2 * bpf.c BPF common code
4 * This program is free software; you can distribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Daniel Borkmann <daniel@iogearbox.net>
10 * Jiri Pirko <jiri@resnulli.us>
11 * Alexei Starovoitov <ast@kernel.org>
31 #include <sys/types.h>
35 #include <sys/mount.h>
36 #include <sys/syscall.h>
37 #include <sys/sendfile.h>
38 #include <sys/resource.h>
40 #include <arpa/inet.h>
43 #include "json_print.h"
49 struct bpf_prog_meta
{
56 static const enum bpf_prog_type __bpf_types
[] = {
57 BPF_PROG_TYPE_SCHED_CLS
,
58 BPF_PROG_TYPE_SCHED_ACT
,
61 BPF_PROG_TYPE_LWT_OUT
,
62 BPF_PROG_TYPE_LWT_XMIT
,
65 static const struct bpf_prog_meta __bpf_prog_meta
[] = {
66 [BPF_PROG_TYPE_SCHED_CLS
] = {
69 .section
= ELF_SECTION_CLASSIFIER
,
70 .may_uds_export
= true,
72 [BPF_PROG_TYPE_SCHED_ACT
] = {
75 .section
= ELF_SECTION_ACTION
,
76 .may_uds_export
= true,
78 [BPF_PROG_TYPE_XDP
] = {
81 .section
= ELF_SECTION_PROG
,
83 [BPF_PROG_TYPE_LWT_IN
] = {
86 .section
= ELF_SECTION_PROG
,
88 [BPF_PROG_TYPE_LWT_OUT
] = {
91 .section
= ELF_SECTION_PROG
,
93 [BPF_PROG_TYPE_LWT_XMIT
] = {
96 .section
= ELF_SECTION_PROG
,
98 [BPF_PROG_TYPE_LWT_SEG6LOCAL
] = {
99 .type
= "lwt_seg6local",
101 .section
= ELF_SECTION_PROG
,
105 static const char *bpf_prog_to_subdir(enum bpf_prog_type type
)
107 assert(type
< ARRAY_SIZE(__bpf_prog_meta
) &&
108 __bpf_prog_meta
[type
].subdir
);
109 return __bpf_prog_meta
[type
].subdir
;
112 const char *bpf_prog_to_default_section(enum bpf_prog_type type
)
114 assert(type
< ARRAY_SIZE(__bpf_prog_meta
) &&
115 __bpf_prog_meta
[type
].section
);
116 return __bpf_prog_meta
[type
].section
;
120 static int bpf_obj_open(const char *path
, enum bpf_prog_type type
,
121 const char *sec
, __u32 ifindex
, bool verbose
);
123 static int bpf_obj_open(const char *path
, enum bpf_prog_type type
,
124 const char *sec
, __u32 ifindex
, bool verbose
)
126 fprintf(stderr
, "No ELF library support compiled in.\n");
132 static inline __u64
bpf_ptr_to_u64(const void *ptr
)
134 return (__u64
)(unsigned long)ptr
;
137 static int bpf(int cmd
, union bpf_attr
*attr
, unsigned int size
)
140 return syscall(__NR_bpf
, cmd
, attr
, size
);
142 fprintf(stderr
, "No bpf syscall, kernel headers too old?\n");
148 static int bpf_map_update(int fd
, const void *key
, const void *value
,
151 union bpf_attr attr
= {};
154 attr
.key
= bpf_ptr_to_u64(key
);
155 attr
.value
= bpf_ptr_to_u64(value
);
158 return bpf(BPF_MAP_UPDATE_ELEM
, &attr
, sizeof(attr
));
161 static int bpf_prog_fd_by_id(uint32_t id
)
163 union bpf_attr attr
= {};
167 return bpf(BPF_PROG_GET_FD_BY_ID
, &attr
, sizeof(attr
));
170 static int bpf_prog_info_by_fd(int fd
, struct bpf_prog_info
*info
,
173 union bpf_attr attr
= {};
176 attr
.info
.bpf_fd
= fd
;
177 attr
.info
.info
= bpf_ptr_to_u64(info
);
178 attr
.info
.info_len
= *info_len
;
181 ret
= bpf(BPF_OBJ_GET_INFO_BY_FD
, &attr
, sizeof(attr
));
183 *info_len
= attr
.info
.info_len
;
188 int bpf_dump_prog_info(FILE *f
, uint32_t id
)
190 struct bpf_prog_info info
= {};
191 uint32_t len
= sizeof(info
);
192 int fd
, ret
, dump_ok
= 0;
195 open_json_object("prog");
196 print_uint(PRINT_ANY
, "id", "id %u ", id
);
198 fd
= bpf_prog_fd_by_id(id
);
202 ret
= bpf_prog_info_by_fd(fd
, &info
, &len
);
204 int jited
= !!info
.jited_prog_len
;
206 print_string(PRINT_ANY
, "tag", "tag %s ",
207 hexstring_n2a(info
.tag
, sizeof(info
.tag
),
209 print_uint(PRINT_JSON
, "jited", NULL
, jited
);
210 if (jited
&& !is_json_context())
211 fprintf(f
, "jited ");
221 static int bpf_parse_string(char *arg
, bool from_file
, __u16
*bpf_len
,
222 char **bpf_string
, bool *need_release
,
223 const char separator
)
228 size_t tmp_len
, op_len
= sizeof("65535 255 255 4294967295,");
229 char *tmp_string
, *pos
, c_prev
= ' ';
233 tmp_len
= sizeof("4096,") + BPF_MAXINSNS
* op_len
;
234 tmp_string
= pos
= calloc(1, tmp_len
);
235 if (tmp_string
== NULL
)
238 fp
= fopen(arg
, "r");
240 perror("Cannot fopen");
245 while ((c
= fgetc(fp
)) != EOF
) {
262 if (pos
- tmp_string
== tmp_len
)
275 *need_release
= true;
276 *bpf_string
= tmp_string
;
278 *need_release
= false;
282 if (sscanf(*bpf_string
, "%hu%c", bpf_len
, &sp
) != 2 ||
292 static int bpf_ops_parse(int argc
, char **argv
, struct sock_filter
*bpf_ops
,
295 char *bpf_string
, *token
, separator
= ',';
302 if (bpf_parse_string(argv
[0], from_file
, &bpf_len
, &bpf_string
,
303 &need_release
, separator
))
305 if (bpf_len
== 0 || bpf_len
> BPF_MAXINSNS
) {
311 while ((token
= strchr(token
, separator
)) && (++token
)[0]) {
313 fprintf(stderr
, "Real program length exceeds encoded length parameter!\n");
318 if (sscanf(token
, "%hu %hhu %hhu %u,",
319 &bpf_ops
[i
].code
, &bpf_ops
[i
].jt
,
320 &bpf_ops
[i
].jf
, &bpf_ops
[i
].k
) != 4) {
321 fprintf(stderr
, "Error at instruction %d!\n", i
);
330 fprintf(stderr
, "Parsed program length is less than encoded length parameter!\n");
342 void bpf_print_ops(struct rtattr
*bpf_ops
, __u16 len
)
344 struct sock_filter
*ops
= RTA_DATA(bpf_ops
);
350 open_json_object("bytecode");
351 print_uint(PRINT_ANY
, "length", "bytecode \'%u,", len
);
352 open_json_array(PRINT_JSON
, "insns");
354 for (i
= 0; i
< len
; i
++) {
355 open_json_object(NULL
);
356 print_hu(PRINT_ANY
, "code", "%hu ", ops
[i
].code
);
357 print_hhu(PRINT_ANY
, "jt", "%hhu ", ops
[i
].jt
);
358 print_hhu(PRINT_ANY
, "jf", "%hhu ", ops
[i
].jf
);
360 print_uint(PRINT_ANY
, "k", "%u\'", ops
[i
].k
);
362 print_uint(PRINT_ANY
, "k", "%u,", ops
[i
].k
);
366 close_json_array(PRINT_JSON
, NULL
);
370 static void bpf_map_pin_report(const struct bpf_elf_map
*pin
,
371 const struct bpf_elf_map
*obj
)
373 fprintf(stderr
, "Map specification differs from pinned file!\n");
375 if (obj
->type
!= pin
->type
)
376 fprintf(stderr
, " - Type: %u (obj) != %u (pin)\n",
377 obj
->type
, pin
->type
);
378 if (obj
->size_key
!= pin
->size_key
)
379 fprintf(stderr
, " - Size key: %u (obj) != %u (pin)\n",
380 obj
->size_key
, pin
->size_key
);
381 if (obj
->size_value
!= pin
->size_value
)
382 fprintf(stderr
, " - Size value: %u (obj) != %u (pin)\n",
383 obj
->size_value
, pin
->size_value
);
384 if (obj
->max_elem
!= pin
->max_elem
)
385 fprintf(stderr
, " - Max elems: %u (obj) != %u (pin)\n",
386 obj
->max_elem
, pin
->max_elem
);
387 if (obj
->flags
!= pin
->flags
)
388 fprintf(stderr
, " - Flags: %#x (obj) != %#x (pin)\n",
389 obj
->flags
, pin
->flags
);
391 fprintf(stderr
, "\n");
394 struct bpf_prog_data
{
400 struct bpf_prog_data owner
;
401 unsigned int btf_id_key
;
402 unsigned int btf_id_val
;
405 static int bpf_derive_elf_map_from_fdinfo(int fd
, struct bpf_elf_map
*map
,
406 struct bpf_map_ext
*ext
)
408 unsigned int val
, owner_type
= 0, owner_jited
= 0;
414 ret
= asprintf(&file
, "/proc/%d/fdinfo/%d", getpid(), fd
);
416 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
420 memset(map
, 0, sizeof(*map
));
422 fp
= fopen(file
, "r");
425 fprintf(stderr
, "No procfs support?!\n");
429 while (fgets(buff
, sizeof(buff
), fp
)) {
430 if (sscanf(buff
, "map_type:\t%u", &val
) == 1)
432 else if (sscanf(buff
, "key_size:\t%u", &val
) == 1)
434 else if (sscanf(buff
, "value_size:\t%u", &val
) == 1)
435 map
->size_value
= val
;
436 else if (sscanf(buff
, "max_entries:\t%u", &val
) == 1)
438 else if (sscanf(buff
, "map_flags:\t%i", &val
) == 1)
440 else if (sscanf(buff
, "owner_prog_type:\t%i", &val
) == 1)
442 else if (sscanf(buff
, "owner_jited:\t%i", &val
) == 1)
448 memset(ext
, 0, sizeof(*ext
));
449 ext
->owner
.type
= owner_type
;
450 ext
->owner
.jited
= owner_jited
;
456 static int bpf_map_selfcheck_pinned(int fd
, const struct bpf_elf_map
*map
,
457 struct bpf_map_ext
*ext
, int length
,
458 enum bpf_prog_type type
)
460 struct bpf_elf_map tmp
, zero
= {};
463 ret
= bpf_derive_elf_map_from_fdinfo(fd
, &tmp
, ext
);
467 /* The decision to reject this is on kernel side eventually, but
468 * at least give the user a chance to know what's wrong.
470 if (ext
->owner
.type
&& ext
->owner
.type
!= type
)
471 fprintf(stderr
, "Program array map owner types differ: %u (obj) != %u (pin)\n",
472 type
, ext
->owner
.type
);
474 if (!memcmp(&tmp
, map
, length
)) {
477 /* If kernel doesn't have eBPF-related fdinfo, we cannot do much,
478 * so just accept it. We know we do have an eBPF fd and in this
479 * case, everything is 0. It is guaranteed that no such map exists
480 * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC.
