libbpf: Convert ELF notes into read-only maps
cover libbpf: Convert ELF notes into read-only maps
While attempting to implement a bpf-based gatekeeper program as was described
https://lore.kernel.org/all/[email protected]/T/#mb10f3112df1a66c725df9d6035c5a68c72a0eb8d
we noticed that relying on IMA and fs-verity signatures alone was insufficient. A user with sufficient privileges could ptrace, ld preload or poke at memory in some other way while using a signed lskel, leaving the signature intact, allowing them to load whatever they wished into the kernel effectively circumventing the gatekeeper. That may be considered insecure in some scenarios.
Here we propose a very simple method of allowing metadata to be stored in skeletons or dynamic libbpf-based loaders, by simply treating note sections as read-only maps that are visible to the gatekeeper program. Gatekeeper programs can then iterate the fd_array and see if there are any relevant maps that they wish to consult. No changes to the kernel-proper are required for this, and this should help facilitate the implementation and design of secure bpf-based gatekeepers, while keeping with the overall philosophy of bpf and not enforcing any obtusive abstractions upon anyone.
Blaise Boscaccy (1):
libbpf: Convert ELF notes into read-only maps
tools/bpf/bpftool/gen.c | 4 ++-- tools/lib/bpf/libbpf.c | 6 ++++++ 2 files changed, 8 insertions(+), 2 deletions(-)
Signature · 1 line
2.48.1
PATCH 1/1 libbpf: Convert ELF notes into read-only maps
Add a flexible mechanism, using existing ELF constructs, to attach additional metadata to BPF programs for possible use by BPF gatekeepers and skeletons.
During object file parsing, note sections are no longer skipped and now treated as read-only data. During libbpf-based loading or skeleton generation, those sections are then transformed into read-only maps which are subsequently passed into the kernel.
--- tools/bpf/bpftool/gen.c | 4 ++-- tools/lib/bpf/libbpf.c | 6 ++++++ 2 files changed, 8 insertions(+), 2 deletions(-)
tools/bpf/bpftool/gen.c +2 -2
@@ -92,7 +92,7 @@ static void get_header_guard(char *guard, const char *obj_name, const char *suffstatic bool get_map_ident (const struct bpf_map *map,char *buf,size_t buf_sz) { -static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };const char *name =bpf_map__name (map);int i, n;
@@ -117,7 +117,7 @@ static bool get_map_ident(const struct bpf_map *map, char *buf, size_t buf_sz)static bool get_datasec_ident (const char *sec_name,char *buf,size_t buf_sz) { -static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };int i, n;/* recognize hard coded LLVM section name */
tools/lib/bpf/libbpf.c +6 -0
@@ -523,6 +523,7 @@ struct bpf_struct_ops {#define STRUCT_OPS_SEC ".struct_ops"#define STRUCT_OPS_LINK_SEC ".struct_ops.link"#define ARENA_SEC ".addr_space.1" +#define NOTE_SEC ".note"enum libbpf_map_type {LIBBPF_MAP_UNSPEC ,
@@ -3977,6 +3978,11 @@ static int bpf_object__elf_collect(struct bpf_object *obj) sec_desc->sec_type =SEC_BSS ; sec_desc->shdr = sh; sec_desc->data = data; + }else if (sh->sh_type ==SHT_NOTE && (strcmp (name,NOTE_SEC ) ==0 || +str_has_pfx (name,NOTE_SEC "." ))) { + sec_desc->sec_type =SEC_RODATA ; + sec_desc->shdr = sh; + sec_desc->data = data; }else {pr_info ("elf: skipping section(%d) %s (size %zu)\n" , idx, name, (size_t )sh->sh_size);
Signature · 1 line
2.48.1
Quoted 9 lines · Blaise Boscaccy
Add a flexible mechanism, using existing ELF constructs, to attach additional metadata to BPF programs for possible use by BPF gatekeepers and skeletons.
During object file parsing, note sections are no longer skipped and now treated as read-only data. During libbpf-based loading or skeleton generation, those sections are then transformed into read-only maps which are subsequently passed into the kernel.
