1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
4 * Common eBPF ELF object loading operations.
6 * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
8 * Copyright (C) 2015 Huawei Inc.
9 * Copyright (C) 2017 Nicira, Inc.
10 * Copyright (C) 2019 Isovalent, Inc.
28 #include <asm/unistd.h>
29 #include <linux/err.h>
30 #include <linux/kernel.h>
31 #include <linux/bpf.h>
32 #include <linux/btf.h>
33 #include <linux/filter.h>
34 #include <linux/limits.h>
35 #include <linux/perf_event.h>
36 #include <linux/ring_buffer.h>
37 #include <sys/epoll.h>
38 #include <sys/ioctl.h>
41 #include <sys/types.h>
43 #include <sys/utsname.h>
44 #include <sys/resource.h>
52 #include "str_error.h"
53 #include "libbpf_internal.h"
55 #include "bpf_gen_internal.h"
59 #define BPF_FS_MAGIC 0xcafe4a11
62 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
64 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
65 * compilation if user enables corresponding warning. Disable it explicitly.
67 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
69 #define __printf(a, b) __attribute__((format(printf, a, b)))
71 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
72 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
74 static const char * const attach_type_name[] = {
75 [BPF_CGROUP_INET_INGRESS] = "cgroup_inet_ingress",
76 [BPF_CGROUP_INET_EGRESS] = "cgroup_inet_egress",
77 [BPF_CGROUP_INET_SOCK_CREATE] = "cgroup_inet_sock_create",
78 [BPF_CGROUP_INET_SOCK_RELEASE] = "cgroup_inet_sock_release",
79 [BPF_CGROUP_SOCK_OPS] = "cgroup_sock_ops",
80 [BPF_CGROUP_DEVICE] = "cgroup_device",
81 [BPF_CGROUP_INET4_BIND] = "cgroup_inet4_bind",
82 [BPF_CGROUP_INET6_BIND] = "cgroup_inet6_bind",
83 [BPF_CGROUP_INET4_CONNECT] = "cgroup_inet4_connect",
84 [BPF_CGROUP_INET6_CONNECT] = "cgroup_inet6_connect",
85 [BPF_CGROUP_INET4_POST_BIND] = "cgroup_inet4_post_bind",
86 [BPF_CGROUP_INET6_POST_BIND] = "cgroup_inet6_post_bind",
87 [BPF_CGROUP_INET4_GETPEERNAME] = "cgroup_inet4_getpeername",
88 [BPF_CGROUP_INET6_GETPEERNAME] = "cgroup_inet6_getpeername",
89 [BPF_CGROUP_INET4_GETSOCKNAME] = "cgroup_inet4_getsockname",
90 [BPF_CGROUP_INET6_GETSOCKNAME] = "cgroup_inet6_getsockname",
91 [BPF_CGROUP_UDP4_SENDMSG] = "cgroup_udp4_sendmsg",
92 [BPF_CGROUP_UDP6_SENDMSG] = "cgroup_udp6_sendmsg",
93 [BPF_CGROUP_SYSCTL] = "cgroup_sysctl",
94 [BPF_CGROUP_UDP4_RECVMSG] = "cgroup_udp4_recvmsg",
95 [BPF_CGROUP_UDP6_RECVMSG] = "cgroup_udp6_recvmsg",
96 [BPF_CGROUP_GETSOCKOPT] = "cgroup_getsockopt",
97 [BPF_CGROUP_SETSOCKOPT] = "cgroup_setsockopt",
98 [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser",
99 [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict",
100 [BPF_SK_SKB_VERDICT] = "sk_skb_verdict",
101 [BPF_SK_MSG_VERDICT] = "sk_msg_verdict",
102 [BPF_LIRC_MODE2] = "lirc_mode2",
103 [BPF_FLOW_DISSECTOR] = "flow_dissector",
104 [BPF_TRACE_RAW_TP] = "trace_raw_tp",
105 [BPF_TRACE_FENTRY] = "trace_fentry",
106 [BPF_TRACE_FEXIT] = "trace_fexit",
107 [BPF_MODIFY_RETURN] = "modify_return",
108 [BPF_LSM_MAC] = "lsm_mac",
109 [BPF_LSM_CGROUP] = "lsm_cgroup",
110 [BPF_SK_LOOKUP] = "sk_lookup",
111 [BPF_TRACE_ITER] = "trace_iter",
112 [BPF_XDP_DEVMAP] = "xdp_devmap",
113 [BPF_XDP_CPUMAP] = "xdp_cpumap",
115 [BPF_SK_REUSEPORT_SELECT] = "sk_reuseport_select",
116 [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_reuseport_select_or_migrate",
117 [BPF_PERF_EVENT] = "perf_event",
118 [BPF_TRACE_KPROBE_MULTI] = "trace_kprobe_multi",
119 [BPF_STRUCT_OPS] = "struct_ops",
122 static const char * const link_type_name[] = {
123 [BPF_LINK_TYPE_UNSPEC] = "unspec",
124 [BPF_LINK_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
125 [BPF_LINK_TYPE_TRACING] = "tracing",
126 [BPF_LINK_TYPE_CGROUP] = "cgroup",
127 [BPF_LINK_TYPE_ITER] = "iter",
128 [BPF_LINK_TYPE_NETNS] = "netns",
129 [BPF_LINK_TYPE_XDP] = "xdp",
130 [BPF_LINK_TYPE_PERF_EVENT] = "perf_event",
131 [BPF_LINK_TYPE_KPROBE_MULTI] = "kprobe_multi",
132 [BPF_LINK_TYPE_STRUCT_OPS] = "struct_ops",
133 [BPF_LINK_TYPE_NETFILTER] = "netfilter",
136 static const char * const map_type_name[] = {
137 [BPF_MAP_TYPE_UNSPEC] = "unspec",
138 [BPF_MAP_TYPE_HASH] = "hash",
139 [BPF_MAP_TYPE_ARRAY] = "array",
140 [BPF_MAP_TYPE_PROG_ARRAY] = "prog_array",
141 [BPF_MAP_TYPE_PERF_EVENT_ARRAY] = "perf_event_array",
142 [BPF_MAP_TYPE_PERCPU_HASH] = "percpu_hash",
143 [BPF_MAP_TYPE_PERCPU_ARRAY] = "percpu_array",
144 [BPF_MAP_TYPE_STACK_TRACE] = "stack_trace",
145 [BPF_MAP_TYPE_CGROUP_ARRAY] = "cgroup_array",
146 [BPF_MAP_TYPE_LRU_HASH] = "lru_hash",
147 [BPF_MAP_TYPE_LRU_PERCPU_HASH] = "lru_percpu_hash",
148 [BPF_MAP_TYPE_LPM_TRIE] = "lpm_trie",
149 [BPF_MAP_TYPE_ARRAY_OF_MAPS] = "array_of_maps",
150 [BPF_MAP_TYPE_HASH_OF_MAPS] = "hash_of_maps",
151 [BPF_MAP_TYPE_DEVMAP] = "devmap",
152 [BPF_MAP_TYPE_DEVMAP_HASH] = "devmap_hash",
153 [BPF_MAP_TYPE_SOCKMAP] = "sockmap",
154 [BPF_MAP_TYPE_CPUMAP] = "cpumap",
155 [BPF_MAP_TYPE_XSKMAP] = "xskmap",
156 [BPF_MAP_TYPE_SOCKHASH] = "sockhash",
157 [BPF_MAP_TYPE_CGROUP_STORAGE] = "cgroup_storage",
158 [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY] = "reuseport_sockarray",
159 [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE] = "percpu_cgroup_storage",
160 [BPF_MAP_TYPE_QUEUE] = "queue",
161 [BPF_MAP_TYPE_STACK] = "stack",
162 [BPF_MAP_TYPE_SK_STORAGE] = "sk_storage",
163 [BPF_MAP_TYPE_STRUCT_OPS] = "struct_ops",
164 [BPF_MAP_TYPE_RINGBUF] = "ringbuf",
165 [BPF_MAP_TYPE_INODE_STORAGE] = "inode_storage",
166 [BPF_MAP_TYPE_TASK_STORAGE] = "task_storage",
167 [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter",
168 [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf",
169 [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage",
172 static const char * const prog_type_name[] = {
173 [BPF_PROG_TYPE_UNSPEC] = "unspec",
174 [BPF_PROG_TYPE_SOCKET_FILTER] = "socket_filter",
175 [BPF_PROG_TYPE_KPROBE] = "kprobe",
176 [BPF_PROG_TYPE_SCHED_CLS] = "sched_cls",
177 [BPF_PROG_TYPE_SCHED_ACT] = "sched_act",
178 [BPF_PROG_TYPE_TRACEPOINT] = "tracepoint",
179 [BPF_PROG_TYPE_XDP] = "xdp",
180 [BPF_PROG_TYPE_PERF_EVENT] = "perf_event",
181 [BPF_PROG_TYPE_CGROUP_SKB] = "cgroup_skb",
182 [BPF_PROG_TYPE_CGROUP_SOCK] = "cgroup_sock",
183 [BPF_PROG_TYPE_LWT_IN] = "lwt_in",
184 [BPF_PROG_TYPE_LWT_OUT] = "lwt_out",
185 [BPF_PROG_TYPE_LWT_XMIT] = "lwt_xmit",
186 [BPF_PROG_TYPE_SOCK_OPS] = "sock_ops",
187 [BPF_PROG_TYPE_SK_SKB] = "sk_skb",
188 [BPF_PROG_TYPE_CGROUP_DEVICE] = "cgroup_device",
189 [BPF_PROG_TYPE_SK_MSG] = "sk_msg",
190 [BPF_PROG_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
191 [BPF_PROG_TYPE_CGROUP_SOCK_ADDR] = "cgroup_sock_addr",
192 [BPF_PROG_TYPE_LWT_SEG6LOCAL] = "lwt_seg6local",
193 [BPF_PROG_TYPE_LIRC_MODE2] = "lirc_mode2",
194 [BPF_PROG_TYPE_SK_REUSEPORT] = "sk_reuseport",
195 [BPF_PROG_TYPE_FLOW_DISSECTOR] = "flow_dissector",
196 [BPF_PROG_TYPE_CGROUP_SYSCTL] = "cgroup_sysctl",
197 [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable",
198 [BPF_PROG_TYPE_CGROUP_SOCKOPT] = "cgroup_sockopt",
199 [BPF_PROG_TYPE_TRACING] = "tracing",
200 [BPF_PROG_TYPE_STRUCT_OPS] = "struct_ops",
201 [BPF_PROG_TYPE_EXT] = "ext",
202 [BPF_PROG_TYPE_LSM] = "lsm",
203 [BPF_PROG_TYPE_SK_LOOKUP] = "sk_lookup",
204 [BPF_PROG_TYPE_SYSCALL] = "syscall",
205 [BPF_PROG_TYPE_NETFILTER] = "netfilter",
208 static int __base_pr(enum libbpf_print_level level, const char *format,
211 if (level == LIBBPF_DEBUG)
214 return vfprintf(stderr, format, args);
217 static libbpf_print_fn_t __libbpf_pr = __base_pr;
219 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
221 libbpf_print_fn_t old_print_fn;
223 old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
229 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
233 libbpf_print_fn_t print_fn;
235 print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
241 va_start(args, format);
242 __libbpf_pr(level, format, args);
248 static void pr_perm_msg(int err)
253 if (err != -EPERM || geteuid() != 0)
256 err = getrlimit(RLIMIT_MEMLOCK, &limit);
260 if (limit.rlim_cur == RLIM_INFINITY)
263 if (limit.rlim_cur < 1024)
264 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
265 else if (limit.rlim_cur < 1024*1024)
266 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
268 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
270 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
274 #define STRERR_BUFSIZE 128
276 /* Copied from tools/perf/util/util.h */
278 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
282 # define zclose(fd) ({ \
285 ___err = close((fd)); \
290 static inline __u64 ptr_to_u64(const void *ptr)
292 return (__u64) (unsigned long) ptr;
295 int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
297 /* as of v1.0 libbpf_set_strict_mode() is a no-op */
301 __u32 libbpf_major_version(void)
303 return LIBBPF_MAJOR_VERSION;
306 __u32 libbpf_minor_version(void)
308 return LIBBPF_MINOR_VERSION;
311 const char *libbpf_version_string(void)
315 return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
331 enum reloc_type type;
334 const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
343 /* stored as sec_def->cookie for all libbpf-supported SEC()s */
346 /* expected_attach_type is optional, if kernel doesn't support that */
347 SEC_EXP_ATTACH_OPT = 1,
348 /* legacy, only used by libbpf_get_type_names() and
349 * libbpf_attach_type_by_name(), not used by libbpf itself at all.
350 * This used to be associated with cgroup (and few other) BPF programs
351 * that were attachable through BPF_PROG_ATTACH command. Pretty
352 * meaningless nowadays, though.
355 SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
356 /* attachment target is specified through BTF ID in either kernel or
357 * other BPF program's BTF object
360 /* BPF program type allows sleeping/blocking in kernel */
362 /* BPF program support non-linear XDP buffer */
368 enum bpf_prog_type prog_type;
369 enum bpf_attach_type expected_attach_type;
373 libbpf_prog_setup_fn_t prog_setup_fn;
374 libbpf_prog_prepare_load_fn_t prog_prepare_load_fn;
375 libbpf_prog_attach_fn_t prog_attach_fn;
379 * bpf_prog should be a better name but it has been used in
386 const struct bpf_sec_def *sec_def;
387 /* this program's instruction offset (in number of instructions)
388 * within its containing ELF section
391 /* number of original instructions in ELF section belonging to this
392 * program, not taking into account subprogram instructions possible
393 * appended later during relocation
396 /* Offset (in number of instructions) of the start of instruction
397 * belonging to this BPF program within its containing main BPF
398 * program. For the entry-point (main) BPF program, this is always
399 * zero. For a sub-program, this gets reset before each of main BPF
400 * programs are processed and relocated and is used to determined
401 * whether sub-program was already appended to the main program, and
402 * if yes, at which instruction offset.
406 /* instructions that belong to BPF program; insns[0] is located at
407 * sec_insn_off instruction within its ELF section in ELF file, so
408 * when mapping ELF file instruction index to the local instruction,
409 * one needs to subtract sec_insn_off; and vice versa.
411 struct bpf_insn *insns;
412 /* actual number of instruction in this BPF program's image; for
413 * entry-point BPF programs this includes the size of main program
414 * itself plus all the used sub-programs, appended at the end
418 struct reloc_desc *reloc_desc;
421 /* BPF verifier log settings */
426 struct bpf_object *obj;
431 bool mark_btf_static;
432 enum bpf_prog_type type;
433 enum bpf_attach_type expected_attach_type;
436 __u32 attach_btf_obj_fd;
438 __u32 attach_prog_fd;
441 __u32 func_info_rec_size;
445 __u32 line_info_rec_size;
450 struct bpf_struct_ops {
452 const struct btf_type *type;
453 struct bpf_program **progs;
454 __u32 *kern_func_off;
455 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
457 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
458 * btf_vmlinux's format.
459 * struct bpf_struct_ops_tcp_congestion_ops {
460 * [... some other kernel fields ...]
461 * struct tcp_congestion_ops data;
463 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
464 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
471 #define DATA_SEC ".data"
472 #define BSS_SEC ".bss"
473 #define RODATA_SEC ".rodata"
474 #define KCONFIG_SEC ".kconfig"
475 #define KSYMS_SEC ".ksyms"
476 #define STRUCT_OPS_SEC ".struct_ops"
477 #define STRUCT_OPS_LINK_SEC ".struct_ops.link"
479 enum libbpf_map_type {
489 unsigned int key_size;
490 unsigned int value_size;
491 unsigned int max_entries;
492 unsigned int map_flags;
496 struct bpf_object *obj;
498 /* real_name is defined for special internal maps (.rodata*,
499 * .data*, .bss, .kconfig) and preserves their original ELF section
500 * name. This is important to be able to find corresponding BTF
501 * DATASEC information.
509 struct bpf_map_def def;
512 __u32 btf_key_type_id;
513 __u32 btf_value_type_id;
514 __u32 btf_vmlinux_value_type_id;
515 enum libbpf_map_type libbpf_type;
517 struct bpf_struct_ops *st_ops;
518 struct bpf_map *inner_map;
544 enum extern_type type;
560 unsigned long long addr;
562 /* target btf_id of the corresponding kernel var. */
563 int kernel_btf_obj_fd;
566 /* local btf_id of the ksym extern's type. */
568 /* BTF fd index to be patched in for insn->off, this is
569 * 0 for vmlinux BTF, index in obj->fd_array for module
593 struct elf_sec_desc {
594 enum sec_type sec_type;
606 Elf_Data *st_ops_data;
607 Elf_Data *st_ops_link_data;
608 size_t shstrndx; /* section index for section name strings */
610 struct elf_sec_desc *secs;
613 __u32 btf_maps_sec_btf_id;
617 int st_ops_link_shndx;
623 char name[BPF_OBJ_NAME_LEN];
627 struct bpf_program *programs;
629 struct bpf_map *maps;
634 struct extern_desc *externs;
642 struct bpf_gen *gen_loader;
644 /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
645 struct elf_state efile;
648 struct btf_ext *btf_ext;
650 /* Parse and load BTF vmlinux if any of the programs in the object need
653 struct btf *btf_vmlinux;
654 /* Path to the custom BTF to be used for BPF CO-RE relocations as an
655 * override for vmlinux BTF.
657 char *btf_custom_path;
658 /* vmlinux BTF override for CO-RE relocations */
659 struct btf *btf_vmlinux_override;
660 /* Lazily initialized kernel module BTFs */
661 struct module_btf *btf_modules;
662 bool btf_modules_loaded;
663 size_t btf_module_cnt;
664 size_t btf_module_cap;
666 /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
675 struct usdt_manager *usdt_man;
680 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
681 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
682 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
683 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
684 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
685 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
686 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
687 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
688 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
690 void bpf_program__unload(struct bpf_program *prog)
697 zfree(&prog->func_info);
698 zfree(&prog->line_info);
701 static void bpf_program__exit(struct bpf_program *prog)
706 bpf_program__unload(prog);
708 zfree(&prog->sec_name);
710 zfree(&prog->reloc_desc);
717 static bool insn_is_subprog_call(const struct bpf_insn *insn)
719 return BPF_CLASS(insn->code) == BPF_JMP &&
720 BPF_OP(insn->code) == BPF_CALL &&
721 BPF_SRC(insn->code) == BPF_K &&
722 insn->src_reg == BPF_PSEUDO_CALL &&
723 insn->dst_reg == 0 &&
727 static bool is_call_insn(const struct bpf_insn *insn)
729 return insn->code == (BPF_JMP | BPF_CALL);
732 static bool insn_is_pseudo_func(struct bpf_insn *insn)
734 return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
738 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
739 const char *name, size_t sec_idx, const char *sec_name,
740 size_t sec_off, void *insn_data, size_t insn_data_sz)
742 if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
743 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
744 sec_name, name, sec_off, insn_data_sz);
748 memset(prog, 0, sizeof(*prog));
751 prog->sec_idx = sec_idx;
752 prog->sec_insn_off = sec_off / BPF_INSN_SZ;
753 prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
754 /* insns_cnt can later be increased by appending used subprograms */
755 prog->insns_cnt = prog->sec_insn_cnt;
757 prog->type = BPF_PROG_TYPE_UNSPEC;
760 /* libbpf's convention for SEC("?abc...") is that it's just like
761 * SEC("abc...") but the corresponding bpf_program starts out with
762 * autoload set to false.
764 if (sec_name[0] == '?') {
765 prog->autoload = false;
766 /* from now on forget there was ? in section name */
769 prog->autoload = true;
772 prog->autoattach = true;
774 /* inherit object's log_level */
775 prog->log_level = obj->log_level;
777 prog->sec_name = strdup(sec_name);
781 prog->name = strdup(name);
785 prog->insns = malloc(insn_data_sz);
788 memcpy(prog->insns, insn_data, insn_data_sz);
792 pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
793 bpf_program__exit(prog);
798 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
799 const char *sec_name, int sec_idx)
801 Elf_Data *symbols = obj->efile.symbols;
802 struct bpf_program *prog, *progs;
803 void *data = sec_data->d_buf;
804 size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
805 int nr_progs, err, i;
809 progs = obj->programs;
810 nr_progs = obj->nr_programs;
811 nr_syms = symbols->d_size / sizeof(Elf64_Sym);
813 for (i = 0; i < nr_syms; i++) {
814 sym = elf_sym_by_idx(obj, i);
816 if (sym->st_shndx != sec_idx)
818 if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
821 prog_sz = sym->st_size;
822 sec_off = sym->st_value;
824 name = elf_sym_str(obj, sym->st_name);
826 pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
828 return -LIBBPF_ERRNO__FORMAT;
831 if (sec_off + prog_sz > sec_sz) {
832 pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
834 return -LIBBPF_ERRNO__FORMAT;
837 if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
838 pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
842 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
843 sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
845 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
848 * In this case the original obj->programs
849 * is still valid, so don't need special treat for
850 * bpf_close_object().
852 pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
856 obj->programs = progs;
858 prog = &progs[nr_progs];
860 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
861 sec_off, data + sec_off, prog_sz);
865 /* if function is a global/weak symbol, but has restricted
866 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
867 * as static to enable more permissive BPF verification mode
868 * with more outside context available to BPF verifier
870 if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL
871 && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
872 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
873 prog->mark_btf_static = true;
876 obj->nr_programs = nr_progs;
882 static const struct btf_member *
883 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
885 struct btf_member *m;
888 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
889 if (btf_member_bit_offset(t, i) == bit_offset)
896 static const struct btf_member *
897 find_member_by_name(const struct btf *btf, const struct btf_type *t,
900 struct btf_member *m;
903 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
904 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
911 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
912 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
913 const char *name, __u32 kind);
916 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
917 const struct btf_type **type, __u32 *type_id,
918 const struct btf_type **vtype, __u32 *vtype_id,
919 const struct btf_member **data_member)
921 const struct btf_type *kern_type, *kern_vtype;
922 const struct btf_member *kern_data_member;
923 __s32 kern_vtype_id, kern_type_id;
926 kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
927 if (kern_type_id < 0) {
928 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
932 kern_type = btf__type_by_id(btf, kern_type_id);
934 /* Find the corresponding "map_value" type that will be used
935 * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example,
936 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
939 kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
940 tname, BTF_KIND_STRUCT);
941 if (kern_vtype_id < 0) {
942 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
943 STRUCT_OPS_VALUE_PREFIX, tname);
944 return kern_vtype_id;
946 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
948 /* Find "struct tcp_congestion_ops" from
949 * struct bpf_struct_ops_tcp_congestion_ops {
951 * struct tcp_congestion_ops data;
954 kern_data_member = btf_members(kern_vtype);
955 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
956 if (kern_data_member->type == kern_type_id)
959 if (i == btf_vlen(kern_vtype)) {
960 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
961 tname, STRUCT_OPS_VALUE_PREFIX, tname);
966 *type_id = kern_type_id;
968 *vtype_id = kern_vtype_id;
969 *data_member = kern_data_member;
974 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
976 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
979 /* Init the map's fields that depend on kern_btf */
980 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
981 const struct btf *btf,
982 const struct btf *kern_btf)
984 const struct btf_member *member, *kern_member, *kern_data_member;
985 const struct btf_type *type, *kern_type, *kern_vtype;
986 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
987 struct bpf_struct_ops *st_ops;
988 void *data, *kern_data;
992 st_ops = map->st_ops;
994 tname = st_ops->tname;
995 err = find_struct_ops_kern_types(kern_btf, tname,
996 &kern_type, &kern_type_id,
997 &kern_vtype, &kern_vtype_id,
1002 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
1003 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
1005 map->def.value_size = kern_vtype->size;
1006 map->btf_vmlinux_value_type_id = kern_vtype_id;
1008 st_ops->kern_vdata = calloc(1, kern_vtype->size);
1009 if (!st_ops->kern_vdata)
1012 data = st_ops->data;
1013 kern_data_off = kern_data_member->offset / 8;
1014 kern_data = st_ops->kern_vdata + kern_data_off;
1016 member = btf_members(type);
1017 for (i = 0; i < btf_vlen(type); i++, member++) {
1018 const struct btf_type *mtype, *kern_mtype;
1019 __u32 mtype_id, kern_mtype_id;
1020 void *mdata, *kern_mdata;
1021 __s64 msize, kern_msize;
1022 __u32 moff, kern_moff;
1023 __u32 kern_member_idx;
1026 mname = btf__name_by_offset(btf, member->name_off);
1027 kern_member = find_member_by_name(kern_btf, kern_type, mname);
1029 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
1034 kern_member_idx = kern_member - btf_members(kern_type);
1035 if (btf_member_bitfield_size(type, i) ||
1036 btf_member_bitfield_size(kern_type, kern_member_idx)) {
1037 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
1042 moff = member->offset / 8;
1043 kern_moff = kern_member->offset / 8;
1045 mdata = data + moff;
1046 kern_mdata = kern_data + kern_moff;
1048 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
1049 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
1051 if (BTF_INFO_KIND(mtype->info) !=
1052 BTF_INFO_KIND(kern_mtype->info)) {
1053 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
1054 map->name, mname, BTF_INFO_KIND(mtype->info),
1055 BTF_INFO_KIND(kern_mtype->info));
1059 if (btf_is_ptr(mtype)) {
1060 struct bpf_program *prog;
1062 prog = st_ops->progs[i];
1066 kern_mtype = skip_mods_and_typedefs(kern_btf,
1070 /* mtype->type must be a func_proto which was
1071 * guaranteed in bpf_object__collect_st_ops_relos(),
1072 * so only check kern_mtype for func_proto here.
1074 if (!btf_is_func_proto(kern_mtype)) {
1075 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1080 prog->attach_btf_id = kern_type_id;
1081 prog->expected_attach_type = kern_member_idx;
1083 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1085 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1086 map->name, mname, prog->name, moff,
1092 msize = btf__resolve_size(btf, mtype_id);
1093 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1094 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
1095 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1096 map->name, mname, (ssize_t)msize,
1097 (ssize_t)kern_msize);
1101 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1102 map->name, mname, (unsigned int)msize,
1104 memcpy(kern_mdata, mdata, msize);
1110 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1112 struct bpf_map *map;
1116 for (i = 0; i < obj->nr_maps; i++) {
1117 map = &obj->maps[i];
1119 if (!bpf_map__is_struct_ops(map))
1122 err = bpf_map__init_kern_struct_ops(map, obj->btf,
1131 static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name,
1132 int shndx, Elf_Data *data, __u32 map_flags)
1134 const struct btf_type *type, *datasec;
1135 const struct btf_var_secinfo *vsi;
1136 struct bpf_struct_ops *st_ops;
1137 const char *tname, *var_name;
1138 __s32 type_id, datasec_id;
1139 const struct btf *btf;
1140 struct bpf_map *map;
1147 datasec_id = btf__find_by_name_kind(btf, sec_name,
1149 if (datasec_id < 0) {
1150 pr_warn("struct_ops init: DATASEC %s not found\n",
1155 datasec = btf__type_by_id(btf, datasec_id);
1156 vsi = btf_var_secinfos(datasec);
1157 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1158 type = btf__type_by_id(obj->btf, vsi->type);
1159 var_name = btf__name_by_offset(obj->btf, type->name_off);
1161 type_id = btf__resolve_type(obj->btf, vsi->type);
1163 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1164 vsi->type, sec_name);
1168 type = btf__type_by_id(obj->btf, type_id);
1169 tname = btf__name_by_offset(obj->btf, type->name_off);
1171 pr_warn("struct_ops init: anonymous type is not supported\n");
1174 if (!btf_is_struct(type)) {
1175 pr_warn("struct_ops init: %s is not a struct\n", tname);
1179 map = bpf_object__add_map(obj);
1181 return PTR_ERR(map);
1183 map->sec_idx = shndx;
1184 map->sec_offset = vsi->offset;
1185 map->name = strdup(var_name);
1189 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1190 map->def.key_size = sizeof(int);
1191 map->def.value_size = type->size;
1192 map->def.max_entries = 1;
1193 map->def.map_flags = map_flags;
1195 map->st_ops = calloc(1, sizeof(*map->st_ops));
1198 st_ops = map->st_ops;
1199 st_ops->data = malloc(type->size);
1200 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1201 st_ops->kern_func_off = malloc(btf_vlen(type) *
1202 sizeof(*st_ops->kern_func_off));
1203 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1206 if (vsi->offset + type->size > data->d_size) {
1207 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1208 var_name, sec_name);
1212 memcpy(st_ops->data,
1213 data->d_buf + vsi->offset,
1215 st_ops->tname = tname;
1216 st_ops->type = type;
1217 st_ops->type_id = type_id;
1219 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1220 tname, type_id, var_name, vsi->offset);
1226 static int bpf_object_init_struct_ops(struct bpf_object *obj)
1230 err = init_struct_ops_maps(obj, STRUCT_OPS_SEC, obj->efile.st_ops_shndx,
1231 obj->efile.st_ops_data, 0);
1232 err = err ?: init_struct_ops_maps(obj, STRUCT_OPS_LINK_SEC,
1233 obj->efile.st_ops_link_shndx,
1234 obj->efile.st_ops_link_data,
1239 static struct bpf_object *bpf_object__new(const char *path,
1240 const void *obj_buf,
1242 const char *obj_name)
1244 struct bpf_object *obj;
1247 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1249 pr_warn("alloc memory failed for %s\n", path);
1250 return ERR_PTR(-ENOMEM);
1253 strcpy(obj->path, path);
1255 libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1257 /* Using basename() GNU version which doesn't modify arg. */
1258 libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1259 end = strchr(obj->name, '.');
1266 * Caller of this function should also call
1267 * bpf_object__elf_finish() after data collection to return
1268 * obj_buf to user. If not, we should duplicate the buffer to
1269 * avoid user freeing them before elf finish.
1271 obj->efile.obj_buf = obj_buf;
1272 obj->efile.obj_buf_sz = obj_buf_sz;
1273 obj->efile.btf_maps_shndx = -1;
1274 obj->efile.st_ops_shndx = -1;
1275 obj->efile.st_ops_link_shndx = -1;
1276 obj->kconfig_map_idx = -1;
1278 obj->kern_version = get_kernel_version();
1279 obj->loaded = false;
1284 static void bpf_object__elf_finish(struct bpf_object *obj)
1286 if (!obj->efile.elf)
1289 elf_end(obj->efile.elf);
1290 obj->efile.elf = NULL;
1291 obj->efile.symbols = NULL;
1292 obj->efile.st_ops_data = NULL;
1293 obj->efile.st_ops_link_data = NULL;
1295 zfree(&obj->efile.secs);
1296 obj->efile.sec_cnt = 0;
1297 zclose(obj->efile.fd);
1298 obj->efile.obj_buf = NULL;
1299 obj->efile.obj_buf_sz = 0;
1302 static int bpf_object__elf_init(struct bpf_object *obj)
1308 if (obj->efile.elf) {
1309 pr_warn("elf: init internal error\n");
1310 return -LIBBPF_ERRNO__LIBELF;
1313 if (obj->efile.obj_buf_sz > 0) {
1314 /* obj_buf should have been validated by bpf_object__open_mem(). */
1315 elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1317 obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1318 if (obj->efile.fd < 0) {
1319 char errmsg[STRERR_BUFSIZE], *cp;
1322 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1323 pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1327 elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1331 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1332 err = -LIBBPF_ERRNO__LIBELF;
1336 obj->efile.elf = elf;
1338 if (elf_kind(elf) != ELF_K_ELF) {
1339 err = -LIBBPF_ERRNO__FORMAT;
1340 pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1344 if (gelf_getclass(elf) != ELFCLASS64) {
1345 err = -LIBBPF_ERRNO__FORMAT;
1346 pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1350 obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1351 if (!obj->efile.ehdr) {
1352 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1353 err = -LIBBPF_ERRNO__FORMAT;
1357 if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1358 pr_warn("elf: failed to get section names section index for %s: %s\n",
1359 obj->path, elf_errmsg(-1));
1360 err = -LIBBPF_ERRNO__FORMAT;
1364 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1365 if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1366 pr_warn("elf: failed to get section names strings from %s: %s\n",
1367 obj->path, elf_errmsg(-1));
1368 err = -LIBBPF_ERRNO__FORMAT;
1372 /* Old LLVM set e_machine to EM_NONE */
1373 if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1374 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1375 err = -LIBBPF_ERRNO__FORMAT;
1381 bpf_object__elf_finish(obj);
1385 static int bpf_object__check_endianness(struct bpf_object *obj)
1387 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1388 if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB)
1390 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1391 if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB)
1394 # error "Unrecognized __BYTE_ORDER__"
1396 pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1397 return -LIBBPF_ERRNO__ENDIAN;
1401 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1404 pr_warn("invalid license section in %s\n", obj->path);
1405 return -LIBBPF_ERRNO__FORMAT;
1407 /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1408 * go over allowed ELF data section buffer
1410 libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1411 pr_debug("license of %s is %s\n", obj->path, obj->license);
1416 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1420 if (!data || size != sizeof(kver)) {
1421 pr_warn("invalid kver section in %s\n", obj->path);
1422 return -LIBBPF_ERRNO__FORMAT;
1424 memcpy(&kver, data, sizeof(kver));
1425 obj->kern_version = kver;
1426 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1430 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1432 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1433 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1438 static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1446 scn = elf_sec_by_name(obj, name);
1447 data = elf_sec_data(obj, scn);
1449 *size = data->d_size;
1450 return 0; /* found it */
1456 static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name)
1458 Elf_Data *symbols = obj->efile.symbols;
1462 for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1463 Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1465 if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1468 if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL &&
1469 ELF64_ST_BIND(sym->st_info) != STB_WEAK)
1472 sname = elf_sym_str(obj, sym->st_name);
1474 pr_warn("failed to get sym name string for var %s\n", name);
1475 return ERR_PTR(-EIO);
1477 if (strcmp(name, sname) == 0)
1481 return ERR_PTR(-ENOENT);
1484 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1486 struct bpf_map *map;
1489 err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap,
1490 sizeof(*obj->maps), obj->nr_maps + 1);
1492 return ERR_PTR(err);
1494 map = &obj->maps[obj->nr_maps++];
1497 map->inner_map_fd = -1;
1498 map->autocreate = true;
1503 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1505 long page_sz = sysconf(_SC_PAGE_SIZE);
1508 map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1509 map_sz = roundup(map_sz, page_sz);
1513 static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1515 char map_name[BPF_OBJ_NAME_LEN], *p;
1516 int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1518 /* This is one of the more confusing parts of libbpf for various
1519 * reasons, some of which are historical. The original idea for naming
1520 * internal names was to include as much of BPF object name prefix as
1521 * possible, so that it can be distinguished from similar internal
1522 * maps of a different BPF object.
