> On Jul 24, 2019, at 12:27 PM, Andrii Nakryiko <andriin@xxxxxx> wrote:
>
> This patch implements the core logic for BPF CO-RE offsets relocations.
> All the details are described in code comments.
>
> Signed-off-by: Andrii Nakryiko <andriin@xxxxxx>
> ---
> tools/lib/bpf/libbpf.c | 866 ++++++++++++++++++++++++++++++++++++++++-
> tools/lib/bpf/libbpf.h | 1 +
> 2 files changed, 861 insertions(+), 6 deletions(-)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index 8741c39adb1c..86d87bf10d46 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -38,6 +38,7 @@
> #include <sys/stat.h>
> #include <sys/types.h>
> #include <sys/vfs.h>
> +#include <sys/utsname.h>
> #include <tools/libc_compat.h>
> #include <libelf.h>
> #include <gelf.h>
> @@ -47,6 +48,7 @@
> #include "btf.h"
> #include "str_error.h"
> #include "libbpf_internal.h"
> +#include "hashmap.h"
>
> #ifndef EM_BPF
> #define EM_BPF 247
> @@ -1013,16 +1015,22 @@ static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
> }
>
> static const struct btf_type *skip_mods_and_typedefs(const struct btf *btf,
> - __u32 id)
> + __u32 id,
> + __u32 *res_id)
> {
> const struct btf_type *t = btf__type_by_id(btf, id);
>
> + if (res_id)
> + *res_id = id;
> +
> while (true) {
> switch (BTF_INFO_KIND(t->info)) {
> case BTF_KIND_VOLATILE:
> case BTF_KIND_CONST:
> case BTF_KIND_RESTRICT:
> case BTF_KIND_TYPEDEF:
> + if (res_id)
> + *res_id = t->type;
> t = btf__type_by_id(btf, t->type);
> break;
> default:
> @@ -1041,7 +1049,7 @@ static const struct btf_type *skip_mods_and_typedefs(const struct btf *btf,
> static bool get_map_field_int(const char *map_name, const struct btf *btf,
> const struct btf_type *def,
> const struct btf_member *m, __u32 *res) {
> - const struct btf_type *t = skip_mods_and_typedefs(btf, m->type);
> + const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
> const char *name = btf__name_by_offset(btf, m->name_off);
> const struct btf_array *arr_info;
> const struct btf_type *arr_t;
> @@ -1107,7 +1115,7 @@ static int bpf_object__init_user_btf_map(struct bpf_object *obj,
> return -EOPNOTSUPP;
> }
>
> - def = skip_mods_and_typedefs(obj->btf, var->type);
> + def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
> if (BTF_INFO_KIND(def->info) != BTF_KIND_STRUCT) {
> pr_warning("map '%s': unexpected def kind %u.\n",
> map_name, BTF_INFO_KIND(var->info));
> @@ -2289,6 +2297,845 @@ bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
> return 0;
> }
>
> +#define BPF_CORE_SPEC_MAX_LEN 64
> +
> +/* represents BPF CO-RE field or array element accessor */
> +struct bpf_core_accessor {
> + __u32 type_id; /* struct/union type or array element type */
> + __u32 idx; /* field index or array index */
> + const char *name; /* field name or NULL for array accessor */
> +};
> +
> +struct bpf_core_spec {
> + const struct btf *btf;
> + /* high-level spec: named fields and array indicies only */
> + struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
> + /* high-level spec length */
> + int len;
> + /* raw, low-level spec: 1-to-1 with accessor spec string */
> + int raw_spec[BPF_CORE_SPEC_MAX_LEN];
> + /* raw spec length */
> + int raw_len;
> + /* field byte offset represented by spec */
> + __u32 offset;
> +};
> +
> +static bool str_is_empty(const char *s)
> +{
> + return !s || !s[0];
> +}
> +
> +static int btf_kind(const struct btf_type *t)
> +{
> + return BTF_INFO_KIND(t->info);
> +}
> +
> +static bool btf_is_composite(const struct btf_type *t)
> +{
> + int kind = btf_kind(t);
> +
> + return kind == BTF_KIND_STRUCT || kind == BTF_KIND_UNION;
> +}
> +
> +static bool btf_is_array(const struct btf_type *t)
> +{
> + return btf_kind(t) == BTF_KIND_ARRAY;
> +}
> +
> +/*
> + * Turn bpf_offset_reloc into a low- and high-level spec representation,
> + * validating correctness along the way, as well as calculating resulting
> + * field offset (in bytes), specified by accessor string. Low-level spec
> + * captures every single level of nestedness, including traversing anonymous
> + * struct/union members. High-level one only captures semantically meaningful
> + * "turning points": named fields and array indicies.