482 if (!memcmp(&tmp
, &zero
, length
))
485 bpf_map_pin_report(&tmp
, map
);
490 static int bpf_mnt_fs(const char *target
)
492 bool bind_done
= false;
494 while (mount("", target
, "none", MS_PRIVATE
| MS_REC
, NULL
)) {
495 if (errno
!= EINVAL
|| bind_done
) {
496 fprintf(stderr
, "mount --make-private %s failed: %s\n",
497 target
, strerror(errno
));
501 if (mount(target
, target
, "none", MS_BIND
, NULL
)) {
502 fprintf(stderr
, "mount --bind %s %s failed: %s\n",
503 target
, target
, strerror(errno
));
510 if (mount("bpf", target
, "bpf", 0, "mode=0700")) {
511 fprintf(stderr
, "mount -t bpf bpf %s failed: %s\n",
512 target
, strerror(errno
));
519 static int bpf_mnt_check_target(const char *target
)
524 ret
= stat(target
, &sb
);
526 ret
= mkdir(target
, S_IRWXU
);
528 fprintf(stderr
, "mkdir %s failed: %s\n", target
,
537 static int bpf_valid_mntpt(const char *mnt
, unsigned long magic
)
541 if (statfs(mnt
, &st_fs
) < 0)
543 if ((unsigned long)st_fs
.f_type
!= magic
)
549 static const char *bpf_find_mntpt_single(unsigned long magic
, char *mnt
,
550 int len
, const char *mntpt
)
554 ret
= bpf_valid_mntpt(mntpt
, magic
);
556 strlcpy(mnt
, mntpt
, len
);
563 static const char *bpf_find_mntpt(const char *fstype
, unsigned long magic
,
565 const char * const *known_mnts
)
567 const char * const *ptr
;
574 if (bpf_find_mntpt_single(magic
, mnt
, len
, *ptr
))
583 fp
= fopen("/proc/mounts", "r");
587 while (fscanf(fp
, "%*s %" textify(PATH_MAX
) "s %99s %*s %*d %*d\n",
589 if (strcmp(type
, fstype
) == 0)
594 if (strcmp(type
, fstype
) != 0)
600 int bpf_trace_pipe(void)
602 char tracefs_mnt
[PATH_MAX
] = TRACE_DIR_MNT
;
603 static const char * const tracefs_known_mnts
[] = {
605 "/sys/kernel/debug/tracing",
610 int fd_in
, fd_out
= STDERR_FILENO
;
615 mnt
= bpf_find_mntpt("tracefs", TRACEFS_MAGIC
, tracefs_mnt
,
616 sizeof(tracefs_mnt
), tracefs_known_mnts
);
618 fprintf(stderr
, "tracefs not mounted?\n");
622 ret
= asprintf(&tpipe
, "%s/trace_pipe", mnt
);
624 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
629 fd_in
= open(tpipe
, O_RDONLY
);
634 fprintf(stderr
, "Running! Hang up with ^C!\n\n");
636 static char buff
[4096];
639 ret
= read(fd_in
, buff
, sizeof(buff
));
640 if (ret
> 0 && write(fd_out
, buff
, ret
) == ret
)
649 static int bpf_gen_global(const char *bpf_sub_dir
)
651 char *bpf_glo_dir
= NULL
;
654 ret
= asprintf(&bpf_glo_dir
, "%s/%s/", bpf_sub_dir
, BPF_DIR_GLOBALS
);
656 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
660 ret
= mkdir(bpf_glo_dir
, S_IRWXU
);
661 if (ret
&& errno
!= EEXIST
) {
662 fprintf(stderr
, "mkdir %s failed: %s\n", bpf_glo_dir
,
673 static int bpf_gen_master(const char *base
, const char *name
)
675 char *bpf_sub_dir
= NULL
;
678 ret
= asprintf(&bpf_sub_dir
, "%s%s/", base
, name
);
680 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
684 ret
= mkdir(bpf_sub_dir
, S_IRWXU
);
685 if (ret
&& errno
!= EEXIST
) {
686 fprintf(stderr
, "mkdir %s failed: %s\n", bpf_sub_dir
,
691 ret
= bpf_gen_global(bpf_sub_dir
);
697 static int bpf_slave_via_bind_mnt(const char *full_name
,
698 const char *full_link
)
702 ret
= mkdir(full_name
, S_IRWXU
);
704 assert(errno
!= EEXIST
);
705 fprintf(stderr
, "mkdir %s failed: %s\n", full_name
,
710 ret
= mount(full_link
, full_name
, "none", MS_BIND
, NULL
);
713 fprintf(stderr
, "mount --bind %s %s failed: %s\n",
714 full_link
, full_name
, strerror(errno
));
720 static int bpf_gen_slave(const char *base
, const char *name
,
723 char *bpf_lnk_dir
= NULL
;
724 char *bpf_sub_dir
= NULL
;
728 ret
= asprintf(&bpf_lnk_dir
, "%s%s/", base
, link
);
730 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
734 ret
= asprintf(&bpf_sub_dir
, "%s%s", base
, name
);
736 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
740 ret
= symlink(bpf_lnk_dir
, bpf_sub_dir
);
742 if (errno
!= EEXIST
) {
743 if (errno
!= EPERM
) {
744 fprintf(stderr
, "symlink %s failed: %s\n",
745 bpf_sub_dir
, strerror(errno
));
749 ret
= bpf_slave_via_bind_mnt(bpf_sub_dir
, bpf_lnk_dir
);
753 ret
= lstat(bpf_sub_dir
, &sb
);
755 fprintf(stderr
, "lstat %s failed: %s\n",
756 bpf_sub_dir
, strerror(errno
));
760 if ((sb
.st_mode
& S_IFMT
) != S_IFLNK
) {
761 ret
= bpf_gen_global(bpf_sub_dir
);
772 static int bpf_gen_hierarchy(const char *base
)
776 ret
= bpf_gen_master(base
, bpf_prog_to_subdir(__bpf_types
[0]));
777 for (i
= 1; i
< ARRAY_SIZE(__bpf_types
) && !ret
; i
++)
778 ret
= bpf_gen_slave(base
,
779 bpf_prog_to_subdir(__bpf_types
[i
]),
780 bpf_prog_to_subdir(__bpf_types
[0]));
784 static const char *bpf_get_work_dir(enum bpf_prog_type type
)
786 static char bpf_tmp
[PATH_MAX
] = BPF_DIR_MNT
;
787 static char *bpf_wrk_dir
;
788 static const char *mnt
;
789 static bool bpf_mnt_cached
;
790 const char *mnt_env
= getenv(BPF_ENV_MNT
);
791 static const char * const bpf_known_mnts
[] = {
798 if (bpf_mnt_cached
) {
799 const char *out
= mnt
;
802 snprintf(bpf_tmp
, sizeof(bpf_tmp
), "%s%s/",
803 out
, bpf_prog_to_subdir(type
));
810 mnt
= bpf_find_mntpt_single(BPF_FS_MAGIC
, bpf_tmp
,
811 sizeof(bpf_tmp
), mnt_env
);
813 mnt
= bpf_find_mntpt("bpf", BPF_FS_MAGIC
, bpf_tmp
,
814 sizeof(bpf_tmp
), bpf_known_mnts
);
816 mnt
= mnt_env
? : BPF_DIR_MNT
;
817 ret
= bpf_mnt_check_target(mnt
);
819 ret
= bpf_mnt_fs(mnt
);
826 ret
= asprintf(&bpf_wrk_dir
, "%s/", mnt
);
828 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
833 ret
= bpf_gen_hierarchy(bpf_wrk_dir
);
841 bpf_mnt_cached
= true;
845 static int bpf_obj_get(const char *pathname
, enum bpf_prog_type type
)
847 union bpf_attr attr
= {};
850 if (strlen(pathname
) > 2 && pathname
[0] == 'm' &&
851 pathname
[1] == ':' && bpf_get_work_dir(type
)) {
852 snprintf(tmp
, sizeof(tmp
), "%s/%s",
853 bpf_get_work_dir(type
), pathname
+ 2);
857 attr
.pathname
= bpf_ptr_to_u64(pathname
);
859 return bpf(BPF_OBJ_GET
, &attr
, sizeof(attr
));
862 static int bpf_obj_pinned(const char *pathname
, enum bpf_prog_type type
)
864 int prog_fd
= bpf_obj_get(pathname
, type
);
867 fprintf(stderr
, "Couldn\'t retrieve pinned program \'%s\': %s\n",
868 pathname
, strerror(errno
));
872 static int bpf_do_parse(struct bpf_cfg_in
*cfg
, const bool *opt_tbl
)
874 const char *file
, *section
, *uds_name
;
875 bool verbose
= false;
882 if (opt_tbl
[CBPF_BYTECODE
] &&
883 (matches(*argv
, "bytecode") == 0 ||
884 strcmp(*argv
, "bc") == 0)) {
885 cfg
->mode
= CBPF_BYTECODE
;
886 } else if (opt_tbl
[CBPF_FILE
] &&
887 (matches(*argv
, "bytecode-file") == 0 ||
888 strcmp(*argv
, "bcf") == 0)) {
889 cfg
->mode
= CBPF_FILE
;
890 } else if (opt_tbl
[EBPF_OBJECT
] &&
891 (matches(*argv
, "object-file") == 0 ||
892 strcmp(*argv
, "obj") == 0)) {
893 cfg
->mode
= EBPF_OBJECT
;
894 } else if (opt_tbl
[EBPF_PINNED
] &&
895 (matches(*argv
, "object-pinned") == 0 ||
896 matches(*argv
, "pinned") == 0 ||
897 matches(*argv
, "fd") == 0)) {
898 cfg
->mode
= EBPF_PINNED
;
900 fprintf(stderr
, "What mode is \"%s\"?\n", *argv
);
905 file
= section
= uds_name
= NULL
;
906 if (cfg
->mode
== EBPF_OBJECT
|| cfg
->mode
== EBPF_PINNED
) {
910 if (cfg
->type
== BPF_PROG_TYPE_UNSPEC
) {
911 if (argc
> 0 && matches(*argv
, "type") == 0) {
913 for (i
= 0; i
< ARRAY_SIZE(__bpf_prog_meta
);
915 if (!__bpf_prog_meta
[i
].type
)
918 __bpf_prog_meta
[i
].type
)) {
924 if (cfg
->type
== BPF_PROG_TYPE_UNSPEC
) {
925 fprintf(stderr
, "What type is \"%s\"?\n",
931 cfg
->type
= BPF_PROG_TYPE_SCHED_CLS
;
935 section
= bpf_prog_to_default_section(cfg
->type
);
936 if (argc
> 0 && matches(*argv
, "section") == 0) {
942 if (__bpf_prog_meta
[cfg
->type
].may_uds_export
) {
943 uds_name
= getenv(BPF_ENV_UDS
);
944 if (argc
> 0 && !uds_name
&&
945 matches(*argv
, "export") == 0) {
952 if (argc
> 0 && matches(*argv
, "verbose") == 0) {
960 if (cfg
->mode
== CBPF_BYTECODE
|| cfg
->mode
== CBPF_FILE
) {
961 ret
= bpf_ops_parse(argc
, argv
, cfg
->opcodes
,
962 cfg
->mode
== CBPF_FILE
);
963 cfg
->n_opcodes
= ret
;
964 } else if (cfg
->mode
== EBPF_OBJECT
) {
965 ret
= 0; /* program will be loaded by load stage */
966 } else if (cfg
->mode
== EBPF_PINNED
) {
967 ret
= bpf_obj_pinned(file
, cfg
->type
);
974 cfg
->section
= section
;
978 cfg
->verbose
= verbose
;
983 static int bpf_do_load(struct bpf_cfg_in
*cfg
)
985 if (cfg
->mode
== EBPF_OBJECT
) {
986 cfg
->prog_fd
= bpf_obj_open(cfg
->object
, cfg
->type
,
987 cfg
->section
, cfg
->ifindex
,
994 int bpf_load_common(struct bpf_cfg_in
*cfg
, const struct bpf_cfg_ops
*ops
,
997 char annotation
[256];
1000 ret
= bpf_do_load(cfg
);
1004 if (cfg
->mode
== CBPF_BYTECODE
|| cfg
->mode
== CBPF_FILE
)
1005 ops
->cbpf_cb(nl
, cfg
->opcodes
, cfg
->n_opcodes
);
1006 if (cfg
->mode
== EBPF_OBJECT
|| cfg
->mode
== EBPF_PINNED
) {
1007 snprintf(annotation
, sizeof(annotation
), "%s:[%s]",
1008 basename(cfg
->object
), cfg
->mode
== EBPF_PINNED
?
