提交 faf6ed32 编写于 作者: A Andrii Nakryiko 提交者: Alexei Starovoitov

libbpf: Add BPF static linker APIs

Introduce BPF static linker APIs to libbpf. BPF static linker allows to
perform static linking of multiple BPF object files into a single combined
resulting object file, preserving all the BPF programs, maps, global
variables, etc.

Data sections (.bss, .data, .rodata, .maps, maps, etc) with the same name are
concatenated together. Similarly, code sections are also concatenated. All the
symbols and ELF relocations are also concatenated in their respective ELF
sections and are adjusted accordingly to the new object file layout.

Static variables and functions are handled correctly as well, adjusting BPF
instructions offsets to reflect new variable/function offset within the
combined ELF section. Such relocations are referencing STT_SECTION symbols and
that stays intact.

Data sections in different files can have different alignment requirements, so
that is taken care of as well, adjusting sizes and offsets as necessary to
satisfy both old and new alignment requirements.

DWARF data sections are stripped out, currently. As well as LLLVM_ADDRSIG
section, which is ignored by libbpf in bpf_object__open() anyways. So, in
a way, BPF static linker is an analogue to `llvm-strip -g`, which is a pretty
nice property, especially if resulting .o file is then used to generate BPF
skeleton.

Original string sections are ignored and instead we construct our own set of
unique strings using libbpf-internal `struct strset` API.

To reduce the size of the patch, all the .BTF and .BTF.ext processing was
moved into a separate patch.

The high-level API consists of just 4 functions:
  - bpf_linker__new() creates an instance of BPF static linker. It accepts
    output filename and (currently empty) options struct;
  - bpf_linker__add_file() takes input filename and appends it to the already
    processed ELF data; it can be called multiple times, one for each BPF
    ELF object file that needs to be linked in;
  - bpf_linker__finalize() needs to be called to dump final ELF contents into
    the output file, specified when bpf_linker was created; after
    bpf_linker__finalize() is called, no more bpf_linker__add_file() and
    bpf_linker__finalize() calls are allowed, they will return error;
  - regardless of whether bpf_linker__finalize() was called or not,
    bpf_linker__free() will free up all the used resources.

Currently, BPF static linker doesn't resolve cross-object file references
(extern variables and/or functions). This will be added in the follow up patch
set.
Signed-off-by: NAndrii Nakryiko <andrii@kernel.org>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210318194036.3521577-7-andrii@kernel.org
上级 9af44bc5
libbpf-y := libbpf.o bpf.o nlattr.o btf.o libbpf_errno.o str_error.o \
netlink.o bpf_prog_linfo.o libbpf_probes.o xsk.o hashmap.o \
btf_dump.o ringbuf.o strset.o
btf_dump.o ringbuf.o strset.o linker.o
......@@ -55,10 +55,6 @@
#include "libbpf_internal.h"
#include "hashmap.h"
#ifndef EM_BPF
#define EM_BPF 247
#endif
#ifndef BPF_FS_MAGIC
#define BPF_FS_MAGIC 0xcafe4a11
#endif
......@@ -1134,11 +1130,6 @@ static void bpf_object__elf_finish(struct bpf_object *obj)
obj->efile.obj_buf_sz = 0;
}
/* if libelf is old and doesn't support mmap(), fall back to read() */
#ifndef ELF_C_READ_MMAP
#define ELF_C_READ_MMAP ELF_C_READ
#endif
static int bpf_object__elf_init(struct bpf_object *obj)
{
int err = 0;
......@@ -2807,7 +2798,7 @@ static bool ignore_elf_section(GElf_Shdr *hdr, const char *name)
return true;
/* ignore .llvm_addrsig section as well */
if (hdr->sh_type == 0x6FFF4C03 /* SHT_LLVM_ADDRSIG */)
if (hdr->sh_type == SHT_LLVM_ADDRSIG)
return true;
/* no subprograms will lead to an empty .text section, ignore it */
......
......@@ -760,6 +760,19 @@ enum libbpf_tristate {
TRI_MODULE = 2,
};
struct bpf_linker_opts {
/* size of this struct, for forward/backward compatiblity */
size_t sz;
};
#define bpf_linker_opts__last_field sz
struct bpf_linker;
LIBBPF_API struct bpf_linker *bpf_linker__new(const char *filename, struct bpf_linker_opts *opts);
LIBBPF_API int bpf_linker__add_file(struct bpf_linker *linker, const char *filename);
LIBBPF_API int bpf_linker__finalize(struct bpf_linker *linker);
LIBBPF_API void bpf_linker__free(struct bpf_linker *linker);
#ifdef __cplusplus
} /* extern "C" */
#endif
......
......@@ -355,4 +355,8 @@ LIBBPF_0.4.0 {
global:
btf__add_float;
btf__add_type;
bpf_linker__add_file;
bpf_linker__finalize;
bpf_linker__free;
bpf_linker__new;
} LIBBPF_0.3.0;
......@@ -20,6 +20,26 @@
#include "libbpf.h"
#ifndef EM_BPF
#define EM_BPF 247
#endif
#ifndef R_BPF_64_64
#define R_BPF_64_64 1
#endif
#ifndef R_BPF_64_32
#define R_BPF_64_32 10
#endif
#ifndef SHT_LLVM_ADDRSIG
#define SHT_LLVM_ADDRSIG 0x6FFF4C03
#endif
/* if libelf is old and doesn't support mmap(), fall back to read() */
#ifndef ELF_C_READ_MMAP
#define ELF_C_READ_MMAP ELF_C_READ
#endif
#define BTF_INFO_ENC(kind, kind_flag, vlen) \
((!!(kind_flag) << 31) | ((kind) << 24) | ((vlen) & BTF_MAX_VLEN))
#define BTF_TYPE_ENC(name, info, size_or_type) (name), (info), (size_or_type)
......
此差异已折叠。
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册