1523 * As an example, let's say we have bpf_object named 'my_object_name'
1524 * and internal map corresponding to '.rodata' ELF section. The final
1525 * map name advertised to user and to the kernel will be
1526 * 'my_objec.rodata', taking first 8 characters of object name and
1527 * entire 7 characters of '.rodata'.
1528 * Somewhat confusingly, if internal map ELF section name is shorter
1529 * than 7 characters, e.g., '.bss', we still reserve 7 characters
1530 * for the suffix, even though we only have 4 actual characters, and
1531 * resulting map will be called 'my_objec.bss', not even using all 15
1532 * characters allowed by the kernel. Oh well, at least the truncated
1533 * object name is somewhat consistent in this case. But if the map
1534 * name is '.kconfig', we'll still have entirety of '.kconfig' added
1535 * (8 chars) and thus will be left with only first 7 characters of the
1536 * object name ('my_obje'). Happy guessing, user, that the final map
1537 * name will be "my_obje.kconfig".
1538 * Now, with libbpf starting to support arbitrarily named .rodata.*
1539 * and .data.* data sections, it's possible that ELF section name is
1540 * longer than allowed 15 chars, so we now need to be careful to take
1541 * only up to 15 first characters of ELF name, taking no BPF object
1542 * name characters at all. So '.rodata.abracadabra' will result in
1543 * '.rodata.abracad' kernel and user-visible name.
1544 * We need to keep this convoluted logic intact for .data, .bss and
1545 * .rodata maps, but for new custom .data.custom and .rodata.custom
1546 * maps we use their ELF names as is, not prepending bpf_object name
1547 * in front. We still need to truncate them to 15 characters for the
1548 * kernel. Full name can be recovered for such maps by using DATASEC
1549 * BTF type associated with such map's value type, though.
1551 if (sfx_len >= BPF_OBJ_NAME_LEN)
1552 sfx_len = BPF_OBJ_NAME_LEN - 1;
1554 /* if there are two or more dots in map name, it's a custom dot map */
1555 if (strchr(real_name + 1, '.') != NULL)
1558 pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1560 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1561 sfx_len, real_name);
1563 /* sanitise map name to characters allowed by kernel */
1564 for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1565 if (!isalnum(*p) && *p != '_' && *p != '.')
1568 return strdup(map_name);
1572 map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map);
1574 /* Internal BPF map is mmap()'able only if at least one of corresponding
1575 * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL
1576 * variable and it's not marked as __hidden (which turns it into, effectively,
1577 * a STATIC variable).
1579 static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map)
1581 const struct btf_type *t, *vt;
1582 struct btf_var_secinfo *vsi;
1585 if (!map->btf_value_type_id)
1588 t = btf__type_by_id(obj->btf, map->btf_value_type_id);
1589 if (!btf_is_datasec(t))
1592 vsi = btf_var_secinfos(t);
1593 for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) {
1594 vt = btf__type_by_id(obj->btf, vsi->type);
1595 if (!btf_is_var(vt))
1598 if (btf_var(vt)->linkage != BTF_VAR_STATIC)
1606 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1607 const char *real_name, int sec_idx, void *data, size_t data_sz)
1609 struct bpf_map_def *def;
1610 struct bpf_map *map;
1613 map = bpf_object__add_map(obj);
1615 return PTR_ERR(map);
1617 map->libbpf_type = type;
1618 map->sec_idx = sec_idx;
1619 map->sec_offset = 0;
1620 map->real_name = strdup(real_name);
1621 map->name = internal_map_name(obj, real_name);
1622 if (!map->real_name || !map->name) {
1623 zfree(&map->real_name);
1629 def->type = BPF_MAP_TYPE_ARRAY;
1630 def->key_size = sizeof(int);
1631 def->value_size = data_sz;
1632 def->max_entries = 1;
1633 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1634 ? BPF_F_RDONLY_PROG : 0;
1636 /* failures are fine because of maps like .rodata.str1.1 */
1637 (void) map_fill_btf_type_info(obj, map);
1639 if (map_is_mmapable(obj, map))
1640 def->map_flags |= BPF_F_MMAPABLE;
1642 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1643 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1645 map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1646 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1647 if (map->mmaped == MAP_FAILED) {
1650 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1652 zfree(&map->real_name);
1658 memcpy(map->mmaped, data, data_sz);
1660 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1664 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1666 struct elf_sec_desc *sec_desc;
1667 const char *sec_name;
1668 int err = 0, sec_idx;
1671 * Populate obj->maps with libbpf internal maps.
1673 for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
1674 sec_desc = &obj->efile.secs[sec_idx];
1676 /* Skip recognized sections with size 0. */
1677 if (!sec_desc->data || sec_desc->data->d_size == 0)
1680 switch (sec_desc->sec_type) {
1682 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1683 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1685 sec_desc->data->d_buf,
1686 sec_desc->data->d_size);
1689 obj->has_rodata = true;
1690 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1691 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1693 sec_desc->data->d_buf,
1694 sec_desc->data->d_size);
1697 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1698 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1701 sec_desc->data->d_size);
1714 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1719 for (i = 0; i < obj->nr_extern; i++) {
1720 if (strcmp(obj->externs[i].name, name) == 0)
1721 return &obj->externs[i];
1726 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1729 switch (ext->kcfg.type) {
1732 pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n",
1736 *(bool *)ext_val = value == 'y' ? true : false;
1740 *(enum libbpf_tristate *)ext_val = TRI_YES;
1741 else if (value == 'm')
1742 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1743 else /* value == 'n' */
1744 *(enum libbpf_tristate *)ext_val = TRI_NO;
1747 *(char *)ext_val = value;
1753 pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n",
1761 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1766 if (ext->kcfg.type != KCFG_CHAR_ARR) {
1767 pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n",
1772 len = strlen(value);
1773 if (value[len - 1] != '"') {
1774 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1781 if (len >= ext->kcfg.sz) {
1782 pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n",
1783 ext->name, value, len, ext->kcfg.sz - 1);
1784 len = ext->kcfg.sz - 1;
1786 memcpy(ext_val, value + 1, len);
1787 ext_val[len] = '\0';
1792 static int parse_u64(const char *value, __u64 *res)
1798 *res = strtoull(value, &value_end, 0);
1801 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1805 pr_warn("failed to parse '%s' as integer completely\n", value);
1811 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1813 int bit_sz = ext->kcfg.sz * 8;
1815 if (ext->kcfg.sz == 8)
1818 /* Validate that value stored in u64 fits in integer of `ext->sz`
1819 * bytes size without any loss of information. If the target integer
1820 * is signed, we rely on the following limits of integer type of
1821 * Y bits and subsequent transformation:
1823 * -2^(Y-1) <= X <= 2^(Y-1) - 1
1824 * 0 <= X + 2^(Y-1) <= 2^Y - 1
1825 * 0 <= X + 2^(Y-1) < 2^Y
1827 * For unsigned target integer, check that all the (64 - Y) bits are
1830 if (ext->kcfg.is_signed)
1831 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1833 return (v >> bit_sz) == 0;
1836 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1839 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR &&
1840 ext->kcfg.type != KCFG_BOOL) {
1841 pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n",
1842 ext->name, (unsigned long long)value);
1845 if (ext->kcfg.type == KCFG_BOOL && value > 1) {
1846 pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n",
1847 ext->name, (unsigned long long)value);
1851 if (!is_kcfg_value_in_range(ext, value)) {
1852 pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n",
1853 ext->name, (unsigned long long)value, ext->kcfg.sz);
1856 switch (ext->kcfg.sz) {
1858 *(__u8 *)ext_val = value;
1861 *(__u16 *)ext_val = value;
1864 *(__u32 *)ext_val = value;
1867 *(__u64 *)ext_val = value;
1876 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1877 char *buf, void *data)
1879 struct extern_desc *ext;
1885 if (!str_has_pfx(buf, "CONFIG_"))
1888 sep = strchr(buf, '=');
1890 pr_warn("failed to parse '%s': no separator\n", buf);
1894 /* Trim ending '\n' */
1896 if (buf[len - 1] == '\n')
1897 buf[len - 1] = '\0';
1898 /* Split on '=' and ensure that a value is present. */
1902 pr_warn("failed to parse '%s': no value\n", buf);
1906 ext = find_extern_by_name(obj, buf);
1907 if (!ext || ext->is_set)
1910 ext_val = data + ext->kcfg.data_off;
1914 case 'y': case 'n': case 'm':
1915 err = set_kcfg_value_tri(ext, ext_val, *value);
1918 err = set_kcfg_value_str(ext, ext_val, value);
1921 /* assume integer */
1922 err = parse_u64(value, &num);
1924 pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value);
1927 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1928 pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value);
1931 err = set_kcfg_value_num(ext, ext_val, num);
1936 pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value);
1940 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1948 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1951 else if (len >= PATH_MAX)
1952 return -ENAMETOOLONG;
1954 /* gzopen also accepts uncompressed files. */
1955 file = gzopen(buf, "r");
1957 file = gzopen("/proc/config.gz", "r");
1960 pr_warn("failed to open system Kconfig\n");
1964 while (gzgets(file, buf, sizeof(buf))) {
1965 err = bpf_object__process_kconfig_line(obj, buf, data);
1967 pr_warn("error parsing system Kconfig line '%s': %d\n",
1978 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1979 const char *config, void *data)
1985 file = fmemopen((void *)config, strlen(config), "r");
1988 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1992 while (fgets(buf, sizeof(buf), file)) {
1993 err = bpf_object__process_kconfig_line(obj, buf, data);
1995 pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
2005 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
2007 struct extern_desc *last_ext = NULL, *ext;
2011 for (i = 0; i < obj->nr_extern; i++) {
2012 ext = &obj->externs[i];
2013 if (ext->type == EXT_KCFG)
2020 map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
2021 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
2022 ".kconfig", obj->efile.symbols_shndx,
2027 obj->kconfig_map_idx = obj->nr_maps - 1;
2032 const struct btf_type *
2033 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2035 const struct btf_type *t = btf__type_by_id(btf, id);
2040 while (btf_is_mod(t) || btf_is_typedef(t)) {
2043 t = btf__type_by_id(btf, t->type);
2049 static const struct btf_type *
2050 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2052 const struct btf_type *t;
2054 t = skip_mods_and_typedefs(btf, id, NULL);
2058 t = skip_mods_and_typedefs(btf, t->type, res_id);
2060 return btf_is_func_proto(t) ? t : NULL;
2063 static const char *__btf_kind_str(__u16 kind)
2066 case BTF_KIND_UNKN: return "void";
2067 case BTF_KIND_INT: return "int";
2068 case BTF_KIND_PTR: return "ptr";
2069 case BTF_KIND_ARRAY: return "array";
2070 case BTF_KIND_STRUCT: return "struct";
2071 case BTF_KIND_UNION: return "union";
2072 case BTF_KIND_ENUM: return "enum";
2073 case BTF_KIND_FWD: return "fwd";
2074 case BTF_KIND_TYPEDEF: return "typedef";
2075 case BTF_KIND_VOLATILE: return "volatile";
2076 case BTF_KIND_CONST: return "const";
2077 case BTF_KIND_RESTRICT: return "restrict";
2078 case BTF_KIND_FUNC: return "func";
2079 case BTF_KIND_FUNC_PROTO: return "func_proto";
2080 case BTF_KIND_VAR: return "var";
2081 case BTF_KIND_DATASEC: return "datasec";
2082 case BTF_KIND_FLOAT: return "float";
2083 case BTF_KIND_DECL_TAG: return "decl_tag";
2084 case BTF_KIND_TYPE_TAG: return "type_tag";
2085 case BTF_KIND_ENUM64: return "enum64";
2086 default: return "unknown";
2090 const char *btf_kind_str(const struct btf_type *t)
2092 return __btf_kind_str(btf_kind(t));
2096 * Fetch integer attribute of BTF map definition. Such attributes are
2097 * represented using a pointer to an array, in which dimensionality of array
2098 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2099 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2100 * type definition, while using only sizeof(void *) space in ELF data section.
2102 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2103 const struct btf_member *m, __u32 *res)
2105 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2106 const char *name = btf__name_by_offset(btf, m->name_off);
2107 const struct btf_array *arr_info;
2108 const struct btf_type *arr_t;
2110 if (!btf_is_ptr(t)) {
2111 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2112 map_name, name, btf_kind_str(t));
2116 arr_t = btf__type_by_id(btf, t->type);
2118 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2119 map_name, name, t->type);
2122 if (!btf_is_array(arr_t)) {
2123 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2124 map_name, name, btf_kind_str(arr_t));
2127 arr_info = btf_array(arr_t);
2128 *res = arr_info->nelems;
2132 static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name)
2136 len = snprintf(buf, buf_sz, "%s/%s", path, name);
2140 return -ENAMETOOLONG;
2145 static int build_map_pin_path(struct bpf_map *map, const char *path)
2151 path = "/sys/fs/bpf";
2153 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
2157 return bpf_map__set_pin_path(map, buf);
2160 /* should match definition in bpf_helpers.h */
2161 enum libbpf_pin_type {
2163 /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
2167 int parse_btf_map_def(const char *map_name, struct btf *btf,
2168 const struct btf_type *def_t, bool strict,
2169 struct btf_map_def *map_def, struct btf_map_def *inner_def)
2171 const struct btf_type *t;
2172 const struct btf_member *m;
2173 bool is_inner = inner_def == NULL;
2176 vlen = btf_vlen(def_t);
2177 m = btf_members(def_t);
2178 for (i = 0; i < vlen; i++, m++) {
2179 const char *name = btf__name_by_offset(btf, m->name_off);
2182 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2185 if (strcmp(name, "type") == 0) {
2186 if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2188 map_def->parts |= MAP_DEF_MAP_TYPE;
2189 } else if (strcmp(name, "max_entries") == 0) {
2190 if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2192 map_def->parts |= MAP_DEF_MAX_ENTRIES;
2193 } else if (strcmp(name, "map_flags") == 0) {
2194 if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2196 map_def->parts |= MAP_DEF_MAP_FLAGS;
2197 } else if (strcmp(name, "numa_node") == 0) {
2198 if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2200 map_def->parts |= MAP_DEF_NUMA_NODE;
2201 } else if (strcmp(name, "key_size") == 0) {
2204 if (!get_map_field_int(map_name, btf, m, &sz))
2206 if (map_def->key_size && map_def->key_size != sz) {
2207 pr_warn("map '%s': conflicting key size %u != %u.\n",
2208 map_name, map_def->key_size, sz);
2211 map_def->key_size = sz;
2212 map_def->parts |= MAP_DEF_KEY_SIZE;
2213 } else if (strcmp(name, "key") == 0) {
2216 t = btf__type_by_id(btf, m->type);
2218 pr_warn("map '%s': key type [%d] not found.\n",
2222 if (!btf_is_ptr(t)) {
2223 pr_warn("map '%s': key spec is not PTR: %s.\n",
2224 map_name, btf_kind_str(t));
2227 sz = btf__resolve_size(btf, t->type);
2229 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2230 map_name, t->type, (ssize_t)sz);
2233 if (map_def->key_size && map_def->key_size != sz) {
2234 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2235 map_name, map_def->key_size, (ssize_t)sz);
2238 map_def->key_size = sz;
2239 map_def->key_type_id = t->type;
2240 map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2241 } else if (strcmp(name, "value_size") == 0) {
2244 if (!get_map_field_int(map_name, btf, m, &sz))
2246 if (map_def->value_size && map_def->value_size != sz) {
2247 pr_warn("map '%s': conflicting value size %u != %u.\n",
2248 map_name, map_def->value_size, sz);
2251 map_def->value_size = sz;
2252 map_def->parts |= MAP_DEF_VALUE_SIZE;
2253 } else if (strcmp(name, "value") == 0) {
2256 t = btf__type_by_id(btf, m->type);
2258 pr_warn("map '%s': value type [%d] not found.\n",
2262 if (!btf_is_ptr(t)) {
2263 pr_warn("map '%s': value spec is not PTR: %s.\n",
2264 map_name, btf_kind_str(t));
2267 sz = btf__resolve_size(btf, t->type);
2269 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2270 map_name, t->type, (ssize_t)sz);
2273 if (map_def->value_size && map_def->value_size != sz) {
2274 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2275 map_name, map_def->value_size, (ssize_t)sz);
2278 map_def->value_size = sz;
2279 map_def->value_type_id = t->type;
2280 map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2282 else if (strcmp(name, "values") == 0) {
2283 bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2284 bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2285 const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2286 char inner_map_name[128];
2290 pr_warn("map '%s': multi-level inner maps not supported.\n",
2294 if (i != vlen - 1) {
2295 pr_warn("map '%s': '%s' member should be last.\n",
2299 if (!is_map_in_map && !is_prog_array) {
2300 pr_warn("map '%s': should be map-in-map or prog-array.\n",
2304 if (map_def->value_size && map_def->value_size != 4) {
2305 pr_warn("map '%s': conflicting value size %u != 4.\n",
2306 map_name, map_def->value_size);
2309 map_def->value_size = 4;
2310 t = btf__type_by_id(btf, m->type);
2312 pr_warn("map '%s': %s type [%d] not found.\n",
2313 map_name, desc, m->type);
2316 if (!btf_is_array(t) || btf_array(t)->nelems) {
2317 pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2321 t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2322 if (!btf_is_ptr(t)) {
2323 pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2324 map_name, desc, btf_kind_str(t));
2327 t = skip_mods_and_typedefs(btf, t->type, NULL);
2328 if (is_prog_array) {
2329 if (!btf_is_func_proto(t)) {
2330 pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2331 map_name, btf_kind_str(t));
2336 if (!btf_is_struct(t)) {
2337 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2338 map_name, btf_kind_str(t));
2342 snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2343 err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2347 map_def->parts |= MAP_DEF_INNER_MAP;
2348 } else if (strcmp(name, "pinning") == 0) {
2352 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2355 if (!get_map_field_int(map_name, btf, m, &val))
2357 if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2358 pr_warn("map '%s': invalid pinning value %u.\n",
2362 map_def->pinning = val;
2363 map_def->parts |= MAP_DEF_PINNING;
2364 } else if (strcmp(name, "map_extra") == 0) {
2367 if (!get_map_field_int(map_name, btf, m, &map_extra))
2369 map_def->map_extra = map_extra;
2370 map_def->parts |= MAP_DEF_MAP_EXTRA;
2373 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2376 pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2380 if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2381 pr_warn("map '%s': map type isn't specified.\n", map_name);
2388 static size_t adjust_ringbuf_sz(size_t sz)
2390 __u32 page_sz = sysconf(_SC_PAGE_SIZE);
2393 /* if user forgot to set any size, make sure they see error */
2396 /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be
2397 * a power-of-2 multiple of kernel's page size. If user diligently
2398 * satisified these conditions, pass the size through.
2400 if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz))
2403 /* Otherwise find closest (page_sz * power_of_2) product bigger than
2404 * user-set size to satisfy both user size request and kernel
2405 * requirements and substitute correct max_entries for map creation.
2407 for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) {
2408 if (mul * page_sz > sz)
2409 return mul * page_sz;
2412 /* if it's impossible to satisfy the conditions (i.e., user size is
2413 * very close to UINT_MAX but is not a power-of-2 multiple of
2414 * page_size) then just return original size and let kernel reject it
2419 static bool map_is_ringbuf(const struct bpf_map *map)
2421 return map->def.type == BPF_MAP_TYPE_RINGBUF ||
2422 map->def.type == BPF_MAP_TYPE_USER_RINGBUF;
2425 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2427 map->def.type = def->map_type;
2428 map->def.key_size = def->key_size;
2429 map->def.value_size = def->value_size;
2430 map->def.max_entries = def->max_entries;
2431 map->def.map_flags = def->map_flags;
2432 map->map_extra = def->map_extra;
2434 map->numa_node = def->numa_node;
2435 map->btf_key_type_id = def->key_type_id;
2436 map->btf_value_type_id = def->value_type_id;
2438 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
2439 if (map_is_ringbuf(map))
2440 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
2442 if (def->parts & MAP_DEF_MAP_TYPE)
2443 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2445 if (def->parts & MAP_DEF_KEY_TYPE)
2446 pr_debug("map '%s': found key [%u], sz = %u.\n",
2447 map->name, def->key_type_id, def->key_size);
2448 else if (def->parts & MAP_DEF_KEY_SIZE)
2449 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2451 if (def->parts & MAP_DEF_VALUE_TYPE)
2452 pr_debug("map '%s': found value [%u], sz = %u.\n",
2453 map->name, def->value_type_id, def->value_size);
2454 else if (def->parts & MAP_DEF_VALUE_SIZE)
2455 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2457 if (def->parts & MAP_DEF_MAX_ENTRIES)
2458 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2459 if (def->parts & MAP_DEF_MAP_FLAGS)
2460 pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2461 if (def->parts & MAP_DEF_MAP_EXTRA)
2462 pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2463 (unsigned long long)def->map_extra);
2464 if (def->parts & MAP_DEF_PINNING)
2465 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2466 if (def->parts & MAP_DEF_NUMA_NODE)
2467 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2469 if (def->parts & MAP_DEF_INNER_MAP)
2470 pr_debug("map '%s': found inner map definition.\n", map->name);
2473 static const char *btf_var_linkage_str(__u32 linkage)
2476 case BTF_VAR_STATIC: return "static";
2477 case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2478 case BTF_VAR_GLOBAL_EXTERN: return "extern";
2479 default: return "unknown";
2483 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2484 const struct btf_type *sec,
2485 int var_idx, int sec_idx,
2486 const Elf_Data *data, bool strict,
2487 const char *pin_root_path)
2489 struct btf_map_def map_def = {}, inner_def = {};
2490 const struct btf_type *var, *def;
2491 const struct btf_var_secinfo *vi;
2492 const struct btf_var *var_extra;
2493 const char *map_name;
2494 struct bpf_map *map;
2497 vi = btf_var_secinfos(sec) + var_idx;
2498 var = btf__type_by_id(obj->btf, vi->type);
2499 var_extra = btf_var(var);
2500 map_name = btf__name_by_offset(obj->btf, var->name_off);
2502 if (map_name == NULL || map_name[0] == '\0') {
2503 pr_warn("map #%d: empty name.\n", var_idx);
2506 if ((__u64)vi->offset + vi->size > data->d_size) {
2507 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2510 if (!btf_is_var(var)) {
2511 pr_warn("map '%s': unexpected var kind %s.\n",
2512 map_name, btf_kind_str(var));
2515 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2516 pr_warn("map '%s': unsupported map linkage %s.\n",
2517 map_name, btf_var_linkage_str(var_extra->linkage));
2521 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2522 if (!btf_is_struct(def)) {
2523 pr_warn("map '%s': unexpected def kind %s.\n",
2524 map_name, btf_kind_str(var));
2527 if (def->size > vi->size) {
2528 pr_warn("map '%s': invalid def size.\n", map_name);
2532 map = bpf_object__add_map(obj);
2534 return PTR_ERR(map);
2535 map->name = strdup(map_name);
2537 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2540 map->libbpf_type = LIBBPF_MAP_UNSPEC;
2541 map->def.type = BPF_MAP_TYPE_UNSPEC;
2542 map->sec_idx = sec_idx;
2543 map->sec_offset = vi->offset;
2544 map->btf_var_idx = var_idx;
2545 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2546 map_name, map->sec_idx, map->sec_offset);
2548 err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2552 fill_map_from_def(map, &map_def);
2554 if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2555 err = build_map_pin_path(map, pin_root_path);
2557 pr_warn("map '%s': couldn't build pin path.\n", map->name);
2562 if (map_def.parts & MAP_DEF_INNER_MAP) {
2563 map->inner_map = calloc(1, sizeof(*map->inner_map));
2564 if (!map->inner_map)
2566 map->inner_map->fd = -1;
2567 map->inner_map->sec_idx = sec_idx;
2568 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2569 if (!map->inner_map->name)
2571 sprintf(map->inner_map->name, "%s.inner", map_name);
2573 fill_map_from_def(map->inner_map, &inner_def);
2576 err = map_fill_btf_type_info(obj, map);
2583 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2584 const char *pin_root_path)
2586 const struct btf_type *sec = NULL;
2587 int nr_types, i, vlen, err;
2588 const struct btf_type *t;
2593 if (obj->efile.btf_maps_shndx < 0)
2596 scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2597 data = elf_sec_data(obj, scn);
2598 if (!scn || !data) {
2599 pr_warn("elf: failed to get %s map definitions for %s\n",
2600 MAPS_ELF_SEC, obj->path);
2604 nr_types = btf__type_cnt(obj->btf);
2605 for (i = 1; i < nr_types; i++) {
2606 t = btf__type_by_id(obj->btf, i);
2607 if (!btf_is_datasec(t))
2609 name = btf__name_by_offset(obj->btf, t->name_off);
2610 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2612 obj->efile.btf_maps_sec_btf_id = i;
2618 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2622 vlen = btf_vlen(sec);
2623 for (i = 0; i < vlen; i++) {
2624 err = bpf_object__init_user_btf_map(obj, sec, i,
2625 obj->efile.btf_maps_shndx,
2635 static int bpf_object__init_maps(struct bpf_object *obj,
2636 const struct bpf_object_open_opts *opts)
2638 const char *pin_root_path;
2642 strict = !OPTS_GET(opts, relaxed_maps, false);
2643 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2645 err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2646 err = err ?: bpf_object__init_global_data_maps(obj);
2647 err = err ?: bpf_object__init_kconfig_map(obj);
2648 err = err ?: bpf_object_init_struct_ops(obj);
2653 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2657 sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
2661 return sh->sh_flags & SHF_EXECINSTR;
2664 static bool btf_needs_sanitization(struct bpf_object *obj)
2666 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2667 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2668 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2669 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2670 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2671 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2672 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
2674 return !has_func || !has_datasec || !has_func_global || !has_float ||
2675 !has_decl_tag || !has_type_tag || !has_enum64;
2678 static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2680 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2681 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2682 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2683 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2684 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2685 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2686 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
2687 int enum64_placeholder_id = 0;
2691 for (i = 1; i < btf__type_cnt(btf); i++) {
2692 t = (struct btf_type *)btf__type_by_id(btf, i);
2694 if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
2695 /* replace VAR/DECL_TAG with INT */
2696 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2698 * using size = 1 is the safest choice, 4 will be too
2699 * big and cause kernel BTF validation failure if
2700 * original variable took less than 4 bytes
2703 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2704 } else if (!has_datasec && btf_is_datasec(t)) {
2705 /* replace DATASEC with STRUCT */
2706 const struct btf_var_secinfo *v = btf_var_secinfos(t);
2707 struct btf_member *m = btf_members(t);
2708 struct btf_type *vt;
2711 name = (char *)btf__name_by_offset(btf, t->name_off);
2719 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2720 for (j = 0; j < vlen; j++, v++, m++) {
2721 /* order of field assignments is important */
2722 m->offset = v->offset * 8;
2724 /* preserve variable name as member name */
2725 vt = (void *)btf__type_by_id(btf, v->type);
2726 m->name_off = vt->name_off;
2728 } else if (!has_func && btf_is_func_proto(t)) {
2729 /* replace FUNC_PROTO with ENUM */
2731 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2732 t->size = sizeof(__u32); /* kernel enforced */
2733 } else if (!has_func && btf_is_func(t)) {
2734 /* replace FUNC with TYPEDEF */
2735 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2736 } else if (!has_func_global && btf_is_func(t)) {
2737 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2738 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2739 } else if (!has_float && btf_is_float(t)) {
2740 /* replace FLOAT with an equally-sized empty STRUCT;
2741 * since C compilers do not accept e.g. "float" as a
2742 * valid struct name, make it anonymous
2745 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
2746 } else if (!has_type_tag && btf_is_type_tag(t)) {
2747 /* replace TYPE_TAG with a CONST */
2749 t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
2750 } else if (!has_enum64 && btf_is_enum(t)) {
2751 /* clear the kflag */
2752 t->info = btf_type_info(btf_kind(t), btf_vlen(t), false);
2753 } else if (!has_enum64 && btf_is_enum64(t)) {
2754 /* replace ENUM64 with a union */
2755 struct btf_member *m;
2757 if (enum64_placeholder_id == 0) {
2758 enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0);
2759 if (enum64_placeholder_id < 0)
2760 return enum64_placeholder_id;
2762 t = (struct btf_type *)btf__type_by_id(btf, i);
2767 t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen);
2768 for (j = 0; j < vlen; j++, m++) {
2769 m->type = enum64_placeholder_id;
2778 static bool libbpf_needs_btf(const struct bpf_object *obj)
2780 return obj->efile.btf_maps_shndx >= 0 ||
2781 obj->efile.st_ops_shndx >= 0 ||
2782 obj->efile.st_ops_link_shndx >= 0 ||
2786 static bool kernel_needs_btf(const struct bpf_object *obj)
2788 return obj->efile.st_ops_shndx >= 0 || obj->efile.st_ops_link_shndx >= 0;
2791 static int bpf_object__init_btf(struct bpf_object *obj,
2793 Elf_Data *btf_ext_data)
2798 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2799 err = libbpf_get_error(obj->btf);
2802 pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err);
2805 /* enforce 8-byte pointers for BPF-targeted BTFs */
2806 btf__set_pointer_size(obj->btf, 8);
2809 struct btf_ext_info *ext_segs[3];
2810 int seg_num, sec_num;
2813 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2814 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2817 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
2818 err = libbpf_get_error(obj->btf_ext);
2820 pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n",
2821 BTF_EXT_ELF_SEC, err);
2822 obj->btf_ext = NULL;
2826 /* setup .BTF.ext to ELF section mapping */
2827 ext_segs[0] = &obj->btf_ext->func_info;
2828 ext_segs[1] = &obj->btf_ext->line_info;
2829 ext_segs[2] = &obj->btf_ext->core_relo_info;
2830 for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) {
2831 struct btf_ext_info *seg = ext_segs[seg_num];
2832 const struct btf_ext_info_sec *sec;
2833 const char *sec_name;
2836 if (seg->sec_cnt == 0)
2839 seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs));
2840 if (!seg->sec_idxs) {
2846 for_each_btf_ext_sec(seg, sec) {
2847 /* preventively increment index to avoid doing
2848 * this before every continue below
2852 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
2853 if (str_is_empty(sec_name))
2855 scn = elf_sec_by_name(obj, sec_name);
2859 seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn);
2864 if (err && libbpf_needs_btf(obj)) {
2865 pr_warn("BTF is required, but is missing or corrupted.\n");
2871 static int compare_vsi_off(const void *_a, const void *_b)
2873 const struct btf_var_secinfo *a = _a;
2874 const struct btf_var_secinfo *b = _b;
2876 return a->offset - b->offset;
2879 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
2882 __u32 size = 0, i, vars = btf_vlen(t);
2883 const char *sec_name = btf__name_by_offset(btf, t->name_off);
2884 struct btf_var_secinfo *vsi;
2885 bool fixup_offsets = false;
2889 pr_debug("No name found in string section for DATASEC kind.\n");
2893 /* Extern-backing datasecs (.ksyms, .kconfig) have their size and
2894 * variable offsets set at the previous step. Further, not every
2895 * extern BTF VAR has corresponding ELF symbol preserved, so we skip
2896 * all fixups altogether for such sections and go straight to sorting
2897 * VARs within their DATASEC.
2899 if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0)
2902 /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to
2903 * fix this up. But BPF static linker already fixes this up and fills
2904 * all the sizes and offsets during static linking. So this step has
2905 * to be optional. But the STV_HIDDEN handling is non-optional for any
2906 * non-extern DATASEC, so the variable fixup loop below handles both
2907 * functions at the same time, paying the cost of BTF VAR <-> ELF
2908 * symbol matching just once.
2911 err = find_elf_sec_sz(obj, sec_name, &size);
2913 pr_debug("sec '%s': failed to determine size from ELF: size %u, err %d\n",
2914 sec_name, size, err);
2919 fixup_offsets = true;
2922 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
2923 const struct btf_type *t_var;
2924 struct btf_var *var;
2925 const char *var_name;
2928 t_var = btf__type_by_id(btf, vsi->type);
2929 if (!t_var || !btf_is_var(t_var)) {
2930 pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name);
2934 var = btf_var(t_var);
2935 if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN)
2938 var_name = btf__name_by_offset(btf, t_var->name_off);
2940 pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n",
2945 sym = find_elf_var_sym(obj, var_name);
2947 pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n",
2948 sec_name, var_name);
2953 vsi->offset = sym->st_value;
2955 /* if variable is a global/weak symbol, but has restricted
2956 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR
2957 * as static. This follows similar logic for functions (BPF
2958 * subprogs) and influences libbpf's further decisions about
2959 * whether to make global data BPF array maps as
2962 if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
2963 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)
2964 var->linkage = BTF_VAR_STATIC;
2968 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
2972 static int bpf_object_fixup_btf(struct bpf_object *obj)
2979 n = btf__type_cnt(obj->btf);
2980 for (i = 1; i < n; i++) {
2981 struct btf_type *t = btf_type_by_id(obj->btf, i);
2983 /* Loader needs to fix up some of the things compiler
2984 * couldn't get its hands on while emitting BTF. This
2985 * is section size and global variable offset. We use
2986 * the info from the ELF itself for this purpose.
2988 if (btf_is_datasec(t)) {
2989 err = btf_fixup_datasec(obj, obj->btf, t);
2998 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
3000 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
3001 prog->type == BPF_PROG_TYPE_LSM)
3004 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
3005 * also need vmlinux BTF
3007 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
3013 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
3015 struct bpf_program *prog;
3018 /* CO-RE relocations need kernel BTF, only when btf_custom_path
3021 if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
3024 /* Support for typed ksyms needs kernel BTF */
3025 for (i = 0; i < obj->nr_extern; i++) {
3026 const struct extern_desc *ext;
3028 ext = &obj->externs[i];
3029 if (ext->type == EXT_KSYM && ext->ksym.type_id)
3033 bpf_object__for_each_program(prog, obj) {
3034 if (!prog->autoload)
3036 if (prog_needs_vmlinux_btf(prog))
3043 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
3047 /* btf_vmlinux could be loaded earlier */
3048 if (obj->btf_vmlinux || obj->gen_loader)
3051 if (!force && !obj_needs_vmlinux_btf(obj))
3054 obj->btf_vmlinux = btf__load_vmlinux_btf();
3055 err = libbpf_get_error(obj->btf_vmlinux);
3057 pr_warn("Error loading vmlinux BTF: %d\n", err);
3058 obj->btf_vmlinux = NULL;
3064 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
3066 struct btf *kern_btf = obj->btf;
3067 bool btf_mandatory, sanitize;
3073 if (!kernel_supports(obj, FEAT_BTF)) {
3074 if (kernel_needs_btf(obj)) {
3078 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
3082 /* Even though some subprogs are global/weak, user might prefer more
3083 * permissive BPF verification process that BPF verifier performs for
3084 * static functions, taking into account more context from the caller
3085 * functions. In such case, they need to mark such subprogs with
3086 * __attribute__((visibility("hidden"))) and libbpf will adjust
3087 * corresponding FUNC BTF type to be marked as static and trigger more
3088 * involved BPF verification process.
