
+	/* zero value should be printed at toplevel */
+	TEST_BTF_DUMP_DATA(btf, d, str, char, BTF_F_COMPACT, "(char)0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, char, BTF_F_COMPACT | BTF_F_NONAME,
+			   "0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, char, BTF_F_COMPACT | BTF_F_ZERO,
+			   "(char)0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, char,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, char, 0, "(char)0\n", 0);
+
+	TEST_BTF_DUMP_DATA_OVER(btf, d, str, char, sizeof(char)-1, "", 100);
+
+	return 0;
+}
+
+int test_btf_dump_typedef_data(struct btf *btf, struct btf_dump *d, char *str)
+{
+	/* simple typedef */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, uint64_t, BTF_F_COMPACT, 100);
+	TEST_BTF_DUMP_DATA(btf, d, str, u64, BTF_F_COMPACT | BTF_F_NONAME,
+			   "1", 1);
+	TEST_BTF_DUMP_DATA(btf, d, str, u64, 0, "(u64)1\n", 1);
+	/* zero value should be printed at toplevel */
+	TEST_BTF_DUMP_DATA(btf, d, str, u64, BTF_F_COMPACT, "(u64)0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, u64, BTF_F_COMPACT | BTF_F_NONAME,
+			   "0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, u64, BTF_F_COMPACT | BTF_F_ZERO,
+			   "(u64)0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, u64,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "0", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, u64, 0, "(u64)0\n", 0);
+
+	/* typedef struct */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, atomic_t, BTF_F_COMPACT,
+			     {.counter = (int)1,});
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, BTF_F_COMPACT | BTF_F_NONAME,
+			   "{1,}", { .counter = 1 });
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, 0,
+			   "(atomic_t){\n\t.counter = (int)1,\n}\n",
+			   {.counter = 1,});
+	/* typedef with 0 value should be printed at toplevel */
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, BTF_F_COMPACT, "(atomic_t){}",
+			   {.counter = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, BTF_F_COMPACT | BTF_F_NONAME,
+			   "{}", {.counter = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, 0,
+			   "(atomic_t){\n}\n", {.counter = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, BTF_F_COMPACT | BTF_F_ZERO,
+			   "(atomic_t){.counter = (int)0,}",
+			   {.counter = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "{0,}", {.counter = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, atomic_t, BTF_F_ZERO,
+			   "(atomic_t){\n\t.counter = (int)0,\n}\n",
+			   { .counter = 0,});
+
+	/* overflow should show type but not value since it overflows */
+	TEST_BTF_DUMP_DATA_OVER(btf, d, str, atomic_t, sizeof(atomic_t)-1,
+				"(atomic_t){\n", { .counter = 1});
+
+	return 0;
+}
+
+int test_btf_dump_enum_data(struct btf *btf, struct btf_dump *d, char *str)
+{
+	/* enum where enum value does (and does not) exist */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, enum bpf_cmd, BTF_F_COMPACT,
+			     BPF_MAP_CREATE);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd, BTF_F_COMPACT,
+			   "(enum bpf_cmd)BPF_MAP_CREATE", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "BPF_MAP_CREATE",
+			   BPF_MAP_CREATE);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd, 0,
+			   "(enum bpf_cmd)BPF_MAP_CREATE\n",
+			   BPF_MAP_CREATE);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "BPF_MAP_CREATE", 0);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd,
+			   BTF_F_COMPACT | BTF_F_ZERO,
+			   "(enum bpf_cmd)BPF_MAP_CREATE",
+			   BPF_MAP_CREATE);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "BPF_MAP_CREATE", BPF_MAP_CREATE);