> + * E.g., for this case:
> + *
> + * struct sample {
> + * int __unimportant;
> + * struct {
> + * int __1;
> + * int __2;
> + * int a[7];
> + * };
> + * };
> + *
> + * struct sample *s = ...;
> + *
> + * int x = &s->a[3]; // access string = '0:1:2:3'
> + *
> + * Low-level spec has 1:1 mapping with each element of access string (it's
> + * just a parsed access string representation): [0, 1, 2, 3].
> + *
> + * High-level spec will capture only 3 points:
> + * - intial zero-index access by pointer (&s->... is the same as &s[0]...);
> + * - field 'a' access (corresponds to '2' in low-level spec);
> + * - array element #3 access (corresponds to '3' in low-level spec).
> + *
> + */
> +static int bpf_core_spec_parse(const struct btf *btf,
> + __u32 type_id,
> + const char *spec_str,
> + struct bpf_core_spec *spec)
> +{
> + int access_idx, parsed_len, i;
> + const struct btf_type *t;
> + __u32 id = type_id;
> + const char *name;
> + __s64 sz;
> +
> + if (str_is_empty(spec_str) || *spec_str == ':')
> + return -EINVAL;
> +
> + memset(spec, 0, sizeof(*spec));
> + spec->btf = btf;
> +
> + /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
> + while (*spec_str) {
> + if (*spec_str == ':')
> + ++spec_str;
> + if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
> + return -EINVAL;
> + if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
> + return -E2BIG;
> + spec_str += parsed_len;
> + spec->raw_spec[spec->raw_len++] = access_idx;
> + }
> +
> + if (spec->raw_len == 0)
> + return -EINVAL;
> +
> + for (i = 0; i < spec->raw_len; i++) {
> + t = skip_mods_and_typedefs(btf, id, &id);
> + if (!t)
> + return -EINVAL;
> +
> + access_idx = spec->raw_spec[i];
> +
> + if (i == 0) {
> + /* first spec value is always reloc type array index */
> + spec->spec[spec->len].type_id = id;
> + spec->spec[spec->len].idx = access_idx;
> + spec->len++;
> +
> + sz = btf__resolve_size(btf, id);
> + if (sz < 0)
> + return sz;
> + spec->offset += access_idx * sz;
> + continue;
> + }
> +
> + if (btf_is_composite(t)) {
> + const struct btf_member *m = (void *)(t + 1);
> + __u32 offset;
> +
> + if (access_idx >= BTF_INFO_VLEN(t->info))
> + return -EINVAL;
> +
> + m = &m[access_idx];
> +
> + if (BTF_INFO_KFLAG(t->info)) {
> + if (BTF_MEMBER_BITFIELD_SIZE(m->offset))
> + return -EINVAL;
> + offset = BTF_MEMBER_BIT_OFFSET(m->offset);
> + } else {
> + offset = m->offset;
> + }
> + if (m->offset % 8)
> + return -EINVAL;
> + spec->offset += offset / 8;
> +
> + if (m->name_off) {
> + name = btf__name_by_offset(btf, m->name_off);
> + if (str_is_empty(name))
> + return -EINVAL;
> +
> + spec->spec[spec->len].type_id = id;
> + spec->spec[spec->len].idx = access_idx;
> + spec->spec[spec->len].name = name;
> + spec->len++;
> + }
> +
> + id = m->type;
> + } else if (btf_is_array(t)) {
> + const struct btf_array *a = (void *)(t + 1);
> +
> + t = skip_mods_and_typedefs(btf, a->type, &id);
> + if (!t || access_idx >= a->nelems)
> + return -EINVAL;
> +
> + spec->spec[spec->len].type_id = id;
> + spec->spec[spec->len].idx = access_idx;
> + spec->len++;
> +
> + sz = btf__resolve_size(btf, id);
> + if (sz < 0)
> + return sz;
> + spec->offset += access_idx * sz;
> + } else {
> + pr_warning("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
> + type_id, spec_str, i, id, btf_kind(t));
> + return -EINVAL;
> + }
> + }
> +
> + if (spec->len == 0)
> + return -EINVAL;
> +
> + return 0;
> +}
> +
> +/* Given 'some_struct_name___with_flavor' return the length of a name prefix
> + * before last triple underscore. Struct name part after last triple
> + * underscore is ignored by BPF CO-RE relocation during relocation matching.