We already have this mechanism, it's .rodata (and .rodata.<customname>) section(s). Adding .note sections as BPF maps make no sense to me. Just piggy-back on .rodata for storing any necessary metadata.
pw-bot: cr
Quoted 56 lines
Signed-off-by Blaise Boscaccy <[email protected]>--- tools/bpf/bpftool/gen.c | 4 ++-- tools/lib/bpf/libbpf.c | 6 ++++++ 2 files changed, 8 insertions(+), 2 deletions(-)tools/bpf/bpftool/gen.c +2
-2@@ -92,7 +92,7 @@ static void get_header_guard(char *guard, const char *obj_name, const char *suffstatic bool get_map_ident (const struct bpf_map *map,char *buf,size_t buf_sz) { -static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };const char *name =bpf_map__name (map);int i, n;@@ -117,7 +117,7 @@ static bool get_map_ident(const struct bpf_map *map, char *buf, size_t buf_sz)static bool get_datasec_ident (const char *sec_name,char *buf,size_t buf_sz) { -static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };int i, n;/* recognize hard coded LLVM section name */ tools/lib/bpf/libbpf.c +6
-1@@ -523,6 +523,7 @@ struct bpf_struct_ops {#define STRUCT_OPS_SEC ".struct_ops"#define STRUCT_OPS_LINK_SEC ".struct_ops.link"#define ARENA_SEC ".addr_space.1" +#define NOTE_SEC ".note"enum libbpf_map_type {LIBBPF_MAP_UNSPEC ,@@ -3977,6 +3978,11 @@ static int bpf_object__elf_collect(struct bpf_object *obj) sec_desc->sec_type =SEC_BSS ; sec_desc->shdr = sh; sec_desc->data = data; + }else if (sh->sh_type ==SHT_NOTE && (strcmp (name,NOTE_SEC ) ==0 || +str_has_pfx (name,NOTE_SEC "." ))) { + sec_desc->sec_type =SEC_RODATA ; + sec_desc->shdr = sh; + sec_desc->data = data; }else {pr_info ("elf: skipping section(%d) %s (size %zu)\n" , idx, name, (size_t )sh->sh_size); --2.48.1
Quoted 12 lines · Andrii Nakryiko
Quoted 9 lines · Blaise Boscaccy
Add a flexible mechanism, using existing ELF constructs, to attach additional metadata to BPF programs for possible use by BPF gatekeepers and skeletons.
During object file parsing, note sections are no longer skipped and now treated as read-only data. During libbpf-based loading or skeleton generation, those sections are then transformed into read-only maps which are subsequently passed into the kernel.
Quoted 7 lines
We already have this mechanism, it's .rodata (and .rodata.<customname>) section(s). Adding .note sections as BPF maps make no sense to me. Just piggy-back on .rodata for storing any necessary metadata.
pw-bot: cr
The ELF specification clearly states:
"Sometimes a vendor or system builder needs to mark an object file with special information that other programs will check for conformance, compatibility, etc. Sections of type SHT_NOTE and program header elements of type PT_NOTE can be used for this purpose."
Which is exactly what we are trying to do. They make no mention of piggy-backing off of .rodata.
Further, one can generally strip away .note sections and be left with an object that's still functioning. The same cannot be said about .rodata.
Quoted 56 lines
Quoted 56 lines
Signed-off-by Blaise Boscaccy <[email protected]>--- tools/bpf/bpftool/gen.c | 4 ++-- tools/lib/bpf/libbpf.c | 6 ++++++ 2 files changed, 8 insertions(+), 2 deletions(-)tools/bpf/bpftool/gen.c +2
-2@@ -92,7 +92,7 @@ static void get_header_guard(char *guard, const char *obj_name, const char *suffstatic bool get_map_ident (const struct bpf_map *map,char *buf,size_t buf_sz) { -static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };const char *name =bpf_map__name (map);int i, n;@@ -117,7 +117,7 @@ static bool get_map_ident(const struct bpf_map *map, char *buf, size_t buf_sz)static bool get_datasec_ident (const char *sec_name,char *buf,size_t buf_sz) { -static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };int i, n;/* recognize hard coded LLVM section name */ tools/lib/bpf/libbpf.c +6
-1@@ -523,6 +523,7 @@ struct bpf_struct_ops {#define STRUCT_OPS_SEC ".struct_ops"#define STRUCT_OPS_LINK_SEC ".struct_ops.link"#define ARENA_SEC ".addr_space.1" +#define NOTE_SEC ".note"enum libbpf_map_type {LIBBPF_MAP_UNSPEC ,@@ -3977,6 +3978,11 @@ static int bpf_object__elf_collect(struct bpf_object *obj) sec_desc->sec_type =SEC_BSS ; sec_desc->shdr = sh; sec_desc->data = data; + }else if (sh->sh_type ==SHT_NOTE && (strcmp (name,NOTE_SEC ) ==0 || +str_has_pfx (name,NOTE_SEC "." ))) { + sec_desc->sec_type =SEC_RODATA ; + sec_desc->shdr = sh; + sec_desc->data = data; }else {pr_info ("elf: skipping section(%d) %s (size %zu)\n" , idx, name, (size_t )sh->sh_size); --2.48.1
Quoted 20 lines · Blaise Boscaccy
Quoted 17 lines · Andrii Nakryiko
Quoted 9 lines · Blaise Boscaccy
Add a flexible mechanism, using existing ELF constructs, to attach additional metadata to BPF programs for possible use by BPF gatekeepers and skeletons.