+	TEST_BTF_DUMP_DATA_C(btf, d, str, enum bpf_cmd, BTF_F_COMPACT, 2000);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "2000", 2000);
+	TEST_BTF_DUMP_DATA(btf, d, str, enum bpf_cmd, 0,
+			   "(enum bpf_cmd)2000\n", 2000);
+
+	TEST_BTF_DUMP_DATA_OVER(btf, d, str, enum bpf_cmd,
+				sizeof(enum bpf_cmd) - 1, "", BPF_MAP_CREATE);
+
+	return 0;
+}
+
+int test_btf_dump_struct_data(struct btf *btf, struct btf_dump *d, char *str)
+{
+	DECLARE_LIBBPF_OPTS(btf_dump_type_data_opts, opts);
+	char skbdata[512] = { };
+	void *skb = skbdata;
+	size_t typesize;
+	__s32 type_id;
+	int ret;
+
+	/* simple struct */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, struct btf_enum, BTF_F_COMPACT,
+			     {.name_off = (__u32)3,.val = (__s32)-1,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "{3,-1,}",
+			   { .name_off = 3, .val = -1,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum, 0,
+			   "(struct btf_enum){\n\t.name_off = (__u32)3,\n\t.val = (__s32)-1,\n}\n",
+			   { .name_off = 3, .val = -1,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "{-1,}",
+			   { .name_off = 0, .val = -1,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum,
+			   BTF_F_COMPACT | BTF_F_NONAME | BTF_F_ZERO,
+			   "{0,-1,}",
+			   { .name_off = 0, .val = -1,});
+	/* empty struct should be printed */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum, BTF_F_COMPACT,
+			   "(struct btf_enum){}",
+			   { .name_off = 0, .val = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "{}",
+			   { .name_off = 0, .val = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum, 0,
+			   "(struct btf_enum){\n}\n",
+			   { .name_off = 0, .val = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum,
+			   BTF_F_COMPACT | BTF_F_ZERO,
+			   "(struct btf_enum){.name_off = (__u32)0,.val = (__s32)0,}",
+			   { .name_off = 0, .val = 0,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct btf_enum,
+			   BTF_F_ZERO,
+			   "(struct btf_enum){\n\t.name_off = (__u32)0,\n\t.val = (__s32)0,\n}\n",
+			   { .name_off = 0, .val = 0,});
+
+	/* struct with pointers */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct list_head, BTF_F_COMPACT,
+			   "(struct list_head){.next = (struct list_head *)0x1,}",
+			   { .next = (struct list_head *)1 });
+	TEST_BTF_DUMP_DATA(btf, d, str, struct list_head, 0,
+			   "(struct list_head){\n\t.next = (struct list_head *)0x1,\n}\n",
+			   { .next = (struct list_head *)1 });
+	/* NULL pointer should not be displayed */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct list_head, BTF_F_COMPACT,
+			   "(struct list_head){}",
+			   { .next = (struct list_head *)0 });
+	TEST_BTF_DUMP_DATA(btf, d, str, struct list_head, 0,
+			   "(struct list_head){\n}\n",
+			   { .next = (struct list_head *)0 });
+
+	/* struct with char array */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_prog_info, BTF_F_COMPACT,
+			   "(struct bpf_prog_info){.name = (char[])['f','o','o',],}",
+			   { .name = "foo",});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_prog_info,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "{['f','o','o',],}",
+			   {.name = "foo",});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_prog_info, 0,
+			   "(struct bpf_prog_info){\n\t.name = (char[])[\n\t\t'f',\n\t\t\'o',\n\t\t'o',\n\t],\n}\n",
+			   {.name = "foo",});
+	/* leading null char means do not display string */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_prog_info, BTF_F_COMPACT,
+			   "(struct bpf_prog_info){}",
+			   {.name = {'\0', 'f', 'o', 'o'}});
+	/* handle non-printable characters */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_prog_info, BTF_F_COMPACT,
