Integrate kernel module BTF support
Subsystems (8)
Andrii Nakryiko
maintainerv2 cover Integrate kernel module BTF support
This patch set adds BTF generation for kernel modules using a compact split BTF approach. Respective patches have all the details.
Kernel module BTFs rely on pahole's split BTF support, which is added in [0] and will be available starting from v1.19. Support for it is detected automatically during kernel build time.
This patch set implements in-kernel support for split BTF loading and validation. It also extends GET_OBJ_INFO API for BTFs to return BTF's module name and a flag whether BTF itself is in-kernel or user-provided. vmlinux BTF is also exposed to user-space through the same BTF object iteration APIs.
Follow up patch set will utilize the fact that vmlinux and module BTFs now have associated ID to provide ability to attach BPF fentry/fexit/etc programs to functions defined in kernel modules.
bpftool is also extended to show module/vmlinux BTF's name.
[0] https://patchwork.kernel.org/project/netdevbpf/list/?series=378699&state=*
v2->v1:
- drop WARNs, add fewer pr_warn()'s instead (Greg);
- properly initialize sysfs binary attribute structure (Greg);
- add __maybe_unused to any_section_objs, used conditionally by module BTF;
rfc->v1:
- CONFIG_DEBUG_INFO_BTF_MODULES is derived automatically (Alexei);
- vmlinux BTF now has explicit "vmlinux" name (Alexei);
- added sysfs ABI documentation for /sys/kernel/btf/<module> (Greg).
Andrii Nakryiko (5):
bpf: add in-kernel split BTF support
bpf: assign ID to vmlinux BTF and return extra info for BTF in
GET_OBJ_INFO
kbuild: build kernel module BTFs if BTF is enabled and pahole supports
it
bpf: load and verify kernel module BTFs
tools/bpftool: add support for in-kernel and named BTF in btf show
Documentation/ABI/testing/sysfs-kernel-btf | 8 + include/linux/bpf.h | 2 + include/linux/module.h | 4 + include/uapi/linux/bpf.h | 3 + kernel/bpf/btf.c | 413 ++++++++++++++++++--- kernel/bpf/sysfs_btf.c | 2 +- kernel/module.c | 32 ++ lib/Kconfig.debug | 9 + scripts/Makefile.modfinal | 20 +- tools/bpf/bpftool/btf.c | 28 +- tools/include/uapi/linux/bpf.h | 3 + 11 files changed, 467 insertions(+), 57 deletions(-)
Signature · 1 line
2.24.1
Andrii Nakryiko
maintainerPATCH v2 1/5 bpf: add in-kernel split BTF support
Adjust in-kernel BTF implementation to support a split BTF mode of operation. Changes are mostly mirroring libbpf split BTF changes, with the exception of start_id being 0 for in-kernel implementation due to simpler read-only mode.
Otherwise, for split BTF logic, most of the logic of jumping to base BTF, where necessary, is encapsulated in few helper functions. Type numbering and string offset in a split BTF are logically continuing where base BTF ends, so most of the high-level logic is kept without changes.
Type verification and size resolution is only doing an added resolution of new split BTF types and relies on already cached size and type resolution results in the base BTF.
--- kernel/bpf/btf.c | 182 +++++++++++++++++++++++++++++++++-------------- 1 file changed, 130 insertions(+), 52 deletions(-)
kernel/bpf/btf.c +130 -52
@@ -204,12 +204,17 @@ struct btf {const char *strings;void *nohdr_data;struct btf_header hdr; -u32 nr_types; +u32 nr_types;/* includes VOID for base BTF */ u32 types_size;u32 data_size;refcount_t refcnt;u32 id;struct rcu_head rcu; + +/* split BTF support */ +struct btf *base_btf; +u32 start_id;/* first type ID in this BTF (0 for base BTF) */ +u32 start_str_off;/* first string offset (0 for base BTF) */ };enum verifier_phase {
@@ -450,14 +455,27 @@ static bool btf_type_is_datasec(const struct btf_type *t)return BTF_INFO_KIND (t->info) ==BTF_KIND_DATASEC ; } +static u32 btf_nr_types_total (const struct btf *btf) +{ +u32 total =0 ; + +while (btf) { + total += btf->nr_types; + btf = btf->base_btf; + } + +return total; +} +s32 btf_find_by_name_kind (const struct btf *btf,const char *name,u8 kind) {const struct btf_type *t;const char *tname; -u32 i; +u32 i, total; -for (i =1 ; i <= btf->nr_types; i++) { - t = btf->types[i]; + total =btf_nr_types_total (btf); +for (i =1 ; i < total; i++) { + t =btf_type_by_id (btf, i);if (BTF_INFO_KIND (t->info) != kind)continue ;
@@ -600,8 +618,15 @@ static const struct btf_kind_operations *btf_type_ops(const struct btf_type *t)static bool btf_name_offset_valid (const struct btf *btf,u32 offset) { -return BTF_STR_OFFSET_VALID (offset) && - offset < btf->hdr.str_len; +if (!BTF_STR_OFFSET_VALID (offset)) +return false; +again: +if (offset < btf->start_str_off) { + btf = btf->base_btf; + goto again; + } + offset -= btf->start_str_off; +return offset < btf->hdr.str_len; }static bool __btf_name_char_ok (char c,bool first,bool dot_ok)