During object file parsing, note sections are no longer skipped and now treated as read-only data. During libbpf-based loading or skeleton generation, those sections are then transformed into read-only maps which are subsequently passed into the kernel.
We already have this mechanism, it's .rodata (and .rodata.<customname>) section(s). Adding .note sections as BPF maps make no sense to me. Just piggy-back on .rodata for storing any necessary metadata.
Quoted 8 lines
Quoted 2 lines
pw-bot: cr
The ELF specification clearly states:
"Sometimes a vendor or system builder needs to mark an object file with special information that other programs will check for conformance, compatibility, etc. Sections of typeSHT_NOTEand program header elements of typePT_NOTEcan be used for this purpose."
Does ELF specification say anything about "and libbpf should create a BPF map for those SHT_NOTE sections"?
Quoted 3 lines
Which is exactly what we are trying to do. They make no mention of piggy-backing off of .rodata.
You are trying to redefine non-loadable SHT_NOTE data into loadable data backed by a BPF map. I'm not sure how your arguments are supporting the hack you are trying to do. We are not going to start creating new BPF maps for any random SHT_NOTE section in the BPF object file. Use .rodata if you want to get some read-only data into the kernel.
Quoted 62 lines
Further, one can generally strip away .note sections and be left with an object that's still functioning. The same cannot be said about .rodata.
Quoted 56 lines
Quoted 56 lines
Signed-off-by Blaise Boscaccy <[email protected]>--- tools/bpf/bpftool/gen.c | 4 ++-- tools/lib/bpf/libbpf.c | 6 ++++++ 2 files changed, 8 insertions(+), 2 deletions(-)tools/bpf/bpftool/gen.c +2
-2@@ -92,7 +92,7 @@ static void get_header_guard(char *guard, const char *obj_name, const char *suffstatic bool get_map_ident (const struct bpf_map *map,char *buf,size_t buf_sz) { -static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *sfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };const char *name =bpf_map__name (map);int i, n;@@ -117,7 +117,7 @@ static bool get_map_ident(const struct bpf_map *map, char *buf, size_t buf_sz)static bool get_datasec_ident (const char *sec_name,char *buf,size_t buf_sz) { -static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" }; +static const char *pfxs[] = {".data" ,".rodata" ,".bss" ,".kconfig" ,".note" };int i, n;/* recognize hard coded LLVM section name */ tools/lib/bpf/libbpf.c +6
-1@@ -523,6 +523,7 @@ struct bpf_struct_ops {#define STRUCT_OPS_SEC ".struct_ops"#define STRUCT_OPS_LINK_SEC ".struct_ops.link"#define ARENA_SEC ".addr_space.1" +#define NOTE_SEC ".note"enum libbpf_map_type {LIBBPF_MAP_UNSPEC ,@@ -3977,6 +3978,11 @@ static int bpf_object__elf_collect(struct bpf_object *obj) sec_desc->sec_type =SEC_BSS ; sec_desc->shdr = sh; sec_desc->data = data; + }else if (sh->sh_type ==SHT_NOTE && (strcmp (name,NOTE_SEC ) ==0 || +str_has_pfx (name,NOTE_SEC "." ))) { + sec_desc->sec_type =SEC_RODATA ; + sec_desc->shdr = sh; + sec_desc->data = data; }else {pr_info ("elf: skipping section(%d) %s (size %zu)\n" , idx, name, (size_t )sh->sh_size); --2.48.1