3090 for (i = 0; i < obj->nr_programs; i++) {
3091 struct bpf_program *prog = &obj->programs[i];
3096 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
3099 n = btf__type_cnt(obj->btf);
3100 for (j = 1; j < n; j++) {
3101 t = btf_type_by_id(obj->btf, j);
3102 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
3105 name = btf__str_by_offset(obj->btf, t->name_off);
3106 if (strcmp(name, prog->name) != 0)
3109 t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
3114 sanitize = btf_needs_sanitization(obj);
3116 const void *raw_data;
3119 /* clone BTF to sanitize a copy and leave the original intact */
3120 raw_data = btf__raw_data(obj->btf, &sz);
3121 kern_btf = btf__new(raw_data, sz);
3122 err = libbpf_get_error(kern_btf);
3126 /* enforce 8-byte pointers for BPF-targeted BTFs */
3127 btf__set_pointer_size(obj->btf, 8);
3128 err = bpf_object__sanitize_btf(obj, kern_btf);
3133 if (obj->gen_loader) {
3135 const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3139 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3140 /* Pretend to have valid FD to pass various fd >= 0 checks.
3141 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3143 btf__set_fd(kern_btf, 0);
3145 /* currently BPF_BTF_LOAD only supports log_level 1 */
3146 err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3147 obj->log_level ? 1 : 0);
3151 /* move fd to libbpf's BTF */
3152 btf__set_fd(obj->btf, btf__fd(kern_btf));
3153 btf__set_fd(kern_btf, -1);
3155 btf__free(kern_btf);
3159 btf_mandatory = kernel_needs_btf(obj);
3160 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
3161 btf_mandatory ? "BTF is mandatory, can't proceed."
3162 : "BTF is optional, ignoring.");
3169 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3173 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3175 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3176 off, obj->path, elf_errmsg(-1));
3183 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3187 name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3189 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3190 off, obj->path, elf_errmsg(-1));
3197 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3201 scn = elf_getscn(obj->efile.elf, idx);
3203 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3204 idx, obj->path, elf_errmsg(-1));
3210 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3212 Elf_Scn *scn = NULL;
3213 Elf *elf = obj->efile.elf;
3214 const char *sec_name;
3216 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3217 sec_name = elf_sec_name(obj, scn);
3221 if (strcmp(sec_name, name) != 0)
3229 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3236 shdr = elf64_getshdr(scn);
3238 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3239 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3246 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3254 sh = elf_sec_hdr(obj, scn);
3258 name = elf_sec_str(obj, sh->sh_name);
3260 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3261 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3268 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3275 data = elf_getdata(scn, 0);
3277 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3278 elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3279 obj->path, elf_errmsg(-1));
3286 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3288 if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3291 return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3294 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3296 if (idx >= data->d_size / sizeof(Elf64_Rel))
3299 return (Elf64_Rel *)data->d_buf + idx;
3302 static bool is_sec_name_dwarf(const char *name)
3304 /* approximation, but the actual list is too long */
3305 return str_has_pfx(name, ".debug_");
3308 static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3310 /* no special handling of .strtab */
3311 if (hdr->sh_type == SHT_STRTAB)
3314 /* ignore .llvm_addrsig section as well */
3315 if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3318 /* no subprograms will lead to an empty .text section, ignore it */
3319 if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3320 strcmp(name, ".text") == 0)
3323 /* DWARF sections */
3324 if (is_sec_name_dwarf(name))
3327 if (str_has_pfx(name, ".rel")) {
3328 name += sizeof(".rel") - 1;
3329 /* DWARF section relocations */
3330 if (is_sec_name_dwarf(name))
3333 /* .BTF and .BTF.ext don't need relocations */
3334 if (strcmp(name, BTF_ELF_SEC) == 0 ||
3335 strcmp(name, BTF_EXT_ELF_SEC) == 0)
3342 static int cmp_progs(const void *_a, const void *_b)
3344 const struct bpf_program *a = _a;
3345 const struct bpf_program *b = _b;
3347 if (a->sec_idx != b->sec_idx)
3348 return a->sec_idx < b->sec_idx ? -1 : 1;
3350 /* sec_insn_off can't be the same within the section */
3351 return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3354 static int bpf_object__elf_collect(struct bpf_object *obj)
3356 struct elf_sec_desc *sec_desc;
3357 Elf *elf = obj->efile.elf;
3358 Elf_Data *btf_ext_data = NULL;
3359 Elf_Data *btf_data = NULL;
3360 int idx = 0, err = 0;
3366 /* ELF section indices are 0-based, but sec #0 is special "invalid"
3367 * section. Since section count retrieved by elf_getshdrnum() does
3368 * include sec #0, it is already the necessary size of an array to keep
3371 if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) {
3372 pr_warn("elf: failed to get the number of sections for %s: %s\n",
3373 obj->path, elf_errmsg(-1));
3374 return -LIBBPF_ERRNO__FORMAT;
3376 obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3377 if (!obj->efile.secs)
3380 /* a bunch of ELF parsing functionality depends on processing symbols,
3381 * so do the first pass and find the symbol table
3384 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3385 sh = elf_sec_hdr(obj, scn);
3387 return -LIBBPF_ERRNO__FORMAT;
3389 if (sh->sh_type == SHT_SYMTAB) {
3390 if (obj->efile.symbols) {
3391 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3392 return -LIBBPF_ERRNO__FORMAT;
3395 data = elf_sec_data(obj, scn);
3397 return -LIBBPF_ERRNO__FORMAT;
3399 idx = elf_ndxscn(scn);
3401 obj->efile.symbols = data;
3402 obj->efile.symbols_shndx = idx;
3403 obj->efile.strtabidx = sh->sh_link;
3407 if (!obj->efile.symbols) {
3408 pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3414 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3415 idx = elf_ndxscn(scn);
3416 sec_desc = &obj->efile.secs[idx];
3418 sh = elf_sec_hdr(obj, scn);
3420 return -LIBBPF_ERRNO__FORMAT;
3422 name = elf_sec_str(obj, sh->sh_name);
3424 return -LIBBPF_ERRNO__FORMAT;
3426 if (ignore_elf_section(sh, name))
3429 data = elf_sec_data(obj, scn);
3431 return -LIBBPF_ERRNO__FORMAT;
3433 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3434 idx, name, (unsigned long)data->d_size,
3435 (int)sh->sh_link, (unsigned long)sh->sh_flags,
3438 if (strcmp(name, "license") == 0) {
3439 err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3442 } else if (strcmp(name, "version") == 0) {
3443 err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3446 } else if (strcmp(name, "maps") == 0) {
3447 pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n");
3449 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3450 obj->efile.btf_maps_shndx = idx;
3451 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3452 if (sh->sh_type != SHT_PROGBITS)
3453 return -LIBBPF_ERRNO__FORMAT;
3455 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3456 if (sh->sh_type != SHT_PROGBITS)
3457 return -LIBBPF_ERRNO__FORMAT;
3458 btf_ext_data = data;
3459 } else if (sh->sh_type == SHT_SYMTAB) {
3460 /* already processed during the first pass above */
3461 } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3462 if (sh->sh_flags & SHF_EXECINSTR) {
3463 if (strcmp(name, ".text") == 0)
3464 obj->efile.text_shndx = idx;
3465 err = bpf_object__add_programs(obj, data, name, idx);
3468 } else if (strcmp(name, DATA_SEC) == 0 ||
3469 str_has_pfx(name, DATA_SEC ".")) {
3470 sec_desc->sec_type = SEC_DATA;
3471 sec_desc->shdr = sh;
3472 sec_desc->data = data;
3473 } else if (strcmp(name, RODATA_SEC) == 0 ||
3474 str_has_pfx(name, RODATA_SEC ".")) {
3475 sec_desc->sec_type = SEC_RODATA;
3476 sec_desc->shdr = sh;
3477 sec_desc->data = data;
3478 } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
3479 obj->efile.st_ops_data = data;
3480 obj->efile.st_ops_shndx = idx;
3481 } else if (strcmp(name, STRUCT_OPS_LINK_SEC) == 0) {
3482 obj->efile.st_ops_link_data = data;
3483 obj->efile.st_ops_link_shndx = idx;
3485 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3488 } else if (sh->sh_type == SHT_REL) {
3489 int targ_sec_idx = sh->sh_info; /* points to other section */
3491 if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3492 targ_sec_idx >= obj->efile.sec_cnt)
3493 return -LIBBPF_ERRNO__FORMAT;
3495 /* Only do relo for section with exec instructions */
3496 if (!section_have_execinstr(obj, targ_sec_idx) &&
3497 strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3498 strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) &&
3499 strcmp(name, ".rel" MAPS_ELF_SEC)) {
3500 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3501 idx, name, targ_sec_idx,
3502 elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
3506 sec_desc->sec_type = SEC_RELO;
3507 sec_desc->shdr = sh;
3508 sec_desc->data = data;
3509 } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 ||
3510 str_has_pfx(name, BSS_SEC "."))) {
3511 sec_desc->sec_type = SEC_BSS;
3512 sec_desc->shdr = sh;
3513 sec_desc->data = data;
3515 pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3516 (size_t)sh->sh_size);
3520 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3521 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3522 return -LIBBPF_ERRNO__FORMAT;
3525 /* sort BPF programs by section name and in-section instruction offset
3528 if (obj->nr_programs)
3529 qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
3531 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
3534 static bool sym_is_extern(const Elf64_Sym *sym)
3536 int bind = ELF64_ST_BIND(sym->st_info);
3537 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
3538 return sym->st_shndx == SHN_UNDEF &&
3539 (bind == STB_GLOBAL || bind == STB_WEAK) &&
3540 ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
3543 static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
3545 int bind = ELF64_ST_BIND(sym->st_info);
3546 int type = ELF64_ST_TYPE(sym->st_info);
3548 /* in .text section */
3549 if (sym->st_shndx != text_shndx)
3552 /* local function */
3553 if (bind == STB_LOCAL && type == STT_SECTION)
3556 /* global function */
3557 return bind == STB_GLOBAL && type == STT_FUNC;
3560 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
3562 const struct btf_type *t;
3569 n = btf__type_cnt(btf);
3570 for (i = 1; i < n; i++) {
3571 t = btf__type_by_id(btf, i);
3573 if (!btf_is_var(t) && !btf_is_func(t))
3576 tname = btf__name_by_offset(btf, t->name_off);
3577 if (strcmp(tname, ext_name))
3580 if (btf_is_var(t) &&
3581 btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
3584 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
3593 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3594 const struct btf_var_secinfo *vs;
3595 const struct btf_type *t;
3601 n = btf__type_cnt(btf);
3602 for (i = 1; i < n; i++) {
3603 t = btf__type_by_id(btf, i);
3605 if (!btf_is_datasec(t))
3608 vs = btf_var_secinfos(t);
3609 for (j = 0; j < btf_vlen(t); j++, vs++) {
3610 if (vs->type == ext_btf_id)
3618 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3621 const struct btf_type *t;
3624 t = skip_mods_and_typedefs(btf, id, NULL);
3625 name = btf__name_by_offset(btf, t->name_off);
3629 switch (btf_kind(t)) {
3630 case BTF_KIND_INT: {
3631 int enc = btf_int_encoding(t);
3633 if (enc & BTF_INT_BOOL)
3634 return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3636 *is_signed = enc & BTF_INT_SIGNED;
3639 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3640 return KCFG_UNKNOWN;
3645 return KCFG_UNKNOWN;
3646 if (strcmp(name, "libbpf_tristate"))
3647 return KCFG_UNKNOWN;
3648 return KCFG_TRISTATE;
3649 case BTF_KIND_ENUM64:
3650 if (strcmp(name, "libbpf_tristate"))
3651 return KCFG_UNKNOWN;
3652 return KCFG_TRISTATE;
3653 case BTF_KIND_ARRAY:
3654 if (btf_array(t)->nelems == 0)
3655 return KCFG_UNKNOWN;
3656 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3657 return KCFG_UNKNOWN;
3658 return KCFG_CHAR_ARR;
3660 return KCFG_UNKNOWN;
3664 static int cmp_externs(const void *_a, const void *_b)
3666 const struct extern_desc *a = _a;
3667 const struct extern_desc *b = _b;
3669 if (a->type != b->type)
3670 return a->type < b->type ? -1 : 1;
3672 if (a->type == EXT_KCFG) {
3673 /* descending order by alignment requirements */
3674 if (a->kcfg.align != b->kcfg.align)
3675 return a->kcfg.align > b->kcfg.align ? -1 : 1;
3676 /* ascending order by size, within same alignment class */
3677 if (a->kcfg.sz != b->kcfg.sz)
3678 return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3681 /* resolve ties by name */
3682 return strcmp(a->name, b->name);
3685 static int find_int_btf_id(const struct btf *btf)
3687 const struct btf_type *t;
3690 n = btf__type_cnt(btf);
3691 for (i = 1; i < n; i++) {
3692 t = btf__type_by_id(btf, i);
3694 if (btf_is_int(t) && btf_int_bits(t) == 32)
3701 static int add_dummy_ksym_var(struct btf *btf)
3703 int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
3704 const struct btf_var_secinfo *vs;
3705 const struct btf_type *sec;
3710 sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
3715 sec = btf__type_by_id(btf, sec_btf_id);
3716 vs = btf_var_secinfos(sec);
3717 for (i = 0; i < btf_vlen(sec); i++, vs++) {
3718 const struct btf_type *vt;
3720 vt = btf__type_by_id(btf, vs->type);
3721 if (btf_is_func(vt))
3725 /* No func in ksyms sec. No need to add dummy var. */
3726 if (i == btf_vlen(sec))
3729 int_btf_id = find_int_btf_id(btf);
3730 dummy_var_btf_id = btf__add_var(btf,
3732 BTF_VAR_GLOBAL_ALLOCATED,
3734 if (dummy_var_btf_id < 0)
3735 pr_warn("cannot create a dummy_ksym var\n");
3737 return dummy_var_btf_id;
3740 static int bpf_object__collect_externs(struct bpf_object *obj)
3742 struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
3743 const struct btf_type *t;
3744 struct extern_desc *ext;
3745 int i, n, off, dummy_var_btf_id;
3746 const char *ext_name, *sec_name;
3750 if (!obj->efile.symbols)
3753 scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
3754 sh = elf_sec_hdr(obj, scn);
3755 if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
3756 return -LIBBPF_ERRNO__FORMAT;
3758 dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
3759 if (dummy_var_btf_id < 0)
3760 return dummy_var_btf_id;
3762 n = sh->sh_size / sh->sh_entsize;
3763 pr_debug("looking for externs among %d symbols...\n", n);
3765 for (i = 0; i < n; i++) {
3766 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
3769 return -LIBBPF_ERRNO__FORMAT;
3770 if (!sym_is_extern(sym))
3772 ext_name = elf_sym_str(obj, sym->st_name);
3773 if (!ext_name || !ext_name[0])
3777 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
3781 ext = &ext[obj->nr_extern];
3782 memset(ext, 0, sizeof(*ext));
3785 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
3786 if (ext->btf_id <= 0) {
3787 pr_warn("failed to find BTF for extern '%s': %d\n",
3788 ext_name, ext->btf_id);
3791 t = btf__type_by_id(obj->btf, ext->btf_id);
3792 ext->name = btf__name_by_offset(obj->btf, t->name_off);
3794 ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
3796 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
3797 if (ext->sec_btf_id <= 0) {
3798 pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
3799 ext_name, ext->btf_id, ext->sec_btf_id);
3800 return ext->sec_btf_id;
3802 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
3803 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
3805 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
3806 if (btf_is_func(t)) {
3807 pr_warn("extern function %s is unsupported under %s section\n",
3808 ext->name, KCONFIG_SEC);
3812 ext->type = EXT_KCFG;
3813 ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
3814 if (ext->kcfg.sz <= 0) {
3815 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
3816 ext_name, ext->kcfg.sz);
3817 return ext->kcfg.sz;
3819 ext->kcfg.align = btf__align_of(obj->btf, t->type);
3820 if (ext->kcfg.align <= 0) {
3821 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
3822 ext_name, ext->kcfg.align);
3825 ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
3826 &ext->kcfg.is_signed);
3827 if (ext->kcfg.type == KCFG_UNKNOWN) {
3828 pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name);
3831 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
3833 ext->type = EXT_KSYM;
3834 skip_mods_and_typedefs(obj->btf, t->type,
3835 &ext->ksym.type_id);
3837 pr_warn("unrecognized extern section '%s'\n", sec_name);
3841 pr_debug("collected %d externs total\n", obj->nr_extern);
3843 if (!obj->nr_extern)
3846 /* sort externs by type, for kcfg ones also by (align, size, name) */
3847 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
3849 /* for .ksyms section, we need to turn all externs into allocated
3850 * variables in BTF to pass kernel verification; we do this by
3851 * pretending that each extern is a 8-byte variable
3854 /* find existing 4-byte integer type in BTF to use for fake
3855 * extern variables in DATASEC
3857 int int_btf_id = find_int_btf_id(obj->btf);
3858 /* For extern function, a dummy_var added earlier
3859 * will be used to replace the vs->type and
3860 * its name string will be used to refill
3861 * the missing param's name.
3863 const struct btf_type *dummy_var;
3865 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
3866 for (i = 0; i < obj->nr_extern; i++) {
3867 ext = &obj->externs[i];
3868 if (ext->type != EXT_KSYM)
3870 pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
3871 i, ext->sym_idx, ext->name);
3876 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
3877 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3878 struct btf_type *vt;
3880 vt = (void *)btf__type_by_id(obj->btf, vs->type);
3881 ext_name = btf__name_by_offset(obj->btf, vt->name_off);
3882 ext = find_extern_by_name(obj, ext_name);
3884 pr_warn("failed to find extern definition for BTF %s '%s'\n",
3885 btf_kind_str(vt), ext_name);
3888 if (btf_is_func(vt)) {
3889 const struct btf_type *func_proto;
3890 struct btf_param *param;
3893 func_proto = btf__type_by_id(obj->btf,
3895 param = btf_params(func_proto);
3896 /* Reuse the dummy_var string if the
3897 * func proto does not have param name.
3899 for (j = 0; j < btf_vlen(func_proto); j++)
3900 if (param[j].type && !param[j].name_off)
3902 dummy_var->name_off;
3903 vs->type = dummy_var_btf_id;
3904 vt->info &= ~0xffff;
3905 vt->info |= BTF_FUNC_GLOBAL;
3907 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3908 vt->type = int_btf_id;
3911 vs->size = sizeof(int);
3918 /* for kcfg externs calculate their offsets within a .kconfig map */
3920 for (i = 0; i < obj->nr_extern; i++) {
3921 ext = &obj->externs[i];
3922 if (ext->type != EXT_KCFG)
3925 ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3926 off = ext->kcfg.data_off + ext->kcfg.sz;
3927 pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3928 i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3932 for (i = 0; i < n; i++) {
3933 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3935 t = btf__type_by_id(obj->btf, vs->type);
3936 ext_name = btf__name_by_offset(obj->btf, t->name_off);
3937 ext = find_extern_by_name(obj, ext_name);
3939 pr_warn("failed to find extern definition for BTF var '%s'\n",
3943 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3944 vs->offset = ext->kcfg.data_off;
3950 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog)
3952 return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
3955 struct bpf_program *
3956 bpf_object__find_program_by_name(const struct bpf_object *obj,
3959 struct bpf_program *prog;
3961 bpf_object__for_each_program(prog, obj) {
3962 if (prog_is_subprog(obj, prog))
3964 if (!strcmp(prog->name, name))
3967 return errno = ENOENT, NULL;
3970 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3973 switch (obj->efile.secs[shndx].sec_type) {
3983 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3986 return shndx == obj->efile.btf_maps_shndx;
3989 static enum libbpf_map_type
3990 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3992 if (shndx == obj->efile.symbols_shndx)
3993 return LIBBPF_MAP_KCONFIG;
3995 switch (obj->efile.secs[shndx].sec_type) {
3997 return LIBBPF_MAP_BSS;
3999 return LIBBPF_MAP_DATA;
4001 return LIBBPF_MAP_RODATA;
4003 return LIBBPF_MAP_UNSPEC;
4007 static int bpf_program__record_reloc(struct bpf_program *prog,
4008 struct reloc_desc *reloc_desc,
4009 __u32 insn_idx, const char *sym_name,
4010 const Elf64_Sym *sym, const Elf64_Rel *rel)
4012 struct bpf_insn *insn = &prog->insns[insn_idx];
4013 size_t map_idx, nr_maps = prog->obj->nr_maps;
4014 struct bpf_object *obj = prog->obj;
4015 __u32 shdr_idx = sym->st_shndx;
4016 enum libbpf_map_type type;
4017 const char *sym_sec_name;
4018 struct bpf_map *map;
4020 if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
4021 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
4022 prog->name, sym_name, insn_idx, insn->code);
4023 return -LIBBPF_ERRNO__RELOC;
4026 if (sym_is_extern(sym)) {
4027 int sym_idx = ELF64_R_SYM(rel->r_info);
4028 int i, n = obj->nr_extern;
4029 struct extern_desc *ext;
4031 for (i = 0; i < n; i++) {
4032 ext = &obj->externs[i];
4033 if (ext->sym_idx == sym_idx)
4037 pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
4038 prog->name, sym_name, sym_idx);
4039 return -LIBBPF_ERRNO__RELOC;
4041 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
4042 prog->name, i, ext->name, ext->sym_idx, insn_idx);
4043 if (insn->code == (BPF_JMP | BPF_CALL))
4044 reloc_desc->type = RELO_EXTERN_CALL;
4046 reloc_desc->type = RELO_EXTERN_LD64;
4047 reloc_desc->insn_idx = insn_idx;
4048 reloc_desc->ext_idx = i;
4052 /* sub-program call relocation */
4053 if (is_call_insn(insn)) {
4054 if (insn->src_reg != BPF_PSEUDO_CALL) {
4055 pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
4056 return -LIBBPF_ERRNO__RELOC;
4058 /* text_shndx can be 0, if no default "main" program exists */
4059 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
4060 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4061 pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
4062 prog->name, sym_name, sym_sec_name);
4063 return -LIBBPF_ERRNO__RELOC;
4065 if (sym->st_value % BPF_INSN_SZ) {
4066 pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
4067 prog->name, sym_name, (size_t)sym->st_value);
4068 return -LIBBPF_ERRNO__RELOC;
4070 reloc_desc->type = RELO_CALL;
4071 reloc_desc->insn_idx = insn_idx;
4072 reloc_desc->sym_off = sym->st_value;
4076 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
4077 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
4078 prog->name, sym_name, shdr_idx);
4079 return -LIBBPF_ERRNO__RELOC;
4082 /* loading subprog addresses */
4083 if (sym_is_subprog(sym, obj->efile.text_shndx)) {
4084 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
4085 * local_func: sym->st_value = 0, insn->imm = offset in the section.
4087 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
4088 pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
4089 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
4090 return -LIBBPF_ERRNO__RELOC;
4093 reloc_desc->type = RELO_SUBPROG_ADDR;
4094 reloc_desc->insn_idx = insn_idx;
4095 reloc_desc->sym_off = sym->st_value;
4099 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
4100 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4102 /* generic map reference relocation */
4103 if (type == LIBBPF_MAP_UNSPEC) {
4104 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
4105 pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
4106 prog->name, sym_name, sym_sec_name);
4107 return -LIBBPF_ERRNO__RELOC;
4109 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4110 map = &obj->maps[map_idx];
4111 if (map->libbpf_type != type ||
4112 map->sec_idx != sym->st_shndx ||
4113 map->sec_offset != sym->st_value)
4115 pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4116 prog->name, map_idx, map->name, map->sec_idx,
4117 map->sec_offset, insn_idx);
4120 if (map_idx >= nr_maps) {
4121 pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4122 prog->name, sym_sec_name, (size_t)sym->st_value);
4123 return -LIBBPF_ERRNO__RELOC;
4125 reloc_desc->type = RELO_LD64;
4126 reloc_desc->insn_idx = insn_idx;
4127 reloc_desc->map_idx = map_idx;
4128 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4132 /* global data map relocation */
4133 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4134 pr_warn("prog '%s': bad data relo against section '%s'\n",
4135 prog->name, sym_sec_name);
4136 return -LIBBPF_ERRNO__RELOC;
4138 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4139 map = &obj->maps[map_idx];
4140 if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4142 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4143 prog->name, map_idx, map->name, map->sec_idx,
4144 map->sec_offset, insn_idx);
4147 if (map_idx >= nr_maps) {
4148 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4149 prog->name, sym_sec_name);
4150 return -LIBBPF_ERRNO__RELOC;
4153 reloc_desc->type = RELO_DATA;
4154 reloc_desc->insn_idx = insn_idx;
4155 reloc_desc->map_idx = map_idx;
4156 reloc_desc->sym_off = sym->st_value;
4160 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4162 return insn_idx >= prog->sec_insn_off &&
4163 insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4166 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4167 size_t sec_idx, size_t insn_idx)
4169 int l = 0, r = obj->nr_programs - 1, m;
4170 struct bpf_program *prog;
4172 if (!obj->nr_programs)
4176 m = l + (r - l + 1) / 2;
4177 prog = &obj->programs[m];
4179 if (prog->sec_idx < sec_idx ||
4180 (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4185 /* matching program could be at index l, but it still might be the
4186 * wrong one, so we need to double check conditions for the last time
4188 prog = &obj->programs[l];
4189 if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4195 bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4197 const char *relo_sec_name, *sec_name;
4198 size_t sec_idx = shdr->sh_info, sym_idx;
4199 struct bpf_program *prog;
4200 struct reloc_desc *relos;
4202 const char *sym_name;
4209 if (sec_idx >= obj->efile.sec_cnt)
4212 scn = elf_sec_by_idx(obj, sec_idx);
4213 scn_data = elf_sec_data(obj, scn);
4215 relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4216 sec_name = elf_sec_name(obj, scn);
4217 if (!relo_sec_name || !sec_name)
4220 pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4221 relo_sec_name, sec_idx, sec_name);
4222 nrels = shdr->sh_size / shdr->sh_entsize;
4224 for (i = 0; i < nrels; i++) {
4225 rel = elf_rel_by_idx(data, i);
4227 pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4228 return -LIBBPF_ERRNO__FORMAT;
4231 sym_idx = ELF64_R_SYM(rel->r_info);
4232 sym = elf_sym_by_idx(obj, sym_idx);
4234 pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4235 relo_sec_name, sym_idx, i);
4236 return -LIBBPF_ERRNO__FORMAT;
4239 if (sym->st_shndx >= obj->efile.sec_cnt) {
4240 pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4241 relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4242 return -LIBBPF_ERRNO__FORMAT;
4245 if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4246 pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4247 relo_sec_name, (size_t)rel->r_offset, i);
4248 return -LIBBPF_ERRNO__FORMAT;
4251 insn_idx = rel->r_offset / BPF_INSN_SZ;
4252 /* relocations against static functions are recorded as
4253 * relocations against the section that contains a function;
4254 * in such case, symbol will be STT_SECTION and sym.st_name
4255 * will point to empty string (0), so fetch section name
4258 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4259 sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4261 sym_name = elf_sym_str(obj, sym->st_name);
4262 sym_name = sym_name ?: "<?";
4264 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4265 relo_sec_name, i, insn_idx, sym_name);
4267 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4269 pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4270 relo_sec_name, i, sec_name, insn_idx);
4274 relos = libbpf_reallocarray(prog->reloc_desc,
4275 prog->nr_reloc + 1, sizeof(*relos));
4278 prog->reloc_desc = relos;
4280 /* adjust insn_idx to local BPF program frame of reference */
4281 insn_idx -= prog->sec_insn_off;
4282 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4283 insn_idx, sym_name, sym, rel);
4292 static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map)
4299 /* if it's BTF-defined map, we don't need to search for type IDs.
4300 * For struct_ops map, it does not need btf_key_type_id and
4301 * btf_value_type_id.
4303 if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map))
4307 * LLVM annotates global data differently in BTF, that is,
4308 * only as '.data', '.bss' or '.rodata'.
4310 if (!bpf_map__is_internal(map))
4313 id = btf__find_by_name(obj->btf, map->real_name);
4317 map->btf_key_type_id = 0;
4318 map->btf_value_type_id = id;
4322 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4324 char file[PATH_MAX], buff[4096];
4329 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4330 memset(info, 0, sizeof(*info));
4332 fp = fopen(file, "r");
4335 pr_warn("failed to open %s: %d. No procfs support?\n", file,
4340 while (fgets(buff, sizeof(buff), fp)) {
4341 if (sscanf(buff, "map_type:\t%u", &val) == 1)
4343 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4344 info->key_size = val;
4345 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4346 info->value_size = val;
4347 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4348 info->max_entries = val;
4349 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4350 info->map_flags = val;
4358 bool bpf_map__autocreate(const struct bpf_map *map)
4360 return map->autocreate;
4363 int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate)
4365 if (map->obj->loaded)
4366 return libbpf_err(-EBUSY);
4368 map->autocreate = autocreate;
4372 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4374 struct bpf_map_info info;
4375 __u32 len = sizeof(info), name_len;
4379 memset(&info, 0, len);
4380 err = bpf_map_get_info_by_fd(fd, &info, &len);
4381 if (err && errno == EINVAL)
4382 err = bpf_get_map_info_from_fdinfo(fd, &info);
4384 return libbpf_err(err);
4386 name_len = strlen(info.name);
4387 if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0)
4388 new_name = strdup(map->name);
4390 new_name = strdup(info.name);
4393 return libbpf_err(-errno);
4395 new_fd = open("/", O_RDONLY | O_CLOEXEC);
4398 goto err_free_new_name;
4401 new_fd = dup3(fd, new_fd, O_CLOEXEC);
4404 goto err_close_new_fd;
4407 err = zclose(map->fd);
4410 goto err_close_new_fd;
4415 map->name = new_name;
4416 map->def.type = info.type;
4417 map->def.key_size = info.key_size;
4418 map->def.value_size = info.value_size;
4419 map->def.max_entries = info.max_entries;
4420 map->def.map_flags = info.map_flags;
4421 map->btf_key_type_id = info.btf_key_type_id;
4422 map->btf_value_type_id = info.btf_value_type_id;
4424 map->map_extra = info.map_extra;
4432 return libbpf_err(err);
4435 __u32 bpf_map__max_entries(const struct bpf_map *map)
4437 return map->def.max_entries;
4440 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
4442 if (!bpf_map_type__is_map_in_map(map->def.type))
4443 return errno = EINVAL, NULL;
4445 return map->inner_map;
4448 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4450 if (map->obj->loaded)
4451 return libbpf_err(-EBUSY);
4453 map->def.max_entries = max_entries;
4455 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
4456 if (map_is_ringbuf(map))
4457 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
4463 bpf_object__probe_loading(struct bpf_object *obj)
4465 char *cp, errmsg[STRERR_BUFSIZE];
4466 struct bpf_insn insns[] = {
4467 BPF_MOV64_IMM(BPF_REG_0, 0),
4470 int ret, insn_cnt = ARRAY_SIZE(insns);
4472 if (obj->gen_loader)
4475 ret = bump_rlimit_memlock();
4477 pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret);
4479 /* make sure basic loading works */
4480 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4482 ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4485 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4486 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
4487 "program. Make sure your kernel supports BPF "
4488 "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
4489 "set to big enough value.\n", __func__, cp, ret);
4497 static int probe_fd(int fd)
4504 static int probe_kern_prog_name(void)
4506 const size_t attr_sz = offsetofend(union bpf_attr, prog_name);
4507 struct bpf_insn insns[] = {
4508 BPF_MOV64_IMM(BPF_REG_0, 0),
4511 union bpf_attr attr;
4514 memset(&attr, 0, attr_sz);
4515 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4516 attr.license = ptr_to_u64("GPL");
4517 attr.insns = ptr_to_u64(insns);
4518 attr.insn_cnt = (__u32)ARRAY_SIZE(insns);
4519 libbpf_strlcpy(attr.prog_name, "libbpf_nametest", sizeof(attr.prog_name));
4521 /* make sure loading with name works */
4522 ret = sys_bpf_prog_load(&attr, attr_sz, PROG_LOAD_ATTEMPTS);
4523 return probe_fd(ret);
4526 static int probe_kern_global_data(void)
4528 char *cp, errmsg[STRERR_BUFSIZE];
4529 struct bpf_insn insns[] = {
4530 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
4531 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
4532 BPF_MOV64_IMM(BPF_REG_0, 0),
4535 int ret, map, insn_cnt = ARRAY_SIZE(insns);
4537 map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_global", sizeof(int), 32, 1, NULL);
4540 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4541 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4542 __func__, cp, -ret);
4548 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4550 return probe_fd(ret);
4553 static int probe_kern_btf(void)
4555 static const char strs[] = "\0int";
4558 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4561 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4562 strs, sizeof(strs)));
4565 static int probe_kern_btf_func(void)
4567 static const char strs[] = "\0int\0x\0a";
4568 /* void x(int a) {} */
4571 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4572 /* FUNC_PROTO */ /* [2] */
4573 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4574 BTF_PARAM_ENC(7, 1),
4575 /* FUNC x */ /* [3] */
4576 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
4579 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4580 strs, sizeof(strs)));
4583 static int probe_kern_btf_func_global(void)
4585 static const char strs[] = "\0int\0x\0a";
4586 /* static void x(int a) {} */
4589 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4590 /* FUNC_PROTO */ /* [2] */
4591 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4592 BTF_PARAM_ENC(7, 1),
4593 /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */
4594 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
4597 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4598 strs, sizeof(strs)));
4601 static int probe_kern_btf_datasec(void)
4603 static const char strs[] = "\0x\0.data";
4607 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4608 /* VAR x */ /* [2] */
4609 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4611 /* DATASEC val */ /* [3] */
4612 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
4613 BTF_VAR_SECINFO_ENC(2, 0, 4),
4616 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4617 strs, sizeof(strs)));
4620 static int probe_kern_btf_float(void)
4622 static const char strs[] = "\0float";
4625 BTF_TYPE_FLOAT_ENC(1, 4),
4628 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4629 strs, sizeof(strs)));
4632 static int probe_kern_btf_decl_tag(void)
4634 static const char strs[] = "\0tag";
4637 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4638 /* VAR x */ /* [2] */
4639 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4642 BTF_TYPE_DECL_TAG_ENC(1, 2, -1),
4645 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4646 strs, sizeof(strs)));
4649 static int probe_kern_btf_type_tag(void)
4651 static const char strs[] = "\0tag";
4654 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4656 BTF_TYPE_TYPE_TAG_ENC(1, 1), /* [2] */
4658 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2), /* [3] */
4661 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4662 strs, sizeof(strs)));
4665 static int probe_kern_array_mmap(void)
4667 LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
4670 fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_mmap", sizeof(int), sizeof(int), 1, &opts);
4671 return probe_fd(fd);
4674 static int probe_kern_exp_attach_type(void)
4676 LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE);
4677 struct bpf_insn insns[] = {
4678 BPF_MOV64_IMM(BPF_REG_0, 0),
4681 int fd, insn_cnt = ARRAY_SIZE(insns);
4683 /* use any valid combination of program type and (optional)
4684 * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
4685 * to see if kernel supports expected_attach_type field for
4686 * BPF_PROG_LOAD command
4688 fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts);
4689 return probe_fd(fd);
4692 static int probe_kern_probe_read_kernel(void)
4694 struct bpf_insn insns[] = {
4695 BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */
4696 BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */
4697 BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */
4698 BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */
4699 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
4702 int fd, insn_cnt = ARRAY_SIZE(insns);
4704 fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4705 return probe_fd(fd);
4708 static int probe_prog_bind_map(void)
4710 char *cp, errmsg[STRERR_BUFSIZE];
4711 struct bpf_insn insns[] = {
4712 BPF_MOV64_IMM(BPF_REG_0, 0),
4715 int ret, map, prog, insn_cnt = ARRAY_SIZE(insns);
4717 map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_det_bind", sizeof(int), 32, 1, NULL);
4720 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4721 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4722 __func__, cp, -ret);
4726 prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4732 ret = bpf_prog_bind_map(prog, map, NULL);
4740 static int probe_module_btf(void)
4742 static const char strs[] = "\0int";
4745 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4747 struct bpf_btf_info info;
4748 __u32 len = sizeof(info);
4752 fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
4754 return 0; /* BTF not supported at all */
4756 memset(&info, 0, sizeof(info));
4757 info.name = ptr_to_u64(name);
4758 info.name_len = sizeof(name);
4760 /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
4761 * kernel's module BTF support coincides with support for
4762 * name/name_len fields in struct bpf_btf_info.