@@ -615,10 +640,25 @@ static bool __btf_name_char_ok(char c, bool first, bool dot_ok)return true; } +static const char *btf_str_by_offset (const struct btf *btf,u32 offset) +{ +again: +if (offset < btf->start_str_off) { + btf = btf->base_btf; + goto again; + } + + offset -= btf->start_str_off; +if (offset < btf->hdr.str_len) +return &btf->strings[offset]; + +return NULL ; +} +static bool __btf_name_valid (const struct btf *btf,u32 offset,bool dot_ok) {/* offset must be valid */ -const char *src = &btf->strings[offset]; +const char *src =btf_str_by_offset (btf, offset);const char *src_limit;if (!__btf_name_char_ok (*src, true, dot_ok))
@@ -651,27 +691,31 @@ static bool btf_name_valid_section(const struct btf *btf, u32 offset)static const char *__btf_name_by_offset (const struct btf *btf,u32 offset) { +const char *name; +if (!offset)return "(anon)" ; -else if (offset < btf->hdr.str_len) -return &btf->strings[offset]; -else -return "(invalid-name-offset)" ; + + name =btf_str_by_offset (btf, offset); +return name ?:"(invalid-name-offset)" ; }const char *btf_name_by_offset (const struct btf *btf,u32 offset) { -if (offset < btf->hdr.str_len) -return &btf->strings[offset]; - -return NULL ; +return btf_str_by_offset (btf, offset); }const struct btf_type *btf_type_by_id (const struct btf *btf,u32 type_id) { -if (type_id > btf->nr_types) -return NULL ; +again: +if (type_id < btf->start_id) { + btf = btf->base_btf; + goto again; + } + type_id -= btf->start_id; +if (type_id >= btf->nr_types) +return NULL ;return btf->types[type_id]; }
@@ -1391,17 +1435,13 @@ static int btf_add_type(struct btf_verifier_env *env, struct btf_type *t) {struct btf *btf = env->btf; -/* < 2 because +1 for btf_void which is always in btf->types[0]. -* btf_void is not accounted in btf->nr_types because btf_void -* does not come from the BTF file. -*/ -if (btf->types_size - btf->nr_types <2 ) { +if (btf->types_size == btf->nr_types) {/* Expand 'types' array */ struct btf_type **new_types;u32 expand_by, new_size; -if (btf->types_size ==BTF_MAX_TYPE ) { +if (btf->start_id + btf->types_size ==BTF_MAX_TYPE ) {btf_verifier_log (env,"Exceeded max num of types" );return -E2BIG ; }
@@ -1415,18 +1455,23 @@ static int btf_add_type(struct btf_verifier_env *env, struct btf_type *t)if (!new_types)return -ENOMEM ; -if (btf->nr_types ==0 ) - new_types[0 ] = &btf_void; -else +if (btf->nr_types ==0 ) { +if (!btf->base_btf) { +/* lazily init VOID type */ + new_types[0 ] = &btf_void; + btf->nr_types++; + } + }else {memcpy (new_types, btf->types, -sizeof (*btf->types) * (btf->nr_types +1 )); +sizeof (*btf->types) * btf->nr_types); + }kvfree (btf->types); btf->types = new_types; btf->types_size = new_size; } - btf->types[++(btf->nr_types)] = t; + btf->types[btf->nr_types++] = t;return 0 ; }
@@ -1499,18 +1544,17 @@ static int env_resolve_init(struct btf_verifier_env *env)u32 *resolved_ids =NULL ;u8 *visit_states =NULL ; -/* +1 for btf_void */ - resolved_sizes =kvcalloc (nr_types +1 ,sizeof (*resolved_sizes), + resolved_sizes =kvcalloc (nr_types,sizeof (*resolved_sizes),GFP_KERNEL | __GFP_NOWARN);if (!resolved_sizes) goto nomem; - resolved_ids =kvcalloc (nr_types +1 ,sizeof (*resolved_ids), + resolved_ids =kvcalloc (nr_types,sizeof (*resolved_ids),GFP_KERNEL | __GFP_NOWARN);if (!resolved_ids) goto nomem; - visit_states =kvcalloc (nr_types +1 ,sizeof (*visit_states), + visit_states =kvcalloc (nr_types,sizeof (*visit_states),GFP_KERNEL | __GFP_NOWARN);if (!visit_states) goto nomem;
@@ -1562,21 +1606,27 @@ static bool env_type_is_resolve_sink(const struct btf_verifier_env *env,static bool env_type_is_resolved (const struct btf_verifier_env *env,u32 type_id) { -return env->visit_states[type_id] ==RESOLVED ; +/* base BTF types should be resolved by now */ +if (type_id < env->btf->start_id) +return true; + +return env->visit_states[type_id - env->btf->start_id] ==RESOLVED ; }static int env_stack_push (struct btf_verifier_env *env,const struct btf_type *t,u32 type_id) { +const struct btf *btf = env->btf;struct resolve_vertex *v;if (env->top_stack ==MAX_RESOLVE_DEPTH )return -E2BIG ; -if (env->visit_states[type_id] !=NOT_VISITED ) +if (type_id < btf->start_id + || env->visit_states[type_id - btf->start_id] !=NOT_VISITED )return -EEXIST ; - env->visit_states[type_id] =VISITED ; + env->visit_states[type_id - btf->start_id] =VISITED ; v = &env->stack[env->top_stack++]; v->t = t;
@@ -1606,6 +1656,7 @@ static void env_stack_pop_resolved(struct btf_verifier_env *env,u32 type_id = env->stack[--(env->top_stack)].type_id;struct btf *btf = env->btf; + type_id -= btf->start_id;/* adjust to local type id */ btf->resolved_sizes[type_id] = resolved_size; btf->resolved_ids[type_id] = resolved_type_id; env->visit_states[type_id] =RESOLVED ;