+			   "(struct bpf_prog_info){.name = (char[])[1,2,3,],}",
+			   { .name = {1, 2, 3, 0}});
+
+	/* struct with non-char array */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct __sk_buff, BTF_F_COMPACT,
+			   "(struct __sk_buff){.cb = (__u32[])[1,2,3,4,5,],}",
+			   { .cb = {1, 2, 3, 4, 5,},});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct __sk_buff,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "{[1,2,3,4,5,],}",
+			   { .cb = { 1, 2, 3, 4, 5},});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct __sk_buff, 0,
+			   "(struct __sk_buff){\n\t.cb = (__u32[])[\n\t\t1,\n\t\t2,\n\t\t3,\n\t\t4,\n\t\t5,\n\t],\n}\n",
+			   { .cb = { 1, 2, 3, 4, 5},});
+	/* For non-char, arrays, show non-zero values only */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct __sk_buff, BTF_F_COMPACT,
+			   "(struct __sk_buff){.cb = (__u32[])[0,0,1,0,0,],}",
+			   { .cb = { 0, 0, 1, 0, 0},});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct __sk_buff, 0,
+			   "(struct __sk_buff){\n\t.cb = (__u32[])[\n\t\t0,\n\t\t0,\n\t\t1,\n\t\t0,\n\t\t0,\n\t],\n}\n",
+			   { .cb = { 0, 0, 1, 0, 0},});
+
+	/* struct with bitfields */
+	TEST_BTF_DUMP_DATA_C(btf, d, str, struct bpf_insn, BTF_F_COMPACT,
+		{.code = (__u8)1,.dst_reg = (__u8)0x2,.src_reg = (__u8)0x3,.off = (__s16)4,.imm = (__s32)5,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_insn,
+			   BTF_F_COMPACT | BTF_F_NONAME,
+			   "{1,0x2,0x3,4,5,}",
+			   { .code = 1, .dst_reg = 0x2, .src_reg = 0x3, .off = 4,
+			     .imm = 5,});
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_insn, 0,
+			   "(struct bpf_insn){\n\t.code = (__u8)1,\n\t.dst_reg = (__u8)0x2,\n\t.src_reg = (__u8)0x3,\n\t.off = (__s16)4,\n\t.imm = (__s32)5,\n}\n",
+			   {.code = 1, .dst_reg = 2, .src_reg = 3, .off = 4, .imm = 5});
+
+	/* zeroed bitfields should not be displayed */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_insn, BTF_F_COMPACT,
+			   "(struct bpf_insn){.dst_reg = (__u8)0x1,}",
+			   { .code = 0, .dst_reg = 1});
+
+	/* struct with nested anon union */
+	TEST_BTF_DUMP_DATA(btf, d, str, struct bpf_sock_ops, BTF_F_COMPACT,
+			   "(struct bpf_sock_ops){.op = (__u32)1,(union){.args = (__u32[])[1,2,3,4,],.reply = (__u32)1,.replylong = (__u32[])[1,2,3,4,],},}",
+			   { .op = 1, .args = { 1, 2, 3, 4}});
+
+	/* union with nested struct */
+	TEST_BTF_DUMP_DATA(btf, d, str, union bpf_iter_link_info, BTF_F_COMPACT,
+			   "(union bpf_iter_link_info){.map = (struct){.map_fd = (__u32)1,},}",
+			   { .map = { .map_fd = 1 }});
+
+	/* struct skb with nested structs/unions; because type output is so
+	 * complex, we don't do a string comparison, just verify we return
+	 * the type size as the amount of data displayed.
+	 */
+	type_id = btf__find_by_name(btf, "sk_buff");
+	if (CHECK(type_id <= 0, "find type id",
+		  "no 'struct sk_buff' in BTF: %d\n", type_id))
+		return -ENOENT;
+	typesize = btf__resolve_size(btf, type_id);
+	str[0] = '\0';
+
+	memset(skb, 255, 512);
+	ret = btf_dump__dump_type_data(d, type_id, skb, typesize, &opts);
+	if (CHECK(ret != typesize,
+		  "dump sk_buff is successful",
+		  "unexpected return value dumping sk_buff '%s': %d\n",
+		  str, ret))
+		return -EINVAL;
+
+	/* overflow bpf_sock_ops struct with final element nonzero/zero.
+	 * Regardless of the value of the final field, we don't have all the
+	 * data we need to display it, so we should trigger an overflow.
+	 * In other words oveflow checking should trump "is field zero?"
+	 * checks because if we've overflowed, it shouldn't matter what the
+	 * field is - we can't trust its value so shouldn't display it.