4764 err = bpf_btf_get_info_by_fd(fd, &info, &len);
4769 static int probe_perf_link(void)
4771 struct bpf_insn insns[] = {
4772 BPF_MOV64_IMM(BPF_REG_0, 0),
4775 int prog_fd, link_fd, err;
4777 prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL",
4778 insns, ARRAY_SIZE(insns), NULL);
4782 /* use invalid perf_event FD to get EBADF, if link is supported;
4783 * otherwise EINVAL should be returned
4785 link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL);
4786 err = -errno; /* close() can clobber errno */
4792 return link_fd < 0 && err == -EBADF;
4795 static int probe_kern_bpf_cookie(void)
4797 struct bpf_insn insns[] = {
4798 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_attach_cookie),
4801 int ret, insn_cnt = ARRAY_SIZE(insns);
4803 ret = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL);
4804 return probe_fd(ret);
4807 static int probe_kern_btf_enum64(void)
4809 static const char strs[] = "\0enum64";
4811 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_ENUM64, 0, 0), 8),
4814 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4815 strs, sizeof(strs)));
4818 static int probe_kern_syscall_wrapper(void);
4820 enum kern_feature_result {
4826 typedef int (*feature_probe_fn)(void);
4828 static struct kern_feature_desc {
4830 feature_probe_fn probe;
4831 enum kern_feature_result res;
4832 } feature_probes[__FEAT_CNT] = {
4833 [FEAT_PROG_NAME] = {
4834 "BPF program name", probe_kern_prog_name,
4836 [FEAT_GLOBAL_DATA] = {
4837 "global variables", probe_kern_global_data,
4840 "minimal BTF", probe_kern_btf,
4843 "BTF functions", probe_kern_btf_func,
4845 [FEAT_BTF_GLOBAL_FUNC] = {
4846 "BTF global function", probe_kern_btf_func_global,
4848 [FEAT_BTF_DATASEC] = {
4849 "BTF data section and variable", probe_kern_btf_datasec,
4851 [FEAT_ARRAY_MMAP] = {
4852 "ARRAY map mmap()", probe_kern_array_mmap,
4854 [FEAT_EXP_ATTACH_TYPE] = {
4855 "BPF_PROG_LOAD expected_attach_type attribute",
4856 probe_kern_exp_attach_type,
4858 [FEAT_PROBE_READ_KERN] = {
4859 "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
4861 [FEAT_PROG_BIND_MAP] = {
4862 "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
4864 [FEAT_MODULE_BTF] = {
4865 "module BTF support", probe_module_btf,
4867 [FEAT_BTF_FLOAT] = {
4868 "BTF_KIND_FLOAT support", probe_kern_btf_float,
4870 [FEAT_PERF_LINK] = {
4871 "BPF perf link support", probe_perf_link,
4873 [FEAT_BTF_DECL_TAG] = {
4874 "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag,
4876 [FEAT_BTF_TYPE_TAG] = {
4877 "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag,
4879 [FEAT_MEMCG_ACCOUNT] = {
4880 "memcg-based memory accounting", probe_memcg_account,
4882 [FEAT_BPF_COOKIE] = {
4883 "BPF cookie support", probe_kern_bpf_cookie,
4885 [FEAT_BTF_ENUM64] = {
4886 "BTF_KIND_ENUM64 support", probe_kern_btf_enum64,
4888 [FEAT_SYSCALL_WRAPPER] = {
4889 "Kernel using syscall wrapper", probe_kern_syscall_wrapper,
4893 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
4895 struct kern_feature_desc *feat = &feature_probes[feat_id];
4898 if (obj && obj->gen_loader)
4899 /* To generate loader program assume the latest kernel
4900 * to avoid doing extra prog_load, map_create syscalls.
4904 if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
4905 ret = feat->probe();
4907 WRITE_ONCE(feat->res, FEAT_SUPPORTED);
4908 } else if (ret == 0) {
4909 WRITE_ONCE(feat->res, FEAT_MISSING);
4911 pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
4912 WRITE_ONCE(feat->res, FEAT_MISSING);
4916 return READ_ONCE(feat->res) == FEAT_SUPPORTED;
4919 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
4921 struct bpf_map_info map_info;
4922 char msg[STRERR_BUFSIZE];
4923 __u32 map_info_len = sizeof(map_info);
4926 memset(&map_info, 0, map_info_len);
4927 err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len);
4928 if (err && errno == EINVAL)
4929 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
4931 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
4932 libbpf_strerror_r(errno, msg, sizeof(msg)));
4936 return (map_info.type == map->def.type &&
4937 map_info.key_size == map->def.key_size &&
4938 map_info.value_size == map->def.value_size &&
4939 map_info.max_entries == map->def.max_entries &&
4940 map_info.map_flags == map->def.map_flags &&
4941 map_info.map_extra == map->map_extra);
4945 bpf_object__reuse_map(struct bpf_map *map)
4947 char *cp, errmsg[STRERR_BUFSIZE];
4950 pin_fd = bpf_obj_get(map->pin_path);
4953 if (err == -ENOENT) {
4954 pr_debug("found no pinned map to reuse at '%s'\n",
4959 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4960 pr_warn("couldn't retrieve pinned map '%s': %s\n",
4965 if (!map_is_reuse_compat(map, pin_fd)) {
4966 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
4972 err = bpf_map__reuse_fd(map, pin_fd);
4978 pr_debug("reused pinned map at '%s'\n", map->pin_path);
4984 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
4986 enum libbpf_map_type map_type = map->libbpf_type;
4987 char *cp, errmsg[STRERR_BUFSIZE];
4990 if (obj->gen_loader) {
4991 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
4992 map->mmaped, map->def.value_size);
4993 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
4994 bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
4997 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
5000 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5001 pr_warn("Error setting initial map(%s) contents: %s\n",
5006 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
5007 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
5008 err = bpf_map_freeze(map->fd);
5011 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5012 pr_warn("Error freezing map(%s) as read-only: %s\n",
5020 static void bpf_map__destroy(struct bpf_map *map);
5022 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
5024 LIBBPF_OPTS(bpf_map_create_opts, create_attr);
5025 struct bpf_map_def *def = &map->def;
5026 const char *map_name = NULL;
5029 if (kernel_supports(obj, FEAT_PROG_NAME))
5030 map_name = map->name;
5031 create_attr.map_ifindex = map->map_ifindex;
5032 create_attr.map_flags = def->map_flags;
5033 create_attr.numa_node = map->numa_node;
5034 create_attr.map_extra = map->map_extra;
5036 if (bpf_map__is_struct_ops(map))
5037 create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
5039 if (obj->btf && btf__fd(obj->btf) >= 0) {
5040 create_attr.btf_fd = btf__fd(obj->btf);
5041 create_attr.btf_key_type_id = map->btf_key_type_id;
5042 create_attr.btf_value_type_id = map->btf_value_type_id;
5045 if (bpf_map_type__is_map_in_map(def->type)) {
5046 if (map->inner_map) {
5047 err = bpf_object__create_map(obj, map->inner_map, true);
5049 pr_warn("map '%s': failed to create inner map: %d\n",
5053 map->inner_map_fd = bpf_map__fd(map->inner_map);
5055 if (map->inner_map_fd >= 0)
5056 create_attr.inner_map_fd = map->inner_map_fd;
5059 switch (def->type) {
5060 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
5061 case BPF_MAP_TYPE_CGROUP_ARRAY:
5062 case BPF_MAP_TYPE_STACK_TRACE:
5063 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
5064 case BPF_MAP_TYPE_HASH_OF_MAPS:
5065 case BPF_MAP_TYPE_DEVMAP:
5066 case BPF_MAP_TYPE_DEVMAP_HASH:
5067 case BPF_MAP_TYPE_CPUMAP:
5068 case BPF_MAP_TYPE_XSKMAP:
5069 case BPF_MAP_TYPE_SOCKMAP:
5070 case BPF_MAP_TYPE_SOCKHASH:
5071 case BPF_MAP_TYPE_QUEUE:
5072 case BPF_MAP_TYPE_STACK:
5073 create_attr.btf_fd = 0;
5074 create_attr.btf_key_type_id = 0;
5075 create_attr.btf_value_type_id = 0;
5076 map->btf_key_type_id = 0;
5077 map->btf_value_type_id = 0;
5082 if (obj->gen_loader) {
5083 bpf_gen__map_create(obj->gen_loader, def->type, map_name,
5084 def->key_size, def->value_size, def->max_entries,
5085 &create_attr, is_inner ? -1 : map - obj->maps);
5086 /* Pretend to have valid FD to pass various fd >= 0 checks.
5087 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
5091 map->fd = bpf_map_create(def->type, map_name,
5092 def->key_size, def->value_size,
5093 def->max_entries, &create_attr);
5095 if (map->fd < 0 && (create_attr.btf_key_type_id ||
5096 create_attr.btf_value_type_id)) {
5097 char *cp, errmsg[STRERR_BUFSIZE];
5100 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5101 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
5102 map->name, cp, err);
5103 create_attr.btf_fd = 0;
5104 create_attr.btf_key_type_id = 0;
5105 create_attr.btf_value_type_id = 0;
5106 map->btf_key_type_id = 0;
5107 map->btf_value_type_id = 0;
5108 map->fd = bpf_map_create(def->type, map_name,
5109 def->key_size, def->value_size,
5110 def->max_entries, &create_attr);
5113 err = map->fd < 0 ? -errno : 0;
5115 if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
5116 if (obj->gen_loader)
5117 map->inner_map->fd = -1;
5118 bpf_map__destroy(map->inner_map);
5119 zfree(&map->inner_map);
5125 static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
5127 const struct bpf_map *targ_map;
5131 for (i = 0; i < map->init_slots_sz; i++) {
5132 if (!map->init_slots[i])
5135 targ_map = map->init_slots[i];
5136 fd = bpf_map__fd(targ_map);
5138 if (obj->gen_loader) {
5139 bpf_gen__populate_outer_map(obj->gen_loader,
5141 targ_map - obj->maps);
5143 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5147 pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
5148 map->name, i, targ_map->name, fd, err);
5151 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5152 map->name, i, targ_map->name, fd);
5155 zfree(&map->init_slots);
5156 map->init_slots_sz = 0;
5161 static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5163 const struct bpf_program *targ_prog;
5167 if (obj->gen_loader)
5170 for (i = 0; i < map->init_slots_sz; i++) {
5171 if (!map->init_slots[i])
5174 targ_prog = map->init_slots[i];
5175 fd = bpf_program__fd(targ_prog);
5177 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5180 pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n",
5181 map->name, i, targ_prog->name, fd, err);
5184 pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5185 map->name, i, targ_prog->name, fd);
5188 zfree(&map->init_slots);
5189 map->init_slots_sz = 0;
5194 static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5196 struct bpf_map *map;
5199 for (i = 0; i < obj->nr_maps; i++) {
5200 map = &obj->maps[i];
5202 if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5205 err = init_prog_array_slots(obj, map);
5214 static int map_set_def_max_entries(struct bpf_map *map)
5216 if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) {
5219 nr_cpus = libbpf_num_possible_cpus();
5221 pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
5222 map->name, nr_cpus);
5225 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
5226 map->def.max_entries = nr_cpus;
5233 bpf_object__create_maps(struct bpf_object *obj)
5235 struct bpf_map *map;
5236 char *cp, errmsg[STRERR_BUFSIZE];
5241 for (i = 0; i < obj->nr_maps; i++) {
5242 map = &obj->maps[i];
5244 /* To support old kernels, we skip creating global data maps
5245 * (.rodata, .data, .kconfig, etc); later on, during program
5246 * loading, if we detect that at least one of the to-be-loaded
5247 * programs is referencing any global data map, we'll error
5248 * out with program name and relocation index logged.
5249 * This approach allows to accommodate Clang emitting
5250 * unnecessary .rodata.str1.1 sections for string literals,
5251 * but also it allows to have CO-RE applications that use
5252 * global variables in some of BPF programs, but not others.
5253 * If those global variable-using programs are not loaded at
5254 * runtime due to bpf_program__set_autoload(prog, false),
5255 * bpf_object loading will succeed just fine even on old
5258 if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA))
5259 map->autocreate = false;
5261 if (!map->autocreate) {
5262 pr_debug("map '%s': skipped auto-creating...\n", map->name);
5266 err = map_set_def_max_entries(map);
5272 if (map->pin_path) {
5273 err = bpf_object__reuse_map(map);
5275 pr_warn("map '%s': error reusing pinned map\n",
5279 if (retried && map->fd < 0) {
5280 pr_warn("map '%s': cannot find pinned map\n",
5288 pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5289 map->name, map->fd);
5291 err = bpf_object__create_map(obj, map, false);
5295 pr_debug("map '%s': created successfully, fd=%d\n",
5296 map->name, map->fd);
5298 if (bpf_map__is_internal(map)) {
5299 err = bpf_object__populate_internal_map(obj, map);
5306 if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5307 err = init_map_in_map_slots(obj, map);
5315 if (map->pin_path && !map->pinned) {
5316 err = bpf_map__pin(map, NULL);
5319 if (!retried && err == -EEXIST) {
5323 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
5324 map->name, map->pin_path, err);
5333 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5334 pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
5336 for (j = 0; j < i; j++)
5337 zclose(obj->maps[j].fd);
5341 static bool bpf_core_is_flavor_sep(const char *s)
5343 /* check X___Y name pattern, where X and Y are not underscores */
5344 return s[0] != '_' && /* X */
5345 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
5346 s[4] != '_'; /* Y */
5349 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
5350 * before last triple underscore. Struct name part after last triple
5351 * underscore is ignored by BPF CO-RE relocation during relocation matching.
5353 size_t bpf_core_essential_name_len(const char *name)
5355 size_t n = strlen(name);
5358 for (i = n - 5; i >= 0; i--) {
5359 if (bpf_core_is_flavor_sep(name + i))
5365 void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5374 int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5375 size_t local_essent_len,
5376 const struct btf *targ_btf,
5377 const char *targ_btf_name,
5379 struct bpf_core_cand_list *cands)
5381 struct bpf_core_cand *new_cands, *cand;
5382 const struct btf_type *t, *local_t;
5383 const char *targ_name, *local_name;
5384 size_t targ_essent_len;
5387 local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5388 local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5390 n = btf__type_cnt(targ_btf);
5391 for (i = targ_start_id; i < n; i++) {
5392 t = btf__type_by_id(targ_btf, i);
5393 if (!btf_kind_core_compat(t, local_t))
5396 targ_name = btf__name_by_offset(targ_btf, t->name_off);
5397 if (str_is_empty(targ_name))
5400 targ_essent_len = bpf_core_essential_name_len(targ_name);
5401 if (targ_essent_len != local_essent_len)
5404 if (strncmp(local_name, targ_name, local_essent_len) != 0)
5407 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5408 local_cand->id, btf_kind_str(local_t),
5409 local_name, i, btf_kind_str(t), targ_name,
5411 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5412 sizeof(*cands->cands));
5416 cand = &new_cands[cands->len];
5417 cand->btf = targ_btf;
5420 cands->cands = new_cands;
5426 static int load_module_btfs(struct bpf_object *obj)
5428 struct bpf_btf_info info;
5429 struct module_btf *mod_btf;
5435 if (obj->btf_modules_loaded)
5438 if (obj->gen_loader)
5441 /* don't do this again, even if we find no module BTFs */
5442 obj->btf_modules_loaded = true;
5444 /* kernel too old to support module BTFs */
5445 if (!kernel_supports(obj, FEAT_MODULE_BTF))
5449 err = bpf_btf_get_next_id(id, &id);
5450 if (err && errno == ENOENT)
5454 pr_warn("failed to iterate BTF objects: %d\n", err);
5458 fd = bpf_btf_get_fd_by_id(id);
5460 if (errno == ENOENT)
5461 continue; /* expected race: BTF was unloaded */
5463 pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
5468 memset(&info, 0, sizeof(info));
5469 info.name = ptr_to_u64(name);
5470 info.name_len = sizeof(name);
5472 err = bpf_btf_get_info_by_fd(fd, &info, &len);
5475 pr_warn("failed to get BTF object #%d info: %d\n", id, err);
5479 /* ignore non-module BTFs */
5480 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5485 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5486 err = libbpf_get_error(btf);
5488 pr_warn("failed to load module [%s]'s BTF object #%d: %d\n",
5493 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5494 sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5498 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
5503 mod_btf->name = strdup(name);
5504 if (!mod_btf->name) {
5518 static struct bpf_core_cand_list *
5519 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5521 struct bpf_core_cand local_cand = {};
5522 struct bpf_core_cand_list *cands;
5523 const struct btf *main_btf;
5524 const struct btf_type *local_t;
5525 const char *local_name;
5526 size_t local_essent_len;
5529 local_cand.btf = local_btf;
5530 local_cand.id = local_type_id;
5531 local_t = btf__type_by_id(local_btf, local_type_id);
5533 return ERR_PTR(-EINVAL);
5535 local_name = btf__name_by_offset(local_btf, local_t->name_off);
5536 if (str_is_empty(local_name))
5537 return ERR_PTR(-EINVAL);
5538 local_essent_len = bpf_core_essential_name_len(local_name);
5540 cands = calloc(1, sizeof(*cands));
5542 return ERR_PTR(-ENOMEM);
5544 /* Attempt to find target candidates in vmlinux BTF first */
5545 main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5546 err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5550 /* if vmlinux BTF has any candidate, don't got for module BTFs */
5554 /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5555 if (obj->btf_vmlinux_override)
5558 /* now look through module BTFs, trying to still find candidates */
5559 err = load_module_btfs(obj);
5563 for (i = 0; i < obj->btf_module_cnt; i++) {
5564 err = bpf_core_add_cands(&local_cand, local_essent_len,
5565 obj->btf_modules[i].btf,
5566 obj->btf_modules[i].name,
5567 btf__type_cnt(obj->btf_vmlinux),
5575 bpf_core_free_cands(cands);
5576 return ERR_PTR(err);
5579 /* Check local and target types for compatibility. This check is used for
5580 * type-based CO-RE relocations and follow slightly different rules than
5581 * field-based relocations. This function assumes that root types were already
5582 * checked for name match. Beyond that initial root-level name check, names
5583 * are completely ignored. Compatibility rules are as follows:
5584 * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5585 * kind should match for local and target types (i.e., STRUCT is not
5586 * compatible with UNION);
5587 * - for ENUMs, the size is ignored;
5588 * - for INT, size and signedness are ignored;
5589 * - for ARRAY, dimensionality is ignored, element types are checked for
5590 * compatibility recursively;
5591 * - CONST/VOLATILE/RESTRICT modifiers are ignored;
5592 * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5593 * - FUNC_PROTOs are compatible if they have compatible signature: same
5594 * number of input args and compatible return and argument types.
5595 * These rules are not set in stone and probably will be adjusted as we get
5596 * more experience with using BPF CO-RE relocations.
5598 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5599 const struct btf *targ_btf, __u32 targ_id)
5601 return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32);
5604 int bpf_core_types_match(const struct btf *local_btf, __u32 local_id,
5605 const struct btf *targ_btf, __u32 targ_id)
5607 return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32);
5610 static size_t bpf_core_hash_fn(const long key, void *ctx)
5615 static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx)
5620 static int record_relo_core(struct bpf_program *prog,
5621 const struct bpf_core_relo *core_relo, int insn_idx)
5623 struct reloc_desc *relos, *relo;
5625 relos = libbpf_reallocarray(prog->reloc_desc,
5626 prog->nr_reloc + 1, sizeof(*relos));
5629 relo = &relos[prog->nr_reloc];
5630 relo->type = RELO_CORE;
5631 relo->insn_idx = insn_idx;
5632 relo->core_relo = core_relo;
5633 prog->reloc_desc = relos;
5638 static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx)
5640 struct reloc_desc *relo;
5643 for (i = 0; i < prog->nr_reloc; i++) {
5644 relo = &prog->reloc_desc[i];
5645 if (relo->type != RELO_CORE || relo->insn_idx != insn_idx)
5648 return relo->core_relo;
5654 static int bpf_core_resolve_relo(struct bpf_program *prog,
5655 const struct bpf_core_relo *relo,
5657 const struct btf *local_btf,
5658 struct hashmap *cand_cache,
5659 struct bpf_core_relo_res *targ_res)
5661 struct bpf_core_spec specs_scratch[3] = {};
5662 struct bpf_core_cand_list *cands = NULL;
5663 const char *prog_name = prog->name;
5664 const struct btf_type *local_type;
5665 const char *local_name;
5666 __u32 local_id = relo->type_id;
5669 local_type = btf__type_by_id(local_btf, local_id);
5673 local_name = btf__name_by_offset(local_btf, local_type->name_off);
5677 if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
5678 !hashmap__find(cand_cache, local_id, &cands)) {
5679 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5680 if (IS_ERR(cands)) {
5681 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5682 prog_name, relo_idx, local_id, btf_kind_str(local_type),
5683 local_name, PTR_ERR(cands));
5684 return PTR_ERR(cands);
5686 err = hashmap__set(cand_cache, local_id, cands, NULL, NULL);
5688 bpf_core_free_cands(cands);
5693 return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch,
5698 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5700 const struct btf_ext_info_sec *sec;
5701 struct bpf_core_relo_res targ_res;
5702 const struct bpf_core_relo *rec;
5703 const struct btf_ext_info *seg;
5704 struct hashmap_entry *entry;
5705 struct hashmap *cand_cache = NULL;
5706 struct bpf_program *prog;
5707 struct bpf_insn *insn;
5708 const char *sec_name;
5709 int i, err = 0, insn_idx, sec_idx, sec_num;
5711 if (obj->btf_ext->core_relo_info.len == 0)
5714 if (targ_btf_path) {
5715 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
5716 err = libbpf_get_error(obj->btf_vmlinux_override);
5718 pr_warn("failed to parse target BTF: %d\n", err);
5723 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
5724 if (IS_ERR(cand_cache)) {
5725 err = PTR_ERR(cand_cache);
5729 seg = &obj->btf_ext->core_relo_info;
5731 for_each_btf_ext_sec(seg, sec) {
5732 sec_idx = seg->sec_idxs[sec_num];
5735 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5736 if (str_is_empty(sec_name)) {
5741 pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info);
5743 for_each_btf_ext_rec(seg, sec, i, rec) {
5744 if (rec->insn_off % BPF_INSN_SZ)
5746 insn_idx = rec->insn_off / BPF_INSN_SZ;
5747 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
5749 /* When __weak subprog is "overridden" by another instance
5750 * of the subprog from a different object file, linker still
5751 * appends all the .BTF.ext info that used to belong to that
5752 * eliminated subprogram.
5753 * This is similar to what x86-64 linker does for relocations.
5754 * So just ignore such relocations just like we ignore
5755 * subprog instructions when discovering subprograms.
5757 pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n",
5758 sec_name, i, insn_idx);
5761 /* no need to apply CO-RE relocation if the program is
5762 * not going to be loaded
5764 if (!prog->autoload)
5767 /* adjust insn_idx from section frame of reference to the local
5768 * program's frame of reference; (sub-)program code is not yet
5769 * relocated, so it's enough to just subtract in-section offset
5771 insn_idx = insn_idx - prog->sec_insn_off;
5772 if (insn_idx >= prog->insns_cnt)
5774 insn = &prog->insns[insn_idx];
5776 err = record_relo_core(prog, rec, insn_idx);
5778 pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n",
5779 prog->name, i, err);
5783 if (prog->obj->gen_loader)
5786 err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res);
5788 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5789 prog->name, i, err);
5793 err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res);
5795 pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n",
5796 prog->name, i, insn_idx, err);
5803 /* obj->btf_vmlinux and module BTFs are freed after object load */
5804 btf__free(obj->btf_vmlinux_override);
5805 obj->btf_vmlinux_override = NULL;
5807 if (!IS_ERR_OR_NULL(cand_cache)) {
5808 hashmap__for_each_entry(cand_cache, entry, i) {
5809 bpf_core_free_cands(entry->pvalue);
5811 hashmap__free(cand_cache);
5816 /* base map load ldimm64 special constant, used also for log fixup logic */
5817 #define POISON_LDIMM64_MAP_BASE 2001000000
5818 #define POISON_LDIMM64_MAP_PFX "200100"
5820 static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx,
5821 int insn_idx, struct bpf_insn *insn,
5822 int map_idx, const struct bpf_map *map)
5826 pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n",
5827 prog->name, relo_idx, insn_idx, map_idx, map->name);
5829 /* we turn single ldimm64 into two identical invalid calls */
5830 for (i = 0; i < 2; i++) {
5831 insn->code = BPF_JMP | BPF_CALL;
5835 /* if this instruction is reachable (not a dead code),
5836 * verifier will complain with something like:
5837 * invalid func unknown#2001000123
5838 * where lower 123 is map index into obj->maps[] array
5840 insn->imm = POISON_LDIMM64_MAP_BASE + map_idx;
5846 /* unresolved kfunc call special constant, used also for log fixup logic */
5847 #define POISON_CALL_KFUNC_BASE 2002000000
5848 #define POISON_CALL_KFUNC_PFX "2002"
5850 static void poison_kfunc_call(struct bpf_program *prog, int relo_idx,
5851 int insn_idx, struct bpf_insn *insn,
5852 int ext_idx, const struct extern_desc *ext)
5854 pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n",
5855 prog->name, relo_idx, insn_idx, ext->name);
5857 /* we turn kfunc call into invalid helper call with identifiable constant */
5858 insn->code = BPF_JMP | BPF_CALL;
5862 /* if this instruction is reachable (not a dead code),
5863 * verifier will complain with something like:
5864 * invalid func unknown#2001000123
5865 * where lower 123 is extern index into obj->externs[] array
5867 insn->imm = POISON_CALL_KFUNC_BASE + ext_idx;
5870 /* Relocate data references within program code:
5872 * - global variable references;
5873 * - extern references.
5876 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
5880 for (i = 0; i < prog->nr_reloc; i++) {
5881 struct reloc_desc *relo = &prog->reloc_desc[i];
5882 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5883 const struct bpf_map *map;
5884 struct extern_desc *ext;
5886 switch (relo->type) {
5888 map = &obj->maps[relo->map_idx];
5889 if (obj->gen_loader) {
5890 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
5891 insn[0].imm = relo->map_idx;
5892 } else if (map->autocreate) {
5893 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
5894 insn[0].imm = map->fd;
5896 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
5897 relo->map_idx, map);
5901 map = &obj->maps[relo->map_idx];
5902 insn[1].imm = insn[0].imm + relo->sym_off;
5903 if (obj->gen_loader) {
5904 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5905 insn[0].imm = relo->map_idx;
5906 } else if (map->autocreate) {
5907 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5908 insn[0].imm = map->fd;
5910 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
5911 relo->map_idx, map);
5914 case RELO_EXTERN_LD64:
5915 ext = &obj->externs[relo->ext_idx];
5916 if (ext->type == EXT_KCFG) {
5917 if (obj->gen_loader) {
5918 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5919 insn[0].imm = obj->kconfig_map_idx;
5921 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5922 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
5924 insn[1].imm = ext->kcfg.data_off;
5925 } else /* EXT_KSYM */ {
5926 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
5927 insn[0].src_reg = BPF_PSEUDO_BTF_ID;
5928 insn[0].imm = ext->ksym.kernel_btf_id;
5929 insn[1].imm = ext->ksym.kernel_btf_obj_fd;
5930 } else { /* typeless ksyms or unresolved typed ksyms */
5931 insn[0].imm = (__u32)ext->ksym.addr;
5932 insn[1].imm = ext->ksym.addr >> 32;
5936 case RELO_EXTERN_CALL:
5937 ext = &obj->externs[relo->ext_idx];
5938 insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
5940 insn[0].imm = ext->ksym.kernel_btf_id;
5941 insn[0].off = ext->ksym.btf_fd_idx;
5942 } else { /* unresolved weak kfunc call */
5943 poison_kfunc_call(prog, i, relo->insn_idx, insn,
5944 relo->ext_idx, ext);
5947 case RELO_SUBPROG_ADDR:
5948 if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
5949 pr_warn("prog '%s': relo #%d: bad insn\n",
5953 /* handled already */
5956 /* handled already */
5959 /* will be handled by bpf_program_record_relos() */
5962 pr_warn("prog '%s': relo #%d: bad relo type %d\n",
5963 prog->name, i, relo->type);
5971 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
5972 const struct bpf_program *prog,
5973 const struct btf_ext_info *ext_info,
5974 void **prog_info, __u32 *prog_rec_cnt,
5977 void *copy_start = NULL, *copy_end = NULL;
5978 void *rec, *rec_end, *new_prog_info;
5979 const struct btf_ext_info_sec *sec;
5980 size_t old_sz, new_sz;
5981 int i, sec_num, sec_idx, off_adj;
5984 for_each_btf_ext_sec(ext_info, sec) {
5985 sec_idx = ext_info->sec_idxs[sec_num];
5987 if (prog->sec_idx != sec_idx)
5990 for_each_btf_ext_rec(ext_info, sec, i, rec) {
5991 __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
5993 if (insn_off < prog->sec_insn_off)
5995 if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
6000 copy_end = rec + ext_info->rec_size;
6006 /* append func/line info of a given (sub-)program to the main
6007 * program func/line info
6009 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
6010 new_sz = old_sz + (copy_end - copy_start);
6011 new_prog_info = realloc(*prog_info, new_sz);
6014 *prog_info = new_prog_info;
6015 *prog_rec_cnt = new_sz / ext_info->rec_size;
6016 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
6018 /* Kernel instruction offsets are in units of 8-byte
6019 * instructions, while .BTF.ext instruction offsets generated
6020 * by Clang are in units of bytes. So convert Clang offsets
6021 * into kernel offsets and adjust offset according to program
6022 * relocated position.
6024 off_adj = prog->sub_insn_off - prog->sec_insn_off;
6025 rec = new_prog_info + old_sz;
6026 rec_end = new_prog_info + new_sz;
6027 for (; rec < rec_end; rec += ext_info->rec_size) {
6028 __u32 *insn_off = rec;
6030 *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
6032 *prog_rec_sz = ext_info->rec_size;
6040 reloc_prog_func_and_line_info(const struct bpf_object *obj,
6041 struct bpf_program *main_prog,
6042 const struct bpf_program *prog)
6046 /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
6047 * supprot func/line info
6049 if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
6052 /* only attempt func info relocation if main program's func_info
6053 * relocation was successful
6055 if (main_prog != prog && !main_prog->func_info)
6058 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
6059 &main_prog->func_info,
6060 &main_prog->func_info_cnt,
6061 &main_prog->func_info_rec_size);
6063 if (err != -ENOENT) {
6064 pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
6068 if (main_prog->func_info) {
6070 * Some info has already been found but has problem
6071 * in the last btf_ext reloc. Must have to error out.
6073 pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
6076 /* Have problem loading the very first info. Ignore the rest. */
6077 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
6082 /* don't relocate line info if main program's relocation failed */
6083 if (main_prog != prog && !main_prog->line_info)
6086 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
6087 &main_prog->line_info,
6088 &main_prog->line_info_cnt,
6089 &main_prog->line_info_rec_size);
6091 if (err != -ENOENT) {
6092 pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
6096 if (main_prog->line_info) {
6098 * Some info has already been found but has problem
6099 * in the last btf_ext reloc. Must have to error out.
6101 pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
6104 /* Have problem loading the very first info. Ignore the rest. */
6105 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
6111 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
6113 size_t insn_idx = *(const size_t *)key;
6114 const struct reloc_desc *relo = elem;
6116 if (insn_idx == relo->insn_idx)
6118 return insn_idx < relo->insn_idx ? -1 : 1;
6121 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
6123 if (!prog->nr_reloc)
6125 return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
6126 sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
6129 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
6131 int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
6132 struct reloc_desc *relos;
6135 if (main_prog == subprog)
6137 relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
6140 if (subprog->nr_reloc)
6141 memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
6142 sizeof(*relos) * subprog->nr_reloc);
6144 for (i = main_prog->nr_reloc; i < new_cnt; i++)
6145 relos[i].insn_idx += subprog->sub_insn_off;
6146 /* After insn_idx adjustment the 'relos' array is still sorted
6147 * by insn_idx and doesn't break bsearch.
6149 main_prog->reloc_desc = relos;
6150 main_prog->nr_reloc = new_cnt;
6155 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6156 struct bpf_program *prog)
6158 size_t sub_insn_idx, insn_idx, new_cnt;
6159 struct bpf_program *subprog;
6160 struct bpf_insn *insns, *insn;
6161 struct reloc_desc *relo;
6164 err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6168 for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6169 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6170 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
6173 relo = find_prog_insn_relo(prog, insn_idx);
6174 if (relo && relo->type == RELO_EXTERN_CALL)
6175 /* kfunc relocations will be handled later
6176 * in bpf_object__relocate_data()
6179 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6180 pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6181 prog->name, insn_idx, relo->type);
6182 return -LIBBPF_ERRNO__RELOC;
6185 /* sub-program instruction index is a combination of
6186 * an offset of a symbol pointed to by relocation and
6187 * call instruction's imm field; for global functions,
6188 * call always has imm = -1, but for static functions
6189 * relocation is against STT_SECTION and insn->imm
6190 * points to a start of a static function
6192 * for subprog addr relocation, the relo->sym_off + insn->imm is
6193 * the byte offset in the corresponding section.