> + */
> +static size_t bpf_core_essential_name_len(const char *name)
> +{
> + size_t n = strlen(name);
> + int i = n - 3;
> +
> + while (i > 0) {
> + if (name[i] == '_' && name[i + 1] == '_' && name[i + 2] == '_')
> + return i;
> + i--;
> + }
> + return n;
> +}
> +
> +/* dynamically sized list of type IDs */
> +struct ids_vec {
> + __u32 *data;
> + int len;
> +};
> +
> +static void bpf_core_free_cands(struct ids_vec *cand_ids)
> +{
> + free(cand_ids->data);
> + free(cand_ids);
> +}
> +
> +static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
> + __u32 local_type_id,
> + const struct btf *targ_btf)
> +{
> + size_t local_essent_len, targ_essent_len;
> + const char *local_name, *targ_name;
> + const struct btf_type *t;
> + struct ids_vec *cand_ids;
> + __u32 *new_ids;
> + int i, err, n;
> +
> + t = btf__type_by_id(local_btf, local_type_id);
> + if (!t)
> + return ERR_PTR(-EINVAL);
> +
> + local_name = btf__name_by_offset(local_btf, t->name_off);
> + if (str_is_empty(local_name))
> + return ERR_PTR(-EINVAL);
> + local_essent_len = bpf_core_essential_name_len(local_name);
> +
> + cand_ids = calloc(1, sizeof(*cand_ids));
> + if (!cand_ids)
> + return ERR_PTR(-ENOMEM);
> +
> + n = btf__get_nr_types(targ_btf);
> + for (i = 1; i <= n; i++) {
> + t = btf__type_by_id(targ_btf, i);
> + targ_name = btf__name_by_offset(targ_btf, t->name_off);
> + if (str_is_empty(targ_name))
> + continue;
> +
> + targ_essent_len = bpf_core_essential_name_len(targ_name);
> + if (targ_essent_len != local_essent_len)
> + continue;
> +
> + if (strncmp(local_name, targ_name, local_essent_len) == 0) {
> + pr_debug("[%d] (%s): found candidate [%d] (%s)\n",
> + local_type_id, local_name, i, targ_name);
> + new_ids = realloc(cand_ids->data, cand_ids->len + 1);
> + if (!new_ids) {
> + err = -ENOMEM;
> + goto err_out;
> + }
> + cand_ids->data = new_ids;
> + cand_ids->data[cand_ids->len++] = i;
> + }
> + }
> + return cand_ids;
> +err_out:
> + bpf_core_free_cands(cand_ids);
> + return ERR_PTR(err);
> +}
> +
> +/* Check two types for compatibility, skipping const/volatile/restrict and
> + * typedefs, to ensure we are relocating offset to the compatible entities:
> + * - any two STRUCTs/UNIONs are compatible and can be mixed;
> + * - any two FWDs are compatible;
> + * - any two PTRs are always compatible;
> + * - for ENUMs, check sizes, names are ignored;
> + * - for INT, size and bitness should match, signedness is ignored;
> + * - for ARRAY, dimensionality is ignored, element types are checked for
> + * compatibility recursively;
> + * - everything else shouldn't be ever a target of relocation.
> + * These rules are not set in stone and probably will be adjusted as we get
> + * more experience with using BPF CO-RE relocations.