+	 */
+	TEST_BTF_DUMP_DATA_OVER(btf, d, str, struct bpf_sock_ops,
+				sizeof(struct bpf_sock_ops) - 1,
+				"(struct bpf_sock_ops){\n\t.op = (__u32)1,\n",
+				{ .op = 1, .skb_tcp_flags = 2});
+	TEST_BTF_DUMP_DATA_OVER(btf, d, str, struct bpf_sock_ops,
+				sizeof(struct bpf_sock_ops) - 1,
+				"(struct bpf_sock_ops){\n\t.op = (__u32)1,\n",
+				{ .op = 1, .skb_tcp_flags = 0});
+
+	return 0;
+}
+
+int test_btf_dump_var_data(struct btf *btf, struct btf_dump *d, char *str)
+{
+
+	TEST_BTF_DUMP_VAR(btf, d, str, "cpu_number", int, BTF_F_COMPACT,
+			  "int cpu_number = (int)100", 100);
+	TEST_BTF_DUMP_VAR(btf, d, str, "cpu_profile_flip", int, BTF_F_COMPACT,
+			  "static int cpu_profile_flip = (int)2", 2);
+
+	return 0;
+}
+
+int test_btf_datasec(struct btf *btf, struct btf_dump *d, char *str,
+		     const char *name, const char *expectedval,
+		     void *data, size_t data_sz)
+{
+	DECLARE_LIBBPF_OPTS(btf_dump_type_data_opts, opts);
+	int ret = 0, cmp;
+	size_t secsize;
+	__s32 type_id;
+
+	opts.compact = true;
+
+	type_id = btf__find_by_name(btf, name);
+	if (CHECK(type_id <= 0, "find type id",
+		  "no '%s' in BTF: %d\n", name, type_id))
+		return -ENOENT;
+
+	secsize = btf__resolve_size(btf, type_id);
+	if (CHECK(secsize != 0, "verify section size",
+		  "unexpected section size %ld for %s\n", secsize, name))
+		return -EINVAL;
+
+	str[0] = '\0';
+	ret = btf_dump__dump_type_data(d, type_id, data, data_sz, &opts);
+	if (CHECK(ret != 0, "btf_dump__dump_type_data",
+		  "failed/unexpected return value: %d\n", ret))
+		return ret;
+
+	cmp = strcmp(str, expectedval);
+	if (CHECK(cmp, "ensure expected/actual match",
+		  "'%s' does not match expected '%s': %d\n",
+		  str, expectedval, cmp))
+		ret = -EFAULT;
+
+	return ret;
+}
+
+int test_btf_dump_datasec_data(char *str)
+{
+	struct btf *btf = btf__parse("xdping_kern.o", NULL);
+	struct btf_dump_opts opts = { .ctx = str };
+	char license[4] = "GPL";
+	struct btf_dump *d;
+
+	if (CHECK(!btf, "get prog BTF", "xdping_kern.o BTF not found"))
+		return -ENOENT;
+
+	d = btf_dump__new(btf, NULL, &opts, btf_dump_snprintf);
+
+	if (CHECK(!d, "new dump", "could not create BTF dump"))
+		return -ENOENT;
+
+	if (test_btf_datasec(btf, d, str, "license",
+			     "SEC(\"license\") char[] _license = (char[])['G','P','L',];",
+			     license, sizeof(license)))
+		return -EINVAL;
+
+	return 0;
+}
+
+void test_btf_dump_data(void)
+{
+	struct btf *btf = libbpf_find_kernel_btf();
+	char str[STRSIZE];
+	struct btf_dump_opts opts = { .ctx = str };
+	struct btf_dump *d;
+
+	if (CHECK(!btf, "get kernel BTF", "no kernel BTF found"))
+		return;
+
+	d = btf_dump__new(btf, NULL, &opts, btf_dump_snprintf);
+
+	if (CHECK(!d, "new dump", "could not create BTF dump"))
+		return;
+
+	/* Verify type display for various types. */
+	if (test_btf_dump_int_data(btf, d, str))
+		return;
+	if (test_btf_dump_char_data(btf, d, str))
+		return;
+	if (test_btf_dump_typedef_data(btf, d, str))
+		return;
+	if (test_btf_dump_enum_data(btf, d, str))
+		return;
+	if (test_btf_dump_struct_data(btf, d, str))
+		return;
+	if (test_btf_dump_var_data(btf, d, str))
+		return;
+	btf_dump__free(d);
+	btf__free(btf);
+
+	/* verify datasec display */
+	if (test_btf_dump_datasec_data(str))
+		return;
+
+}
+
 void test_btf_dump() {
 	int i;
 
@@ -245,4 +767,6 @@ void test_btf_dump() {
 	}
 	if (test__start_subtest("btf_dump: incremental"))
 		test_btf_dump_incremental();
+	if (test__start_subtest("btf_dump: data"))
+		test_btf_dump_data();
 }
-- 
1.8.3.1