@@ -1710,14 +1761,34 @@ btf_resolve_size(const struct btf *btf, const struct btf_type *type,return __btf_resolve_size (btf, type, type_size,NULL ,NULL ,NULL ,NULL ); } +static u32 btf_resolved_type_id (const struct btf *btf,u32 type_id) +{ +again: +if (type_id < btf->start_id) { + btf = btf->base_btf; + goto again; + } +return btf->resolved_ids[type_id - btf->start_id]; +} +/* The input param "type_id" must point to a needs_resolve type */ static const struct btf_type *btf_type_id_resolve (const struct btf *btf,u32 *type_id) { - *type_id = btf->resolved_ids[*type_id]; + *type_id =btf_resolved_type_id (btf, *type_id);return btf_type_by_id (btf, *type_id); } +static u32 btf_resolved_type_size (const struct btf *btf,u32 type_id) +{ +again: +if (type_id < btf->start_id) { + btf = btf->base_btf; + goto again; + } +return btf->resolved_sizes[type_id - btf->start_id]; +} +const struct btf_type *btf_type_id_size (const struct btf *btf,u32 *type_id,u32 *ret_size) {
@@ -1732,7 +1803,7 @@ const struct btf_type *btf_type_id_size(const struct btf *btf,if (btf_type_has_size (size_type)) { size = size_type->size; }else if (btf_type_is_array (size_type)) { - size = btf->resolved_sizes[size_type_id]; + size =btf_resolved_type_size (btf, size_type_id); }else if (btf_type_is_ptr (size_type)) { size =sizeof (void *); }else {
@@ -1740,14 +1811,14 @@ const struct btf_type *btf_type_id_size(const struct btf *btf, !btf_type_is_var(size_type)))return NULL ; - size_type_id = btf->resolved_ids[size_type_id]; + size_type_id =btf_resolved_type_id (btf, size_type_id); size_type =btf_type_by_id (btf, size_type_id);if (btf_type_nosize_or_null (size_type))return NULL ;else if (btf_type_has_size (size_type)) size = size_type->size;else if (btf_type_is_array (size_type)) - size = btf->resolved_sizes[size_type_id]; + size =btf_resolved_type_size (btf, size_type_id);else if (btf_type_is_ptr (size_type)) size =sizeof (void *);else
@@ -3799,7 +3870,7 @@ static int btf_check_all_metas(struct btf_verifier_env *env) cur = btf->nohdr_data + hdr->type_off; end = cur + hdr->type_len; - env->log_type_id =1 ; + env->log_type_id = btf->base_btf ? btf->start_id :1 ;while (cur < end) {struct btf_type *t = cur;s32 meta_size;
@@ -3826,8 +3897,8 @@ static bool btf_resolve_valid(struct btf_verifier_env *env,return false;if (btf_type_is_struct (t) ||btf_type_is_datasec (t)) -return !btf->resolved_ids[type_id] && - !btf->resolved_sizes[type_id]; +return !btf_resolved_type_id (btf, type_id) && + !btf_resolved_type_size (btf, type_id);if (btf_type_is_modifier (t) ||btf_type_is_ptr (t) ||btf_type_is_var (t)) {
@@ -3847,7 +3918,7 @@ static bool btf_resolve_valid(struct btf_verifier_env *env, elem_type =btf_type_id_size (btf, &elem_type_id, &elem_size);return elem_type && !btf_type_is_modifier (elem_type) && (array->nelems * elem_size == - btf->resolved_sizes[type_id]); +btf_resolved_type_size (btf, type_id)); }return false;
@@ -3889,7 +3960,8 @@ static int btf_resolve(struct btf_verifier_env *env,static int btf_check_all_types (struct btf_verifier_env *env) {struct btf *btf = env->btf; -u32 type_id; +const struct btf_type *t; +u32 type_id, i;int err; err =env_resolve_init (env);
@@ -3897,8 +3969,9 @@ static int btf_check_all_types(struct btf_verifier_env *env)return err; env->phase++; -for (type_id =1 ; type_id <= btf->nr_types; type_id++) { -const struct btf_type *t =btf_type_by_id (btf, type_id); +for (i = btf->base_btf ?0 :1 ; i < btf->nr_types; i++) { + type_id = btf->start_id + i; + t =btf_type_by_id (btf, type_id); env->log_type_id = type_id; if (btf_type_needs_resolve(t) &&
@@ -3935,7 +4008,7 @@ static int btf_parse_type_sec(struct btf_verifier_env *env)return -EINVAL ; } -if (!hdr->type_len) { +if (!env->btf->base_btf && !hdr->type_len) {btf_verifier_log (env,"No type found" );return -EINVAL ; }
@@ -3962,13 +4035,18 @@ static int btf_parse_str_sec(struct btf_verifier_env *env)return -EINVAL ; } -if (!hdr->str_len || hdr->str_len -1 >BTF_MAX_NAME_OFFSET || - start[0 ] || end[-1 ]) { + btf->strings = start; + +if (btf->base_btf && !hdr->str_len) +return 0 ; +if (!hdr->str_len || hdr->str_len -1 >BTF_MAX_NAME_OFFSET || end[-1 ]) { +btf_verifier_log (env,"Invalid string section" ); +return -EINVAL ; + } +if (!btf->base_btf && start[0 ]) {btf_verifier_log (env,"Invalid string section" );return -EINVAL ; } - - btf->strings = start;return 0 ; }
@@ -4909,7 +4987,7 @@ static int __get_type_size(struct btf *btf, u32 btf_id,while (t &&btf_type_is_modifier (t)) t =btf_type_by_id (btf, t->type);if (!t) { - *bad_type = btf->types[0 ]; + *bad_type =btf_type_by_id (btf,0 );return -EINVAL ; }if (btf_type_is_ptr (t))
Signature · 1 line
2.24.1
Re: [PATCH v2 bpf-next 1/5] bpf: add in-kernel split BTF support
Quoted 18 lines · Andrii Nakryiko
Adjust in-kernel BTF implementation to support a split BTF mode of operation. Changes are mostly mirroring libbpf split BTF changes, with the exception of
start_idbeing 0 for in-kernel implementation due to simpler read-only mode.Otherwise, for split BTF logic, most of the logic of jumping to base BTF, where necessary, is encapsulated in few helper functions. Type numbering and string offset in a split BTF are logically continuing where base BTF ends, so most of the high-level logic is kept without changes.