6195 if (relo->type == RELO_CALL)
6196 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6198 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6199 } else if (insn_is_pseudo_func(insn)) {
6201 * RELO_SUBPROG_ADDR relo is always emitted even if both
6202 * functions are in the same section, so it shouldn't reach here.
6204 pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6205 prog->name, insn_idx);
6206 return -LIBBPF_ERRNO__RELOC;
6208 /* if subprogram call is to a static function within
6209 * the same ELF section, there won't be any relocation
6210 * emitted, but it also means there is no additional
6211 * offset necessary, insns->imm is relative to
6212 * instruction's original position within the section
6214 sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6217 /* we enforce that sub-programs should be in .text section */
6218 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6220 pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6222 return -LIBBPF_ERRNO__RELOC;
6225 /* if it's the first call instruction calling into this
6226 * subprogram (meaning this subprog hasn't been processed
6227 * yet) within the context of current main program:
6228 * - append it at the end of main program's instructions blog;
6229 * - process is recursively, while current program is put on hold;
6230 * - if that subprogram calls some other not yet processes
6231 * subprogram, same thing will happen recursively until
6232 * there are no more unprocesses subprograms left to append
6235 if (subprog->sub_insn_off == 0) {
6236 subprog->sub_insn_off = main_prog->insns_cnt;
6238 new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6239 insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6241 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6244 main_prog->insns = insns;
6245 main_prog->insns_cnt = new_cnt;
6247 memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6248 subprog->insns_cnt * sizeof(*insns));
6250 pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6251 main_prog->name, subprog->insns_cnt, subprog->name);
6253 /* The subprog insns are now appended. Append its relos too. */
6254 err = append_subprog_relos(main_prog, subprog);
6257 err = bpf_object__reloc_code(obj, main_prog, subprog);
6262 /* main_prog->insns memory could have been re-allocated, so
6263 * calculate pointer again
6265 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6266 /* calculate correct instruction position within current main
6267 * prog; each main prog can have a different set of
6268 * subprograms appended (potentially in different order as
6269 * well), so position of any subprog can be different for
6270 * different main programs
6272 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6274 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6275 prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6282 * Relocate sub-program calls.
6284 * Algorithm operates as follows. Each entry-point BPF program (referred to as
6285 * main prog) is processed separately. For each subprog (non-entry functions,
6286 * that can be called from either entry progs or other subprogs) gets their
6287 * sub_insn_off reset to zero. This serves as indicator that this subprogram
6288 * hasn't been yet appended and relocated within current main prog. Once its
6289 * relocated, sub_insn_off will point at the position within current main prog
6290 * where given subprog was appended. This will further be used to relocate all
6291 * the call instructions jumping into this subprog.
6293 * We start with main program and process all call instructions. If the call
6294 * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6295 * is zero), subprog instructions are appended at the end of main program's
6296 * instruction array. Then main program is "put on hold" while we recursively
6297 * process newly appended subprogram. If that subprogram calls into another
6298 * subprogram that hasn't been appended, new subprogram is appended again to
6299 * the *main* prog's instructions (subprog's instructions are always left
6300 * untouched, as they need to be in unmodified state for subsequent main progs
6301 * and subprog instructions are always sent only as part of a main prog) and
6302 * the process continues recursively. Once all the subprogs called from a main
6303 * prog or any of its subprogs are appended (and relocated), all their
6304 * positions within finalized instructions array are known, so it's easy to
6305 * rewrite call instructions with correct relative offsets, corresponding to
6306 * desired target subprog.
6308 * Its important to realize that some subprogs might not be called from some
6309 * main prog and any of its called/used subprogs. Those will keep their
6310 * subprog->sub_insn_off as zero at all times and won't be appended to current
6311 * main prog and won't be relocated within the context of current main prog.
6312 * They might still be used from other main progs later.
6314 * Visually this process can be shown as below. Suppose we have two main
6315 * programs mainA and mainB and BPF object contains three subprogs: subA,
6316 * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6317 * subC both call subB:
6319 * +--------+ +-------+
6321 * +--+---+ +--+-+-+ +---+--+
6322 * | subA | | subB | | subC |
6323 * +--+---+ +------+ +---+--+
6326 * +---+-------+ +------+----+
6327 * | mainA | | mainB |
6328 * +-----------+ +-----------+
6330 * We'll start relocating mainA, will find subA, append it and start
6331 * processing sub A recursively:
6333 * +-----------+------+
6335 * +-----------+------+
6337 * At this point we notice that subB is used from subA, so we append it and
6338 * relocate (there are no further subcalls from subB):
6340 * +-----------+------+------+
6341 * | mainA | subA | subB |
6342 * +-----------+------+------+
6344 * At this point, we relocate subA calls, then go one level up and finish with
6345 * relocatin mainA calls. mainA is done.
6347 * For mainB process is similar but results in different order. We start with
6348 * mainB and skip subA and subB, as mainB never calls them (at least
6349 * directly), but we see subC is needed, so we append and start processing it:
6351 * +-----------+------+
6353 * +-----------+------+
6354 * Now we see subC needs subB, so we go back to it, append and relocate it:
6356 * +-----------+------+------+
6357 * | mainB | subC | subB |
6358 * +-----------+------+------+
6360 * At this point we unwind recursion, relocate calls in subC, then in mainB.
6363 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6365 struct bpf_program *subprog;
6368 /* mark all subprogs as not relocated (yet) within the context of
6369 * current main program
6371 for (i = 0; i < obj->nr_programs; i++) {
6372 subprog = &obj->programs[i];
6373 if (!prog_is_subprog(obj, subprog))
6376 subprog->sub_insn_off = 0;
6379 err = bpf_object__reloc_code(obj, prog, prog);
6387 bpf_object__free_relocs(struct bpf_object *obj)
6389 struct bpf_program *prog;
6392 /* free up relocation descriptors */
6393 for (i = 0; i < obj->nr_programs; i++) {
6394 prog = &obj->programs[i];
6395 zfree(&prog->reloc_desc);
6400 static int cmp_relocs(const void *_a, const void *_b)
6402 const struct reloc_desc *a = _a;
6403 const struct reloc_desc *b = _b;
6405 if (a->insn_idx != b->insn_idx)
6406 return a->insn_idx < b->insn_idx ? -1 : 1;
6408 /* no two relocations should have the same insn_idx, but ... */
6409 if (a->type != b->type)
6410 return a->type < b->type ? -1 : 1;
6415 static void bpf_object__sort_relos(struct bpf_object *obj)
6419 for (i = 0; i < obj->nr_programs; i++) {
6420 struct bpf_program *p = &obj->programs[i];
6425 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6430 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
6432 struct bpf_program *prog;
6437 err = bpf_object__relocate_core(obj, targ_btf_path);
6439 pr_warn("failed to perform CO-RE relocations: %d\n",
6443 bpf_object__sort_relos(obj);
6446 /* Before relocating calls pre-process relocations and mark
6447 * few ld_imm64 instructions that points to subprogs.
6448 * Otherwise bpf_object__reloc_code() later would have to consider
6449 * all ld_imm64 insns as relocation candidates. That would
6450 * reduce relocation speed, since amount of find_prog_insn_relo()
6451 * would increase and most of them will fail to find a relo.
6453 for (i = 0; i < obj->nr_programs; i++) {
6454 prog = &obj->programs[i];
6455 for (j = 0; j < prog->nr_reloc; j++) {
6456 struct reloc_desc *relo = &prog->reloc_desc[j];
6457 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6459 /* mark the insn, so it's recognized by insn_is_pseudo_func() */
6460 if (relo->type == RELO_SUBPROG_ADDR)
6461 insn[0].src_reg = BPF_PSEUDO_FUNC;
6465 /* relocate subprogram calls and append used subprograms to main
6466 * programs; each copy of subprogram code needs to be relocated
6467 * differently for each main program, because its code location might
6469 * Append subprog relos to main programs to allow data relos to be
6470 * processed after text is completely relocated.
6472 for (i = 0; i < obj->nr_programs; i++) {
6473 prog = &obj->programs[i];
6474 /* sub-program's sub-calls are relocated within the context of
6475 * its main program only
6477 if (prog_is_subprog(obj, prog))
6479 if (!prog->autoload)
6482 err = bpf_object__relocate_calls(obj, prog);
6484 pr_warn("prog '%s': failed to relocate calls: %d\n",
6489 /* Process data relos for main programs */
6490 for (i = 0; i < obj->nr_programs; i++) {
6491 prog = &obj->programs[i];
6492 if (prog_is_subprog(obj, prog))
6494 if (!prog->autoload)
6496 err = bpf_object__relocate_data(obj, prog);
6498 pr_warn("prog '%s': failed to relocate data references: %d\n",
6507 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
6508 Elf64_Shdr *shdr, Elf_Data *data);
6510 static int bpf_object__collect_map_relos(struct bpf_object *obj,
6511 Elf64_Shdr *shdr, Elf_Data *data)
6513 const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
6514 int i, j, nrels, new_sz;
6515 const struct btf_var_secinfo *vi = NULL;
6516 const struct btf_type *sec, *var, *def;
6517 struct bpf_map *map = NULL, *targ_map = NULL;
6518 struct bpf_program *targ_prog = NULL;
6519 bool is_prog_array, is_map_in_map;
6520 const struct btf_member *member;
6521 const char *name, *mname, *type;
6527 if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
6529 sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
6533 nrels = shdr->sh_size / shdr->sh_entsize;
6534 for (i = 0; i < nrels; i++) {
6535 rel = elf_rel_by_idx(data, i);
6537 pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
6538 return -LIBBPF_ERRNO__FORMAT;
6541 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
6543 pr_warn(".maps relo #%d: symbol %zx not found\n",
6544 i, (size_t)ELF64_R_SYM(rel->r_info));
6545 return -LIBBPF_ERRNO__FORMAT;
6547 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
6549 pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
6550 i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
6551 (size_t)rel->r_offset, sym->st_name, name);
6553 for (j = 0; j < obj->nr_maps; j++) {
6554 map = &obj->maps[j];
6555 if (map->sec_idx != obj->efile.btf_maps_shndx)
6558 vi = btf_var_secinfos(sec) + map->btf_var_idx;
6559 if (vi->offset <= rel->r_offset &&
6560 rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
6563 if (j == obj->nr_maps) {
6564 pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
6565 i, name, (size_t)rel->r_offset);
6569 is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
6570 is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
6571 type = is_map_in_map ? "map" : "prog";
6572 if (is_map_in_map) {
6573 if (sym->st_shndx != obj->efile.btf_maps_shndx) {
6574 pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
6576 return -LIBBPF_ERRNO__RELOC;
6578 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
6579 map->def.key_size != sizeof(int)) {
6580 pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
6581 i, map->name, sizeof(int));
6584 targ_map = bpf_object__find_map_by_name(obj, name);
6586 pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
6590 } else if (is_prog_array) {
6591 targ_prog = bpf_object__find_program_by_name(obj, name);
6593 pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
6597 if (targ_prog->sec_idx != sym->st_shndx ||
6598 targ_prog->sec_insn_off * 8 != sym->st_value ||
6599 prog_is_subprog(obj, targ_prog)) {
6600 pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
6602 return -LIBBPF_ERRNO__RELOC;
6608 var = btf__type_by_id(obj->btf, vi->type);
6609 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
6610 if (btf_vlen(def) == 0)
6612 member = btf_members(def) + btf_vlen(def) - 1;
6613 mname = btf__name_by_offset(obj->btf, member->name_off);
6614 if (strcmp(mname, "values"))
6617 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
6618 if (rel->r_offset - vi->offset < moff)
6621 moff = rel->r_offset - vi->offset - moff;
6622 /* here we use BPF pointer size, which is always 64 bit, as we
6623 * are parsing ELF that was built for BPF target
6625 if (moff % bpf_ptr_sz)
6628 if (moff >= map->init_slots_sz) {
6630 tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
6633 map->init_slots = tmp;
6634 memset(map->init_slots + map->init_slots_sz, 0,
6635 (new_sz - map->init_slots_sz) * host_ptr_sz);
6636 map->init_slots_sz = new_sz;
6638 map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
6640 pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
6641 i, map->name, moff, type, name);
6647 static int bpf_object__collect_relos(struct bpf_object *obj)
6651 for (i = 0; i < obj->efile.sec_cnt; i++) {
6652 struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
6657 if (sec_desc->sec_type != SEC_RELO)
6660 shdr = sec_desc->shdr;
6661 data = sec_desc->data;
6662 idx = shdr->sh_info;
6664 if (shdr->sh_type != SHT_REL) {
6665 pr_warn("internal error at %d\n", __LINE__);
6666 return -LIBBPF_ERRNO__INTERNAL;
6669 if (idx == obj->efile.st_ops_shndx || idx == obj->efile.st_ops_link_shndx)
6670 err = bpf_object__collect_st_ops_relos(obj, shdr, data);
6671 else if (idx == obj->efile.btf_maps_shndx)
6672 err = bpf_object__collect_map_relos(obj, shdr, data);
6674 err = bpf_object__collect_prog_relos(obj, shdr, data);
6679 bpf_object__sort_relos(obj);
6683 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
6685 if (BPF_CLASS(insn->code) == BPF_JMP &&
6686 BPF_OP(insn->code) == BPF_CALL &&
6687 BPF_SRC(insn->code) == BPF_K &&
6688 insn->src_reg == 0 &&
6689 insn->dst_reg == 0) {
6690 *func_id = insn->imm;
6696 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
6698 struct bpf_insn *insn = prog->insns;
6699 enum bpf_func_id func_id;
6702 if (obj->gen_loader)
6705 for (i = 0; i < prog->insns_cnt; i++, insn++) {
6706 if (!insn_is_helper_call(insn, &func_id))
6709 /* on kernels that don't yet support
6710 * bpf_probe_read_{kernel,user}[_str] helpers, fall back
6711 * to bpf_probe_read() which works well for old kernels
6714 case BPF_FUNC_probe_read_kernel:
6715 case BPF_FUNC_probe_read_user:
6716 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6717 insn->imm = BPF_FUNC_probe_read;
6719 case BPF_FUNC_probe_read_kernel_str:
6720 case BPF_FUNC_probe_read_user_str:
6721 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6722 insn->imm = BPF_FUNC_probe_read_str;
6731 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
6732 int *btf_obj_fd, int *btf_type_id);
6734 /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */
6735 static int libbpf_prepare_prog_load(struct bpf_program *prog,
6736 struct bpf_prog_load_opts *opts, long cookie)
6738 enum sec_def_flags def = cookie;
6740 /* old kernels might not support specifying expected_attach_type */
6741 if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
6742 opts->expected_attach_type = 0;
6744 if (def & SEC_SLEEPABLE)
6745 opts->prog_flags |= BPF_F_SLEEPABLE;
6747 if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
6748 opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
6750 if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) {
6751 int btf_obj_fd = 0, btf_type_id = 0, err;
6752 const char *attach_name;
6754 attach_name = strchr(prog->sec_name, '/');
6756 /* if BPF program is annotated with just SEC("fentry")
6757 * (or similar) without declaratively specifying
6758 * target, then it is expected that target will be
6759 * specified with bpf_program__set_attach_target() at
6760 * runtime before BPF object load step. If not, then
6761 * there is nothing to load into the kernel as BPF
6762 * verifier won't be able to validate BPF program
6763 * correctness anyways.
6765 pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n",
6769 attach_name++; /* skip over / */
6771 err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
6775 /* cache resolved BTF FD and BTF type ID in the prog */
6776 prog->attach_btf_obj_fd = btf_obj_fd;
6777 prog->attach_btf_id = btf_type_id;
6779 /* but by now libbpf common logic is not utilizing
6780 * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
6781 * this callback is called after opts were populated by
6782 * libbpf, so this callback has to update opts explicitly here
6784 opts->attach_btf_obj_fd = btf_obj_fd;
6785 opts->attach_btf_id = btf_type_id;
6790 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz);
6792 static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
6793 struct bpf_insn *insns, int insns_cnt,
6794 const char *license, __u32 kern_version, int *prog_fd)
6796 LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
6797 const char *prog_name = NULL;
6798 char *cp, errmsg[STRERR_BUFSIZE];
6799 size_t log_buf_size = 0;
6800 char *log_buf = NULL, *tmp;
6801 int btf_fd, ret, err;
6802 bool own_log_buf = true;
6803 __u32 log_level = prog->log_level;
6805 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6807 * The program type must be set. Most likely we couldn't find a proper
6808 * section definition at load time, and thus we didn't infer the type.
6810 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
6811 prog->name, prog->sec_name);
6815 if (!insns || !insns_cnt)
6818 load_attr.expected_attach_type = prog->expected_attach_type;
6819 if (kernel_supports(obj, FEAT_PROG_NAME))
6820 prog_name = prog->name;
6821 load_attr.attach_prog_fd = prog->attach_prog_fd;
6822 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
6823 load_attr.attach_btf_id = prog->attach_btf_id;
6824 load_attr.kern_version = kern_version;
6825 load_attr.prog_ifindex = prog->prog_ifindex;
6827 /* specify func_info/line_info only if kernel supports them */
6828 btf_fd = bpf_object__btf_fd(obj);
6829 if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
6830 load_attr.prog_btf_fd = btf_fd;
6831 load_attr.func_info = prog->func_info;
6832 load_attr.func_info_rec_size = prog->func_info_rec_size;
6833 load_attr.func_info_cnt = prog->func_info_cnt;
6834 load_attr.line_info = prog->line_info;
6835 load_attr.line_info_rec_size = prog->line_info_rec_size;
6836 load_attr.line_info_cnt = prog->line_info_cnt;
6838 load_attr.log_level = log_level;
6839 load_attr.prog_flags = prog->prog_flags;
6840 load_attr.fd_array = obj->fd_array;
6842 /* adjust load_attr if sec_def provides custom preload callback */
6843 if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) {
6844 err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie);
6846 pr_warn("prog '%s': failed to prepare load attributes: %d\n",
6850 insns = prog->insns;
6851 insns_cnt = prog->insns_cnt;
6854 if (obj->gen_loader) {
6855 bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
6856 license, insns, insns_cnt, &load_attr,
6857 prog - obj->programs);
6863 /* if log_level is zero, we don't request logs initially even if
6864 * custom log_buf is specified; if the program load fails, then we'll
6865 * bump log_level to 1 and use either custom log_buf or we'll allocate
6866 * our own and retry the load to get details on what failed
6869 if (prog->log_buf) {
6870 log_buf = prog->log_buf;
6871 log_buf_size = prog->log_size;
6872 own_log_buf = false;
6873 } else if (obj->log_buf) {
6874 log_buf = obj->log_buf;
6875 log_buf_size = obj->log_size;
6876 own_log_buf = false;
6878 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
6879 tmp = realloc(log_buf, log_buf_size);
6890 load_attr.log_buf = log_buf;
6891 load_attr.log_size = log_buf_size;
6892 load_attr.log_level = log_level;
6894 ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
6896 if (log_level && own_log_buf) {
6897 pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
6898 prog->name, log_buf);
6901 if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
6902 struct bpf_map *map;
6905 for (i = 0; i < obj->nr_maps; i++) {
6906 map = &prog->obj->maps[i];
6907 if (map->libbpf_type != LIBBPF_MAP_RODATA)
6910 if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) {
6911 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6912 pr_warn("prog '%s': failed to bind map '%s': %s\n",
6913 prog->name, map->real_name, cp);
6914 /* Don't fail hard if can't bind rodata. */
6924 if (log_level == 0) {
6928 /* On ENOSPC, increase log buffer size and retry, unless custom
6929 * log_buf is specified.
6930 * Be careful to not overflow u32, though. Kernel's log buf size limit
6931 * isn't part of UAPI so it can always be bumped to full 4GB. So don't
6932 * multiply by 2 unless we are sure we'll fit within 32 bits.
6933 * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
6935 if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
6940 /* post-process verifier log to improve error descriptions */
6941 fixup_verifier_log(prog, log_buf, log_buf_size);
6943 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6944 pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp);
6947 if (own_log_buf && log_buf && log_buf[0] != '\0') {
6948 pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
6949 prog->name, log_buf);
6958 static char *find_prev_line(char *buf, char *cur)
6962 if (cur == buf) /* end of a log buf */
6966 while (p - 1 >= buf && *(p - 1) != '\n')
6972 static void patch_log(char *buf, size_t buf_sz, size_t log_sz,
6973 char *orig, size_t orig_sz, const char *patch)
6975 /* size of the remaining log content to the right from the to-be-replaced part */
6976 size_t rem_sz = (buf + log_sz) - (orig + orig_sz);
6977 size_t patch_sz = strlen(patch);
6979 if (patch_sz != orig_sz) {
6980 /* If patch line(s) are longer than original piece of verifier log,
6981 * shift log contents by (patch_sz - orig_sz) bytes to the right
6982 * starting from after to-be-replaced part of the log.
6984 * If patch line(s) are shorter than original piece of verifier log,
6985 * shift log contents by (orig_sz - patch_sz) bytes to the left
6986 * starting from after to-be-replaced part of the log
6988 * We need to be careful about not overflowing available
6989 * buf_sz capacity. If that's the case, we'll truncate the end
6990 * of the original log, as necessary.
6992 if (patch_sz > orig_sz) {
6993 if (orig + patch_sz >= buf + buf_sz) {
6994 /* patch is big enough to cover remaining space completely */
6995 patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1;
6997 } else if (patch_sz - orig_sz > buf_sz - log_sz) {
6998 /* patch causes part of remaining log to be truncated */
6999 rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz);
7002 /* shift remaining log to the right by calculated amount */
7003 memmove(orig + patch_sz, orig + orig_sz, rem_sz);
7006 memcpy(orig, patch, patch_sz);
7009 static void fixup_log_failed_core_relo(struct bpf_program *prog,
7010 char *buf, size_t buf_sz, size_t log_sz,
7011 char *line1, char *line2, char *line3)
7013 /* Expected log for failed and not properly guarded CO-RE relocation:
7014 * line1 -> 123: (85) call unknown#195896080
7015 * line2 -> invalid func unknown#195896080
7016 * line3 -> <anything else or end of buffer>
7018 * "123" is the index of the instruction that was poisoned. We extract
7019 * instruction index to find corresponding CO-RE relocation and
7020 * replace this part of the log with more relevant information about
7021 * failed CO-RE relocation.
7023 const struct bpf_core_relo *relo;
7024 struct bpf_core_spec spec;
7025 char patch[512], spec_buf[256];
7026 int insn_idx, err, spec_len;
7028 if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1)
7031 relo = find_relo_core(prog, insn_idx);
7035 err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec);
7039 spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec);
7040 snprintf(patch, sizeof(patch),
7041 "%d: <invalid CO-RE relocation>\n"
7042 "failed to resolve CO-RE relocation %s%s\n",
7043 insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : "");
7045 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7048 static void fixup_log_missing_map_load(struct bpf_program *prog,
7049 char *buf, size_t buf_sz, size_t log_sz,
7050 char *line1, char *line2, char *line3)
7052 /* Expected log for failed and not properly guarded map reference:
7053 * line1 -> 123: (85) call unknown#2001000345
7054 * line2 -> invalid func unknown#2001000345
7055 * line3 -> <anything else or end of buffer>
7057 * "123" is the index of the instruction that was poisoned.
7058 * "345" in "2001000345" is a map index in obj->maps to fetch map name.
7060 struct bpf_object *obj = prog->obj;
7061 const struct bpf_map *map;
7062 int insn_idx, map_idx;
7065 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2)
7068 map_idx -= POISON_LDIMM64_MAP_BASE;
7069 if (map_idx < 0 || map_idx >= obj->nr_maps)
7071 map = &obj->maps[map_idx];
7073 snprintf(patch, sizeof(patch),
7074 "%d: <invalid BPF map reference>\n"
7075 "BPF map '%s' is referenced but wasn't created\n",
7076 insn_idx, map->name);
7078 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7081 static void fixup_log_missing_kfunc_call(struct bpf_program *prog,
7082 char *buf, size_t buf_sz, size_t log_sz,
7083 char *line1, char *line2, char *line3)
7085 /* Expected log for failed and not properly guarded kfunc call:
7086 * line1 -> 123: (85) call unknown#2002000345
7087 * line2 -> invalid func unknown#2002000345
7088 * line3 -> <anything else or end of buffer>
7090 * "123" is the index of the instruction that was poisoned.
7091 * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name.
7093 struct bpf_object *obj = prog->obj;
7094 const struct extern_desc *ext;
7095 int insn_idx, ext_idx;
7098 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2)
7101 ext_idx -= POISON_CALL_KFUNC_BASE;
7102 if (ext_idx < 0 || ext_idx >= obj->nr_extern)
7104 ext = &obj->externs[ext_idx];
7106 snprintf(patch, sizeof(patch),
7107 "%d: <invalid kfunc call>\n"
7108 "kfunc '%s' is referenced but wasn't resolved\n",
7109 insn_idx, ext->name);
7111 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7114 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz)
7116 /* look for familiar error patterns in last N lines of the log */
7117 const size_t max_last_line_cnt = 10;
7118 char *prev_line, *cur_line, *next_line;
7125 log_sz = strlen(buf) + 1;
7126 next_line = buf + log_sz - 1;
7128 for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) {
7129 cur_line = find_prev_line(buf, next_line);
7133 if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) {
7134 prev_line = find_prev_line(buf, cur_line);
7138 /* failed CO-RE relocation case */
7139 fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz,
7140 prev_line, cur_line, next_line);
7142 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) {
7143 prev_line = find_prev_line(buf, cur_line);
7147 /* reference to uncreated BPF map */
7148 fixup_log_missing_map_load(prog, buf, buf_sz, log_sz,
7149 prev_line, cur_line, next_line);
7151 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) {
7152 prev_line = find_prev_line(buf, cur_line);
7156 /* reference to unresolved kfunc */
7157 fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz,
7158 prev_line, cur_line, next_line);
7164 static int bpf_program_record_relos(struct bpf_program *prog)
7166 struct bpf_object *obj = prog->obj;
7169 for (i = 0; i < prog->nr_reloc; i++) {
7170 struct reloc_desc *relo = &prog->reloc_desc[i];
7171 struct extern_desc *ext = &obj->externs[relo->ext_idx];
7174 switch (relo->type) {
7175 case RELO_EXTERN_LD64:
7176 if (ext->type != EXT_KSYM)
7178 kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ?
7179 BTF_KIND_VAR : BTF_KIND_FUNC;
7180 bpf_gen__record_extern(obj->gen_loader, ext->name,
7181 ext->is_weak, !ext->ksym.type_id,
7182 true, kind, relo->insn_idx);
7184 case RELO_EXTERN_CALL:
7185 bpf_gen__record_extern(obj->gen_loader, ext->name,
7186 ext->is_weak, false, false, BTF_KIND_FUNC,
7190 struct bpf_core_relo cr = {
7191 .insn_off = relo->insn_idx * 8,
7192 .type_id = relo->core_relo->type_id,
7193 .access_str_off = relo->core_relo->access_str_off,
7194 .kind = relo->core_relo->kind,
7197 bpf_gen__record_relo_core(obj->gen_loader, &cr);
7208 bpf_object__load_progs(struct bpf_object *obj, int log_level)
7210 struct bpf_program *prog;
7214 for (i = 0; i < obj->nr_programs; i++) {
7215 prog = &obj->programs[i];
7216 err = bpf_object__sanitize_prog(obj, prog);
7221 for (i = 0; i < obj->nr_programs; i++) {
7222 prog = &obj->programs[i];
7223 if (prog_is_subprog(obj, prog))
7225 if (!prog->autoload) {
7226 pr_debug("prog '%s': skipped loading\n", prog->name);
7229 prog->log_level |= log_level;
7231 if (obj->gen_loader)
7232 bpf_program_record_relos(prog);
7234 err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt,
7235 obj->license, obj->kern_version, &prog->fd);
7237 pr_warn("prog '%s': failed to load: %d\n", prog->name, err);
7242 bpf_object__free_relocs(obj);
7246 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
7248 static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
7250 struct bpf_program *prog;
7253 bpf_object__for_each_program(prog, obj) {
7254 prog->sec_def = find_sec_def(prog->sec_name);
7255 if (!prog->sec_def) {
7256 /* couldn't guess, but user might manually specify */
7257 pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
7258 prog->name, prog->sec_name);
7262 prog->type = prog->sec_def->prog_type;
7263 prog->expected_attach_type = prog->sec_def->expected_attach_type;
7265 /* sec_def can have custom callback which should be called
7266 * after bpf_program is initialized to adjust its properties
7268 if (prog->sec_def->prog_setup_fn) {
7269 err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie);
7271 pr_warn("prog '%s': failed to initialize: %d\n",
7281 static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
7282 const struct bpf_object_open_opts *opts)
7284 const char *obj_name, *kconfig, *btf_tmp_path;
7285 struct bpf_object *obj;
7292 if (elf_version(EV_CURRENT) == EV_NONE) {
7293 pr_warn("failed to init libelf for %s\n",
7294 path ? : "(mem buf)");
7295 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
7298 if (!OPTS_VALID(opts, bpf_object_open_opts))
7299 return ERR_PTR(-EINVAL);
7301 obj_name = OPTS_GET(opts, object_name, NULL);
7304 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
7305 (unsigned long)obj_buf,
7306 (unsigned long)obj_buf_sz);
7307 obj_name = tmp_name;
7310 pr_debug("loading object '%s' from buffer\n", obj_name);
7313 log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
7314 log_size = OPTS_GET(opts, kernel_log_size, 0);
7315 log_level = OPTS_GET(opts, kernel_log_level, 0);
7316 if (log_size > UINT_MAX)
7317 return ERR_PTR(-EINVAL);
7318 if (log_size && !log_buf)
7319 return ERR_PTR(-EINVAL);
7321 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
7325 obj->log_buf = log_buf;
7326 obj->log_size = log_size;
7327 obj->log_level = log_level;
7329 btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
7331 if (strlen(btf_tmp_path) >= PATH_MAX) {
7332 err = -ENAMETOOLONG;
7335 obj->btf_custom_path = strdup(btf_tmp_path);
7336 if (!obj->btf_custom_path) {
7342 kconfig = OPTS_GET(opts, kconfig, NULL);
7344 obj->kconfig = strdup(kconfig);
7345 if (!obj->kconfig) {
7351 err = bpf_object__elf_init(obj);
7352 err = err ? : bpf_object__check_endianness(obj);
7353 err = err ? : bpf_object__elf_collect(obj);
7354 err = err ? : bpf_object__collect_externs(obj);
7355 err = err ? : bpf_object_fixup_btf(obj);
7356 err = err ? : bpf_object__init_maps(obj, opts);
7357 err = err ? : bpf_object_init_progs(obj, opts);
7358 err = err ? : bpf_object__collect_relos(obj);
7362 bpf_object__elf_finish(obj);
7366 bpf_object__close(obj);
7367 return ERR_PTR(err);
7371 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
7374 return libbpf_err_ptr(-EINVAL);
7376 pr_debug("loading %s\n", path);
7378 return libbpf_ptr(bpf_object_open(path, NULL, 0, opts));
7381 struct bpf_object *bpf_object__open(const char *path)
7383 return bpf_object__open_file(path, NULL);
7387 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
7388 const struct bpf_object_open_opts *opts)
7390 if (!obj_buf || obj_buf_sz == 0)
7391 return libbpf_err_ptr(-EINVAL);
7393 return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts));
7396 static int bpf_object_unload(struct bpf_object *obj)
7401 return libbpf_err(-EINVAL);
7403 for (i = 0; i < obj->nr_maps; i++) {
7404 zclose(obj->maps[i].fd);
7405 if (obj->maps[i].st_ops)
7406 zfree(&obj->maps[i].st_ops->kern_vdata);
7409 for (i = 0; i < obj->nr_programs; i++)
7410 bpf_program__unload(&obj->programs[i]);
7415 static int bpf_object__sanitize_maps(struct bpf_object *obj)
7419 bpf_object__for_each_map(m, obj) {
7420 if (!bpf_map__is_internal(m))
7422 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
7423 m->def.map_flags &= ~BPF_F_MMAPABLE;
7429 int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx)
7431 char sym_type, sym_name[500];
7432 unsigned long long sym_addr;
7436 f = fopen("/proc/kallsyms", "r");
7439 pr_warn("failed to open /proc/kallsyms: %d\n", err);
7444 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
7445 &sym_addr, &sym_type, sym_name);
7446 if (ret == EOF && feof(f))
7449 pr_warn("failed to read kallsyms entry: %d\n", ret);
7454 err = cb(sym_addr, sym_type, sym_name, ctx);
7463 static int kallsyms_cb(unsigned long long sym_addr, char sym_type,
7464 const char *sym_name, void *ctx)
7466 struct bpf_object *obj = ctx;
7467 const struct btf_type *t;
7468 struct extern_desc *ext;
7470 ext = find_extern_by_name(obj, sym_name);
7471 if (!ext || ext->type != EXT_KSYM)
7474 t = btf__type_by_id(obj->btf, ext->btf_id);
7478 if (ext->is_set && ext->ksym.addr != sym_addr) {
7479 pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n",
7480 sym_name, ext->ksym.addr, sym_addr);
7485 ext->ksym.addr = sym_addr;
7486 pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr);
7491 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
7493 return libbpf_kallsyms_parse(kallsyms_cb, obj);
7496 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
7497 __u16 kind, struct btf **res_btf,
7498 struct module_btf **res_mod_btf)
7500 struct module_btf *mod_btf;
7504 btf = obj->btf_vmlinux;
7506 id = btf__find_by_name_kind(btf, ksym_name, kind);
7508 if (id == -ENOENT) {
7509 err = load_module_btfs(obj);
7513 for (i = 0; i < obj->btf_module_cnt; i++) {
7514 /* we assume module_btf's BTF FD is always >0 */
7515 mod_btf = &obj->btf_modules[i];
7517 id = btf__find_by_name_kind_own(btf, ksym_name, kind);
7526 *res_mod_btf = mod_btf;
7530 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
7531 struct extern_desc *ext)
7533 const struct btf_type *targ_var, *targ_type;
7534 __u32 targ_type_id, local_type_id;
7535 struct module_btf *mod_btf = NULL;
7536 const char *targ_var_name;
7537 struct btf *btf = NULL;
7540 id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
7542 if (id == -ESRCH && ext->is_weak)
7544 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
7549 /* find local type_id */
7550 local_type_id = ext->ksym.type_id;
7552 /* find target type_id */
7553 targ_var = btf__type_by_id(btf, id);
7554 targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
7555 targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
7557 err = bpf_core_types_are_compat(obj->btf, local_type_id,
7560 const struct btf_type *local_type;
7561 const char *targ_name, *local_name;
7563 local_type = btf__type_by_id(obj->btf, local_type_id);
7564 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
7565 targ_name = btf__name_by_offset(btf, targ_type->name_off);
7567 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
7568 ext->name, local_type_id,
7569 btf_kind_str(local_type), local_name, targ_type_id,
7570 btf_kind_str(targ_type), targ_name);
7575 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7576 ext->ksym.kernel_btf_id = id;
7577 pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
7578 ext->name, id, btf_kind_str(targ_var), targ_var_name);
7583 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
7584 struct extern_desc *ext)
7586 int local_func_proto_id, kfunc_proto_id, kfunc_id;
7587 struct module_btf *mod_btf = NULL;
7588 const struct btf_type *kern_func;
7589 struct btf *kern_btf = NULL;
7592 local_func_proto_id = ext->ksym.type_id;
7594 kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf);
7596 if (kfunc_id == -ESRCH && ext->is_weak)
7598 pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
7603 kern_func = btf__type_by_id(kern_btf, kfunc_id);
7604 kfunc_proto_id = kern_func->type;
7606 ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
7607 kern_btf, kfunc_proto_id);
7609 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n",
7610 ext->name, local_func_proto_id,
7611 mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id);
7615 /* set index for module BTF fd in fd_array, if unset */
7616 if (mod_btf && !mod_btf->fd_array_idx) {
7617 /* insn->off is s16 */
7618 if (obj->fd_array_cnt == INT16_MAX) {
7619 pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
7620 ext->name, mod_btf->fd_array_idx);
7623 /* Cannot use index 0 for module BTF fd */
7624 if (!obj->fd_array_cnt)
7625 obj->fd_array_cnt = 1;
7627 ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
7628 obj->fd_array_cnt + 1);
7631 mod_btf->fd_array_idx = obj->fd_array_cnt;
7632 /* we assume module BTF FD is always >0 */
7633 obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
7637 ext->ksym.kernel_btf_id = kfunc_id;
7638 ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
7639 /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data()
7640 * populates FD into ld_imm64 insn when it's used to point to kfunc.