1009 "*fsobj" : cfg
->section
);
1010 ops
->ebpf_cb(nl
, cfg
->prog_fd
, annotation
);
1016 int bpf_parse_common(struct bpf_cfg_in
*cfg
, const struct bpf_cfg_ops
*ops
)
1018 bool opt_tbl
[BPF_MODE_MAX
] = {};
1021 opt_tbl
[CBPF_BYTECODE
] = true;
1022 opt_tbl
[CBPF_FILE
] = true;
1026 opt_tbl
[EBPF_OBJECT
] = true;
1027 opt_tbl
[EBPF_PINNED
] = true;
1030 return bpf_do_parse(cfg
, opt_tbl
);
1033 int bpf_parse_and_load_common(struct bpf_cfg_in
*cfg
,
1034 const struct bpf_cfg_ops
*ops
, void *nl
)
1038 ret
= bpf_parse_common(cfg
, ops
);
1042 return bpf_load_common(cfg
, ops
, nl
);
1045 int bpf_graft_map(const char *map_path
, uint32_t *key
, int argc
, char **argv
)
1047 const bool opt_tbl
[BPF_MODE_MAX
] = {
1048 [EBPF_OBJECT
] = true,
1049 [EBPF_PINNED
] = true,
1051 const struct bpf_elf_map test
= {
1052 .type
= BPF_MAP_TYPE_PROG_ARRAY
,
1053 .size_key
= sizeof(int),
1054 .size_value
= sizeof(int),
1056 struct bpf_cfg_in cfg
= {
1057 .type
= BPF_PROG_TYPE_UNSPEC
,
1061 struct bpf_map_ext ext
= {};
1062 int ret
, prog_fd
, map_fd
;
1065 ret
= bpf_do_parse(&cfg
, opt_tbl
);
1069 ret
= bpf_do_load(&cfg
);
1073 prog_fd
= cfg
.prog_fd
;
1078 ret
= sscanf(cfg
.section
, "%*i/%i", &map_key
);
1080 fprintf(stderr
, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n");
1086 map_fd
= bpf_obj_get(map_path
, cfg
.type
);
1088 fprintf(stderr
, "Couldn\'t retrieve pinned map \'%s\': %s\n",
1089 map_path
, strerror(errno
));
1094 ret
= bpf_map_selfcheck_pinned(map_fd
, &test
, &ext
,
1095 offsetof(struct bpf_elf_map
, max_elem
),
1098 fprintf(stderr
, "Map \'%s\' self-check failed!\n", map_path
);
1102 ret
= bpf_map_update(map_fd
, &map_key
, &prog_fd
, BPF_ANY
);
1104 fprintf(stderr
, "Map update failed: %s\n", strerror(errno
));
1112 int bpf_prog_attach_fd(int prog_fd
, int target_fd
, enum bpf_attach_type type
)
1114 union bpf_attr attr
= {};
1116 attr
.target_fd
= target_fd
;
1117 attr
.attach_bpf_fd
= prog_fd
;
1118 attr
.attach_type
= type
;
1120 return bpf(BPF_PROG_ATTACH
, &attr
, sizeof(attr
));
1123 int bpf_prog_detach_fd(int target_fd
, enum bpf_attach_type type
)
1125 union bpf_attr attr
= {};
1127 attr
.target_fd
= target_fd
;
1128 attr
.attach_type
= type
;
1130 return bpf(BPF_PROG_DETACH
, &attr
, sizeof(attr
));
1133 static int bpf_prog_load_dev(enum bpf_prog_type type
,
1134 const struct bpf_insn
*insns
, size_t size_insns
,
1135 const char *license
, __u32 ifindex
,
1136 char *log
, size_t size_log
)
1138 union bpf_attr attr
= {};
1140 attr
.prog_type
= type
;
1141 attr
.insns
= bpf_ptr_to_u64(insns
);
1142 attr
.insn_cnt
= size_insns
/ sizeof(struct bpf_insn
);
1143 attr
.license
= bpf_ptr_to_u64(license
);
1144 attr
.prog_ifindex
= ifindex
;
1147 attr
.log_buf
= bpf_ptr_to_u64(log
);
1148 attr
.log_size
= size_log
;
1152 return bpf(BPF_PROG_LOAD
, &attr
, sizeof(attr
));
1155 int bpf_prog_load(enum bpf_prog_type type
, const struct bpf_insn
*insns
,
1156 size_t size_insns
, const char *license
, char *log
,
1159 return bpf_prog_load_dev(type
, insns
, size_insns
, license
, 0,
1164 struct bpf_elf_prog
{
1165 enum bpf_prog_type type
;
1166 struct bpf_insn
*insns
;
1167 unsigned int insns_num
;
1169 const char *license
;
1172 struct bpf_hash_entry
{
1173 unsigned int pinning
;
1174 const char *subpath
;
1175 struct bpf_hash_entry
*next
;
1179 unsigned int jit_enabled
;
1183 const struct btf_header
*hdr
;
1185 const char *strings
;
1186 const struct btf_type
**types
;
1190 struct bpf_elf_ctx
{
1191 struct bpf_config cfg
;
1200 int map_fds
[ELF_MAX_MAPS
];
1201 struct bpf_elf_map maps
[ELF_MAX_MAPS
];
1202 struct bpf_map_ext maps_ext
[ELF_MAX_MAPS
];
1203 struct bpf_elf_prog prog_text
;
1212 char license
[ELF_MAX_LICENSE_LEN
];
1213 enum bpf_prog_type type
;
1217 struct bpf_elf_st stat
;
1218 struct bpf_hash_entry
*ht
[256];
1223 struct bpf_elf_sec_data
{
1226 const char *sec_name
;
1229 struct bpf_map_data
{
1232 struct bpf_elf_st
*st
;
1233 struct bpf_elf_map
*ent
;
1236 static bool bpf_log_has_data(struct bpf_elf_ctx
*ctx
)
1238 return ctx
->log
&& ctx
->log
[0];
1241 static __check_format_string(2, 3) void
1242 bpf_dump_error(struct bpf_elf_ctx
*ctx
, const char *format
, ...)
1246 va_start(vl
, format
);
1247 vfprintf(stderr
, format
, vl
);
1250 if (bpf_log_has_data(ctx
)) {
1252 fprintf(stderr
, "%s\n", ctx
->log
);
1254 unsigned int off
= 0, len
= strlen(ctx
->log
);
1256 if (len
> BPF_MAX_LOG
) {
1257 off
= len
- BPF_MAX_LOG
;
1258 fprintf(stderr
, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n",
1261 fprintf(stderr
, "%s\n", ctx
->log
+ off
);
1264 memset(ctx
->log
, 0, ctx
->log_size
);
1268 static int bpf_log_realloc(struct bpf_elf_ctx
*ctx
)
1270 const size_t log_max
= UINT_MAX
>> 8;
1271 size_t log_size
= ctx
->log_size
;
1276 } else if (log_size
< log_max
) {
1278 if (log_size
> log_max
)
1284 ptr
= realloc(ctx
->log
, log_size
);
1290 ctx
->log_size
= log_size
;
1295 static int bpf_map_create(enum bpf_map_type type
, uint32_t size_key
,
1296 uint32_t size_value
, uint32_t max_elem
,
1297 uint32_t flags
, int inner_fd
, int btf_fd
,
1298 uint32_t ifindex
, uint32_t btf_id_key
,
1299 uint32_t btf_id_val
)
1301 union bpf_attr attr
= {};
1303 attr
.map_type
= type
;
1304 attr
.key_size
= size_key
;
1305 attr
.value_size
= inner_fd
? sizeof(int) : size_value
;
1306 attr
.max_entries
= max_elem
;
1307 attr
.map_flags
= flags
;
1308 attr
.inner_map_fd
= inner_fd
;
1309 attr
.map_ifindex
= ifindex
;
1310 attr
.btf_fd
= btf_fd
;
1311 attr
.btf_key_type_id
= btf_id_key
;
1312 attr
.btf_value_type_id
= btf_id_val
;
1314 return bpf(BPF_MAP_CREATE
, &attr
, sizeof(attr
));
1317 static int bpf_btf_load(void *btf
, size_t size_btf
,
1318 char *log
, size_t size_log
)
1320 union bpf_attr attr
= {};
1322 attr
.btf
= bpf_ptr_to_u64(btf
);
1323 attr
.btf_size
= size_btf
;
1326 attr
.btf_log_buf
= bpf_ptr_to_u64(log
);
1327 attr
.btf_log_size
= size_log
;
1328 attr
.btf_log_level
= 1;
1331 return bpf(BPF_BTF_LOAD
, &attr
, sizeof(attr
));
1334 static int bpf_obj_pin(int fd
, const char *pathname
)
1336 union bpf_attr attr
= {};
1338 attr
.pathname
= bpf_ptr_to_u64(pathname
);
1341 return bpf(BPF_OBJ_PIN
, &attr
, sizeof(attr
));
1344 static int bpf_obj_hash(const char *object
, uint8_t *out
, size_t len
)
1346 struct sockaddr_alg alg
= {
1347 .salg_family
= AF_ALG
,
1348 .salg_type
= "hash",
1349 .salg_name
= "sha1",
1351 int ret
, cfd
, ofd
, ffd
;
1355 if (!object
|| len
!= 20)
1358 cfd
= socket(AF_ALG
, SOCK_SEQPACKET
, 0);
1362 ret
= bind(cfd
, (struct sockaddr
*)&alg
, sizeof(alg
));
1366 ofd
= accept(cfd
, NULL
, 0);
1372 ffd
= open(object
, O_RDONLY
);
1374 fprintf(stderr
, "Error opening object %s: %s\n",
1375 object
, strerror(errno
));
1380 ret
= fstat(ffd
, &stbuff
);
1382 fprintf(stderr
, "Error doing fstat: %s\n",
1387 size
= sendfile(ofd
, ffd
, NULL
, stbuff
.st_size
);
1388 if (size
!= stbuff
.st_size
) {
1389 fprintf(stderr
, "Error from sendfile (%zd vs %zu bytes): %s\n",
1390 size
, stbuff
.st_size
, strerror(errno
));
1395 size
= read(ofd
, out
, len
);
1397 fprintf(stderr
, "Error from read (%zd vs %zu bytes): %s\n",
1398 size
, len
, strerror(errno
));
1412 static void bpf_init_env(void)
1414 struct rlimit limit
= {
1415 .rlim_cur
= RLIM_INFINITY
,
1416 .rlim_max
= RLIM_INFINITY
,