> + */
> +static int bpf_core_fields_are_compat(const struct btf *local_btf,
> + __u32 local_id,
> + const struct btf *targ_btf,
> + __u32 targ_id)
> +{
> + const struct btf_type *local_type, *targ_type;
> + __u16 kind;
> +
> +recur:
> + local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
> + targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
> + if (!local_type || !targ_type)
> + return -EINVAL;
> +
> + if (btf_is_composite(local_type) && btf_is_composite(targ_type))
> + return 1;
> + if (BTF_INFO_KIND(local_type->info) != BTF_INFO_KIND(targ_type->info))
> + return 0;
> +
> + kind = BTF_INFO_KIND(local_type->info);
> + switch (kind) {
> + case BTF_KIND_FWD:
> + case BTF_KIND_PTR:
> + return 1;
> + case BTF_KIND_ENUM:
> + return local_type->size == targ_type->size;
> + case BTF_KIND_INT: {
> + __u32 loc_int = *(__u32 *)(local_type + 1);
> + __u32 targ_int = *(__u32 *)(targ_type + 1);
> +
> + return BTF_INT_OFFSET(loc_int) == 0 &&
> + BTF_INT_OFFSET(targ_int) == 0 &&
> + local_type->size == targ_type->size &&
> + BTF_INT_BITS(loc_int) == BTF_INT_BITS(targ_int);
> + }
> + case BTF_KIND_ARRAY: {
> + const struct btf_array *loc_a, *targ_a;
> +
> + loc_a = (void *)(local_type + 1);
> + targ_a = (void *)(targ_type + 1);
> + local_id = loc_a->type;
> + targ_id = targ_a->type;
> + goto recur;
> + }
> + default:
> + pr_warning("unexpected kind %d relocated, local [%d], target [%d]\n",
> + kind, local_id, targ_id);
> + return 0;
> + }
> +}
> +
> +/*
> + * Given single high-level accessor (either named field or array index) in
> + * local type, find corresponding high-level accessor for a target type. Along
> + * the way, maintain low-level spec for target as well. Also keep updating
> + * target offset.
> + */
> +static int bpf_core_match_member(const struct btf *local_btf,
> + const struct bpf_core_accessor *local_acc,
> + const struct btf *targ_btf,
> + __u32 targ_id,
> + struct bpf_core_spec *spec,
> + __u32 *next_targ_id)
> +{
> + const struct btf_type *local_type, *targ_type;
> + const struct btf_member *local_member, *m;
> + const char *local_name, *targ_name;
> + __u32 local_id;
> + int i, n, found;
> +
> + targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
> + if (!targ_type)
> + return -EINVAL;
> + if (!btf_is_composite(targ_type))
> + return 0;
> +
> + local_id = local_acc->type_id;
> + local_type = btf__type_by_id(local_btf, local_id);
> + local_member = (void *)(local_type + 1);
> + local_member += local_acc->idx;
> + local_name = btf__name_by_offset(local_btf, local_member->name_off);
> +
> + n = BTF_INFO_VLEN(targ_type->info);
> + m = (void *)(targ_type + 1);
> + for (i = 0; i < n; i++, m++) {
> + __u32 offset;
> +
> + /* bitfield relocations not supported */
> + if (BTF_INFO_KFLAG(targ_type->info)) {
> + if (BTF_MEMBER_BITFIELD_SIZE(m->offset))
> + continue;
> + offset = BTF_MEMBER_BIT_OFFSET(m->offset);
> + } else {
> + offset = m->offset;
> + }
> + if (offset % 8)
> + continue;
> +
> + /* too deep struct/union/array nesting */
> + if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
> + return -E2BIG;
> +
> + /* speculate this member will be the good one */
> + spec->offset += offset / 8;
> + spec->raw_spec[spec->raw_len++] = i;
> +
> + targ_name = btf__name_by_offset(targ_btf, m->name_off);
> + if (str_is_empty(targ_name)) {
> + /* embedded struct/union, we need to go deeper */
> + found = bpf_core_match_member(local_btf, local_acc,
> + targ_btf, m->type,
> + spec, next_targ_id);
> + if (found) /* either found or error */
> + return found;
> + } else if (strcmp(local_name, targ_name) == 0) {
> + /* matching named field */
> + struct bpf_core_accessor *targ_acc;
> +
> + targ_acc = &spec->spec[spec->len++];
> + targ_acc->type_id = targ_id;
> + targ_acc->idx = i;
> + targ_acc->name = targ_name;
> +
> + *next_targ_id = m->type;
> + found = bpf_core_fields_are_compat(local_btf,
> + local_member->type,
> + targ_btf, m->type);
> + if (!found)
> + spec->len--; /* pop accessor */
> + return found;
> + }
> + /* member turned out to be not we looked for */
/* member turned out not to be what we looked for */
Or something similar.
The rest of this patch looks good to me.
Thanks,
Song