7641 * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call.
7642 * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64.
7644 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7645 pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n",
7646 ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id);
7651 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
7653 const struct btf_type *t;
7654 struct extern_desc *ext;
7657 for (i = 0; i < obj->nr_extern; i++) {
7658 ext = &obj->externs[i];
7659 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
7662 if (obj->gen_loader) {
7664 ext->ksym.kernel_btf_obj_fd = 0;
7665 ext->ksym.kernel_btf_id = 0;
7668 t = btf__type_by_id(obj->btf, ext->btf_id);
7670 err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
7672 err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
7679 static int bpf_object__resolve_externs(struct bpf_object *obj,
7680 const char *extra_kconfig)
7682 bool need_config = false, need_kallsyms = false;
7683 bool need_vmlinux_btf = false;
7684 struct extern_desc *ext;
7685 void *kcfg_data = NULL;
7688 if (obj->nr_extern == 0)
7691 if (obj->kconfig_map_idx >= 0)
7692 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
7694 for (i = 0; i < obj->nr_extern; i++) {
7695 ext = &obj->externs[i];
7697 if (ext->type == EXT_KSYM) {
7698 if (ext->ksym.type_id)
7699 need_vmlinux_btf = true;
7701 need_kallsyms = true;
7703 } else if (ext->type == EXT_KCFG) {
7704 void *ext_ptr = kcfg_data + ext->kcfg.data_off;
7707 /* Kconfig externs need actual /proc/config.gz */
7708 if (str_has_pfx(ext->name, "CONFIG_")) {
7713 /* Virtual kcfg externs are customly handled by libbpf */
7714 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
7715 value = get_kernel_version();
7717 pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name);
7720 } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) {
7721 value = kernel_supports(obj, FEAT_BPF_COOKIE);
7722 } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) {
7723 value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER);
7724 } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) {
7725 /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed
7726 * __kconfig externs, where LINUX_ ones are virtual and filled out
7727 * customly by libbpf (their values don't come from Kconfig).
7728 * If LINUX_xxx variable is not recognized by libbpf, but is marked
7729 * __weak, it defaults to zero value, just like for CONFIG_xxx
7732 pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name);
7736 err = set_kcfg_value_num(ext, ext_ptr, value);
7739 pr_debug("extern (kcfg) '%s': set to 0x%llx\n",
7740 ext->name, (long long)value);
7742 pr_warn("extern '%s': unrecognized extern kind\n", ext->name);
7746 if (need_config && extra_kconfig) {
7747 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
7750 need_config = false;
7751 for (i = 0; i < obj->nr_extern; i++) {
7752 ext = &obj->externs[i];
7753 if (ext->type == EXT_KCFG && !ext->is_set) {
7760 err = bpf_object__read_kconfig_file(obj, kcfg_data);
7764 if (need_kallsyms) {
7765 err = bpf_object__read_kallsyms_file(obj);
7769 if (need_vmlinux_btf) {
7770 err = bpf_object__resolve_ksyms_btf_id(obj);
7774 for (i = 0; i < obj->nr_extern; i++) {
7775 ext = &obj->externs[i];
7777 if (!ext->is_set && !ext->is_weak) {
7778 pr_warn("extern '%s' (strong): not resolved\n", ext->name);
7780 } else if (!ext->is_set) {
7781 pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n",
7789 static void bpf_map_prepare_vdata(const struct bpf_map *map)
7791 struct bpf_struct_ops *st_ops;
7794 st_ops = map->st_ops;
7795 for (i = 0; i < btf_vlen(st_ops->type); i++) {
7796 struct bpf_program *prog = st_ops->progs[i];
7803 prog_fd = bpf_program__fd(prog);
7804 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
7805 *(unsigned long *)kern_data = prog_fd;
7809 static int bpf_object_prepare_struct_ops(struct bpf_object *obj)
7813 for (i = 0; i < obj->nr_maps; i++)
7814 if (bpf_map__is_struct_ops(&obj->maps[i]))
7815 bpf_map_prepare_vdata(&obj->maps[i]);
7820 static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
7825 return libbpf_err(-EINVAL);
7828 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
7829 return libbpf_err(-EINVAL);
7832 if (obj->gen_loader)
7833 bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
7835 err = bpf_object__probe_loading(obj);
7836 err = err ? : bpf_object__load_vmlinux_btf(obj, false);
7837 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
7838 err = err ? : bpf_object__sanitize_and_load_btf(obj);
7839 err = err ? : bpf_object__sanitize_maps(obj);
7840 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
7841 err = err ? : bpf_object__create_maps(obj);
7842 err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
7843 err = err ? : bpf_object__load_progs(obj, extra_log_level);
7844 err = err ? : bpf_object_init_prog_arrays(obj);
7845 err = err ? : bpf_object_prepare_struct_ops(obj);
7847 if (obj->gen_loader) {
7850 btf__set_fd(obj->btf, -1);
7851 for (i = 0; i < obj->nr_maps; i++)
7852 obj->maps[i].fd = -1;
7854 err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
7857 /* clean up fd_array */
7858 zfree(&obj->fd_array);
7860 /* clean up module BTFs */
7861 for (i = 0; i < obj->btf_module_cnt; i++) {
7862 close(obj->btf_modules[i].fd);
7863 btf__free(obj->btf_modules[i].btf);
7864 free(obj->btf_modules[i].name);
7866 free(obj->btf_modules);
7868 /* clean up vmlinux BTF */
7869 btf__free(obj->btf_vmlinux);
7870 obj->btf_vmlinux = NULL;
7872 obj->loaded = true; /* doesn't matter if successfully or not */
7879 /* unpin any maps that were auto-pinned during load */
7880 for (i = 0; i < obj->nr_maps; i++)
7881 if (obj->maps[i].pinned && !obj->maps[i].reused)
7882 bpf_map__unpin(&obj->maps[i], NULL);
7884 bpf_object_unload(obj);
7885 pr_warn("failed to load object '%s'\n", obj->path);
7886 return libbpf_err(err);
7889 int bpf_object__load(struct bpf_object *obj)
7891 return bpf_object_load(obj, 0, NULL);
7894 static int make_parent_dir(const char *path)
7896 char *cp, errmsg[STRERR_BUFSIZE];
7900 dname = strdup(path);
7904 dir = dirname(dname);
7905 if (mkdir(dir, 0700) && errno != EEXIST)
7910 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7911 pr_warn("failed to mkdir %s: %s\n", path, cp);
7916 static int check_path(const char *path)
7918 char *cp, errmsg[STRERR_BUFSIZE];
7919 struct statfs st_fs;
7926 dname = strdup(path);
7930 dir = dirname(dname);
7931 if (statfs(dir, &st_fs)) {
7932 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7933 pr_warn("failed to statfs %s: %s\n", dir, cp);
7938 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
7939 pr_warn("specified path %s is not on BPF FS\n", path);
7946 int bpf_program__pin(struct bpf_program *prog, const char *path)
7948 char *cp, errmsg[STRERR_BUFSIZE];
7952 pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name);
7953 return libbpf_err(-EINVAL);
7956 err = make_parent_dir(path);
7958 return libbpf_err(err);
7960 err = check_path(path);
7962 return libbpf_err(err);
7964 if (bpf_obj_pin(prog->fd, path)) {
7966 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
7967 pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, cp);
7968 return libbpf_err(err);
7971 pr_debug("prog '%s': pinned at '%s'\n", prog->name, path);
7975 int bpf_program__unpin(struct bpf_program *prog, const char *path)
7980 pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name);
7981 return libbpf_err(-EINVAL);
7984 err = check_path(path);
7986 return libbpf_err(err);
7990 return libbpf_err(-errno);
7992 pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path);
7996 int bpf_map__pin(struct bpf_map *map, const char *path)
7998 char *cp, errmsg[STRERR_BUFSIZE];
8002 pr_warn("invalid map pointer\n");
8003 return libbpf_err(-EINVAL);
8006 if (map->pin_path) {
8007 if (path && strcmp(path, map->pin_path)) {
8008 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8009 bpf_map__name(map), map->pin_path, path);
8010 return libbpf_err(-EINVAL);
8011 } else if (map->pinned) {
8012 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
8013 bpf_map__name(map), map->pin_path);
8018 pr_warn("missing a path to pin map '%s' at\n",
8019 bpf_map__name(map));
8020 return libbpf_err(-EINVAL);
8021 } else if (map->pinned) {
8022 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
8023 return libbpf_err(-EEXIST);
8026 map->pin_path = strdup(path);
8027 if (!map->pin_path) {
8033 err = make_parent_dir(map->pin_path);
8035 return libbpf_err(err);
8037 err = check_path(map->pin_path);
8039 return libbpf_err(err);
8041 if (bpf_obj_pin(map->fd, map->pin_path)) {
8047 pr_debug("pinned map '%s'\n", map->pin_path);
8052 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
8053 pr_warn("failed to pin map: %s\n", cp);
8054 return libbpf_err(err);
8057 int bpf_map__unpin(struct bpf_map *map, const char *path)
8062 pr_warn("invalid map pointer\n");
8063 return libbpf_err(-EINVAL);
8066 if (map->pin_path) {
8067 if (path && strcmp(path, map->pin_path)) {
8068 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8069 bpf_map__name(map), map->pin_path, path);
8070 return libbpf_err(-EINVAL);
8072 path = map->pin_path;
8074 pr_warn("no path to unpin map '%s' from\n",
8075 bpf_map__name(map));
8076 return libbpf_err(-EINVAL);
8079 err = check_path(path);
8081 return libbpf_err(err);
8085 return libbpf_err(-errno);
8087 map->pinned = false;
8088 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
8093 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
8100 return libbpf_err(-errno);
8103 free(map->pin_path);
8104 map->pin_path = new;
8108 __alias(bpf_map__pin_path)
8109 const char *bpf_map__get_pin_path(const struct bpf_map *map);
8111 const char *bpf_map__pin_path(const struct bpf_map *map)
8113 return map->pin_path;
8116 bool bpf_map__is_pinned(const struct bpf_map *map)
8121 static void sanitize_pin_path(char *s)
8123 /* bpffs disallows periods in path names */
8131 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
8133 struct bpf_map *map;
8137 return libbpf_err(-ENOENT);
8140 pr_warn("object not yet loaded; load it first\n");
8141 return libbpf_err(-ENOENT);
8144 bpf_object__for_each_map(map, obj) {
8145 char *pin_path = NULL;
8148 if (!map->autocreate)
8152 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8154 goto err_unpin_maps;
8155 sanitize_pin_path(buf);
8157 } else if (!map->pin_path) {
8161 err = bpf_map__pin(map, pin_path);
8163 goto err_unpin_maps;
8169 while ((map = bpf_object__prev_map(obj, map))) {
8173 bpf_map__unpin(map, NULL);
8176 return libbpf_err(err);
8179 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
8181 struct bpf_map *map;
8185 return libbpf_err(-ENOENT);
8187 bpf_object__for_each_map(map, obj) {
8188 char *pin_path = NULL;
8192 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8194 return libbpf_err(err);
8195 sanitize_pin_path(buf);
8197 } else if (!map->pin_path) {
8201 err = bpf_map__unpin(map, pin_path);
8203 return libbpf_err(err);
8209 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
8211 struct bpf_program *prog;
8216 return libbpf_err(-ENOENT);
8219 pr_warn("object not yet loaded; load it first\n");
8220 return libbpf_err(-ENOENT);
8223 bpf_object__for_each_program(prog, obj) {
8224 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8226 goto err_unpin_programs;
8228 err = bpf_program__pin(prog, buf);
8230 goto err_unpin_programs;
8236 while ((prog = bpf_object__prev_program(obj, prog))) {
8237 if (pathname_concat(buf, sizeof(buf), path, prog->name))
8240 bpf_program__unpin(prog, buf);
8243 return libbpf_err(err);
8246 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8248 struct bpf_program *prog;
8252 return libbpf_err(-ENOENT);
8254 bpf_object__for_each_program(prog, obj) {
8257 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8259 return libbpf_err(err);
8261 err = bpf_program__unpin(prog, buf);
8263 return libbpf_err(err);
8269 int bpf_object__pin(struct bpf_object *obj, const char *path)
8273 err = bpf_object__pin_maps(obj, path);
8275 return libbpf_err(err);
8277 err = bpf_object__pin_programs(obj, path);
8279 bpf_object__unpin_maps(obj, path);
8280 return libbpf_err(err);
8286 static void bpf_map__destroy(struct bpf_map *map)
8288 if (map->inner_map) {
8289 bpf_map__destroy(map->inner_map);
8290 zfree(&map->inner_map);
8293 zfree(&map->init_slots);
8294 map->init_slots_sz = 0;
8297 munmap(map->mmaped, bpf_map_mmap_sz(map));
8302 zfree(&map->st_ops->data);
8303 zfree(&map->st_ops->progs);
8304 zfree(&map->st_ops->kern_func_off);
8305 zfree(&map->st_ops);
8309 zfree(&map->real_name);
8310 zfree(&map->pin_path);
8316 void bpf_object__close(struct bpf_object *obj)
8320 if (IS_ERR_OR_NULL(obj))
8323 usdt_manager_free(obj->usdt_man);
8324 obj->usdt_man = NULL;
8326 bpf_gen__free(obj->gen_loader);
8327 bpf_object__elf_finish(obj);
8328 bpf_object_unload(obj);
8329 btf__free(obj->btf);
8330 btf_ext__free(obj->btf_ext);
8332 for (i = 0; i < obj->nr_maps; i++)
8333 bpf_map__destroy(&obj->maps[i]);
8335 zfree(&obj->btf_custom_path);
8336 zfree(&obj->kconfig);
8337 zfree(&obj->externs);
8343 if (obj->programs && obj->nr_programs) {
8344 for (i = 0; i < obj->nr_programs; i++)
8345 bpf_program__exit(&obj->programs[i]);
8347 zfree(&obj->programs);
8352 const char *bpf_object__name(const struct bpf_object *obj)
8354 return obj ? obj->name : libbpf_err_ptr(-EINVAL);
8357 unsigned int bpf_object__kversion(const struct bpf_object *obj)
8359 return obj ? obj->kern_version : 0;
8362 struct btf *bpf_object__btf(const struct bpf_object *obj)
8364 return obj ? obj->btf : NULL;
8367 int bpf_object__btf_fd(const struct bpf_object *obj)
8369 return obj->btf ? btf__fd(obj->btf) : -1;
8372 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
8375 return libbpf_err(-EINVAL);
8377 obj->kern_version = kern_version;
8382 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
8384 struct bpf_gen *gen;
8388 if (!OPTS_VALID(opts, gen_loader_opts))
8390 gen = calloc(sizeof(*gen), 1);
8394 obj->gen_loader = gen;
8398 static struct bpf_program *
8399 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
8402 size_t nr_programs = obj->nr_programs;
8409 /* Iter from the beginning */
8410 return forward ? &obj->programs[0] :
8411 &obj->programs[nr_programs - 1];
8413 if (p->obj != obj) {
8414 pr_warn("error: program handler doesn't match object\n");
8415 return errno = EINVAL, NULL;
8418 idx = (p - obj->programs) + (forward ? 1 : -1);
8419 if (idx >= obj->nr_programs || idx < 0)
8421 return &obj->programs[idx];
8424 struct bpf_program *
8425 bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
8427 struct bpf_program *prog = prev;
8430 prog = __bpf_program__iter(prog, obj, true);
8431 } while (prog && prog_is_subprog(obj, prog));
8436 struct bpf_program *
8437 bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
8439 struct bpf_program *prog = next;
8442 prog = __bpf_program__iter(prog, obj, false);
8443 } while (prog && prog_is_subprog(obj, prog));
8448 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
8450 prog->prog_ifindex = ifindex;
8453 const char *bpf_program__name(const struct bpf_program *prog)
8458 const char *bpf_program__section_name(const struct bpf_program *prog)
8460 return prog->sec_name;
8463 bool bpf_program__autoload(const struct bpf_program *prog)
8465 return prog->autoload;
8468 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
8470 if (prog->obj->loaded)
8471 return libbpf_err(-EINVAL);
8473 prog->autoload = autoload;
8477 bool bpf_program__autoattach(const struct bpf_program *prog)
8479 return prog->autoattach;
8482 void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach)
8484 prog->autoattach = autoattach;
8487 const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
8492 size_t bpf_program__insn_cnt(const struct bpf_program *prog)
8494 return prog->insns_cnt;
8497 int bpf_program__set_insns(struct bpf_program *prog,
8498 struct bpf_insn *new_insns, size_t new_insn_cnt)
8500 struct bpf_insn *insns;
8502 if (prog->obj->loaded)
8505 insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns));
8507 pr_warn("prog '%s': failed to realloc prog code\n", prog->name);
8510 memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns));
8512 prog->insns = insns;
8513 prog->insns_cnt = new_insn_cnt;
8517 int bpf_program__fd(const struct bpf_program *prog)
8520 return libbpf_err(-EINVAL);
8523 return libbpf_err(-ENOENT);
8528 __alias(bpf_program__type)
8529 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
8531 enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
8536 int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
8538 if (prog->obj->loaded)
8539 return libbpf_err(-EBUSY);
8542 prog->sec_def = NULL;
8546 __alias(bpf_program__expected_attach_type)
8547 enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
8549 enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
8551 return prog->expected_attach_type;
8554 int bpf_program__set_expected_attach_type(struct bpf_program *prog,
8555 enum bpf_attach_type type)
8557 if (prog->obj->loaded)
8558 return libbpf_err(-EBUSY);
8560 prog->expected_attach_type = type;
8564 __u32 bpf_program__flags(const struct bpf_program *prog)
8566 return prog->prog_flags;
8569 int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
8571 if (prog->obj->loaded)
8572 return libbpf_err(-EBUSY);
8574 prog->prog_flags = flags;
8578 __u32 bpf_program__log_level(const struct bpf_program *prog)
8580 return prog->log_level;
8583 int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
8585 if (prog->obj->loaded)
8586 return libbpf_err(-EBUSY);
8588 prog->log_level = log_level;
8592 const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
8594 *log_size = prog->log_size;
8595 return prog->log_buf;
8598 int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
8600 if (log_size && !log_buf)
8602 if (prog->log_size > UINT_MAX)
8604 if (prog->obj->loaded)
8607 prog->log_buf = log_buf;
8608 prog->log_size = log_size;
8612 #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
8613 .sec = (char *)sec_pfx, \
8614 .prog_type = BPF_PROG_TYPE_##ptype, \
8615 .expected_attach_type = atype, \
8616 .cookie = (long)(flags), \
8617 .prog_prepare_load_fn = libbpf_prepare_prog_load, \
8621 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8622 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8623 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8624 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8625 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8626 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8627 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8628 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8629 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8630 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8632 static const struct bpf_sec_def section_defs[] = {
8633 SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE),
8634 SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE),
8635 SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE),
8636 SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
8637 SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
8638 SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
8639 SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
8640 SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
8641 SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
8642 SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8643 SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8644 SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
8645 SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
8646 SEC_DEF("usdt+", KPROBE, 0, SEC_NONE, attach_usdt),
8647 SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE),
8648 SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE),
8649 SEC_DEF("action", SCHED_ACT, 0, SEC_NONE),
8650 SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp),
8651 SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp),
8652 SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8653 SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8654 SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8655 SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8656 SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
8657 SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
8658 SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
8659 SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
8660 SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8661 SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8662 SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8663 SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace),
8664 SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
8665 SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
8666 SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF),
8667 SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
8668 SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
8669 SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE),
8670 SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
8671 SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
8672 SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
8673 SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
8674 SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS),
8675 SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT),
8676 SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE),
8677 SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE),
8678 SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE),
8679 SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE),
8680 SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE),
8681 SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT),
8682 SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT),
8683 SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT),
8684 SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE),
8685 SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT),
8686 SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT),
8687 SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT),
8688 SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT),
8689 SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT),
8690 SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE),
8691 SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE),
8692 SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE),
8693 SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT),
8694 SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE),
8695 SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE),
8696 SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE),
8697 SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE),
8698 SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE),
8699 SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE),
8700 SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE),
8701 SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE),
8702 SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE),
8703 SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE),
8704 SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE),
8705 SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE),
8706 SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE),
8707 SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE),
8708 SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE),
8709 SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE),
8710 SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE),
8711 SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT),
8712 SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE),
8713 SEC_DEF("struct_ops.s+", STRUCT_OPS, 0, SEC_SLEEPABLE),
8714 SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE),
8715 SEC_DEF("netfilter", NETFILTER, 0, SEC_NONE),
8718 static size_t custom_sec_def_cnt;
8719 static struct bpf_sec_def *custom_sec_defs;
8720 static struct bpf_sec_def custom_fallback_def;
8721 static bool has_custom_fallback_def;
8723 static int last_custom_sec_def_handler_id;
8725 int libbpf_register_prog_handler(const char *sec,
8726 enum bpf_prog_type prog_type,
8727 enum bpf_attach_type exp_attach_type,
8728 const struct libbpf_prog_handler_opts *opts)
8730 struct bpf_sec_def *sec_def;
8732 if (!OPTS_VALID(opts, libbpf_prog_handler_opts))
8733 return libbpf_err(-EINVAL);
8735 if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */
8736 return libbpf_err(-E2BIG);
8739 sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1,
8742 return libbpf_err(-ENOMEM);
8744 custom_sec_defs = sec_def;
8745 sec_def = &custom_sec_defs[custom_sec_def_cnt];
8747 if (has_custom_fallback_def)
8748 return libbpf_err(-EBUSY);
8750 sec_def = &custom_fallback_def;
8753 sec_def->sec = sec ? strdup(sec) : NULL;
8754 if (sec && !sec_def->sec)
8755 return libbpf_err(-ENOMEM);
8757 sec_def->prog_type = prog_type;
8758 sec_def->expected_attach_type = exp_attach_type;
8759 sec_def->cookie = OPTS_GET(opts, cookie, 0);
8761 sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL);
8762 sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL);
8763 sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL);
8765 sec_def->handler_id = ++last_custom_sec_def_handler_id;
8768 custom_sec_def_cnt++;
8770 has_custom_fallback_def = true;
8772 return sec_def->handler_id;
8775 int libbpf_unregister_prog_handler(int handler_id)
8777 struct bpf_sec_def *sec_defs;
8780 if (handler_id <= 0)
8781 return libbpf_err(-EINVAL);
8783 if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) {
8784 memset(&custom_fallback_def, 0, sizeof(custom_fallback_def));
8785 has_custom_fallback_def = false;
8789 for (i = 0; i < custom_sec_def_cnt; i++) {
8790 if (custom_sec_defs[i].handler_id == handler_id)
8794 if (i == custom_sec_def_cnt)
8795 return libbpf_err(-ENOENT);
8797 free(custom_sec_defs[i].sec);
8798 for (i = i + 1; i < custom_sec_def_cnt; i++)
8799 custom_sec_defs[i - 1] = custom_sec_defs[i];
8800 custom_sec_def_cnt--;
8802 /* try to shrink the array, but it's ok if we couldn't */
8803 sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs));
8805 custom_sec_defs = sec_defs;
8810 static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name)
8812 size_t len = strlen(sec_def->sec);
8814 /* "type/" always has to have proper SEC("type/extras") form */
8815 if (sec_def->sec[len - 1] == '/') {
8816 if (str_has_pfx(sec_name, sec_def->sec))
8821 /* "type+" means it can be either exact SEC("type") or
8822 * well-formed SEC("type/extras") with proper '/' separator
8824 if (sec_def->sec[len - 1] == '+') {
8826 /* not even a prefix */
8827 if (strncmp(sec_name, sec_def->sec, len) != 0)
8829 /* exact match or has '/' separator */
8830 if (sec_name[len] == '\0' || sec_name[len] == '/')
8835 return strcmp(sec_name, sec_def->sec) == 0;
8838 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
8840 const struct bpf_sec_def *sec_def;
8843 n = custom_sec_def_cnt;
8844 for (i = 0; i < n; i++) {
8845 sec_def = &custom_sec_defs[i];
8846 if (sec_def_matches(sec_def, sec_name))
8850 n = ARRAY_SIZE(section_defs);
8851 for (i = 0; i < n; i++) {
8852 sec_def = §ion_defs[i];
8853 if (sec_def_matches(sec_def, sec_name))
8857 if (has_custom_fallback_def)
8858 return &custom_fallback_def;
8863 #define MAX_TYPE_NAME_SIZE 32
8865 static char *libbpf_get_type_names(bool attach_type)
8867 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
8875 /* Forge string buf with all available names */
8876 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8877 const struct bpf_sec_def *sec_def = §ion_defs[i];
8880 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
8883 if (!(sec_def->cookie & SEC_ATTACHABLE))
8887 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
8892 strcat(buf, section_defs[i].sec);
8898 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
8899 enum bpf_attach_type *expected_attach_type)
8901 const struct bpf_sec_def *sec_def;
8905 return libbpf_err(-EINVAL);
8907 sec_def = find_sec_def(name);
8909 *prog_type = sec_def->prog_type;
8910 *expected_attach_type = sec_def->expected_attach_type;
8914 pr_debug("failed to guess program type from ELF section '%s'\n", name);
8915 type_names = libbpf_get_type_names(false);
8916 if (type_names != NULL) {
8917 pr_debug("supported section(type) names are:%s\n", type_names);
8921 return libbpf_err(-ESRCH);
8924 const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t)
8926 if (t < 0 || t >= ARRAY_SIZE(attach_type_name))
8929 return attach_type_name[t];
8932 const char *libbpf_bpf_link_type_str(enum bpf_link_type t)
8934 if (t < 0 || t >= ARRAY_SIZE(link_type_name))
8937 return link_type_name[t];
8940 const char *libbpf_bpf_map_type_str(enum bpf_map_type t)
8942 if (t < 0 || t >= ARRAY_SIZE(map_type_name))
8945 return map_type_name[t];
8948 const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t)
8950 if (t < 0 || t >= ARRAY_SIZE(prog_type_name))
8953 return prog_type_name[t];
8956 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
8960 struct bpf_map *map;
8963 for (i = 0; i < obj->nr_maps; i++) {
8964 map = &obj->maps[i];
8965 if (!bpf_map__is_struct_ops(map))
8967 if (map->sec_idx == sec_idx &&
8968 map->sec_offset <= offset &&
8969 offset - map->sec_offset < map->def.value_size)
8976 /* Collect the reloc from ELF and populate the st_ops->progs[] */
8977 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
8978 Elf64_Shdr *shdr, Elf_Data *data)
8980 const struct btf_member *member;
8981 struct bpf_struct_ops *st_ops;
8982 struct bpf_program *prog;
8983 unsigned int shdr_idx;
8984 const struct btf *btf;
8985 struct bpf_map *map;
8986 unsigned int moff, insn_idx;
8994 nrels = shdr->sh_size / shdr->sh_entsize;
8995 for (i = 0; i < nrels; i++) {
8996 rel = elf_rel_by_idx(data, i);
8998 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
8999 return -LIBBPF_ERRNO__FORMAT;
9002 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
9004 pr_warn("struct_ops reloc: symbol %zx not found\n",
9005 (size_t)ELF64_R_SYM(rel->r_info));
9006 return -LIBBPF_ERRNO__FORMAT;
9009 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
9010 map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset);
9012 pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
9013 (size_t)rel->r_offset);
9017 moff = rel->r_offset - map->sec_offset;
9018 shdr_idx = sym->st_shndx;
9019 st_ops = map->st_ops;
9020 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
9022 (long long)(rel->r_info >> 32),
9023 (long long)sym->st_value,
9024 shdr_idx, (size_t)rel->r_offset,
9025 map->sec_offset, sym->st_name, name);
9027 if (shdr_idx >= SHN_LORESERVE) {
9028 pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
9029 map->name, (size_t)rel->r_offset, shdr_idx);
9030 return -LIBBPF_ERRNO__RELOC;
9032 if (sym->st_value % BPF_INSN_SZ) {
9033 pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
9034 map->name, (unsigned long long)sym->st_value);
9035 return -LIBBPF_ERRNO__FORMAT;
9037 insn_idx = sym->st_value / BPF_INSN_SZ;
9039 member = find_member_by_offset(st_ops->type, moff * 8);
9041 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
9045 member_idx = member - btf_members(st_ops->type);
9046 name = btf__name_by_offset(btf, member->name_off);
9048 if (!resolve_func_ptr(btf, member->type, NULL)) {
9049 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
9054 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
9056 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
9057 map->name, shdr_idx, name);
9061 /* prevent the use of BPF prog with invalid type */
9062 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
9063 pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
9064 map->name, prog->name);
9068 /* if we haven't yet processed this BPF program, record proper
9069 * attach_btf_id and member_idx
9071 if (!prog->attach_btf_id) {
9072 prog->attach_btf_id = st_ops->type_id;
9073 prog->expected_attach_type = member_idx;
9076 /* struct_ops BPF prog can be re-used between multiple
9077 * .struct_ops & .struct_ops.link as long as it's the
9078 * same struct_ops struct definition and the same
9079 * function pointer field
9081 if (prog->attach_btf_id != st_ops->type_id ||
9082 prog->expected_attach_type != member_idx) {
9083 pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
9084 map->name, prog->name, prog->sec_name, prog->type,
9085 prog->attach_btf_id, prog->expected_attach_type, name);
9089 st_ops->progs[member_idx] = prog;
9095 #define BTF_TRACE_PREFIX "btf_trace_"
9096 #define BTF_LSM_PREFIX "bpf_lsm_"
9097 #define BTF_ITER_PREFIX "bpf_iter_"
9098 #define BTF_MAX_NAME_SIZE 128
9100 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
9101 const char **prefix, int *kind)
9103 switch (attach_type) {
9104 case BPF_TRACE_RAW_TP:
9105 *prefix = BTF_TRACE_PREFIX;
9106 *kind = BTF_KIND_TYPEDEF;
9109 case BPF_LSM_CGROUP:
9110 *prefix = BTF_LSM_PREFIX;
9111 *kind = BTF_KIND_FUNC;
9113 case BPF_TRACE_ITER:
9114 *prefix = BTF_ITER_PREFIX;
9115 *kind = BTF_KIND_FUNC;
9119 *kind = BTF_KIND_FUNC;
9123 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
9124 const char *name, __u32 kind)
9126 char btf_type_name[BTF_MAX_NAME_SIZE];
9129 ret = snprintf(btf_type_name, sizeof(btf_type_name),
9130 "%s%s", prefix, name);
9131 /* snprintf returns the number of characters written excluding the
9132 * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
9133 * indicates truncation.
9135 if (ret < 0 || ret >= sizeof(btf_type_name))
9136 return -ENAMETOOLONG;
9137 return btf__find_by_name_kind(btf, btf_type_name, kind);
9140 static inline int find_attach_btf_id(struct btf *btf, const char *name,
9141 enum bpf_attach_type attach_type)
9146 btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
9147 return find_btf_by_prefix_kind(btf, prefix, name, kind);
9150 int libbpf_find_vmlinux_btf_id(const char *name,
9151 enum bpf_attach_type attach_type)
9156 btf = btf__load_vmlinux_btf();
9157 err = libbpf_get_error(btf);
9159 pr_warn("vmlinux BTF is not found\n");
9160 return libbpf_err(err);
9163 err = find_attach_btf_id(btf, name, attach_type);
9165 pr_warn("%s is not found in vmlinux BTF\n", name);
9168 return libbpf_err(err);
9171 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
9173 struct bpf_prog_info info;
9174 __u32 info_len = sizeof(info);
9178 memset(&info, 0, info_len);
9179 err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len);
9181 pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %d\n",
9182 attach_prog_fd, err);
9188 pr_warn("The target program doesn't have BTF\n");
9191 btf = btf__load_from_kernel_by_id(info.btf_id);
9192 err = libbpf_get_error(btf);
9194 pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err);
9197 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
9200 pr_warn("%s is not found in prog's BTF\n", name);
9207 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
9208 enum bpf_attach_type attach_type,
9209 int *btf_obj_fd, int *btf_type_id)
9213 ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
9215 *btf_obj_fd = 0; /* vmlinux BTF */
9222 ret = load_module_btfs(obj);
9226 for (i = 0; i < obj->btf_module_cnt; i++) {
9227 const struct module_btf *mod = &obj->btf_modules[i];
9229 ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
9231 *btf_obj_fd = mod->fd;
9244 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
9245 int *btf_obj_fd, int *btf_type_id)
9247 enum bpf_attach_type attach_type = prog->expected_attach_type;
9248 __u32 attach_prog_fd = prog->attach_prog_fd;
9251 /* BPF program's BTF ID */
9252 if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) {
9253 if (!attach_prog_fd) {
9254 pr_warn("prog '%s': attach program FD is not set\n", prog->name);
9257 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
9259 pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
9260 prog->name, attach_prog_fd, attach_name, err);
9268 /* kernel/module BTF ID */
9269 if (prog->obj->gen_loader) {
9270 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
9274 err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
9277 pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %d\n",
9278 prog->name, attach_name, err);
9284 int libbpf_attach_type_by_name(const char *name,
9285 enum bpf_attach_type *attach_type)
9288 const struct bpf_sec_def *sec_def;
9291 return libbpf_err(-EINVAL);
9293 sec_def = find_sec_def(name);
9295 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
9296 type_names = libbpf_get_type_names(true);
9297 if (type_names != NULL) {
9298 pr_debug("attachable section(type) names are:%s\n", type_names);
9302 return libbpf_err(-EINVAL);
9305 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9306 return libbpf_err(-EINVAL);
9307 if (!(sec_def->cookie & SEC_ATTACHABLE))
9308 return libbpf_err(-EINVAL);
9310 *attach_type = sec_def->expected_attach_type;
9314 int bpf_map__fd(const struct bpf_map *map)
9316 return map ? map->fd : libbpf_err(-EINVAL);
9319 static bool map_uses_real_name(const struct bpf_map *map)
9321 /* Since libbpf started to support custom .data.* and .rodata.* maps,
9322 * their user-visible name differs from kernel-visible name. Users see
9323 * such map's corresponding ELF section name as a map name.