1419 /* Don't bother in case we fail! */
1420 setrlimit(RLIMIT_MEMLOCK
, &limit
);
1422 if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC
))
1423 fprintf(stderr
, "Continuing without mounted eBPF fs. Too old kernel?\n");
1426 static const char *bpf_custom_pinning(const struct bpf_elf_ctx
*ctx
,
1429 struct bpf_hash_entry
*entry
;
1431 entry
= ctx
->ht
[pinning
& (ARRAY_SIZE(ctx
->ht
) - 1)];
1432 while (entry
&& entry
->pinning
!= pinning
)
1433 entry
= entry
->next
;
1435 return entry
? entry
->subpath
: NULL
;
1438 static bool bpf_no_pinning(const struct bpf_elf_ctx
*ctx
,
1448 return !bpf_custom_pinning(ctx
, pinning
);
1452 static void bpf_make_pathname(char *pathname
, size_t len
, const char *name
,
1453 const struct bpf_elf_ctx
*ctx
, uint32_t pinning
)
1457 snprintf(pathname
, len
, "%s/%s/%s",
1458 bpf_get_work_dir(ctx
->type
),
1459 ctx
->obj_uid
, name
);
1462 snprintf(pathname
, len
, "%s/%s/%s",
1463 bpf_get_work_dir(ctx
->type
),
1464 BPF_DIR_GLOBALS
, name
);
1467 snprintf(pathname
, len
, "%s/../%s/%s",
1468 bpf_get_work_dir(ctx
->type
),
1469 bpf_custom_pinning(ctx
, pinning
), name
);
1474 static int bpf_probe_pinned(const char *name
, const struct bpf_elf_ctx
*ctx
,
1477 char pathname
[PATH_MAX
];
1479 if (bpf_no_pinning(ctx
, pinning
) || !bpf_get_work_dir(ctx
->type
))
1482 bpf_make_pathname(pathname
, sizeof(pathname
), name
, ctx
, pinning
);
1483 return bpf_obj_get(pathname
, ctx
->type
);
1486 static int bpf_make_obj_path(const struct bpf_elf_ctx
*ctx
)
1491 ret
= asprintf(&tmp
, "%s/%s", bpf_get_work_dir(ctx
->type
), ctx
->obj_uid
);
1493 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
1497 ret
= mkdir(tmp
, S_IRWXU
);
1498 if (ret
&& errno
!= EEXIST
) {
1499 fprintf(stderr
, "mkdir %s failed: %s\n", tmp
, strerror(errno
));
1509 static int bpf_make_custom_path(const struct bpf_elf_ctx
*ctx
,
1517 ret
= asprintf(&tmp
, "%s/../", bpf_get_work_dir(ctx
->type
));
1519 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
1523 ret
= asprintf(&rem
, "%s/", todo
);
1525 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
1529 sub
= strtok(rem
, "/");
1531 if (strlen(tmp
) + strlen(sub
) + 2 > PATH_MAX
)
1537 ret
= mkdir(tmp
, S_IRWXU
);
1538 if (ret
&& errno
!= EEXIST
) {
1539 fprintf(stderr
, "mkdir %s failed: %s\n", tmp
,
1544 sub
= strtok(NULL
, "/");
1554 static int bpf_place_pinned(int fd
, const char *name
,
1555 const struct bpf_elf_ctx
*ctx
, uint32_t pinning
)
1557 char pathname
[PATH_MAX
];
1561 if (bpf_no_pinning(ctx
, pinning
) || !bpf_get_work_dir(ctx
->type
))
1564 if (pinning
== PIN_OBJECT_NS
)
1565 ret
= bpf_make_obj_path(ctx
);
1566 else if ((tmp
= bpf_custom_pinning(ctx
, pinning
)))
1567 ret
= bpf_make_custom_path(ctx
, tmp
);
1571 bpf_make_pathname(pathname
, sizeof(pathname
), name
, ctx
, pinning
);
1572 return bpf_obj_pin(fd
, pathname
);
1575 static void bpf_prog_report(int fd
, const char *section
,
1576 const struct bpf_elf_prog
*prog
,
1577 struct bpf_elf_ctx
*ctx
)
1579 unsigned int insns
= prog
->size
/ sizeof(struct bpf_insn
);
1581 fprintf(stderr
, "\nProg section \'%s\' %s%s (%d)!\n", section
,
1582 fd
< 0 ? "rejected: " : "loaded",
1583 fd
< 0 ? strerror(errno
) : "",
1584 fd
< 0 ? errno
: fd
);
1586 fprintf(stderr
, " - Type: %u\n", prog
->type
);
1587 fprintf(stderr
, " - Instructions: %u (%u over limit)\n",
1588 insns
, insns
> BPF_MAXINSNS
? insns
- BPF_MAXINSNS
: 0);
1589 fprintf(stderr
, " - License: %s\n\n", prog
->license
);
1591 bpf_dump_error(ctx
, "Verifier analysis:\n\n");
1594 static int bpf_prog_attach(const char *section
,
1595 const struct bpf_elf_prog
*prog
,
1596 struct bpf_elf_ctx
*ctx
)
1601 fd
= bpf_prog_load_dev(prog
->type
, prog
->insns
, prog
->size
,
1602 prog
->license
, ctx
->ifindex
,
1603 ctx
->log
, ctx
->log_size
);
1604 if (fd
< 0 || ctx
->verbose
) {
1605 /* The verifier log is pretty chatty, sometimes so chatty
1606 * on larger programs, that we could fail to dump everything
1607 * into our buffer. Still, try to give a debuggable error
1608 * log for the user, so enlarge it and re-fail.
1610 if (fd
< 0 && (errno
== ENOSPC
|| !ctx
->log_size
)) {
1611 if (tries
++ < 10 && !bpf_log_realloc(ctx
))
1614 fprintf(stderr
, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n",
1615 ctx
->log_size
, tries
);
1619 bpf_prog_report(fd
, section
, prog
, ctx
);
1625 static void bpf_map_report(int fd
, const char *name
,
1626 const struct bpf_elf_map
*map
,
1627 struct bpf_elf_ctx
*ctx
, int inner_fd
)
1629 fprintf(stderr
, "Map object \'%s\' %s%s (%d)!\n", name
,
1630 fd
< 0 ? "rejected: " : "loaded",
1631 fd
< 0 ? strerror(errno
) : "",
1632 fd
< 0 ? errno
: fd
);
1634 fprintf(stderr
, " - Type: %u\n", map
->type
);
1635 fprintf(stderr
, " - Identifier: %u\n", map
->id
);
1636 fprintf(stderr
, " - Pinning: %u\n", map
->pinning
);
1637 fprintf(stderr
, " - Size key: %u\n", map
->size_key
);
1638 fprintf(stderr
, " - Size value: %u\n",
1639 inner_fd
? (int)sizeof(int) : map
->size_value
);
1640 fprintf(stderr
, " - Max elems: %u\n", map
->max_elem
);
1641 fprintf(stderr
, " - Flags: %#x\n\n", map
->flags
);
1644 static int bpf_find_map_id(const struct bpf_elf_ctx
*ctx
, uint32_t id
)
1648 for (i
= 0; i
< ctx
->map_num
; i
++) {
1649 if (ctx
->maps
[i
].id
!= id
)
1651 if (ctx
->map_fds
[i
] < 0)
1654 return ctx
->map_fds
[i
];
1660 static void bpf_report_map_in_map(int outer_fd
, uint32_t idx
)
1662 struct bpf_elf_map outer_map
;
1665 fprintf(stderr
, "Cannot insert map into map! ");
1667 ret
= bpf_derive_elf_map_from_fdinfo(outer_fd
, &outer_map
, NULL
);
1669 if (idx
>= outer_map
.max_elem
&&
1670 outer_map
.type
== BPF_MAP_TYPE_ARRAY_OF_MAPS
) {
1671 fprintf(stderr
, "Outer map has %u elements, index %u is invalid!\n",
1672 outer_map
.max_elem
, idx
);
1677 fprintf(stderr
, "Different map specs used for outer and inner map?\n");
1680 static bool bpf_is_map_in_map_type(const struct bpf_elf_map
*map
)
1682 return map
->type
== BPF_MAP_TYPE_ARRAY_OF_MAPS
||
1683 map
->type
== BPF_MAP_TYPE_HASH_OF_MAPS
;
1686 static bool bpf_map_offload_neutral(enum bpf_map_type type
)
1688 return type
== BPF_MAP_TYPE_PERF_EVENT_ARRAY
;
1691 static int bpf_map_attach(const char *name
, struct bpf_elf_ctx
*ctx
,
1692 const struct bpf_elf_map
*map
, struct bpf_map_ext
*ext
,
1693 int *have_map_in_map
)
1695 int fd
, ifindex
, ret
, map_inner_fd
= 0;
1696 bool retried
= false;
1699 fd
= bpf_probe_pinned(name
, ctx
, map
->pinning
);
1701 ret
= bpf_map_selfcheck_pinned(fd
, map
, ext
,
1702 offsetof(struct bpf_elf_map
,
1706 fprintf(stderr
, "Map \'%s\' self-check failed!\n",
1711 fprintf(stderr
, "Map \'%s\' loaded as pinned!\n",
1716 if (have_map_in_map
&& bpf_is_map_in_map_type(map
)) {
1717 (*have_map_in_map
)++;
1720 fprintf(stderr
, "Map \'%s\' cannot be created since no inner map ID defined!\n",
1725 if (!have_map_in_map
&& bpf_is_map_in_map_type(map
)) {
1726 map_inner_fd
= bpf_find_map_id(ctx
, map
->inner_id
);
1727 if (map_inner_fd
< 0) {
1728 fprintf(stderr
, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n",
1729 name
, map
->inner_id
);
1734 ifindex
= bpf_map_offload_neutral(map
->type
) ? 0 : ctx
->ifindex
;
1736 fd
= bpf_map_create(map
->type
, map
->size_key
, map
->size_value
,
1737 map
->max_elem
, map
->flags
, map_inner_fd
, ctx
->btf_fd