Type verification and size resolution is only doing an added resolution of new split BTF types and relies on already cached size and type resolution results in the base BTF.
Signed-off-by Andrii Nakryiko <[email protected]>
[...]
kernel/bpf/btf.c static const struct btf_kind_operations *btf_type_ops(const struct btf_type *t) · line 626
Quoted 13 lines
kernel/bpf/btf.c +6
-2@@ -600,8 +618,15 @@ static const struct btf_kind_operations *btf_type_ops(const struct btf_type *t)static bool btf_name_offset_valid (const struct btf *btf,u32 offset) { -return BTF_STR_OFFSET_VALID (offset) && - offset < btf->hdr.str_len; +if (!BTF_STR_OFFSET_VALID (offset)) +return false; +again: +if (offset < btf->start_str_off) { + btf = btf->base_btf; + goto again;
Can we do a while loop instead of "goto again;"?
Quoted 7 lines
kernel/bpf/btf.c +3
-0+ } + offset -= btf->start_str_off; +return offset < btf->hdr.str_len; }static bool __btf_name_char_ok (char c,bool first,bool dot_ok)
kernel/bpf/btf.c static bool __btf_name_char_ok(char c, bool first, bool dot_ok) · line 649
Quoted 10 lines
kernel/bpf/btf.c +7
-0@@ -615,10 +640,25 @@ static bool __btf_name_char_ok(char c, bool first, bool dot_ok)return true; } +static const char *btf_str_by_offset (const struct btf *btf,u32 offset) +{ +again: +if (offset < btf->start_str_off) { + btf = btf->base_btf; + goto again; + }
Maybe add a btf_find_base_btf(btf, offset) helper for this logic?
Quoted 8 lines
kernel/bpf/btf.c +8
-0+ + offset -= btf->start_str_off; +if (offset < btf->hdr.str_len) +return &btf->strings[offset]; + +return NULL ; +} +
[...]
Quoted 10 lines
}
const char *btf_name_by_offset(const struct btf *btf, u32 offset) {
kernel/bpf/btf.c +1
-4-if (offset < btf->hdr.str_len) -return &btf->strings[offset]; - -return NULL ; +return btf_str_by_offset (btf, offset);}
IIUC, btf_str_by_offset() and btf_name_by_offset() are identical. Can we just keep btf_name_by_offset()?
kernel/bpf/btf.c static bool btf_name_valid_section(const struct btf *btf, u32 offset) · line 714
Quoted 10 lines
const struct
btf_type*btf_type_by_id(const struct btf *btf, u32type_id) {kernel/bpf/btf.c +5
-2-if (type_id > btf->nr_types) -return NULL ; +again: +if (type_id < btf->start_id) { + btf = btf->base_btf; + goto again; + }
ditto, goto again..
[...]
Re: [PATCH v2 bpf-next 1/5] bpf: add in-kernel split BTF support
Quoted 25 lines · Song Liu
Quoted 19 lines · Andrii Nakryiko
Adjust in-kernel BTF implementation to support a split BTF mode of operation. Changes are mostly mirroring libbpf split BTF changes, with the exception of
start_idbeing 0 for in-kernel implementation due to simpler read-only mode.Otherwise, for split BTF logic, most of the logic of jumping to base BTF, where necessary, is encapsulated in few helper functions. Type numbering and string offset in a split BTF are logically continuing where base BTF ends, so most of the high-level logic is kept without changes.
Type verification and size resolution is only doing an added resolution of new split BTF types and relies on already cached size and type resolution results in the base BTF.
Signed-off-by Andrii Nakryiko <[email protected]>[...]
Quoted 3 lines
(quoted diff) +0
-0@@ -600,8 +618,15 @@ static const struct btf_kind_operations *btf_type_ops(const struct btf_type *t)
Quoted 12 lines
kernel/bpf/btf.c static const struct btf_kind_operations *btf_type_ops(const struct btf_type *t) · line 626
Quoted 10 lines
kernel/bpf/btf.c +6
-2static bool btf_name_offset_valid (const struct btf *btf,u32 offset) { -return BTF_STR_OFFSET_VALID (offset) && - offset < btf->hdr.str_len; +if (!BTF_STR_OFFSET_VALID (offset)) +return false; +again: +if (offset < btf->start_str_off) { + btf = btf->base_btf; + goto again;Can we do a while loop instead of "goto again;"?
yep, not sure why I went with goto...
while (offset < btf->start_str_off)
btf = btf->base_btf;
Shorter.
Quoted 11 lines
Quoted 11 lines
kernel/bpf/btf.c +3
-0+ } + offset -= btf->start_str_off; +return offset < btf->hdr.str_len; }static bool __btf_name_char_ok (char c,bool first,bool dot_ok)@@ -615,10 +640,25 @@ static bool __btf_name_char_ok(char c, bool first, bool dot_ok)return true; }
Quoted 9 lines
kernel/bpf/btf.c static bool __btf_name_char_ok(char c, bool first, bool dot_ok) · line 649
Quoted 7 lines
kernel/bpf/btf.c +7
-0+static const char *btf_str_by_offset (const struct btf *btf,u32 offset) +{ +again: +if (offset < btf->start_str_off) { + btf = btf->base_btf; + goto again; + }Maybe add a
btf_find_base_btf(btf, offset)helper for this logic?
No strong feelings about this, but given it's a two-line loop might not be worth it. I'd also need a pretty verbose btf_find_base_btf_for_str_offset() and btf_find_base_btf_for_type_id(). I feel like loop might be less distracting actually.
Quoted 23 lines
Quoted 9 lines
kernel/bpf/btf.c +8
-0+ + offset -= btf->start_str_off; +if (offset < btf->hdr.str_len) +return &btf->strings[offset]; + +return NULL ; +} +[...]