9324 * This check distinguishes .data/.rodata from .data.* and .rodata.*
9325 * maps to know which name has to be returned to the user.
9327 if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
9329 if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
9334 const char *bpf_map__name(const struct bpf_map *map)
9339 if (map_uses_real_name(map))
9340 return map->real_name;
9345 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
9347 return map->def.type;
9350 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
9353 return libbpf_err(-EBUSY);
9354 map->def.type = type;
9358 __u32 bpf_map__map_flags(const struct bpf_map *map)
9360 return map->def.map_flags;
9363 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
9366 return libbpf_err(-EBUSY);
9367 map->def.map_flags = flags;
9371 __u64 bpf_map__map_extra(const struct bpf_map *map)
9373 return map->map_extra;
9376 int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
9379 return libbpf_err(-EBUSY);
9380 map->map_extra = map_extra;
9384 __u32 bpf_map__numa_node(const struct bpf_map *map)
9386 return map->numa_node;
9389 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
9392 return libbpf_err(-EBUSY);
9393 map->numa_node = numa_node;
9397 __u32 bpf_map__key_size(const struct bpf_map *map)
9399 return map->def.key_size;
9402 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
9405 return libbpf_err(-EBUSY);
9406 map->def.key_size = size;
9410 __u32 bpf_map__value_size(const struct bpf_map *map)
9412 return map->def.value_size;
9415 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
9418 return libbpf_err(-EBUSY);
9419 map->def.value_size = size;
9423 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
9425 return map ? map->btf_key_type_id : 0;
9428 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
9430 return map ? map->btf_value_type_id : 0;
9433 int bpf_map__set_initial_value(struct bpf_map *map,
9434 const void *data, size_t size)
9436 if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
9437 size != map->def.value_size || map->fd >= 0)
9438 return libbpf_err(-EINVAL);
9440 memcpy(map->mmaped, data, size);
9444 const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize)
9448 *psize = map->def.value_size;
9452 bool bpf_map__is_internal(const struct bpf_map *map)
9454 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
9457 __u32 bpf_map__ifindex(const struct bpf_map *map)
9459 return map->map_ifindex;
9462 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
9465 return libbpf_err(-EBUSY);
9466 map->map_ifindex = ifindex;
9470 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
9472 if (!bpf_map_type__is_map_in_map(map->def.type)) {
9473 pr_warn("error: unsupported map type\n");
9474 return libbpf_err(-EINVAL);
9476 if (map->inner_map_fd != -1) {
9477 pr_warn("error: inner_map_fd already specified\n");
9478 return libbpf_err(-EINVAL);
9480 if (map->inner_map) {
9481 bpf_map__destroy(map->inner_map);
9482 zfree(&map->inner_map);
9484 map->inner_map_fd = fd;
9488 static struct bpf_map *
9489 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9492 struct bpf_map *s, *e;
9494 if (!obj || !obj->maps)
9495 return errno = EINVAL, NULL;
9498 e = obj->maps + obj->nr_maps;
9500 if ((m < s) || (m >= e)) {
9501 pr_warn("error in %s: map handler doesn't belong to object\n",
9503 return errno = EINVAL, NULL;
9506 idx = (m - obj->maps) + i;
9507 if (idx >= obj->nr_maps || idx < 0)
9509 return &obj->maps[idx];
9513 bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
9518 return __bpf_map__iter(prev, obj, 1);
9522 bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
9527 return obj->maps + obj->nr_maps - 1;
9530 return __bpf_map__iter(next, obj, -1);
9534 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
9536 struct bpf_map *pos;
9538 bpf_object__for_each_map(pos, obj) {
9539 /* if it's a special internal map name (which always starts
9540 * with dot) then check if that special name matches the
9541 * real map name (ELF section name)
9543 if (name[0] == '.') {
9544 if (pos->real_name && strcmp(pos->real_name, name) == 0)
9548 /* otherwise map name has to be an exact match */
9549 if (map_uses_real_name(pos)) {
9550 if (strcmp(pos->real_name, name) == 0)
9554 if (strcmp(pos->name, name) == 0)
9557 return errno = ENOENT, NULL;
9561 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
9563 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
9566 static int validate_map_op(const struct bpf_map *map, size_t key_sz,
9567 size_t value_sz, bool check_value_sz)
9572 if (map->def.key_size != key_sz) {
9573 pr_warn("map '%s': unexpected key size %zu provided, expected %u\n",
9574 map->name, key_sz, map->def.key_size);
9578 if (!check_value_sz)
9581 switch (map->def.type) {
9582 case BPF_MAP_TYPE_PERCPU_ARRAY:
9583 case BPF_MAP_TYPE_PERCPU_HASH:
9584 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
9585 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: {
9586 int num_cpu = libbpf_num_possible_cpus();
9587 size_t elem_sz = roundup(map->def.value_size, 8);
9589 if (value_sz != num_cpu * elem_sz) {
9590 pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n",
9591 map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz);
9597 if (map->def.value_size != value_sz) {
9598 pr_warn("map '%s': unexpected value size %zu provided, expected %u\n",
9599 map->name, value_sz, map->def.value_size);
9607 int bpf_map__lookup_elem(const struct bpf_map *map,
9608 const void *key, size_t key_sz,
9609 void *value, size_t value_sz, __u64 flags)
9613 err = validate_map_op(map, key_sz, value_sz, true);
9615 return libbpf_err(err);
9617 return bpf_map_lookup_elem_flags(map->fd, key, value, flags);
9620 int bpf_map__update_elem(const struct bpf_map *map,
9621 const void *key, size_t key_sz,
9622 const void *value, size_t value_sz, __u64 flags)
9626 err = validate_map_op(map, key_sz, value_sz, true);
9628 return libbpf_err(err);
9630 return bpf_map_update_elem(map->fd, key, value, flags);
9633 int bpf_map__delete_elem(const struct bpf_map *map,
9634 const void *key, size_t key_sz, __u64 flags)
9638 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
9640 return libbpf_err(err);
9642 return bpf_map_delete_elem_flags(map->fd, key, flags);
9645 int bpf_map__lookup_and_delete_elem(const struct bpf_map *map,
9646 const void *key, size_t key_sz,
9647 void *value, size_t value_sz, __u64 flags)
9651 err = validate_map_op(map, key_sz, value_sz, true);
9653 return libbpf_err(err);
9655 return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags);
9658 int bpf_map__get_next_key(const struct bpf_map *map,
9659 const void *cur_key, void *next_key, size_t key_sz)
9663 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
9665 return libbpf_err(err);
9667 return bpf_map_get_next_key(map->fd, cur_key, next_key);
9670 long libbpf_get_error(const void *ptr)
9672 if (!IS_ERR_OR_NULL(ptr))
9676 errno = -PTR_ERR(ptr);
9678 /* If ptr == NULL, then errno should be already set by the failing
9679 * API, because libbpf never returns NULL on success and it now always
9680 * sets errno on error. So no extra errno handling for ptr == NULL
9686 /* Replace link's underlying BPF program with the new one */
9687 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
9691 ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
9692 return libbpf_err_errno(ret);
9695 /* Release "ownership" of underlying BPF resource (typically, BPF program
9696 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
9697 * link, when destructed through bpf_link__destroy() call won't attempt to
9698 * detach/unregisted that BPF resource. This is useful in situations where,
9699 * say, attached BPF program has to outlive userspace program that attached it
9700 * in the system. Depending on type of BPF program, though, there might be
9701 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
9702 * exit of userspace program doesn't trigger automatic detachment and clean up
9703 * inside the kernel.
9705 void bpf_link__disconnect(struct bpf_link *link)
9707 link->disconnected = true;
9710 int bpf_link__destroy(struct bpf_link *link)
9714 if (IS_ERR_OR_NULL(link))
9717 if (!link->disconnected && link->detach)
9718 err = link->detach(link);
9720 free(link->pin_path);
9722 link->dealloc(link);
9726 return libbpf_err(err);
9729 int bpf_link__fd(const struct bpf_link *link)
9734 const char *bpf_link__pin_path(const struct bpf_link *link)
9736 return link->pin_path;
9739 static int bpf_link__detach_fd(struct bpf_link *link)
9741 return libbpf_err_errno(close(link->fd));
9744 struct bpf_link *bpf_link__open(const char *path)
9746 struct bpf_link *link;
9749 fd = bpf_obj_get(path);
9752 pr_warn("failed to open link at %s: %d\n", path, fd);
9753 return libbpf_err_ptr(fd);
9756 link = calloc(1, sizeof(*link));
9759 return libbpf_err_ptr(-ENOMEM);
9761 link->detach = &bpf_link__detach_fd;
9764 link->pin_path = strdup(path);
9765 if (!link->pin_path) {
9766 bpf_link__destroy(link);
9767 return libbpf_err_ptr(-ENOMEM);
9773 int bpf_link__detach(struct bpf_link *link)
9775 return bpf_link_detach(link->fd) ? -errno : 0;
9778 int bpf_link__pin(struct bpf_link *link, const char *path)
9783 return libbpf_err(-EBUSY);
9784 err = make_parent_dir(path);
9786 return libbpf_err(err);
9787 err = check_path(path);
9789 return libbpf_err(err);
9791 link->pin_path = strdup(path);
9792 if (!link->pin_path)
9793 return libbpf_err(-ENOMEM);
9795 if (bpf_obj_pin(link->fd, link->pin_path)) {
9797 zfree(&link->pin_path);
9798 return libbpf_err(err);
9801 pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
9805 int bpf_link__unpin(struct bpf_link *link)
9809 if (!link->pin_path)
9810 return libbpf_err(-EINVAL);
9812 err = unlink(link->pin_path);
9816 pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
9817 zfree(&link->pin_path);
9821 struct bpf_link_perf {
9822 struct bpf_link link;
9824 /* legacy kprobe support: keep track of probe identifier and type */
9825 char *legacy_probe_name;
9826 bool legacy_is_kprobe;
9827 bool legacy_is_retprobe;
9830 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
9831 static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
9833 static int bpf_link_perf_detach(struct bpf_link *link)
9835 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9838 if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
9841 if (perf_link->perf_event_fd != link->fd)
9842 close(perf_link->perf_event_fd);
9845 /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
9846 if (perf_link->legacy_probe_name) {
9847 if (perf_link->legacy_is_kprobe) {
9848 err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
9849 perf_link->legacy_is_retprobe);
9851 err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
9852 perf_link->legacy_is_retprobe);
9859 static void bpf_link_perf_dealloc(struct bpf_link *link)
9861 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9863 free(perf_link->legacy_probe_name);
9867 struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
9868 const struct bpf_perf_event_opts *opts)
9870 char errmsg[STRERR_BUFSIZE];
9871 struct bpf_link_perf *link;
9872 int prog_fd, link_fd = -1, err;
9873 bool force_ioctl_attach;
9875 if (!OPTS_VALID(opts, bpf_perf_event_opts))
9876 return libbpf_err_ptr(-EINVAL);
9879 pr_warn("prog '%s': invalid perf event FD %d\n",
9881 return libbpf_err_ptr(-EINVAL);
9883 prog_fd = bpf_program__fd(prog);
9885 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
9887 return libbpf_err_ptr(-EINVAL);
9890 link = calloc(1, sizeof(*link));
9892 return libbpf_err_ptr(-ENOMEM);
9893 link->link.detach = &bpf_link_perf_detach;
9894 link->link.dealloc = &bpf_link_perf_dealloc;
9895 link->perf_event_fd = pfd;
9897 force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false);
9898 if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) {
9899 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
9900 .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
9902 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
9905 pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n",
9907 err, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9910 link->link.fd = link_fd;
9912 if (OPTS_GET(opts, bpf_cookie, 0)) {
9913 pr_warn("prog '%s': user context value is not supported\n", prog->name);
9918 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
9920 pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
9921 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9923 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
9927 link->link.fd = pfd;
9929 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
9931 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
9932 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9941 return libbpf_err_ptr(err);
9944 struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
9946 return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
9950 * this function is expected to parse integer in the range of [0, 2^31-1] from
9951 * given file using scanf format string fmt. If actual parsed value is
9952 * negative, the result might be indistinguishable from error
9954 static int parse_uint_from_file(const char *file, const char *fmt)
9956 char buf[STRERR_BUFSIZE];
9960 f = fopen(file, "r");
9963 pr_debug("failed to open '%s': %s\n", file,
9964 libbpf_strerror_r(err, buf, sizeof(buf)));
9967 err = fscanf(f, fmt, &ret);
9969 err = err == EOF ? -EIO : -errno;
9970 pr_debug("failed to parse '%s': %s\n", file,
9971 libbpf_strerror_r(err, buf, sizeof(buf)));
9979 static int determine_kprobe_perf_type(void)
9981 const char *file = "/sys/bus/event_source/devices/kprobe/type";
9983 return parse_uint_from_file(file, "%d\n");
9986 static int determine_uprobe_perf_type(void)
9988 const char *file = "/sys/bus/event_source/devices/uprobe/type";
9990 return parse_uint_from_file(file, "%d\n");
9993 static int determine_kprobe_retprobe_bit(void)
9995 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
9997 return parse_uint_from_file(file, "config:%d\n");
10000 static int determine_uprobe_retprobe_bit(void)
10002 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
10004 return parse_uint_from_file(file, "config:%d\n");
10007 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
10008 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
10010 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
10011 uint64_t offset, int pid, size_t ref_ctr_off)
10013 const size_t attr_sz = sizeof(struct perf_event_attr);
10014 struct perf_event_attr attr;
10015 char errmsg[STRERR_BUFSIZE];
10018 if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
10021 memset(&attr, 0, attr_sz);
10023 type = uprobe ? determine_uprobe_perf_type()
10024 : determine_kprobe_perf_type();
10026 pr_warn("failed to determine %s perf type: %s\n",
10027 uprobe ? "uprobe" : "kprobe",
10028 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
10032 int bit = uprobe ? determine_uprobe_retprobe_bit()
10033 : determine_kprobe_retprobe_bit();
10036 pr_warn("failed to determine %s retprobe bit: %s\n",
10037 uprobe ? "uprobe" : "kprobe",
10038 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
10041 attr.config |= 1 << bit;
10043 attr.size = attr_sz;
10045 attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
10046 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
10047 attr.config2 = offset; /* kprobe_addr or probe_offset */
10049 /* pid filter is meaningful only for uprobes */
10050 pfd = syscall(__NR_perf_event_open, &attr,
10051 pid < 0 ? -1 : pid /* pid */,
10052 pid == -1 ? 0 : -1 /* cpu */,
10053 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10054 return pfd >= 0 ? pfd : -errno;
10057 static int append_to_file(const char *file, const char *fmt, ...)
10059 int fd, n, err = 0;
10064 n = vsnprintf(buf, sizeof(buf), fmt, ap);
10067 if (n < 0 || n >= sizeof(buf))
10070 fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
10074 if (write(fd, buf, n) < 0)
10081 #define DEBUGFS "/sys/kernel/debug/tracing"
10082 #define TRACEFS "/sys/kernel/tracing"
10084 static bool use_debugfs(void)
10086 static int has_debugfs = -1;
10088 if (has_debugfs < 0)
10089 has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0;
10091 return has_debugfs == 1;
10094 static const char *tracefs_path(void)
10096 return use_debugfs() ? DEBUGFS : TRACEFS;
10099 static const char *tracefs_kprobe_events(void)
10101 return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events";
10104 static const char *tracefs_uprobe_events(void)
10106 return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events";
10109 static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz,
10110 const char *kfunc_name, size_t offset)
10112 static int index = 0;
10115 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset,
10116 __sync_fetch_and_add(&index, 1));
10118 /* sanitize binary_path in the probe name */
10119 for (i = 0; buf[i]; i++) {
10120 if (!isalnum(buf[i]))
10125 static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
10126 const char *kfunc_name, size_t offset)
10128 return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx",
10129 retprobe ? 'r' : 'p',
10130 retprobe ? "kretprobes" : "kprobes",
10131 probe_name, kfunc_name, offset);
10134 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
10136 return append_to_file(tracefs_kprobe_events(), "-:%s/%s",
10137 retprobe ? "kretprobes" : "kprobes", probe_name);
10140 static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10144 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
10145 tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name);
10147 return parse_uint_from_file(file, "%d\n");
10150 static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
10151 const char *kfunc_name, size_t offset, int pid)
10153 const size_t attr_sz = sizeof(struct perf_event_attr);
10154 struct perf_event_attr attr;
10155 char errmsg[STRERR_BUFSIZE];
10156 int type, pfd, err;
10158 err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
10160 pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
10161 kfunc_name, offset,
10162 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10165 type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
10168 pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
10169 kfunc_name, offset,
10170 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10171 goto err_clean_legacy;
10174 memset(&attr, 0, attr_sz);
10175 attr.size = attr_sz;
10176 attr.config = type;
10177 attr.type = PERF_TYPE_TRACEPOINT;
10179 pfd = syscall(__NR_perf_event_open, &attr,
10180 pid < 0 ? -1 : pid, /* pid */
10181 pid == -1 ? 0 : -1, /* cpu */
10182 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10185 pr_warn("legacy kprobe perf_event_open() failed: %s\n",
10186 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10187 goto err_clean_legacy;
10192 /* Clear the newly added legacy kprobe_event */
10193 remove_kprobe_event_legacy(probe_name, retprobe);
10197 static const char *arch_specific_syscall_pfx(void)
10199 #if defined(__x86_64__)
10201 #elif defined(__i386__)
10203 #elif defined(__s390x__)
10205 #elif defined(__s390__)
10207 #elif defined(__arm__)
10209 #elif defined(__aarch64__)
10211 #elif defined(__mips__)
10213 #elif defined(__riscv)
10215 #elif defined(__powerpc__)
10217 #elif defined(__powerpc64__)
10218 return "powerpc64";
10224 static int probe_kern_syscall_wrapper(void)
10226 char syscall_name[64];
10227 const char *ksys_pfx;
10229 ksys_pfx = arch_specific_syscall_pfx();
10233 snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx);
10235 if (determine_kprobe_perf_type() >= 0) {
10238 pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0);
10242 return pfd >= 0 ? 1 : 0;
10243 } else { /* legacy mode */
10244 char probe_name[128];
10246 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0);
10247 if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0)
10250 (void)remove_kprobe_event_legacy(probe_name, false);
10256 bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
10257 const char *func_name,
10258 const struct bpf_kprobe_opts *opts)
10260 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10261 enum probe_attach_mode attach_mode;
10262 char errmsg[STRERR_BUFSIZE];
10263 char *legacy_probe = NULL;
10264 struct bpf_link *link;
10266 bool retprobe, legacy;
10269 if (!OPTS_VALID(opts, bpf_kprobe_opts))
10270 return libbpf_err_ptr(-EINVAL);
10272 attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
10273 retprobe = OPTS_GET(opts, retprobe, false);
10274 offset = OPTS_GET(opts, offset, 0);
10275 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10277 legacy = determine_kprobe_perf_type() < 0;
10278 switch (attach_mode) {
10279 case PROBE_ATTACH_MODE_LEGACY:
10281 pe_opts.force_ioctl_attach = true;
10283 case PROBE_ATTACH_MODE_PERF:
10285 return libbpf_err_ptr(-ENOTSUP);
10286 pe_opts.force_ioctl_attach = true;
10288 case PROBE_ATTACH_MODE_LINK:
10289 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
10290 return libbpf_err_ptr(-ENOTSUP);
10292 case PROBE_ATTACH_MODE_DEFAULT:
10295 return libbpf_err_ptr(-EINVAL);
10299 pfd = perf_event_open_probe(false /* uprobe */, retprobe,
10301 -1 /* pid */, 0 /* ref_ctr_off */);
10303 char probe_name[256];
10305 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name),
10306 func_name, offset);
10308 legacy_probe = strdup(probe_name);
10310 return libbpf_err_ptr(-ENOMEM);
10312 pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
10313 offset, -1 /* pid */);
10317 pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
10318 prog->name, retprobe ? "kretprobe" : "kprobe",
10320 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10323 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10324 err = libbpf_get_error(link);
10327 pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
10328 prog->name, retprobe ? "kretprobe" : "kprobe",
10330 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10331 goto err_clean_legacy;
10334 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10336 perf_link->legacy_probe_name = legacy_probe;
10337 perf_link->legacy_is_kprobe = true;
10338 perf_link->legacy_is_retprobe = retprobe;
10345 remove_kprobe_event_legacy(legacy_probe, retprobe);
10347 free(legacy_probe);
10348 return libbpf_err_ptr(err);
10351 struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
10353 const char *func_name)
10355 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
10356 .retprobe = retprobe,
10359 return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
10362 struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog,
10363 const char *syscall_name,
10364 const struct bpf_ksyscall_opts *opts)
10366 LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts);
10367 char func_name[128];
10369 if (!OPTS_VALID(opts, bpf_ksyscall_opts))
10370 return libbpf_err_ptr(-EINVAL);
10372 if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) {
10373 /* arch_specific_syscall_pfx() should never return NULL here
10374 * because it is guarded by kernel_supports(). However, since
10375 * compiler does not know that we have an explicit conditional
10378 snprintf(func_name, sizeof(func_name), "__%s_sys_%s",
10379 arch_specific_syscall_pfx() ? : "", syscall_name);
10381 snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name);
10384 kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false);
10385 kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10387 return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts);
10390 /* Adapted from perf/util/string.c */
10391 static bool glob_match(const char *str, const char *pat)
10393 while (*str && *pat && *pat != '*') {
10394 if (*pat == '?') { /* Matches any single character */
10404 /* Check wild card */
10406 while (*pat == '*')
10408 if (!*pat) /* Tail wild card matches all */
10411 if (glob_match(str++, pat))
10414 return !*str && !*pat;
10417 struct kprobe_multi_resolve {
10418 const char *pattern;
10419 unsigned long *addrs;
10425 resolve_kprobe_multi_cb(unsigned long long sym_addr, char sym_type,
10426 const char *sym_name, void *ctx)
10428 struct kprobe_multi_resolve *res = ctx;
10431 if (!glob_match(sym_name, res->pattern))
10434 err = libbpf_ensure_mem((void **) &res->addrs, &res->cap, sizeof(unsigned long),
10439 res->addrs[res->cnt++] = (unsigned long) sym_addr;
10444 bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog,
10445 const char *pattern,
10446 const struct bpf_kprobe_multi_opts *opts)
10448 LIBBPF_OPTS(bpf_link_create_opts, lopts);
10449 struct kprobe_multi_resolve res = {
10450 .pattern = pattern,
10452 struct bpf_link *link = NULL;
10453 char errmsg[STRERR_BUFSIZE];
10454 const unsigned long *addrs;
10455 int err, link_fd, prog_fd;
10456 const __u64 *cookies;
10461 if (!OPTS_VALID(opts, bpf_kprobe_multi_opts))
10462 return libbpf_err_ptr(-EINVAL);
10464 syms = OPTS_GET(opts, syms, false);
10465 addrs = OPTS_GET(opts, addrs, false);
10466 cnt = OPTS_GET(opts, cnt, false);
10467 cookies = OPTS_GET(opts, cookies, false);
10469 if (!pattern && !addrs && !syms)
10470 return libbpf_err_ptr(-EINVAL);
10471 if (pattern && (addrs || syms || cookies || cnt))
10472 return libbpf_err_ptr(-EINVAL);
10473 if (!pattern && !cnt)
10474 return libbpf_err_ptr(-EINVAL);
10476 return libbpf_err_ptr(-EINVAL);
10479 err = libbpf_kallsyms_parse(resolve_kprobe_multi_cb, &res);
10490 retprobe = OPTS_GET(opts, retprobe, false);
10492 lopts.kprobe_multi.syms = syms;
10493 lopts.kprobe_multi.addrs = addrs;
10494 lopts.kprobe_multi.cookies = cookies;
10495 lopts.kprobe_multi.cnt = cnt;
10496 lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0;
10498 link = calloc(1, sizeof(*link));
10503 link->detach = &bpf_link__detach_fd;
10505 prog_fd = bpf_program__fd(prog);
10506 link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_KPROBE_MULTI, &lopts);
10509 pr_warn("prog '%s': failed to attach: %s\n",
10510 prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10513 link->fd = link_fd;
10520 return libbpf_err_ptr(err);
10523 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10525 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
10526 unsigned long offset = 0;
10527 const char *func_name;
10533 /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */
10534 if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0)
10537 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
10539 func_name = prog->sec_name + sizeof("kretprobe/") - 1;
10541 func_name = prog->sec_name + sizeof("kprobe/") - 1;
10543 n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
10545 pr_warn("kprobe name is invalid: %s\n", func_name);
10548 if (opts.retprobe && offset != 0) {
10550 pr_warn("kretprobes do not support offset specification\n");
10554 opts.offset = offset;
10555 *link = bpf_program__attach_kprobe_opts(prog, func, &opts);
10557 return libbpf_get_error(*link);
10560 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10562 LIBBPF_OPTS(bpf_ksyscall_opts, opts);
10563 const char *syscall_name;
10567 /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */
10568 if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0)
10571 opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/");
10573 syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1;
10575 syscall_name = prog->sec_name + sizeof("ksyscall/") - 1;
10577 *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts);
10578 return *link ? 0 : -errno;
10581 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10583 LIBBPF_OPTS(bpf_kprobe_multi_opts, opts);
10590 /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */
10591 if (strcmp(prog->sec_name, "kprobe.multi") == 0 ||
10592 strcmp(prog->sec_name, "kretprobe.multi") == 0)
10595 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/");
10597 spec = prog->sec_name + sizeof("kretprobe.multi/") - 1;
10599 spec = prog->sec_name + sizeof("kprobe.multi/") - 1;
10601 n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
10603 pr_warn("kprobe multi pattern is invalid: %s\n", pattern);
10607 *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
10609 return libbpf_get_error(*link);
10612 static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz,
10613 const char *binary_path, uint64_t offset)
10617 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset);
10619 /* sanitize binary_path in the probe name */
10620 for (i = 0; buf[i]; i++) {
10621 if (!isalnum(buf[i]))
10626 static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
10627 const char *binary_path, size_t offset)
10629 return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx",
10630 retprobe ? 'r' : 'p',
10631 retprobe ? "uretprobes" : "uprobes",
10632 probe_name, binary_path, offset);
10635 static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
10637 return append_to_file(tracefs_uprobe_events(), "-:%s/%s",
10638 retprobe ? "uretprobes" : "uprobes", probe_name);
10641 static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10645 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
10646 tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name);
10648 return parse_uint_from_file(file, "%d\n");
10651 static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
10652 const char *binary_path, size_t offset, int pid)
10654 const size_t attr_sz = sizeof(struct perf_event_attr);
10655 struct perf_event_attr attr;
10656 int type, pfd, err;
10658 err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
10660 pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n",
10661 binary_path, (size_t)offset, err);
10664 type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
10667 pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n",
10668 binary_path, offset, err);
10669 goto err_clean_legacy;
10672 memset(&attr, 0, attr_sz);
10673 attr.size = attr_sz;
10674 attr.config = type;
10675 attr.type = PERF_TYPE_TRACEPOINT;
10677 pfd = syscall(__NR_perf_event_open, &attr,
10678 pid < 0 ? -1 : pid, /* pid */
10679 pid == -1 ? 0 : -1, /* cpu */
10680 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10683 pr_warn("legacy uprobe perf_event_open() failed: %d\n", err);
10684 goto err_clean_legacy;
10689 /* Clear the newly added legacy uprobe_event */
10690 remove_uprobe_event_legacy(probe_name, retprobe);
10694 /* Return next ELF section of sh_type after scn, or first of that type if scn is NULL. */
10695 static Elf_Scn *elf_find_next_scn_by_type(Elf *elf, int sh_type, Elf_Scn *scn)
10697 while ((scn = elf_nextscn(elf, scn)) != NULL) {
10700 if (!gelf_getshdr(scn, &sh))
10702 if (sh.sh_type == sh_type)
10708 /* Find offset of function name in the provided ELF object. "binary_path" is
10709 * the path to the ELF binary represented by "elf", and only used for error
10710 * reporting matters. "name" matches symbol name or name@@LIB for library
10713 static long elf_find_func_offset(Elf *elf, const char *binary_path, const char *name)
10715 int i, sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB };
10716 bool is_shared_lib, is_name_qualified;
10717 long ret = -ENOENT;
10721 if (!gelf_getehdr(elf, &ehdr)) {
10722 pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1));
10723 ret = -LIBBPF_ERRNO__FORMAT;
10726 /* for shared lib case, we do not need to calculate relative offset */
10727 is_shared_lib = ehdr.e_type == ET_DYN;
10729 name_len = strlen(name);
10730 /* Does name specify "@@LIB"? */
10731 is_name_qualified = strstr(name, "@@") != NULL;
10733 /* Search SHT_DYNSYM, SHT_SYMTAB for symbol. This search order is used because if
10734 * a binary is stripped, it may only have SHT_DYNSYM, and a fully-statically
10735 * linked binary may not have SHT_DYMSYM, so absence of a section should not be
10736 * reported as a warning/error.
10738 for (i = 0; i < ARRAY_SIZE(sh_types); i++) {
10739 size_t nr_syms, strtabidx, idx;
10740 Elf_Data *symbols = NULL;
10741 Elf_Scn *scn = NULL;
10742 int last_bind = -1;
10746 scn = elf_find_next_scn_by_type(elf, sh_types[i], NULL);
10748 pr_debug("elf: failed to find symbol table ELF sections in '%s'\n",
10752 if (!gelf_getshdr(scn, &sh))
10754 strtabidx = sh.sh_link;
10755 symbols = elf_getdata(scn, 0);
10757 pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n",
10758 binary_path, elf_errmsg(-1));
10759 ret = -LIBBPF_ERRNO__FORMAT;
10762 nr_syms = symbols->d_size / sh.sh_entsize;
10764 for (idx = 0; idx < nr_syms; idx++) {
10770 if (!gelf_getsym(symbols, idx, &sym))
10773 if (GELF_ST_TYPE(sym.st_info) != STT_FUNC)
10776 sname = elf_strptr(elf, strtabidx, sym.st_name);
10780 curr_bind = GELF_ST_BIND(sym.st_info);
10782 /* User can specify func, func@@LIB or func@@LIB_VERSION. */
10783 if (strncmp(sname, name, name_len) != 0)
10785 /* ...but we don't want a search for "foo" to match 'foo2" also, so any
10786 * additional characters in sname should be of the form "@@LIB".
10788 if (!is_name_qualified && sname[name_len] != '\0' && sname[name_len] != '@')
10792 /* handle multiple matches */
10793 if (last_bind != STB_WEAK && curr_bind != STB_WEAK) {
10794 /* Only accept one non-weak bind. */
10795 pr_warn("elf: ambiguous match for '%s', '%s' in '%s'\n",
10796 sname, name, binary_path);
10797 ret = -LIBBPF_ERRNO__FORMAT;
10799 } else if (curr_bind == STB_WEAK) {
10800 /* already have a non-weak bind, and
10801 * this is a weak bind, so ignore.
10807 /* Transform symbol's virtual address (absolute for
10808 * binaries and relative for shared libs) into file
10809 * offset, which is what kernel is expecting for
10810 * uprobe/uretprobe attachment.
10811 * See Documentation/trace/uprobetracer.rst for more
10813 * This is done by looking up symbol's containing
10814 * section's header and using it's virtual address
10815 * (sh_addr) and corresponding file offset (sh_offset)
10816 * to transform sym.st_value (virtual address) into
10817 * desired final file offset.
10819 sym_scn = elf_getscn(elf, sym.st_shndx);
10822 if (!gelf_getshdr(sym_scn, &sym_sh))
10825 ret = sym.st_value - sym_sh.sh_addr + sym_sh.sh_offset;
10826 last_bind = curr_bind;
10833 pr_debug("elf: symbol address match for '%s' in '%s': 0x%lx\n", name, binary_path,
10837 pr_warn("elf: '%s' is 0 in symtab for '%s': %s\n", name, binary_path,
10838 is_shared_lib ? "should not be 0 in a shared library" :
10839 "try using shared library path instead");
10842 pr_warn("elf: failed to find symbol '%s' in '%s'\n", name, binary_path);
10849 /* Find offset of function name in ELF object specified by path. "name" matches
10850 * symbol name or name@@LIB for library functions.
10852 static long elf_find_func_offset_from_file(const char *binary_path, const char *name)
10854 char errmsg[STRERR_BUFSIZE];
10855 long ret = -ENOENT;
10859 fd = open(binary_path, O_RDONLY | O_CLOEXEC);
10862 pr_warn("failed to open %s: %s\n", binary_path,
10863 libbpf_strerror_r(ret, errmsg, sizeof(errmsg)));
10866 elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
10868 pr_warn("elf: could not read elf from %s: %s\n", binary_path, elf_errmsg(-1));
10870 return -LIBBPF_ERRNO__FORMAT;
10873 ret = elf_find_func_offset(elf, binary_path, name);
10879 /* Find offset of function name in archive specified by path. Currently
10880 * supported are .zip files that do not compress their contents, as used on
10881 * Android in the form of APKs, for example. "file_name" is the name of the ELF
10882 * file inside the archive. "func_name" matches symbol name or name@@LIB for
10883 * library functions.
10885 * An overview of the APK format specifically provided here:
10886 * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents
10888 static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name,
10889 const char *func_name)
10891 struct zip_archive *archive;
10892 struct zip_entry entry;
10896 archive = zip_archive_open(archive_path);
10897 if (IS_ERR(archive)) {
10898 ret = PTR_ERR(archive);
10899 pr_warn("zip: failed to open %s: %ld\n", archive_path, ret);
10903 ret = zip_archive_find_entry(archive, file_name, &entry);
10905 pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name,
10906 archive_path, ret);
10909 pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path,
10910 (unsigned long)entry.data_offset);
10912 if (entry.compression) {
10913 pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name,
10915 ret = -LIBBPF_ERRNO__FORMAT;
10919 elf = elf_memory((void *)entry.data, entry.data_length);
10921 pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path,
10923 ret = -LIBBPF_ERRNO__LIBELF;
10927 ret = elf_find_func_offset(elf, file_name, func_name);
10929 pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n",
10930 func_name, file_name, archive_path, entry.data_offset, ret,
10931 ret + entry.data_offset);
10932 ret += entry.data_offset;
10937 zip_archive_close(archive);
10941 static const char *arch_specific_lib_paths(void)
10944 * Based on https://packages.debian.org/sid/libc6.