,
1738 ifindex
, ext
->btf_id_key
, ext
->btf_id_val
);
1740 if (fd
< 0 || ctx
->verbose
) {
1741 bpf_map_report(fd
, name
, map
, ctx
, map_inner_fd
);
1746 ret
= bpf_place_pinned(fd
, name
, ctx
, map
->pinning
);
1749 if (!retried
&& errno
== EEXIST
) {
1753 fprintf(stderr
, "Could not pin %s map: %s\n", name
,
1761 static const char *bpf_str_tab_name(const struct bpf_elf_ctx
*ctx
,
1762 const GElf_Sym
*sym
)
1764 return ctx
->str_tab
->d_buf
+ sym
->st_name
;
1767 static int bpf_btf_find(struct bpf_elf_ctx
*ctx
, const char *name
)
1769 const struct btf_type
*type
;
1773 for (id
= 1; id
< ctx
->btf
.types_num
; id
++) {
1774 type
= ctx
->btf
.types
[id
];
1775 if (type
->name_off
>= ctx
->btf
.hdr
->str_len
)
1777 res
= &ctx
->btf
.strings
[type
->name_off
];
1778 if (!strcmp(res
, name
))
1785 static int bpf_btf_find_kv(struct bpf_elf_ctx
*ctx
, const struct bpf_elf_map
*map
,
1786 const char *name
, uint32_t *id_key
, uint32_t *id_val
)
1788 const struct btf_member
*key
, *val
;
1789 const struct btf_type
*type
;
1794 snprintf(btf_name
, sizeof(btf_name
), "____btf_map_%s", name
);
1795 id
= bpf_btf_find(ctx
, btf_name
);
1799 type
= ctx
->btf
.types
[id
];
1800 if (BTF_INFO_KIND(type
->info
) != BTF_KIND_STRUCT
)
1802 if (BTF_INFO_VLEN(type
->info
) != 2)
1805 key
= ((void *) type
) + sizeof(*type
);
1807 if (!key
->type
|| key
->type
>= ctx
->btf
.types_num
||
1808 !val
->type
|| val
->type
>= ctx
->btf
.types_num
)
1811 if (key
->name_off
>= ctx
->btf
.hdr
->str_len
||
1812 val
->name_off
>= ctx
->btf
.hdr
->str_len
)
1815 res
= &ctx
->btf
.strings
[key
->name_off
];
1816 if (strcmp(res
, "key"))
1819 res
= &ctx
->btf
.strings
[val
->name_off
];
1820 if (strcmp(res
, "value"))
1823 *id_key
= key
->type
;
1824 *id_val
= val
->type
;
1828 static void bpf_btf_annotate(struct bpf_elf_ctx
*ctx
, int which
, const char *name
)
1830 uint32_t id_key
= 0, id_val
= 0;
1832 if (!bpf_btf_find_kv(ctx
, &ctx
->maps
[which
], name
, &id_key
, &id_val
)) {
1833 ctx
->maps_ext
[which
].btf_id_key
= id_key
;
1834 ctx
->maps_ext
[which
].btf_id_val
= id_val
;
1838 static const char *bpf_map_fetch_name(struct bpf_elf_ctx
*ctx
, int which
)
1844 for (i
= 0; i
< ctx
->sym_num
; i
++) {
1847 if (gelf_getsym(ctx
->sym_tab
, i
, &sym
) != &sym
)
1850 type
= GELF_ST_TYPE(sym
.st_info
);
1851 if (GELF_ST_BIND(sym
.st_info
) != STB_GLOBAL
||
1852 (type
!= STT_NOTYPE
&& type
!= STT_OBJECT
) ||
1853 sym
.st_shndx
!= ctx
->sec_maps
||
1854 sym
.st_value
/ ctx
->map_len
!= which
)
1857 name
= bpf_str_tab_name(ctx
, &sym
);
1858 bpf_btf_annotate(ctx
, which
, name
);
1865 static int bpf_maps_attach_all(struct bpf_elf_ctx
*ctx
)
1867 int i
, j
, ret
, fd
, inner_fd
, inner_idx
, have_map_in_map
= 0;
1868 const char *map_name
;
1870 for (i
= 0; i
< ctx
->map_num
; i
++) {
1871 if (ctx
->maps
[i
].pinning
== PIN_OBJECT_NS
&&
1873 fprintf(stderr
, "Missing kernel AF_ALG support for PIN_OBJECT_NS!\n");
1877 map_name
= bpf_map_fetch_name(ctx
, i
);
1881 fd
= bpf_map_attach(map_name
, ctx
, &ctx
->maps
[i
],
1882 &ctx
->maps_ext
[i
], &have_map_in_map
);
1886 ctx
->map_fds
[i
] = !fd
? -1 : fd
;
1889 for (i
= 0; have_map_in_map
&& i
< ctx
->map_num
; i
++) {
1890 if (ctx
->map_fds
[i
] >= 0)
1893 map_name
= bpf_map_fetch_name(ctx
, i
);
1897 fd
= bpf_map_attach(map_name
, ctx
, &ctx
->maps
[i
],
1898 &ctx
->maps_ext
[i
], NULL
);
1902 ctx
->map_fds
[i
] = fd
;
1905 for (i
= 0; have_map_in_map
&& i
< ctx
->map_num
; i
++) {
1906 if (!ctx
->maps
[i
].id
||
1907 ctx
->maps
[i
].inner_id
||
1908 ctx
->maps
[i
].inner_idx
== -1)
1911 inner_fd
= ctx
->map_fds
[i
];
1912 inner_idx
= ctx
->maps
[i
].inner_idx
;
1914 for (j
= 0; j
< ctx
->map_num
; j
++) {
1915 if (!bpf_is_map_in_map_type(&ctx
->maps
[j
]))
1917 if (ctx
->maps
[j
].inner_id
!= ctx
->maps
[i
].id
)
1920 ret
= bpf_map_update(ctx
->map_fds
[j
], &inner_idx
,
1921 &inner_fd
, BPF_ANY
);
1923 bpf_report_map_in_map(ctx
->map_fds
[j
],
1933 static int bpf_map_num_sym(struct bpf_elf_ctx
*ctx
)
1938 for (i
= 0; i
< ctx
->sym_num
; i
++) {
1941 if (gelf_getsym(ctx
->sym_tab
, i
, &sym
) != &sym
)
1944 type
= GELF_ST_TYPE(sym
.st_info
);
1945 if (GELF_ST_BIND(sym
.st_info
) != STB_GLOBAL
||
1946 (type
!= STT_NOTYPE
&& type
!= STT_OBJECT
) ||
1947 sym
.st_shndx
!= ctx
->sec_maps
)
1955 static int bpf_fill_section_data(struct bpf_elf_ctx
*ctx
, int section
,
1956 struct bpf_elf_sec_data
*data
)
1958 Elf_Data
*sec_edata
;
1963 memset(data
, 0, sizeof(*data
));
1965 sec_fd
= elf_getscn(ctx
->elf_fd
, section
);
1968 if (gelf_getshdr(sec_fd
, &sec_hdr
) != &sec_hdr
)
1971 sec_name
= elf_strptr(ctx
->elf_fd
, ctx
->elf_hdr
.e_shstrndx
,
1973 if (!sec_name
|| !sec_hdr
.sh_size
)
1976 sec_edata
= elf_getdata(sec_fd
, NULL
);
1977 if (!sec_edata
|| elf_getdata(sec_fd
, sec_edata
))
1980 memcpy(&data
->sec_hdr
, &sec_hdr
, sizeof(sec_hdr
));
1982 data
->sec_name
= sec_name
;
1983 data
->sec_data
= sec_edata
;
1987 struct bpf_elf_map_min
{
1994 static int bpf_fetch_maps_begin(struct bpf_elf_ctx
*ctx
, int section
,
1995 struct bpf_elf_sec_data
*data
)
1997 ctx
->map_num
= data
->sec_data
->d_size
;
1998 ctx
->sec_maps
= section
;
1999 ctx
->sec_done
[section
] = true;
2001 if (ctx
->map_num
> sizeof(ctx
->maps
)) {
2002 fprintf(stderr
, "Too many BPF maps in ELF section!\n");
2006 memcpy(ctx
->maps
, data
->sec_data
->d_buf
, ctx
->map_num
);
2010 static int bpf_map_verify_all_offs(struct bpf_elf_ctx
*ctx
, int end
)
2015 for (off
= 0; off
< end
; off
+= ctx
->map_len
) {
2016 /* Order doesn't need to be linear here, hence we walk
2019 for (i
= 0; i
< ctx
->sym_num
; i
++) {
2022 if (gelf_getsym(ctx
->sym_tab
, i
, &sym
) != &sym
)
2025 type
= GELF_ST_TYPE(sym
.st_info
);
2026 if (GELF_ST_BIND(sym
.st_info
) != STB_GLOBAL
||
2027 (type
!= STT_NOTYPE
&& type
!= STT_OBJECT
) ||
2028 sym
.st_shndx
!= ctx
->sec_maps
)
2030 if (sym
.st_value
== off
)
2032 if (i
== ctx
->sym_num
- 1)
2037 return off
== end
? 0 : -1;
2040 static int bpf_fetch_maps_end(struct bpf_elf_ctx
*ctx
)
2042 struct bpf_elf_map fixup
[ARRAY_SIZE(ctx
->maps
)] = {};
2043 int i
, sym_num
= bpf_map_num_sym(ctx
);
2046 if (sym_num
== 0 || sym_num
> ARRAY_SIZE(ctx
->maps
)) {
2047 fprintf(stderr
, "%u maps not supported in current map section!\n",
2052 if (ctx
->map_num
% sym_num
!= 0 ||
2053 ctx
->map_num
% sizeof(__u32
) != 0) {
2054 fprintf(stderr
, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n");
2058 ctx
->map_len
= ctx
->map_num
/ sym_num
;
2059 if (bpf_map_verify_all_offs(ctx
, ctx
->map_num
)) {
2060 fprintf(stderr
, "Different struct bpf_elf_map in use!\n");
2064 if (ctx
->map_len
== sizeof(struct bpf_elf_map
)) {
2065 ctx
->map_num
= sym_num
;
2067 } else if (ctx
->map_len
> sizeof(struct bpf_elf_map
)) {
2068 fprintf(stderr
, "struct bpf_elf_map not supported, coming from future version?\n");
2070 } else if (ctx
->map_len
< sizeof(struct bpf_elf_map_min
)) {
2071 fprintf(stderr
, "struct bpf_elf_map too small, not supported!\n");
2075 ctx
->map_num
= sym_num
;
2076 for (i
= 0, buff
= (void *)ctx
->maps
; i
< ctx
->map_num
;
2077 i
++, buff
+= ctx
->map_len
) {
2078 /* The fixup leaves the rest of the members as zero, which
2079 * is fine currently, but option exist to set some other
2080 * default value as well when needed in future.