Quoted 10 lines
}
const char *btf_name_by_offset(const struct btf *btf, u32 offset) {
kernel/bpf/btf.c +1
-4-if (offset < btf->hdr.str_len) -return &btf->strings[offset]; - -return NULL ; +return btf_str_by_offset (btf, offset);}
Quoted 2 lines
IIUC,
btf_str_by_offset()andbtf_name_by_offset()are identical. Can we just keepbtf_name_by_offset()?
btf_str_by_offset() is static, so should be inlinable, while btf_name_by_offset() is a global function, I was worrying about regressing performance for __btf_name_valid() and __btf_name_by_offset(). Premature optimization you think?
Quoted 17 lines
kernel/bpf/btf.c static bool btf_name_valid_section(const struct btf *btf, u32 offset) · line 714
Quoted 11 lines
const struct
btf_type*btf_type_by_id(const struct btf *btf, u32type_id) {kernel/bpf/btf.c +5
-2-if (type_id > btf->nr_types) -return NULL ; +again: +if (type_id < btf->start_id) { + btf = btf->base_btf; + goto again; + }ditto, goto again..
[...]
Andrii Nakryiko
maintainerPATCH v2 2/5 bpf: assign ID to vmlinux BTF and return extra info for BTF in GET_OBJ_INFO
In reply to Andrii NakryikoAllocate ID for vmlinux BTF. This makes it visible when iterating over all BTF objects in the system. To allow distinguishing vmlinux BTF (and later kernel module BTF) from user-provided BTFs, expose extra kernel_btf flag, as well as BTF name ("vmlinux" for vmlinux BTF, will equal to module's name for module BTF). We might want to later allow specifying BTF name for user-provided BTFs as well, if that makes sense. But currently this is reserved only for in-kernel BTFs.
Having in-kernel BTFs exposed IDs will allow to extend BPF APIs that require in-kernel BTF type with ability to specify BTF types from kernel modules, not just vmlinux BTF. This will be implemented in a follow up patch set for fentry/fexit/fmod_ret/lsm/etc.
--- include/uapi/linux/bpf.h | 3 +++ kernel/bpf/btf.c | 39 ++++++++++++++++++++++++++++++++-- tools/include/uapi/linux/bpf.h | 3 +++ 3 files changed, 43 insertions(+), 2 deletions(-)
include/uapi/linux/bpf.h +3 -0
@@ -4466,6 +4466,9 @@ struct bpf_btf_info {__aligned_u64 btf;__u32 btf_size;__u32 id; +__aligned_u64 name; +__u32 name_len; +__u32 kernel_btf; } __attribute__((aligned(8)));struct bpf_link_info {
kernel/bpf/btf.c +37 -2
@@ -215,6 +215,8 @@ struct btf {struct btf *base_btf;u32 start_id;/* first type ID in this BTF (0 for base BTF) */ u32 start_str_off;/* first string offset (0 for base BTF) */ +char name[MODULE_NAME_LEN ]; +bool kernel_btf; };enum verifier_phase {
@@ -4441,6 +4443,8 @@ struct btf *btf_parse_vmlinux(void) btf->data = __start_BTF; btf->data_size = __stop_BTF - __start_BTF; + btf->kernel_btf = true; +snprintf (btf->name,sizeof (btf->name),"vmlinux" ); err =btf_parse_hdr (env);if (err)
@@ -4466,8 +4470,13 @@ struct btf *btf_parse_vmlinux(void)bpf_struct_ops_init (btf, log); -btf_verifier_env_free (env);refcount_set (&btf->refcnt,1 ); + + err =btf_alloc_id (btf); +if (err) + goto errout; + +btf_verifier_env_free (env);return btf; errout:
@@ -5565,7 +5574,8 @@ int btf_get_info_by_fd(const struct btf *btf,struct bpf_btf_info info;u32 info_copy, btf_copy;void __user *ubtf; -u32 uinfo_len; +char __user *uname; +u32 uinfo_len, uname_len, name_len; uinfo =u64_to_user_ptr (attr->info.info); uinfo_len = attr->info.info_len;
@@ -5582,6 +5592,31 @@ int btf_get_info_by_fd(const struct btf *btf,return -EFAULT ; info.btf_size = btf->data_size; + info.kernel_btf = btf->kernel_btf; + + uname =u64_to_user_ptr (info.name); + uname_len = info.name_len; +if (!uname ^ !uname_len) +return -EINVAL ; + + name_len =strlen (btf->name); + info.name_len = name_len; + +if (uname) { +if (uname_len >= name_len +1 ) { +if (copy_to_user (uname, btf->name, name_len +1 )) +return -EFAULT ; + }else { +char zero ='\0' ; + +if (copy_to_user (uname, btf->name, uname_len -1 )) +return -EFAULT ; +if (put_user (zero, uname + uname_len -1 )) +return -EFAULT ; +return -ENOSPC ; + } + } +if (copy_to_user (uinfo, &info, info_copy) ||put_user (info_copy, &uattr->info.info_len))return -EFAULT ;
tools/include/uapi/linux/bpf.h +3 -0
@@ -4466,6 +4466,9 @@ struct bpf_btf_info {__aligned_u64 btf;__u32 btf_size;__u32 id; +__aligned_u64 name; +__u32 name_len; +__u32 kernel_btf; } __attribute__((aligned(8)));struct bpf_link_info {
Signature · 1 line
2.24.1
Andrii Nakryiko
maintainerPATCH v2 3/5 kbuild: build kernel module BTFs if BTF is enabled and pahole supports it
In reply to Andrii NakryikoDetect if pahole supports split BTF generation, and generate BTF for each selected kernel module, if it does. This is exposed to Makefiles and C code as CONFIG_DEBUG_INFO_BTF_MODULES flag.
Kernel module BTF has to be re-generated if either vmlinux's BTF changes or module's .ko changes. To achieve that, I needed a helper similar to if_changed, but that would allow to filter out vmlinux from the list of updated dependencies for .ko building. I've put it next to the only place that uses and needs it, but it might be a better idea to just add it along the other if_changed variants into scripts/Kbuild.include.
Each kernel module's BTF deduplication is pretty fast, as it does only incremental BTF deduplication on top of already deduplicated vmlinux BTF. To show the added build time, I've first ran make only just built kernel (to establish the baseline) and then forced only BTF re-generation, without regenerating .ko files. The build was performed with -j60 parallelization on 56-core machine. The final time also includes bzImage building, so it's not a pure BTF overhead.
$ time make -j60 ... make -j60 27.65s user 10.96s system 782% cpu 4.933 total $ touch ~/linux-build/default/vmlinux && time make -j60 ... make -j60 123.69s user 27.85s system 1566% cpu 9.675 total
So 4.6 seconds real time, with noticeable part spent in compressed vmlinux and bzImage building.
To show size savings, I've built my kernel configuration with about 700 kernel modules with full BTF per each kernel module (without deduplicating against vmlinux) and with split BTF against deduplicated vmlinux (approach in this patch). Below are top 10 modules with biggest BTF sizes. And total size of BTF data across all kernel modules.
It shows that split BTF "compresses" 115MB down to 5MB total. And the biggest kernel modules get a downsize from 500-570KB down to 200-300KB.
FULL BTF
========
$ for f in $(find . -name '*.ko'); do size -A -d $f | grep BTF | awk '{print $2}'; done | awk '{ s += $1 } END { print s }'
115710691$ for f in $(find . -name '*.ko'); do printf "%s %d\n" $f $(size -A -d $f | grep BTF | awk '{print $2}'); done | sort -nr -k2 | head -n10
./drivers/gpu/drm/i915/i915.ko 570570
./drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.ko 520240
./drivers/gpu/drm/radeon/radeon.ko 503849
./drivers/infiniband/hw/mlx5/mlx5_ib.ko 491777
./fs/xfs/xfs.ko 411544
./drivers/net/ethernet/intel/i40e/i40e.ko 403904
./drivers/net/ethernet/broadcom/bnx2x/bnx2x.ko 398754
./drivers/infiniband/core/ib_core.ko 397224
./fs/cifs/cifs.ko 386249
./fs/nfsd/nfsd.ko 379738SPLIT BTF
=========
$ for f in $(find . -name '*.ko'); do size -A -d $f | grep BTF | awk '{print $2}'; done | awk '{ s += $1 } END { print s }'
5194047$ for f in $(find . -name '*.ko'); do printf "%s %d\n" $f $(size -A -d $f | grep BTF | awk '{print $2}'); done | sort -nr -k2 | head -n10
./drivers/gpu/drm/i915/i915.ko 293206
./drivers/gpu/drm/radeon/radeon.ko 282103
./fs/xfs/xfs.ko 222150
./drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.ko 198503
./drivers/infiniband/hw/mlx5/mlx5_ib.ko 198356
./drivers/net/ethernet/broadcom/bnx2x/bnx2x.ko 113444
./fs/cifs/cifs.ko 109379
./arch/x86/kvm/kvm.ko 100225
./drivers/gpu/drm/drm.ko 94827
./drivers/infiniband/core/ib_core.ko 91188--- lib/Kconfig.debug | 9 +++++++++ scripts/Makefile.modfinal | 20 ++++++++++++++++++-- 2 files changed, 27 insertions(+), 2 deletions(-)
lib/Kconfig.debug +9 -0
@@ -274,6 +274,15 @@ config DEBUG_INFO_BTF Turning this on expects presence of pahole tool, which will convert DWARF type info into equivalent deduplicated BTF type info. +config PAHOLE_HAS_SPLIT_BTF + def_bool $(success, test `$(PAHOLE) --version | sed -E 's/v([0-9]+)\.([0-9]+)/\1\2/'` -ge "119") + +config DEBUG_INFO_BTF_MODULES + def_bool y + depends on DEBUG_INFO_BTF && MODULES && PAHOLE_HAS_SPLIT_BTF + help + Generate compact split BTF type information for kernel modules. + config GDB_SCRIPTS bool "Provide GDB scripts for kernel debugging" help
scripts/Makefile.modfinal +18 -2
@@ -6,6 +6,7 @@ PHONY := __modfinal __modfinal: +include include/config/auto.conf include $(srctree)/scripts/Kbuild.include # for c_flags
@@ -36,8 +37,23 @@ quiet_cmd_ld_ko_o = LD [M] $@ -T scripts/module.lds -o $@ $(filter %.o, $^); \ $(if $(ARCH_POSTLINK), $(MAKE) -f $(ARCH_POSTLINK) $@, true) -$(modules): %.ko: %.o %.mod.o scripts/module.lds FORCE - +$(call if_changed,ld_ko_o) +quiet_cmd_btf_ko = BTF [M] $@ + cmd_btf_ko = LLVM_OBJCOPY=$(OBJCOPY) $(PAHOLE) -J --btf_base vmlinux $@ + +# Same as newer-prereqs, but allows to exclude specified extra dependencies +newer_prereqs_except = $(filter-out $(PHONY) $(1),$?) + +# Same as if_changed, but allows to exclude specified extra dependencies +if_changed_except = $(if $(call newer_prereqs_except,$(2))$(cmd-check), \ + $(cmd); \ + printf '%s\n' 'cmd_$@ := $(make-cmd)' > $(dot-target).cmd, @:) + +# Re-generate module BTFs if either module's .ko or vmlinux changed +$(modules): %.ko: %.o %.mod.o scripts/module.lds vmlinux FORCE + +$(call if_changed_except,ld_ko_o,vmlinux) +ifdef CONFIG_DEBUG_INFO_BTF_MODULES + +$(if $(newer-prereqs),$(call cmd,btf_ko)) +endif targets += $(modules) $(modules:.ko=.mod.o)
Signature · 1 line
2.24.1
Andrii Nakryiko
maintainerPATCH v2 4/5 bpf: load and verify kernel module BTFs
In reply to Andrii NakryikoAdd kernel module listener that will load/validate and unload module BTF. Module BTFs gets ID generated for them, which makes it possible to iterate them with existing BTF iteration API. They are given their respective module's names, which will get reported through GET_OBJ_INFO API. They are also marked as in-kernel BTFs for tooling to distinguish them from user-provided BTFs.