2082 memcpy(&fixup
[i
], buff
, ctx
->map_len
);
2085 memcpy(ctx
->maps
, fixup
, sizeof(fixup
));
2087 printf("%zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n",
2088 sizeof(struct bpf_elf_map
) - ctx
->map_len
);
2092 static int bpf_fetch_license(struct bpf_elf_ctx
*ctx
, int section
,
2093 struct bpf_elf_sec_data
*data
)
2095 if (data
->sec_data
->d_size
> sizeof(ctx
->license
))
2098 memcpy(ctx
->license
, data
->sec_data
->d_buf
, data
->sec_data
->d_size
);
2099 ctx
->sec_done
[section
] = true;
2103 static int bpf_fetch_symtab(struct bpf_elf_ctx
*ctx
, int section
,
2104 struct bpf_elf_sec_data
*data
)
2106 ctx
->sym_tab
= data
->sec_data
;
2107 ctx
->sym_num
= data
->sec_hdr
.sh_size
/ data
->sec_hdr
.sh_entsize
;
2108 ctx
->sec_done
[section
] = true;
2112 static int bpf_fetch_strtab(struct bpf_elf_ctx
*ctx
, int section
,
2113 struct bpf_elf_sec_data
*data
)
2115 ctx
->str_tab
= data
->sec_data
;
2116 ctx
->sec_done
[section
] = true;
2120 static int bpf_fetch_text(struct bpf_elf_ctx
*ctx
, int section
,
2121 struct bpf_elf_sec_data
*data
)
2123 ctx
->sec_text
= section
;
2124 ctx
->sec_done
[section
] = true;
2128 static void bpf_btf_report(int fd
, struct bpf_elf_ctx
*ctx
)
2130 fprintf(stderr
, "\nBTF debug data section \'.BTF\' %s%s (%d)!\n",
2131 fd
< 0 ? "rejected: " : "loaded",
2132 fd
< 0 ? strerror(errno
) : "",
2133 fd
< 0 ? errno
: fd
);
2135 fprintf(stderr
, " - Length: %zu\n", ctx
->btf_data
->d_size
);
2137 bpf_dump_error(ctx
, "Verifier analysis:\n\n");
2140 static int bpf_btf_attach(struct bpf_elf_ctx
*ctx
)
2145 fd
= bpf_btf_load(ctx
->btf_data
->d_buf
, ctx
->btf_data
->d_size
,
2146 ctx
->log
, ctx
->log_size
);
2147 if (fd
< 0 || ctx
->verbose
) {
2148 if (fd
< 0 && (errno
== ENOSPC
|| !ctx
->log_size
)) {
2149 if (tries
++ < 10 && !bpf_log_realloc(ctx
))
2152 fprintf(stderr
, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n",
2153 ctx
->log_size
, tries
);
2157 if (bpf_log_has_data(ctx
))
2158 bpf_btf_report(fd
, ctx
);
2164 static int bpf_fetch_btf_begin(struct bpf_elf_ctx
*ctx
, int section
,
2165 struct bpf_elf_sec_data
*data
)
2167 ctx
->btf_data
= data
->sec_data
;
2168 ctx
->sec_btf
= section
;
2169 ctx
->sec_done
[section
] = true;
2173 static int bpf_btf_check_header(struct bpf_elf_ctx
*ctx
)
2175 const struct btf_header
*hdr
= ctx
->btf_data
->d_buf
;
2176 const char *str_start
, *str_end
;
2177 unsigned int data_len
;
2179 if (hdr
->magic
!= BTF_MAGIC
) {
2180 fprintf(stderr
, "Object has wrong BTF magic: %x, expected: %x!\n",
2181 hdr
->magic
, BTF_MAGIC
);
2185 if (hdr
->version
!= BTF_VERSION
) {
2186 fprintf(stderr
, "Object has wrong BTF version: %u, expected: %u!\n",
2187 hdr
->version
, BTF_VERSION
);
2192 fprintf(stderr
, "Object has unsupported BTF flags %x!\n",
2197 data_len
= ctx
->btf_data
->d_size
- sizeof(*hdr
);
2198 if (data_len
< hdr
->type_off
||
2199 data_len
< hdr
->str_off
||
2200 data_len
< hdr
->type_len
+ hdr
->str_len
||
2201 hdr
->type_off
>= hdr
->str_off
||
2202 hdr
->type_off
+ hdr
->type_len
!= hdr
->str_off
||
2203 hdr
->str_off
+ hdr
->str_len
!= data_len
||
2204 (hdr
->type_off
& (sizeof(uint32_t) - 1))) {
2205 fprintf(stderr
, "Object has malformed BTF data!\n");
2210 ctx
->btf
.raw
= hdr
+ 1;
2212 str_start
= ctx
->btf
.raw
+ hdr
->str_off
;
2213 str_end
= str_start
+ hdr
->str_len
;
2214 if (!hdr
->str_len
||
2215 hdr
->str_len
- 1 > BTF_MAX_NAME_OFFSET
||
2216 str_start
[0] || str_end
[-1]) {
2217 fprintf(stderr
, "Object has malformed BTF string data!\n");
2221 ctx
->btf
.strings
= str_start
;
2225 static int bpf_btf_register_type(struct bpf_elf_ctx
*ctx
,
2226 const struct btf_type
*type
)
2228 int cur
= ctx
->btf
.types_num
, num
= cur
+ 1;
2229 const struct btf_type
**types
;
2231 types
= realloc(ctx
->btf
.types
, num
* sizeof(type
));
2233 free(ctx
->btf
.types
);
2234 ctx
->btf
.types
= NULL
;
2235 ctx
->btf
.types_num
= 0;
2239 ctx
->btf
.types
= types
;
2240 ctx
->btf
.types
[cur
] = type
;
2241 ctx
->btf
.types_num
= num
;
2245 static struct btf_type btf_type_void
;
2247 static int bpf_btf_prep_type_data(struct bpf_elf_ctx
*ctx
)
2249 const void *type_cur
= ctx
->btf
.raw
+ ctx
->btf
.hdr
->type_off
;
2250 const void *type_end
= ctx
->btf
.raw
+ ctx
->btf
.hdr
->str_off
;
2251 const struct btf_type
*type
;
2255 ret
= bpf_btf_register_type(ctx
, &btf_type_void
);
2259 while (type_cur
< type_end
) {
2261 type_cur
+= sizeof(*type
);
2263 var_len
= BTF_INFO_VLEN(type
->info
);
2264 kind
= BTF_INFO_KIND(type
->info
);
2268 type_cur
+= sizeof(int);
2270 case BTF_KIND_ARRAY
:
2271 type_cur
+= sizeof(struct btf_array
);
2273 case BTF_KIND_STRUCT
:
2274 case BTF_KIND_UNION
:
2275 type_cur
+= var_len
* sizeof(struct btf_member
);
2278 type_cur
+= var_len
* sizeof(struct btf_enum
);
2280 case BTF_KIND_FUNC_PROTO
:
2281 type_cur
+= var_len
* sizeof(struct btf_param
);
2283 case BTF_KIND_TYPEDEF
:
2286 case BTF_KIND_VOLATILE
:
2287 case BTF_KIND_CONST
:
2288 case BTF_KIND_RESTRICT
:
2292 fprintf(stderr
, "Object has unknown BTF type: %u!\n", kind
);
2296 ret
= bpf_btf_register_type(ctx
, type
);
2304 static int bpf_btf_prep_data(struct bpf_elf_ctx
*ctx
)
2306 int ret
= bpf_btf_check_header(ctx
);
2309 return bpf_btf_prep_type_data(ctx
);
2313 static void bpf_fetch_btf_end(struct bpf_elf_ctx
*ctx
)
2315 int fd
= bpf_btf_attach(ctx
);
2320 if (bpf_btf_prep_data(ctx
) < 0) {
2326 static bool bpf_has_map_data(const struct bpf_elf_ctx
*ctx
)
2328 return ctx
->sym_tab
&& ctx
->str_tab
&& ctx
->sec_maps
;
2331 static bool bpf_has_btf_data(const struct bpf_elf_ctx
*ctx
)
2333 return ctx
->sec_btf
;
2336 static bool bpf_has_call_data(const struct bpf_elf_ctx
*ctx
)
2338 return ctx
->sec_text
;
2341 static int bpf_fetch_ancillary(struct bpf_elf_ctx
*ctx
, bool check_text_sec
)
2343 struct bpf_elf_sec_data data
;
2346 for (i
= 1; i
< ctx
->elf_hdr
.e_shnum
; i
++) {
2347 ret
= bpf_fill_section_data(ctx
, i
, &data
);
2351 if (data
.sec_hdr
.sh_type
== SHT_PROGBITS
&&
2352 !strcmp(data
.sec_name
, ELF_SECTION_MAPS
))
2353 ret
= bpf_fetch_maps_begin(ctx
, i
, &data
);
2354 else if (data
.sec_hdr
.sh_type
== SHT_PROGBITS
&&
2355 !strcmp(data
.sec_name
, ELF_SECTION_LICENSE
))
2356 ret
= bpf_fetch_license(ctx
, i
, &data
);
2357 else if (data
.sec_hdr
.sh_type
== SHT_PROGBITS
&&
2358 (data
.sec_hdr
.sh_flags
& SHF_EXECINSTR
) &&
2359 !strcmp(data
.sec_name
, ".text") &&
2361 ret
= bpf_fetch_text(ctx
, i
, &data
);
2362 else if (data
.sec_hdr
.sh_type
== SHT_SYMTAB
&&
2363 !strcmp(data
.sec_name
, ".symtab"))
2364 ret
= bpf_fetch_symtab(ctx
, i
, &data
);
2365 else if (data
.sec_hdr
.sh_type
== SHT_STRTAB
&&
2366 !strcmp(data
.sec_name
, ".strtab"))
2367 ret
= bpf_fetch_strtab(ctx
, i
, &data
);
2368 else if (data
.sec_hdr
.sh_type
== SHT_PROGBITS
&&
2369 !strcmp(data
.sec_name
, ".BTF"))
2370 ret
= bpf_fetch_btf_begin(ctx
, i
, &data
);
2372 fprintf(stderr
, "Error parsing section %d! Perhaps check with readelf -a?\n",
2378 if (bpf_has_btf_data(ctx
))
2379 bpf_fetch_btf_end(ctx
);
2380 if (bpf_has_map_data(ctx
)) {
2381 ret
= bpf_fetch_maps_end(ctx
);
2383 fprintf(stderr
, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n");
2387 ret
= bpf_maps_attach_all(ctx
);
2389 fprintf(stderr
, "Error loading maps into kernel!\n");
2397 static int bpf_fetch_prog(struct bpf_elf_ctx
*ctx
, const char *section
,
2400 struct bpf_elf_sec_data data
;
2401 struct bpf_elf_prog prog
;
2402 int ret
, i
, fd
= -1;
2404 for (i
= 1; i
< ctx
->elf_hdr
.e_shnum
; i
++) {
2405 if (ctx
->sec_done
[i
])
2408 ret
= bpf_fill_section_data(ctx
, i
, &data
);
2410 !(data
.sec_hdr
.sh_type
== SHT_PROGBITS
&&
2411 (data
.sec_hdr
.sh_flags
& SHF_EXECINSTR
) &&
2412 !strcmp(data
.sec_name
, section
)))
2417 memset(&prog
, 0, sizeof(prog
));
2418 prog
.type
= ctx
->type
;
2419 prog
.license
= ctx
->license
;
2420 prog
.size
= data
.sec_data
->d_size
;
2421 prog
.insns_num
= prog
.size