Also, similarly to vmlinux BTF, kernel module BTFs are exposed through sysfs as /sys/kernel/btf/<module-name>. This is convenient for user-space tools to inspect module BTF contents and dump their types with existing tools:
[vmuser@archvm bpf]$ ls -la /sys/kernel/btf
total 0
drwxr-xr-x 2 root root 0 Nov 4 19:46 . drwxr-xr-x 13 root root 0 Nov 4 19:46 ..
...
-r--r--r-- 1 root root 888 Nov 4 19:46 irqbypass -r--r--r-- 1 root root 100225 Nov 4 19:46 kvm -r--r--r-- 1 root root 35401 Nov 4 19:46 kvm_intel -r--r--r-- 1 root root 120 Nov 4 19:46 pcspkr -r--r--r-- 1 root root 399 Nov 4 19:46 serio_raw -r--r--r-- 1 root root 4094095 Nov 4 19:46 vmlinux
--- Documentation/ABI/testing/sysfs-kernel-btf | 8 + include/linux/bpf.h | 2 + include/linux/module.h | 4 + kernel/bpf/btf.c | 194 +++++++++++++++++++++ kernel/bpf/sysfs_btf.c | 2 +- kernel/module.c | 32 ++++ 6 files changed, 241 insertions(+), 1 deletion(-)
Documentation/ABI/testing/sysfs-kernel-btf +8 -0
@@ -15,3 +15,11 @@ Description: information with description of all internal kernel types. See Documentation/bpf/btf.rst for detailed description of format itself. + +What: /sys/kernel/btf/<module-name> +Date: Nov 2020 +KernelVersion: 5.11 +Contact: [email protected] +Description: + Read-only binary attribute exposing kernel module's BTF type + information as an add-on to the kernel's BTF (/sys/kernel/btf/vmlinux).
include/linux/bpf.h +2 -0
@@ -36,9 +36,11 @@ struct seq_operations;struct bpf_iter_aux_info ;struct bpf_local_storage ;struct bpf_local_storage_map ; +struct kobject ;extern struct idr btf_idr;extern spinlock_t btf_idr_lock; +extern struct kobject *btf_kobj;typedef int (*bpf_iter_init_seq_priv_t )(void *private_data,struct bpf_iter_aux_info *aux);
include/linux/module.h +4 -0
@@ -475,6 +475,10 @@ struct module {unsigned int num_bpf_raw_events;struct bpf_raw_event_map *bpf_raw_events;#endif +#ifdef CONFIG_DEBUG_INFO_BTF_MODULES +unsigned int btf_data_size; +void *btf_data; +#endif #ifdef CONFIG_JUMP_LABEL struct jump_entry *jump_entries;unsigned int num_jump_entries;
kernel/bpf/btf.c +194 -0
209 lines not on this page · raw message
kernel/bpf/sysfs_btf.c +1 -1
@@ -26,7 +26,7 @@ static struct bin_attribute bin_attr_btf_vmlinux __ro_after_init = { .read = btf_vmlinux_read, }; -static struct kobject *btf_kobj; +struct kobject *btf_kobj;static int __init btf_vmlinux_init (void ) {
kernel/module.c +32 -0
@@ -380,6 +380,35 @@ static void *section_objs(const struct load_info *info,return (void *)info->sechdrs[sec].sh_addr; } +/* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */ +static unsigned int find_any_sec (const struct load_info *info,const char *name) +{ +unsigned int i; + +for (i =1 ; i < info->hdr->e_shnum; i++) { +Elf_Shdr *shdr = &info->sechdrs[i]; +if (strcmp (info->secstrings + shdr->sh_name, name) ==0 ) +return i; + } +return 0 ; +} + +/* +* Find a module section, or NULL. Fill in number of "objects" in section. +* Ignores SHF_ALLOC flag. +*/ +static __maybe_unused void *any_section_objs (const struct load_info *info, +const char *name, +size_t object_size, +unsigned int *num) +{ +unsigned int sec =find_any_sec (info, name); + +/* Section 0 has sh_addr 0 and sh_size 0. */ + *num = info->sechdrs[sec].sh_size / object_size; +return (void *)info->sechdrs[sec].sh_addr; +} +/* Provided by the linker */ extern const struct kernel_symbol __start___ksymtab[];extern const struct kernel_symbol __stop___ksymtab[];
@@ -3250,6 +3279,9 @@ static int find_module_sections(struct module *mod, struct load_info *info)sizeof (*mod->bpf_raw_events), &mod->num_bpf_raw_events);#endif +#ifdef CONFIG_DEBUG_INFO_BTF_MODULES + mod->btf_data =any_section_objs (info,".BTF" ,1 , &mod->btf_data_size); +#endif #ifdef CONFIG_JUMP_LABEL mod->jump_entries = section_objs(info, "__jump_table", sizeof(*mod->jump_entries),
Signature · 1 line
2.24.1
Re: [PATCH v2 bpf-next 4/5] bpf: load and verify kernel module BTFs
Quoted 25 lines · Andrii Nakryiko
Add kernel module listener that will load/validate and unload module BTF. Module BTFs gets ID generated for them, which makes it possible to iterate them with existing BTF iteration API. They are given their respective module's names, which will get reported through
GET_OBJ_INFOAPI. They are also marked as in-kernel BTFs for tooling to distinguish them from user-provided BTFs.Also, similarly to vmlinux BTF, kernel module BTFs are exposed through sysfs as /sys/kernel/btf/<module-name>. This is convenient for user-space tools to inspect module BTF contents and dump their types with existing tools:
[vmuser@archvm bpf]$ ls -la
/sys/kernel/btf
total 0drwxr-xr-x 2 root root 0 Nov 4 19:46 . drwxr-xr-x 13 root root 0 Nov 4 19:46 .....