/ sizeof(struct bpf_insn
);
2422 prog
.insns
= data
.sec_data
->d_buf
;
2424 fd
= bpf_prog_attach(section
, &prog
, ctx
);
2428 ctx
->sec_done
[i
] = true;
2435 struct bpf_relo_props
{
2436 struct bpf_tail_call
{
2443 static int bpf_apply_relo_map(struct bpf_elf_ctx
*ctx
, struct bpf_elf_prog
*prog
,
2444 GElf_Rel
*relo
, GElf_Sym
*sym
,
2445 struct bpf_relo_props
*props
)
2447 unsigned int insn_off
= relo
->r_offset
/ sizeof(struct bpf_insn
);
2448 unsigned int map_idx
= sym
->st_value
/ ctx
->map_len
;
2450 if (insn_off
>= prog
->insns_num
)
2452 if (prog
->insns
[insn_off
].code
!= (BPF_LD
| BPF_IMM
| BPF_DW
)) {
2453 fprintf(stderr
, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n",
2458 if (map_idx
>= ARRAY_SIZE(ctx
->map_fds
))
2460 if (!ctx
->map_fds
[map_idx
])
2462 if (ctx
->maps
[map_idx
].type
== BPF_MAP_TYPE_PROG_ARRAY
) {
2464 if (ctx
->maps_ext
[map_idx
].owner
.jited
||
2465 (ctx
->maps_ext
[map_idx
].owner
.type
== 0 &&
2466 ctx
->cfg
.jit_enabled
))
2470 prog
->insns
[insn_off
].src_reg
= BPF_PSEUDO_MAP_FD
;
2471 prog
->insns
[insn_off
].imm
= ctx
->map_fds
[map_idx
];
2475 static int bpf_apply_relo_call(struct bpf_elf_ctx
*ctx
, struct bpf_elf_prog
*prog
,
2476 GElf_Rel
*relo
, GElf_Sym
*sym
,
2477 struct bpf_relo_props
*props
)
2479 unsigned int insn_off
= relo
->r_offset
/ sizeof(struct bpf_insn
);
2480 struct bpf_elf_prog
*prog_text
= &ctx
->prog_text
;
2482 if (insn_off
>= prog
->insns_num
)
2484 if (prog
->insns
[insn_off
].code
!= (BPF_JMP
| BPF_CALL
) &&
2485 prog
->insns
[insn_off
].src_reg
!= BPF_PSEUDO_CALL
) {
2486 fprintf(stderr
, "ELF contains relo data for non call instruction at offset %u! Compiler bug?!\n",
2491 if (!props
->main_num
) {
2492 struct bpf_insn
*insns
= realloc(prog
->insns
,
2493 prog
->size
+ prog_text
->size
);
2497 memcpy(insns
+ prog
->insns_num
, prog_text
->insns
,
2499 props
->main_num
= prog
->insns_num
;
2500 prog
->insns
= insns
;
2501 prog
->insns_num
+= prog_text
->insns_num
;
2502 prog
->size
+= prog_text
->size
;
2505 prog
->insns
[insn_off
].imm
+= props
->main_num
- insn_off
;
2509 static int bpf_apply_relo_data(struct bpf_elf_ctx
*ctx
,
2510 struct bpf_elf_sec_data
*data_relo
,
2511 struct bpf_elf_prog
*prog
,
2512 struct bpf_relo_props
*props
)
2514 GElf_Shdr
*rhdr
= &data_relo
->sec_hdr
;
2515 int relo_ent
, relo_num
= rhdr
->sh_size
/ rhdr
->sh_entsize
;
2517 for (relo_ent
= 0; relo_ent
< relo_num
; relo_ent
++) {
2522 if (gelf_getrel(data_relo
->sec_data
, relo_ent
, &relo
) != &relo
)
2524 if (gelf_getsym(ctx
->sym_tab
, GELF_R_SYM(relo
.r_info
), &sym
) != &sym
)
2527 if (sym
.st_shndx
== ctx
->sec_maps
)
2528 ret
= bpf_apply_relo_map(ctx
, prog
, &relo
, &sym
, props
);
2529 else if (sym
.st_shndx
== ctx
->sec_text
)
2530 ret
= bpf_apply_relo_call(ctx
, prog
, &relo
, &sym
, props
);
2532 fprintf(stderr
, "ELF contains non-{map,call} related relo data in entry %u pointing to section %u! Compiler bug?!\n",
2533 relo_ent
, sym
.st_shndx
);
2541 static int bpf_fetch_prog_relo(struct bpf_elf_ctx
*ctx
, const char *section
,
2542 bool *lderr
, bool *sseen
, struct bpf_elf_prog
*prog
)
2544 struct bpf_elf_sec_data data_relo
, data_insn
;
2545 int ret
, idx
, i
, fd
= -1;
2547 for (i
= 1; i
< ctx
->elf_hdr
.e_shnum
; i
++) {
2548 struct bpf_relo_props props
= {};
2550 ret
= bpf_fill_section_data(ctx
, i
, &data_relo
);
2551 if (ret
< 0 || data_relo
.sec_hdr
.sh_type
!= SHT_REL
)
2554 idx
= data_relo
.sec_hdr
.sh_info
;
2556 ret
= bpf_fill_section_data(ctx
, idx
, &data_insn
);
2558 !(data_insn
.sec_hdr
.sh_type
== SHT_PROGBITS
&&
2559 (data_insn
.sec_hdr
.sh_flags
& SHF_EXECINSTR
) &&
2560 !strcmp(data_insn
.sec_name
, section
)))
2565 memset(prog
, 0, sizeof(*prog
));
2566 prog
->type
= ctx
->type
;
2567 prog
->license
= ctx
->license
;
2568 prog
->size
= data_insn
.sec_data
->d_size
;
2569 prog
->insns_num
= prog
->size
/ sizeof(struct bpf_insn
);
2570 prog
->insns
= malloc(prog
->size
);
2576 memcpy(prog
->insns
, data_insn
.sec_data
->d_buf
, prog
->size
);
2578 ret
= bpf_apply_relo_data(ctx
, &data_relo
, prog
, &props
);
2581 if (ctx
->sec_text
!= idx
)
2585 if (ctx
->sec_text
== idx
) {
2590 fd
= bpf_prog_attach(section
, prog
, ctx
);
2594 if (props
.tc
.total
) {
2595 if (ctx
->cfg
.jit_enabled
&&
2596 props
.tc
.total
!= props
.tc
.jited
)
2597 fprintf(stderr
, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n",
2598 props
.tc
.jited
, props
.tc
.total
);
2599 if (!ctx
->cfg
.jit_enabled
&&
2601 fprintf(stderr
, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n",
2602 props
.tc
.jited
, props
.tc
.total
);
2607 ctx
->sec_done
[i
] = true;
2608 ctx
->sec_done
[idx
] = true;
2615 static int bpf_fetch_prog_sec(struct bpf_elf_ctx
*ctx
, const char *section
)
2617 bool lderr
= false, sseen
= false;
2618 struct bpf_elf_prog prog
;
2621 if (bpf_has_call_data(ctx
)) {
2622 ret
= bpf_fetch_prog_relo(ctx
, ".text", &lderr
, NULL
,
2628 if (bpf_has_map_data(ctx
) || bpf_has_call_data(ctx
))
2629 ret
= bpf_fetch_prog_relo(ctx
, section
, &lderr
, &sseen
, &prog
);
2630 if (ret
< 0 && !lderr
)
2631 ret
= bpf_fetch_prog(ctx
, section
, &sseen
);
2632 if (ret
< 0 && !sseen
)
2633 fprintf(stderr
, "Program section \'%s\' not found in ELF file!\n",
2638 static int bpf_find_map_by_id(struct bpf_elf_ctx
*ctx
, uint32_t id
)
2642 for (i
= 0; i
< ARRAY_SIZE(ctx
->map_fds
); i
++)
2643 if (ctx
->map_fds
[i
] && ctx
->maps
[i
].id
== id
&&
2644 ctx
->maps
[i
].type
== BPF_MAP_TYPE_PROG_ARRAY
)
2649 struct bpf_jited_aux
{
2652 struct bpf_prog_data prog
;
2653 struct bpf_map_ext map
;
2656 static int bpf_derive_prog_from_fdinfo(int fd
, struct bpf_prog_data
*prog
)
2664 ret
= asprintf(&file
, "/proc/%d/fdinfo/%d", getpid(), fd
);
2666 fprintf(stderr
, "asprintf failed: %s\n", strerror(errno
));
2671 memset(prog
, 0, sizeof(*prog
));
2673 fp
= fopen(file
, "r");
2676 fprintf(stderr
, "No procfs support?!\n");
2680 while (fgets(buff
, sizeof(buff
), fp
)) {
2681 if (sscanf(buff
, "prog_type:\t%u", &val
) == 1)
2683 else if (sscanf(buff
, "prog_jited:\t%u", &val
) == 1)
2691 static int bpf_tail_call_get_aux(struct bpf_jited_aux
*aux
)
2693 struct bpf_elf_map tmp
;
2696 ret
= bpf_derive_elf_map_from_fdinfo(aux
->map_fd
, &tmp
, &aux
->map
);
2698 ret
= bpf_derive_prog_from_fdinfo(aux
->prog_fd
, &aux
->prog
);
2703 static int bpf_fill_prog_arrays(struct bpf_elf_ctx
*ctx
)
2705 struct bpf_elf_sec_data data
;
2706 uint32_t map_id
, key_id
;
2707 int fd
, i
, ret
, idx
;
2709 for (i
= 1; i
< ctx
->elf_hdr
.e_shnum
; i
++) {
2710 if (ctx
->sec_done
[i
])
2713 ret
= bpf_fill_section_data(ctx
, i
, &data
);
2717 ret
= sscanf(data
.sec_name
, "%i/%i", &map_id
, &key_id
);
2721 idx
= bpf_find_map_by_id(ctx
, map_id
);
2725 fd
= bpf_fetch_prog_sec(ctx
, data
.sec_name
);
2729 ret
= bpf_map_update(ctx
->map_fds
[idx
], &key_id
,
2732 struct bpf_jited_aux aux
= {};
2735 if (errno
== E2BIG
) {
2736 fprintf(stderr
, "Tail call key %u for map %u out of bounds?\n",
2741 aux
.map_fd
= ctx
->map_fds
[idx
];
2744 if (bpf_tail_call_get_aux(&aux
))
2746 if (!aux
.map
.owner
.type
)
2749 if (aux
.prog
.type
!= aux
.map
.owner
.type
)
2750 fprintf(stderr
, "Tail call map owned by prog type %u, but prog type is %u!\n",
2751 aux
.map
.owner
.type
, aux
.prog
.type
);
2752 if (aux
.prog
.jited
!= aux
.map
.owner
.jited
)
2753 fprintf(stderr
, "Tail call map %s jited, but prog %s!\n",
2754 aux
.map
.owner
.jited
? "is" : "not",
2755 aux
.prog
.jited
? "is" : "not");
2759 ctx
->sec_done
[i
] = true;
2765 static void bpf_save_finfo(struct bpf_elf_ctx
*ctx
)
2770 memset(&ctx
->stat
, 0, sizeof(ctx
->stat
));
2772 ret
= fstat(ctx
->obj_fd
, &st
);
2774 fprintf(stderr
, "Stat of elf file failed: %s\n",
2779 ctx
->stat
.st_dev
= st
.st_dev
;
2780 ctx
->stat
.st_ino
= st
.st_ino
;
2783 static int bpf_read_pin_mapping(FILE *fp
, uint32_t *id
, char *path
)
2785 char buff
[PATH_MAX
];
2787 while (fgets(buff
, sizeof(buff
), fp
)) {
2790 while (*ptr
== ' ' || *ptr
== '\t')
2793 if (*ptr
== '#' || *ptr
== '\n' || *ptr
== 0)
2796 if (sscanf(ptr
, "%i %s\n", id
, path
) != 2 &&
2797 sscanf(ptr
, "%i %s #", id
, path
) != 2) {
2808 static bool bpf_pinning_reserved(uint32_t pinning
)
2820 static void bpf_hash_init(struct bpf_elf_ctx
*ctx
, const char *db_file
)
2822 struct bpf_hash_entry
*entry
;
2823 char subpath
[PATH_MAX
] = {};
2828 fp
= fopen(db_file
, "r");
2832 while ((ret
= bpf_read_pin_mapping(fp
, &pinning
, subpath
))) {
2834 fprintf(stderr
, "Database %s is corrupted at: %s\n",
2840 if (bpf_pinning_reserved(pinning
)) {
2841 fprintf(stderr
, "Database %s, id %u is reserved - ignoring!\n",
2846 entry
= malloc(sizeof(*entry
));
2848 fprintf(stderr
, "No memory left for db entry!\n");
2852 entry
->pinning
= pinning
;
2853 entry
->subpath
= strdup(subpath
);
2854 if (!entry
->subpath
) {
2855 fprintf(stderr
, "No memory left for db entry!\n");
2860 entry
->next
= ctx
->ht
[pinning
& (ARRAY_SIZE(ctx
->ht
) - 1)];
2861 ctx
->ht
[pinning
& (ARRAY_SIZE(ctx
->ht
) - 1)] = entry
;
2867 static void bpf_hash_destroy(struct bpf_elf_ctx
*ctx
)
2869 struct bpf_hash_entry
*entry
;
2872 for (i
= 0; i
< ARRAY_SIZE(ctx
->ht
); i
++) {
2873 while ((entry
= ctx
->ht
[i
]) != NULL
) {
2874 ctx
->ht
[i
] = entry
->next
;
2875 free((char *)entry
->subpath
);
2881 static int bpf_elf_check_ehdr(const struct bpf_elf_ctx
*ctx
)
2883 if (ctx
->elf_hdr
.e_type
!= ET_REL
||
2884 (ctx
->elf_hdr
.e_machine
!= EM_NONE
&&
2885 ctx
->elf_hdr
.e_machine
!= EM_BPF
) ||
2886 ctx
->elf_hdr
.e_version
!= EV_CURRENT
) {
2887 fprintf(stderr
, "ELF format error, ELF file not for eBPF?\n");
2891 switch (ctx
->elf_hdr
.e_ident
[EI_DATA
]) {
2893 fprintf(stderr
, "ELF format error, wrong endianness info?\n");
2896 if (htons(1) == 1) {
2898 "We are big endian, eBPF object is little endian!\n");
2903 if (htons(1) != 1) {
2905 "We are little endian, eBPF object is big endian!\n");
2914 static void bpf_get_cfg(struct bpf_elf_ctx
*ctx
)
2916 static const char *path_jit
= "/proc/sys/net/core/bpf_jit_enable";
2919 fd
= open(path_jit
, O_RDONLY
);
2923 if (read(fd
, tmp
, sizeof(tmp
)) > 0)
2924 ctx
->cfg
.jit_enabled
= atoi(tmp
);
2929 static int bpf_elf_ctx_init(struct bpf_elf_ctx
*ctx
, const char *pathname
,
2930 enum bpf_prog_type type
, __u32 ifindex
,
2936 if (elf_version(EV_CURRENT
) == EV_NONE
)
2941 memset(ctx
, 0, sizeof(*ctx
));
2944 ret
= bpf_obj_hash(pathname
, tmp
, sizeof(tmp
));
2946 ctx
->noafalg
= true;
2948 hexstring_n2a(tmp
, sizeof(tmp
), ctx
->obj_uid
,
2949 sizeof(ctx
->obj_uid
));
2951 ctx
->verbose
= verbose
;
2953 ctx
->ifindex
= ifindex
;
2955 ctx
->obj_fd
= open(pathname
, O_RDONLY
);
2956 if (ctx
->obj_fd
< 0)
2959 ctx
->elf_fd
= elf_begin(ctx
->obj_fd
, ELF_C_READ
, NULL
);
2965 if (elf_kind(ctx
->elf_fd
) != ELF_K_ELF
) {
2970 if (gelf_getehdr(ctx
->elf_fd
, &ctx
->elf_hdr
) !=
2976 ret
= bpf_elf_check_ehdr(ctx
);
2980 ctx
->sec_done
= calloc(ctx
->elf_hdr
.e_shnum
,
2981 sizeof(*(ctx
->sec_done
)));
2982 if (!ctx
->sec_done
) {
2987 if (ctx
->verbose
&& bpf_log_realloc(ctx
)) {
2992 bpf_save_finfo(ctx
);
2993 bpf_hash_init(ctx
, CONFDIR
"/bpf_pinning");
2997 free(ctx
->sec_done
);
2999 elf_end(ctx
->elf_fd
);
3005 static int bpf_maps_count(struct bpf_elf_ctx
*ctx
)
3009 for (i
= 0; i
< ARRAY_SIZE(ctx
->map_fds
); i
++) {
3010 if (!ctx
->map_fds
[i
])
3018 static void bpf_maps_teardown(struct bpf_elf_ctx
*ctx
)
3022 for (i
= 0; i
< ARRAY_SIZE(ctx
->map_fds
); i
++) {
3023 if (ctx
->map_fds
[i
])
3024 close(ctx
->map_fds
[i
]);
3029 free(ctx
->btf
.types
);
3032 static void bpf_elf_ctx_destroy(struct bpf_elf_ctx
*ctx
, bool failure
)
3035 bpf_maps_teardown(ctx
);
3037 bpf_hash_destroy(ctx
);
3039 free(ctx
->prog_text
.insns
);
3040 free(ctx
->sec_done
);
3043 elf_end(ctx
->elf_fd
);
3047 static struct bpf_elf_ctx __ctx
;
3049 static int bpf_obj_open(const char *pathname
, enum bpf_prog_type type
,
3050 const char *section
, __u32 ifindex
, bool verbose
)
3052 struct bpf_elf_ctx
*ctx
= &__ctx
;
3055 ret
= bpf_elf_ctx_init(ctx
, pathname
, type
, ifindex
, verbose
);
3057 fprintf(stderr
, "Cannot initialize ELF context!\n");
3061 ret
= bpf_fetch_ancillary(ctx
, strcmp(section
, ".text"));
3063 fprintf(stderr
, "Error fetching ELF ancillary data!\n");
3067 fd
= bpf_fetch_prog_sec(ctx
, section
);
3069 fprintf(stderr
, "Error fetching program/map!\n");
3074 ret
= bpf_fill_prog_arrays(ctx
);
3076 fprintf(stderr
, "Error filling program arrays!\n");
3078 bpf_elf_ctx_destroy(ctx
, ret
< 0);
3089 bpf_map_set_send(int fd
, struct sockaddr_un
*addr
, unsigned int addr_len
,
3090 const struct bpf_map_data
*aux
, unsigned int entries
)
3092 struct bpf_map_set_msg msg
= {
3093 .aux
.uds_ver
= BPF_SCM_AUX_VER
,
3094 .aux
.num_ent
= entries
,
3096 int *cmsg_buf
, min_fd
;
3100 strlcpy(msg
.aux
.obj_name
, aux
->obj
, sizeof(msg
.aux
.obj_name
));
3101 memcpy(&msg
.aux
.obj_st
, aux
->st
, sizeof(msg
.aux
.obj_st
));
3103 cmsg_buf
= bpf_map_set_init(&msg
, addr
, addr_len
);
3104 amsg_buf
= (char *)msg
.aux
.ent
;
3106 for (i
= 0; i
< entries
; i
+= min_fd
) {
3109 min_fd
= min(BPF_SCM_MAX_FDS
* 1U, entries
- i
);
3110 bpf_map_set_init_single(&msg
, min_fd
);
3112 memcpy(cmsg_buf
, &aux
->fds
[i
], sizeof(aux
->fds
[0]) * min_fd
);
3113 memcpy(amsg_buf
, &aux
->ent
[i
], sizeof(aux
->ent
[0]) * min_fd
);
3115 ret
= sendmsg(fd
, &msg
.hdr
, 0);
3124 bpf_map_set_recv(int fd
, int *fds
, struct bpf_map_aux
*aux
,
3125 unsigned int entries
)
3127 struct bpf_map_set_msg msg
;
3128 int *cmsg_buf
, min_fd
;
3129 char *amsg_buf
, *mmsg_buf
;
3130 unsigned int needed
= 1;
3133 cmsg_buf
= bpf_map_set_init(&msg
, NULL
, 0);
3134 amsg_buf
= (char *)msg
.aux
.ent
;
3135 mmsg_buf
= (char *)&msg
.aux
;
3137 for (i
= 0; i
< min(entries
, needed
); i
+= min_fd
) {
3138 struct cmsghdr
*cmsg
;
3141 min_fd
= min(entries
, entries
- i
);
3142 bpf_map_set_init_single(&msg
, min_fd
);
3144 ret
= recvmsg(fd
, &msg
.hdr
, 0);
3148 cmsg
= CMSG_FIRSTHDR(&msg
.hdr
);
3149 if (!cmsg
|| cmsg
->cmsg_type
!= SCM_RIGHTS
)
3151 if (msg
.hdr
.msg_flags
& MSG_CTRUNC
)
3153 if (msg
.aux
.uds_ver
!= BPF_SCM_AUX_VER
)
3156 min_fd
= (cmsg
->cmsg_len
- sizeof(*cmsg
)) / sizeof(fd
);
3157 if (min_fd
> entries
|| min_fd
<= 0)
3160 memcpy(&fds
[i
], cmsg_buf
, sizeof(fds
[0]) * min_fd
);
3161 memcpy(&aux
->ent
[i
], amsg_buf
, sizeof(aux
->ent
[0]) * min_fd
);
3162 memcpy(aux
, mmsg_buf
, offsetof(struct bpf_map_aux
, ent
));
3164 needed
= aux
->num_ent
;
3170 int bpf_send_map_fds(const char *path
, const char *obj
)
3172 struct bpf_elf_ctx
*ctx
= &__ctx
;
3173 struct sockaddr_un addr
= { .sun_family
= AF_UNIX
};
3174 struct bpf_map_data bpf_aux
= {
3175 .fds
= ctx
->map_fds
,
3182 fd
= socket(AF_UNIX
, SOCK_DGRAM
, 0);
3184 fprintf(stderr
, "Cannot open socket: %s\n",
3189 strlcpy(addr
.sun_path
, path
, sizeof(addr
.sun_path
));
3191 ret
= connect(fd
, (struct sockaddr
*)&addr
, sizeof(addr
));
3193 fprintf(stderr
, "Cannot connect to %s: %s\n",
3194 path
, strerror(errno
));
3198 ret
= bpf_map_set_send(fd
, &addr
, sizeof(addr
), &bpf_aux
,
3199 bpf_maps_count(ctx
));
3201 fprintf(stderr
, "Cannot send fds to %s: %s\n",
3202 path
, strerror(errno
));
3204 bpf_maps_teardown(ctx
);
3209 int bpf_recv_map_fds(const char *path
, int *fds
, struct bpf_map_aux
*aux
,
3210 unsigned int entries
)
3212 struct sockaddr_un addr
= { .sun_family
= AF_UNIX
};
3215 fd
= socket(AF_UNIX
, SOCK_DGRAM
, 0);
3217 fprintf(stderr
, "Cannot open socket: %s\n",
3222 strlcpy(addr
.sun_path
, path
, sizeof(addr
.sun_path
));
3224 ret
= bind(fd
, (struct sockaddr
*)&addr
, sizeof(addr
));
3226 fprintf(stderr
, "Cannot bind to socket: %s\n",
3231 ret
= bpf_map_set_recv(fd
, fds
, aux
, entries
);
3233 fprintf(stderr
, "Cannot recv fds from %s: %s\n",
3234 path
, strerror(errno
));
3236 unlink(addr
.sun_path
);
3240 #endif /* HAVE_ELF */