(quoted diff) +0
-6-r--r--r-- 1 root root 888 Nov 4 19:46 irqbypass -r--r--r-- 1 root root 100225 Nov 4 19:46 kvm -r--r--r-- 1 root root 35401 Nov 4 19:46 kvm_intel -r--r--r-- 1 root root 120 Nov 4 19:46 pcspkr -r--r--r-- 1 root root 399 Nov 4 19:46 serio_raw -r--r--r-- 1 root root 4094095 Nov 4 19:46 vmlinuxSigned-off-by Andrii Nakryiko <[email protected]>
Quoted 8 lines
--- Documentation/ABI/testing/sysfs-kernel-btf | 8 + include/linux/bpf.h | 2 + include/linux/module.h | 4 + kernel/bpf/btf.c | 194 +++++++++++++++++++++ kernel/bpf/sysfs_btf.c | 2 +- kernel/module.c | 32 ++++ 6 files changed, 241 insertions(+), 1 deletion(-)
Andrii Nakryiko
maintainerPATCH v2 5/5 tools/bpftool: add support for in-kernel and named BTF in `btf show`
In reply to Andrii NakryikoDisplay vmlinux BTF name and kernel module names when listing available BTFs on the system.
In human-readable output mode, module BTFs are reported with "name [module-name]", while vmlinux BTF will be reported as "name [vmlinux]". Square brackets are added by bpftool and follow kernel convention when displaying modules in human-readable text outputs.
[vmuser@archvm bpf]$ sudo ../../../bpf/bpftool/bpftool btf s 1: name [vmlinux] size 4082281B
6: size 2365B prog_ids 8,6 map_ids 3
7: name [button] size 46895B
8: name [pcspkr] size 42328B
9: name [serio_raw] size 39375B
10: name [floppy] size 57185B
11: name [i2c_core] size 76186B
12: name [crc32c_intel] size 16036B
13: name [i2c_piix4] size 50497B
14: name [irqbypass] size 14124B
15: name [kvm] size 197985B
16: name [kvm_intel] size 123564B
17: name [cryptd] size 42466B
18: name [crypto_simd] size 17187B
19: name [glue_helper] size 39205B
20: name [aesni_intel] size 41034B
25: size 36150B
pids bpftool(2519)In JSON mode, two fields (boolean "kernel" and string "name") are reported for each BTF object. vmlinux BTF is reported with name "vmlinux" (kernel itself returns and empty name for vmlinux BTF).
[vmuser@archvm bpf]$ sudo ../../../bpf/bpftool/bpftool btf s -jp
[{
"id": 1,
"size": 4082281,
"prog_ids": [],
"map_ids": [],
"kernel": true,
"name": "vmlinux"
},{
"id": 6,
"size": 2365,
"prog_ids": [8,6
],
"map_ids": [3
],
"kernel": false
},{
"id": 7,
"size": 46895,
"prog_ids": [],
"map_ids": [],
"kernel": true,
"name": "button"
},{...
--- tools/bpf/bpftool/btf.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-)
tools/bpf/bpftool/btf.c +27 -1
@@ -742,9 +742,14 @@ show_btf_plain(struct bpf_btf_info *info, int fd, struct btf_attach_table *btf_map_table) {struct btf_attach_point *obj; +const char *name =u64_to_ptr (info->name);int n;printf ("%u: " , info->id); +if (info->kernel_btf) +printf ("name [%s] " , name); +else if (name && name[0 ]) +printf ("name %s " , name);printf ("size %uB" , info->btf_size); n =0 ;
@@ -771,6 +776,7 @@ show_btf_json(struct bpf_btf_info *info, int fd, struct btf_attach_table *btf_map_table) {struct btf_attach_point *obj; +const char *name =u64_to_ptr (info->name);jsonw_start_object (json_wtr);/* btf object */ jsonw_uint_field (json_wtr,"id" , info->id);
@@ -796,6 +802,11 @@ show_btf_json(struct bpf_btf_info *info, int fd,emit_obj_refs_json (&refs_table, info->id, json_wtr);/* pids */ +jsonw_bool_field (json_wtr,"kernel" , info->kernel_btf); + +if (name && name[0 ]) +jsonw_string_field (json_wtr,"name" , name); +jsonw_end_object (json_wtr);/* btf object */ }
@@ -803,15 +814,30 @@ static int show_btf(int fd, struct btf_attach_table *btf_prog_table, struct btf_attach_table *btf_map_table) { - struct bpf_btf_info info = {}; + struct bpf_btf_info info;__u32 len =sizeof (info); +char name[64 ];int err; +memset (&info,0 ,sizeof (info)); err =bpf_obj_get_info_by_fd (fd, &info, &len);if (err) {p_err ("can't get BTF object info: %s" ,strerror (errno));return -1 ; } +/* if kernel support emitting BTF object name, pass name pointer */ +if (info.name_len) { +memset (&info,0 ,sizeof (info)); + info.name_len =sizeof (name); + info.name =ptr_to_u64 (name); + len =sizeof (info); + + err =bpf_obj_get_info_by_fd (fd, &info, &len); +if (err) { +p_err ("can't get BTF object info: %s" ,strerror (errno)); +return -1 ; + } + }if (json_output)show_btf_json (&info, fd, btf_prog_table, btf_map_table);
Signature · 1 line
2.24.1