10946 * Assume that the traced program is built for the same architecture
10947 * as libbpf, which should cover the vast majority of cases.
10949 #if defined(__x86_64__)
10950 return "/lib/x86_64-linux-gnu";
10951 #elif defined(__i386__)
10952 return "/lib/i386-linux-gnu";
10953 #elif defined(__s390x__)
10954 return "/lib/s390x-linux-gnu";
10955 #elif defined(__s390__)
10956 return "/lib/s390-linux-gnu";
10957 #elif defined(__arm__) && defined(__SOFTFP__)
10958 return "/lib/arm-linux-gnueabi";
10959 #elif defined(__arm__) && !defined(__SOFTFP__)
10960 return "/lib/arm-linux-gnueabihf";
10961 #elif defined(__aarch64__)
10962 return "/lib/aarch64-linux-gnu";
10963 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64
10964 return "/lib/mips64el-linux-gnuabi64";
10965 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32
10966 return "/lib/mipsel-linux-gnu";
10967 #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
10968 return "/lib/powerpc64le-linux-gnu";
10969 #elif defined(__sparc__) && defined(__arch64__)
10970 return "/lib/sparc64-linux-gnu";
10971 #elif defined(__riscv) && __riscv_xlen == 64
10972 return "/lib/riscv64-linux-gnu";
10978 /* Get full path to program/shared library. */
10979 static int resolve_full_path(const char *file, char *result, size_t result_sz)
10981 const char *search_paths[3] = {};
10984 if (str_has_sfx(file, ".so") || strstr(file, ".so.")) {
10985 search_paths[0] = getenv("LD_LIBRARY_PATH");
10986 search_paths[1] = "/usr/lib64:/usr/lib";
10987 search_paths[2] = arch_specific_lib_paths();
10990 search_paths[0] = getenv("PATH");
10991 search_paths[1] = "/usr/bin:/usr/sbin";
10992 perm = R_OK | X_OK;
10995 for (i = 0; i < ARRAY_SIZE(search_paths); i++) {
10998 if (!search_paths[i])
11000 for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) {
11006 next_path = strchr(s, ':');
11007 seg_len = next_path ? next_path - s : strlen(s);
11010 snprintf(result, result_sz, "%.*s/%s", seg_len, s, file);
11011 /* ensure it has required permissions */
11012 if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0)
11014 pr_debug("resolved '%s' to '%s'\n", file, result);
11021 LIBBPF_API struct bpf_link *
11022 bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
11023 const char *binary_path, size_t func_offset,
11024 const struct bpf_uprobe_opts *opts)
11026 const char *archive_path = NULL, *archive_sep = NULL;
11027 char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL;
11028 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11029 enum probe_attach_mode attach_mode;
11030 char full_path[PATH_MAX];
11031 struct bpf_link *link;
11032 size_t ref_ctr_off;
11034 bool retprobe, legacy;
11035 const char *func_name;
11037 if (!OPTS_VALID(opts, bpf_uprobe_opts))
11038 return libbpf_err_ptr(-EINVAL);
11040 attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
11041 retprobe = OPTS_GET(opts, retprobe, false);
11042 ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
11043 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11046 return libbpf_err_ptr(-EINVAL);
11048 /* Check if "binary_path" refers to an archive. */
11049 archive_sep = strstr(binary_path, "!/");
11051 full_path[0] = '\0';
11052 libbpf_strlcpy(full_path, binary_path,
11053 min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1)));
11054 archive_path = full_path;
11055 binary_path = archive_sep + 2;
11056 } else if (!strchr(binary_path, '/')) {
11057 err = resolve_full_path(binary_path, full_path, sizeof(full_path));
11059 pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11060 prog->name, binary_path, err);
11061 return libbpf_err_ptr(err);
11063 binary_path = full_path;
11065 func_name = OPTS_GET(opts, func_name, NULL);
11069 if (archive_path) {
11070 sym_off = elf_find_func_offset_from_archive(archive_path, binary_path,
11072 binary_path = archive_path;
11074 sym_off = elf_find_func_offset_from_file(binary_path, func_name);
11077 return libbpf_err_ptr(sym_off);
11078 func_offset += sym_off;
11081 legacy = determine_uprobe_perf_type() < 0;
11082 switch (attach_mode) {
11083 case PROBE_ATTACH_MODE_LEGACY:
11085 pe_opts.force_ioctl_attach = true;
11087 case PROBE_ATTACH_MODE_PERF:
11089 return libbpf_err_ptr(-ENOTSUP);
11090 pe_opts.force_ioctl_attach = true;
11092 case PROBE_ATTACH_MODE_LINK:
11093 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
11094 return libbpf_err_ptr(-ENOTSUP);
11096 case PROBE_ATTACH_MODE_DEFAULT:
11099 return libbpf_err_ptr(-EINVAL);
11103 pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
11104 func_offset, pid, ref_ctr_off);
11106 char probe_name[PATH_MAX + 64];
11109 return libbpf_err_ptr(-EINVAL);
11111 gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name),
11112 binary_path, func_offset);
11114 legacy_probe = strdup(probe_name);
11116 return libbpf_err_ptr(-ENOMEM);
11118 pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
11119 binary_path, func_offset, pid);
11123 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
11124 prog->name, retprobe ? "uretprobe" : "uprobe",
11125 binary_path, func_offset,
11126 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11130 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11131 err = libbpf_get_error(link);
11134 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
11135 prog->name, retprobe ? "uretprobe" : "uprobe",
11136 binary_path, func_offset,
11137 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11138 goto err_clean_legacy;
11141 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11143 perf_link->legacy_probe_name = legacy_probe;
11144 perf_link->legacy_is_kprobe = false;
11145 perf_link->legacy_is_retprobe = retprobe;
11151 remove_uprobe_event_legacy(legacy_probe, retprobe);
11153 free(legacy_probe);
11154 return libbpf_err_ptr(err);
11157 /* Format of u[ret]probe section definition supporting auto-attach:
11158 * u[ret]probe/binary:function[+offset]
11160 * binary can be an absolute/relative path or a filename; the latter is resolved to a
11161 * full binary path via bpf_program__attach_uprobe_opts.
11163 * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be
11164 * specified (and auto-attach is not possible) or the above format is specified for
11167 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11169 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts);
11170 char *probe_type = NULL, *binary_path = NULL, *func_name = NULL;
11171 int n, ret = -EINVAL;
11176 n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[a-zA-Z0-9_.]+%li",
11177 &probe_type, &binary_path, &func_name, &offset);
11180 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
11184 pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n",
11185 prog->name, prog->sec_name);
11189 opts.retprobe = strcmp(probe_type, "uretprobe") == 0 ||
11190 strcmp(probe_type, "uretprobe.s") == 0;
11191 if (opts.retprobe && offset != 0) {
11192 pr_warn("prog '%s': uretprobes do not support offset specification\n",
11196 opts.func_name = func_name;
11197 *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts);
11198 ret = libbpf_get_error(*link);
11201 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
11212 struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
11213 bool retprobe, pid_t pid,
11214 const char *binary_path,
11215 size_t func_offset)
11217 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
11219 return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
11222 struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog,
11223 pid_t pid, const char *binary_path,
11224 const char *usdt_provider, const char *usdt_name,
11225 const struct bpf_usdt_opts *opts)
11227 char resolved_path[512];
11228 struct bpf_object *obj = prog->obj;
11229 struct bpf_link *link;
11233 if (!OPTS_VALID(opts, bpf_uprobe_opts))
11234 return libbpf_err_ptr(-EINVAL);
11236 if (bpf_program__fd(prog) < 0) {
11237 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
11239 return libbpf_err_ptr(-EINVAL);
11243 return libbpf_err_ptr(-EINVAL);
11245 if (!strchr(binary_path, '/')) {
11246 err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path));
11248 pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11249 prog->name, binary_path, err);
11250 return libbpf_err_ptr(err);
11252 binary_path = resolved_path;
11255 /* USDT manager is instantiated lazily on first USDT attach. It will
11256 * be destroyed together with BPF object in bpf_object__close().
11258 if (IS_ERR(obj->usdt_man))
11259 return libbpf_ptr(obj->usdt_man);
11260 if (!obj->usdt_man) {
11261 obj->usdt_man = usdt_manager_new(obj);
11262 if (IS_ERR(obj->usdt_man))
11263 return libbpf_ptr(obj->usdt_man);
11266 usdt_cookie = OPTS_GET(opts, usdt_cookie, 0);
11267 link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path,
11268 usdt_provider, usdt_name, usdt_cookie);
11269 err = libbpf_get_error(link);
11271 return libbpf_err_ptr(err);
11275 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11277 char *path = NULL, *provider = NULL, *name = NULL;
11278 const char *sec_name;
11281 sec_name = bpf_program__section_name(prog);
11282 if (strcmp(sec_name, "usdt") == 0) {
11283 /* no auto-attach for just SEC("usdt") */
11288 n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name);
11290 pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n",
11294 *link = bpf_program__attach_usdt(prog, -1 /* any process */, path,
11295 provider, name, NULL);
11296 err = libbpf_get_error(*link);
11304 static int determine_tracepoint_id(const char *tp_category,
11305 const char *tp_name)
11307 char file[PATH_MAX];
11310 ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11311 tracefs_path(), tp_category, tp_name);
11314 if (ret >= sizeof(file)) {
11315 pr_debug("tracepoint %s/%s path is too long\n",
11316 tp_category, tp_name);
11319 return parse_uint_from_file(file, "%d\n");
11322 static int perf_event_open_tracepoint(const char *tp_category,
11323 const char *tp_name)
11325 const size_t attr_sz = sizeof(struct perf_event_attr);
11326 struct perf_event_attr attr;
11327 char errmsg[STRERR_BUFSIZE];
11328 int tp_id, pfd, err;
11330 tp_id = determine_tracepoint_id(tp_category, tp_name);
11332 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
11333 tp_category, tp_name,
11334 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
11338 memset(&attr, 0, attr_sz);
11339 attr.type = PERF_TYPE_TRACEPOINT;
11340 attr.size = attr_sz;
11341 attr.config = tp_id;
11343 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
11344 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11347 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
11348 tp_category, tp_name,
11349 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11355 struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
11356 const char *tp_category,
11357 const char *tp_name,
11358 const struct bpf_tracepoint_opts *opts)
11360 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11361 char errmsg[STRERR_BUFSIZE];
11362 struct bpf_link *link;
11365 if (!OPTS_VALID(opts, bpf_tracepoint_opts))
11366 return libbpf_err_ptr(-EINVAL);
11368 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11370 pfd = perf_event_open_tracepoint(tp_category, tp_name);
11372 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
11373 prog->name, tp_category, tp_name,
11374 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11375 return libbpf_err_ptr(pfd);
11377 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11378 err = libbpf_get_error(link);
11381 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
11382 prog->name, tp_category, tp_name,
11383 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11384 return libbpf_err_ptr(err);
11389 struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
11390 const char *tp_category,
11391 const char *tp_name)
11393 return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
11396 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11398 char *sec_name, *tp_cat, *tp_name;
11402 /* no auto-attach for SEC("tp") or SEC("tracepoint") */
11403 if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0)
11406 sec_name = strdup(prog->sec_name);
11410 /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
11411 if (str_has_pfx(prog->sec_name, "tp/"))
11412 tp_cat = sec_name + sizeof("tp/") - 1;
11414 tp_cat = sec_name + sizeof("tracepoint/") - 1;
11415 tp_name = strchr(tp_cat, '/');
11423 *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
11425 return libbpf_get_error(*link);
11428 struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
11429 const char *tp_name)
11431 char errmsg[STRERR_BUFSIZE];
11432 struct bpf_link *link;
11435 prog_fd = bpf_program__fd(prog);
11437 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11438 return libbpf_err_ptr(-EINVAL);
11441 link = calloc(1, sizeof(*link));
11443 return libbpf_err_ptr(-ENOMEM);
11444 link->detach = &bpf_link__detach_fd;
11446 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
11450 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
11451 prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11452 return libbpf_err_ptr(pfd);
11458 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11460 static const char *const prefixes[] = {
11464 "raw_tracepoint.w",
11467 const char *tp_name = NULL;
11471 for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
11474 if (!str_has_pfx(prog->sec_name, prefixes[i]))
11477 pfx_len = strlen(prefixes[i]);
11478 /* no auto-attach case of, e.g., SEC("raw_tp") */
11479 if (prog->sec_name[pfx_len] == '\0')
11482 if (prog->sec_name[pfx_len] != '/')
11485 tp_name = prog->sec_name + pfx_len + 1;
11490 pr_warn("prog '%s': invalid section name '%s'\n",
11491 prog->name, prog->sec_name);
11495 *link = bpf_program__attach_raw_tracepoint(prog, tp_name);
11496 return libbpf_get_error(*link);
11499 /* Common logic for all BPF program types that attach to a btf_id */
11500 static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog,
11501 const struct bpf_trace_opts *opts)
11503 LIBBPF_OPTS(bpf_link_create_opts, link_opts);
11504 char errmsg[STRERR_BUFSIZE];
11505 struct bpf_link *link;
11508 if (!OPTS_VALID(opts, bpf_trace_opts))
11509 return libbpf_err_ptr(-EINVAL);
11511 prog_fd = bpf_program__fd(prog);
11513 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11514 return libbpf_err_ptr(-EINVAL);
11517 link = calloc(1, sizeof(*link));
11519 return libbpf_err_ptr(-ENOMEM);
11520 link->detach = &bpf_link__detach_fd;
11522 /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */
11523 link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0);
11524 pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts);
11528 pr_warn("prog '%s': failed to attach: %s\n",
11529 prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11530 return libbpf_err_ptr(pfd);
11536 struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
11538 return bpf_program__attach_btf_id(prog, NULL);
11541 struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog,
11542 const struct bpf_trace_opts *opts)
11544 return bpf_program__attach_btf_id(prog, opts);
11547 struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
11549 return bpf_program__attach_btf_id(prog, NULL);
11552 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11554 *link = bpf_program__attach_trace(prog);
11555 return libbpf_get_error(*link);
11558 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11560 *link = bpf_program__attach_lsm(prog);
11561 return libbpf_get_error(*link);
11564 static struct bpf_link *
11565 bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id,
11566 const char *target_name)
11568 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts,
11569 .target_btf_id = btf_id);
11570 enum bpf_attach_type attach_type;
11571 char errmsg[STRERR_BUFSIZE];
11572 struct bpf_link *link;
11573 int prog_fd, link_fd;
11575 prog_fd = bpf_program__fd(prog);
11577 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11578 return libbpf_err_ptr(-EINVAL);
11581 link = calloc(1, sizeof(*link));
11583 return libbpf_err_ptr(-ENOMEM);
11584 link->detach = &bpf_link__detach_fd;
11586 attach_type = bpf_program__expected_attach_type(prog);
11587 link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts);
11591 pr_warn("prog '%s': failed to attach to %s: %s\n",
11592 prog->name, target_name,
11593 libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
11594 return libbpf_err_ptr(link_fd);
11596 link->fd = link_fd;
11601 bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
11603 return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup");
11607 bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
11609 return bpf_program__attach_fd(prog, netns_fd, 0, "netns");
11612 struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
11614 /* target_fd/target_ifindex use the same field in LINK_CREATE */
11615 return bpf_program__attach_fd(prog, ifindex, 0, "xdp");
11618 struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
11620 const char *attach_func_name)
11624 if (!!target_fd != !!attach_func_name) {
11625 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
11627 return libbpf_err_ptr(-EINVAL);
11630 if (prog->type != BPF_PROG_TYPE_EXT) {
11631 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
11633 return libbpf_err_ptr(-EINVAL);
11637 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
11639 return libbpf_err_ptr(btf_id);
11641 return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace");
11643 /* no target, so use raw_tracepoint_open for compatibility
11646 return bpf_program__attach_trace(prog);
11651 bpf_program__attach_iter(const struct bpf_program *prog,
11652 const struct bpf_iter_attach_opts *opts)
11654 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
11655 char errmsg[STRERR_BUFSIZE];
11656 struct bpf_link *link;
11657 int prog_fd, link_fd;
11658 __u32 target_fd = 0;
11660 if (!OPTS_VALID(opts, bpf_iter_attach_opts))
11661 return libbpf_err_ptr(-EINVAL);
11663 link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
11664 link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
11666 prog_fd = bpf_program__fd(prog);
11668 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11669 return libbpf_err_ptr(-EINVAL);
11672 link = calloc(1, sizeof(*link));
11674 return libbpf_err_ptr(-ENOMEM);
11675 link->detach = &bpf_link__detach_fd;
11677 link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
11678 &link_create_opts);
11682 pr_warn("prog '%s': failed to attach to iterator: %s\n",
11683 prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
11684 return libbpf_err_ptr(link_fd);
11686 link->fd = link_fd;
11690 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11692 *link = bpf_program__attach_iter(prog, NULL);
11693 return libbpf_get_error(*link);
11696 struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
11698 struct bpf_link *link = NULL;
11701 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
11702 return libbpf_err_ptr(-EOPNOTSUPP);
11704 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link);
11706 return libbpf_err_ptr(err);
11708 /* When calling bpf_program__attach() explicitly, auto-attach support
11709 * is expected to work, so NULL returned link is considered an error.
11710 * This is different for skeleton's attach, see comment in
11711 * bpf_object__attach_skeleton().
11714 return libbpf_err_ptr(-EOPNOTSUPP);
11719 struct bpf_link_struct_ops {
11720 struct bpf_link link;
11724 static int bpf_link__detach_struct_ops(struct bpf_link *link)
11726 struct bpf_link_struct_ops *st_link;
11729 st_link = container_of(link, struct bpf_link_struct_ops, link);
11731 if (st_link->map_fd < 0)
11732 /* w/o a real link */
11733 return bpf_map_delete_elem(link->fd, &zero);
11735 return close(link->fd);
11738 struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
11740 struct bpf_link_struct_ops *link;
11744 if (!bpf_map__is_struct_ops(map) || map->fd == -1)
11745 return libbpf_err_ptr(-EINVAL);
11747 link = calloc(1, sizeof(*link));
11749 return libbpf_err_ptr(-EINVAL);
11751 /* kern_vdata should be prepared during the loading phase. */
11752 err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
11753 /* It can be EBUSY if the map has been used to create or
11754 * update a link before. We don't allow updating the value of
11755 * a struct_ops once it is set. That ensures that the value
11756 * never changed. So, it is safe to skip EBUSY.
11758 if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) {
11760 return libbpf_err_ptr(err);
11763 link->link.detach = bpf_link__detach_struct_ops;
11765 if (!(map->def.map_flags & BPF_F_LINK)) {
11766 /* w/o a real link */
11767 link->link.fd = map->fd;
11769 return &link->link;
11772 fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL);
11775 return libbpf_err_ptr(fd);
11778 link->link.fd = fd;
11779 link->map_fd = map->fd;
11781 return &link->link;
11785 * Swap the back struct_ops of a link with a new struct_ops map.
11787 int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map)
11789 struct bpf_link_struct_ops *st_ops_link;
11793 if (!bpf_map__is_struct_ops(map) || map->fd < 0)
11796 st_ops_link = container_of(link, struct bpf_link_struct_ops, link);
11797 /* Ensure the type of a link is correct */
11798 if (st_ops_link->map_fd < 0)
11801 err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
11802 /* It can be EBUSY if the map has been used to create or
11803 * update a link before. We don't allow updating the value of
11804 * a struct_ops once it is set. That ensures that the value
11805 * never changed. So, it is safe to skip EBUSY.
11807 if (err && err != -EBUSY)
11810 err = bpf_link_update(link->fd, map->fd, NULL);
11814 st_ops_link->map_fd = map->fd;
11819 typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
11820 void *private_data);
11822 static enum bpf_perf_event_ret
11823 perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
11824 void **copy_mem, size_t *copy_size,
11825 bpf_perf_event_print_t fn, void *private_data)
11827 struct perf_event_mmap_page *header = mmap_mem;
11828 __u64 data_head = ring_buffer_read_head(header);
11829 __u64 data_tail = header->data_tail;
11830 void *base = ((__u8 *)header) + page_size;
11831 int ret = LIBBPF_PERF_EVENT_CONT;
11832 struct perf_event_header *ehdr;
11835 while (data_head != data_tail) {
11836 ehdr = base + (data_tail & (mmap_size - 1));
11837 ehdr_size = ehdr->size;
11839 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
11840 void *copy_start = ehdr;
11841 size_t len_first = base + mmap_size - copy_start;
11842 size_t len_secnd = ehdr_size - len_first;
11844 if (*copy_size < ehdr_size) {
11846 *copy_mem = malloc(ehdr_size);
11849 ret = LIBBPF_PERF_EVENT_ERROR;
11852 *copy_size = ehdr_size;
11855 memcpy(*copy_mem, copy_start, len_first);
11856 memcpy(*copy_mem + len_first, base, len_secnd);
11860 ret = fn(ehdr, private_data);
11861 data_tail += ehdr_size;
11862 if (ret != LIBBPF_PERF_EVENT_CONT)
11866 ring_buffer_write_tail(header, data_tail);
11867 return libbpf_err(ret);
11870 struct perf_buffer;
11872 struct perf_buffer_params {
11873 struct perf_event_attr *attr;
11874 /* if event_cb is specified, it takes precendence */
11875 perf_buffer_event_fn event_cb;
11876 /* sample_cb and lost_cb are higher-level common-case callbacks */
11877 perf_buffer_sample_fn sample_cb;
11878 perf_buffer_lost_fn lost_cb;
11885 struct perf_cpu_buf {
11886 struct perf_buffer *pb;
11887 void *base; /* mmap()'ed memory */
11888 void *buf; /* for reconstructing segmented data */
11895 struct perf_buffer {
11896 perf_buffer_event_fn event_cb;
11897 perf_buffer_sample_fn sample_cb;
11898 perf_buffer_lost_fn lost_cb;
11899 void *ctx; /* passed into callbacks */
11903 struct perf_cpu_buf **cpu_bufs;
11904 struct epoll_event *events;
11905 int cpu_cnt; /* number of allocated CPU buffers */
11906 int epoll_fd; /* perf event FD */
11907 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
11910 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
11911 struct perf_cpu_buf *cpu_buf)
11915 if (cpu_buf->base &&
11916 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
11917 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
11918 if (cpu_buf->fd >= 0) {
11919 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
11920 close(cpu_buf->fd);
11922 free(cpu_buf->buf);
11926 void perf_buffer__free(struct perf_buffer *pb)
11930 if (IS_ERR_OR_NULL(pb))
11932 if (pb->cpu_bufs) {
11933 for (i = 0; i < pb->cpu_cnt; i++) {
11934 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
11939 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
11940 perf_buffer__free_cpu_buf(pb, cpu_buf);
11942 free(pb->cpu_bufs);
11944 if (pb->epoll_fd >= 0)
11945 close(pb->epoll_fd);
11950 static struct perf_cpu_buf *
11951 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
11952 int cpu, int map_key)
11954 struct perf_cpu_buf *cpu_buf;
11955 char msg[STRERR_BUFSIZE];
11958 cpu_buf = calloc(1, sizeof(*cpu_buf));
11960 return ERR_PTR(-ENOMEM);
11963 cpu_buf->cpu = cpu;
11964 cpu_buf->map_key = map_key;
11966 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
11967 -1, PERF_FLAG_FD_CLOEXEC);
11968 if (cpu_buf->fd < 0) {
11970 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
11971 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
11975 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
11976 PROT_READ | PROT_WRITE, MAP_SHARED,
11978 if (cpu_buf->base == MAP_FAILED) {
11979 cpu_buf->base = NULL;
11981 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
11982 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
11986 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
11988 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
11989 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
11996 perf_buffer__free_cpu_buf(pb, cpu_buf);
11997 return (struct perf_cpu_buf *)ERR_PTR(err);
12000 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
12001 struct perf_buffer_params *p);
12003 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
12004 perf_buffer_sample_fn sample_cb,
12005 perf_buffer_lost_fn lost_cb,
12007 const struct perf_buffer_opts *opts)
12009 const size_t attr_sz = sizeof(struct perf_event_attr);
12010 struct perf_buffer_params p = {};
12011 struct perf_event_attr attr;
12012 __u32 sample_period;
12014 if (!OPTS_VALID(opts, perf_buffer_opts))
12015 return libbpf_err_ptr(-EINVAL);
12017 sample_period = OPTS_GET(opts, sample_period, 1);
12018 if (!sample_period)
12021 memset(&attr, 0, attr_sz);
12022 attr.size = attr_sz;
12023 attr.config = PERF_COUNT_SW_BPF_OUTPUT;
12024 attr.type = PERF_TYPE_SOFTWARE;
12025 attr.sample_type = PERF_SAMPLE_RAW;
12026 attr.sample_period = sample_period;
12027 attr.wakeup_events = sample_period;
12030 p.sample_cb = sample_cb;
12031 p.lost_cb = lost_cb;
12034 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
12037 struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt,
12038 struct perf_event_attr *attr,
12039 perf_buffer_event_fn event_cb, void *ctx,
12040 const struct perf_buffer_raw_opts *opts)
12042 struct perf_buffer_params p = {};
12045 return libbpf_err_ptr(-EINVAL);
12047 if (!OPTS_VALID(opts, perf_buffer_raw_opts))
12048 return libbpf_err_ptr(-EINVAL);
12051 p.event_cb = event_cb;
12053 p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
12054 p.cpus = OPTS_GET(opts, cpus, NULL);
12055 p.map_keys = OPTS_GET(opts, map_keys, NULL);
12057 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
12060 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
12061 struct perf_buffer_params *p)
12063 const char *online_cpus_file = "/sys/devices/system/cpu/online";
12064 struct bpf_map_info map;
12065 char msg[STRERR_BUFSIZE];
12066 struct perf_buffer *pb;
12067 bool *online = NULL;
12068 __u32 map_info_len;
12071 if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) {
12072 pr_warn("page count should be power of two, but is %zu\n",
12074 return ERR_PTR(-EINVAL);
12077 /* best-effort sanity checks */
12078 memset(&map, 0, sizeof(map));
12079 map_info_len = sizeof(map);
12080 err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len);
12083 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
12084 * -EBADFD, -EFAULT, or -E2BIG on real error
12086 if (err != -EINVAL) {
12087 pr_warn("failed to get map info for map FD %d: %s\n",
12088 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
12089 return ERR_PTR(err);
12091 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
12094 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
12095 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
12097 return ERR_PTR(-EINVAL);
12101 pb = calloc(1, sizeof(*pb));
12103 return ERR_PTR(-ENOMEM);
12105 pb->event_cb = p->event_cb;
12106 pb->sample_cb = p->sample_cb;
12107 pb->lost_cb = p->lost_cb;
12110 pb->page_size = getpagesize();
12111 pb->mmap_size = pb->page_size * page_cnt;
12112 pb->map_fd = map_fd;
12114 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
12115 if (pb->epoll_fd < 0) {
12117 pr_warn("failed to create epoll instance: %s\n",
12118 libbpf_strerror_r(err, msg, sizeof(msg)));
12122 if (p->cpu_cnt > 0) {
12123 pb->cpu_cnt = p->cpu_cnt;
12125 pb->cpu_cnt = libbpf_num_possible_cpus();
12126 if (pb->cpu_cnt < 0) {
12130 if (map.max_entries && map.max_entries < pb->cpu_cnt)
12131 pb->cpu_cnt = map.max_entries;
12134 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
12137 pr_warn("failed to allocate events: out of memory\n");
12140 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
12141 if (!pb->cpu_bufs) {
12143 pr_warn("failed to allocate buffers: out of memory\n");
12147 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
12149 pr_warn("failed to get online CPU mask: %d\n", err);
12153 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
12154 struct perf_cpu_buf *cpu_buf;
12157 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
12158 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
12160 /* in case user didn't explicitly requested particular CPUs to
12161 * be attached to, skip offline/not present CPUs
12163 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
12166 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
12167 if (IS_ERR(cpu_buf)) {
12168 err = PTR_ERR(cpu_buf);
12172 pb->cpu_bufs[j] = cpu_buf;
12174 err = bpf_map_update_elem(pb->map_fd, &map_key,
12178 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
12179 cpu, map_key, cpu_buf->fd,
12180 libbpf_strerror_r(err, msg, sizeof(msg)));
12184 pb->events[j].events = EPOLLIN;
12185 pb->events[j].data.ptr = cpu_buf;
12186 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
12187 &pb->events[j]) < 0) {
12189 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
12191 libbpf_strerror_r(err, msg, sizeof(msg)));
12204 perf_buffer__free(pb);
12205 return ERR_PTR(err);
12208 struct perf_sample_raw {
12209 struct perf_event_header header;
12214 struct perf_sample_lost {
12215 struct perf_event_header header;
12218 uint64_t sample_id;
12221 static enum bpf_perf_event_ret
12222 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
12224 struct perf_cpu_buf *cpu_buf = ctx;
12225 struct perf_buffer *pb = cpu_buf->pb;
12228 /* user wants full control over parsing perf event */
12230 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
12233 case PERF_RECORD_SAMPLE: {
12234 struct perf_sample_raw *s = data;
12237 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
12240 case PERF_RECORD_LOST: {
12241 struct perf_sample_lost *s = data;
12244 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
12248 pr_warn("unknown perf sample type %d\n", e->type);
12249 return LIBBPF_PERF_EVENT_ERROR;
12251 return LIBBPF_PERF_EVENT_CONT;
12254 static int perf_buffer__process_records(struct perf_buffer *pb,
12255 struct perf_cpu_buf *cpu_buf)
12257 enum bpf_perf_event_ret ret;
12259 ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
12260 pb->page_size, &cpu_buf->buf,
12261 &cpu_buf->buf_size,
12262 perf_buffer__process_record, cpu_buf);
12263 if (ret != LIBBPF_PERF_EVENT_CONT)
12268 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
12270 return pb->epoll_fd;
12273 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
12277 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
12281 for (i = 0; i < cnt; i++) {
12282 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
12284 err = perf_buffer__process_records(pb, cpu_buf);
12286 pr_warn("error while processing records: %d\n", err);
12287 return libbpf_err(err);
12293 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
12296 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
12298 return pb->cpu_cnt;
12302 * Return perf_event FD of a ring buffer in *buf_idx* slot of
12303 * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
12304 * select()/poll()/epoll() Linux syscalls.
12306 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
12308 struct perf_cpu_buf *cpu_buf;
12310 if (buf_idx >= pb->cpu_cnt)
12311 return libbpf_err(-EINVAL);
12313 cpu_buf = pb->cpu_bufs[buf_idx];
12315 return libbpf_err(-ENOENT);
12317 return cpu_buf->fd;
12320 int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size)
12322 struct perf_cpu_buf *cpu_buf;
12324 if (buf_idx >= pb->cpu_cnt)
12325 return libbpf_err(-EINVAL);
12327 cpu_buf = pb->cpu_bufs[buf_idx];
12329 return libbpf_err(-ENOENT);
12331 *buf = cpu_buf->base;
12332 *buf_size = pb->mmap_size;
12337 * Consume data from perf ring buffer corresponding to slot *buf_idx* in
12338 * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
12339 * consume, do nothing and return success.
12344 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
12346 struct perf_cpu_buf *cpu_buf;
12348 if (buf_idx >= pb->cpu_cnt)
12349 return libbpf_err(-EINVAL);
12351 cpu_buf = pb->cpu_bufs[buf_idx];
12353 return libbpf_err(-ENOENT);
12355 return perf_buffer__process_records(pb, cpu_buf);
12358 int perf_buffer__consume(struct perf_buffer *pb)
12362 for (i = 0; i < pb->cpu_cnt; i++) {
12363 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
12368 err = perf_buffer__process_records(pb, cpu_buf);
12370 pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
12371 return libbpf_err(err);
12377 int bpf_program__set_attach_target(struct bpf_program *prog,
12378 int attach_prog_fd,
12379 const char *attach_func_name)
12381 int btf_obj_fd = 0, btf_id = 0, err;
12383 if (!prog || attach_prog_fd < 0)
12384 return libbpf_err(-EINVAL);
12386 if (prog->obj->loaded)
12387 return libbpf_err(-EINVAL);
12389 if (attach_prog_fd && !attach_func_name) {
12390 /* remember attach_prog_fd and let bpf_program__load() find
12391 * BTF ID during the program load
12393 prog->attach_prog_fd = attach_prog_fd;
12397 if (attach_prog_fd) {
12398 btf_id = libbpf_find_prog_btf_id(attach_func_name,
12401 return libbpf_err(btf_id);
12403 if (!attach_func_name)
12404 return libbpf_err(-EINVAL);
12406 /* load btf_vmlinux, if not yet */
12407 err = bpf_object__load_vmlinux_btf(prog->obj, true);
12409 return libbpf_err(err);
12410 err = find_kernel_btf_id(prog->obj, attach_func_name,
12411 prog->expected_attach_type,
12412 &btf_obj_fd, &btf_id);
12414 return libbpf_err(err);
12417 prog->attach_btf_id = btf_id;
12418 prog->attach_btf_obj_fd = btf_obj_fd;
12419 prog->attach_prog_fd = attach_prog_fd;
12423 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
12425 int err = 0, n, len, start, end = -1;
12431 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
12433 if (*s == ',' || *s == '\n') {
12437 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
12438 if (n <= 0 || n > 2) {
12439 pr_warn("Failed to get CPU range %s: %d\n", s, n);
12442 } else if (n == 1) {
12445 if (start < 0 || start > end) {
12446 pr_warn("Invalid CPU range [%d,%d] in %s\n",
12451 tmp = realloc(*mask, end + 1);
12457 memset(tmp + *mask_sz, 0, start - *mask_sz);
12458 memset(tmp + start, 1, end - start + 1);
12459 *mask_sz = end + 1;
12463 pr_warn("Empty CPU range\n");
12473 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
12475 int fd, err = 0, len;
12478 fd = open(fcpu, O_RDONLY | O_CLOEXEC);
12481 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
12484 len = read(fd, buf, sizeof(buf));
12487 err = len ? -errno : -EINVAL;
12488 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
12491 if (len >= sizeof(buf)) {
12492 pr_warn("CPU mask is too big in file %s\n", fcpu);
12497 return parse_cpu_mask_str(buf, mask, mask_sz);
12500 int libbpf_num_possible_cpus(void)
12502 static const char *fcpu = "/sys/devices/system/cpu/possible";
12504 int err, n, i, tmp_cpus;
12507 tmp_cpus = READ_ONCE(cpus);
12511 err = parse_cpu_mask_file(fcpu, &mask, &n);
12513 return libbpf_err(err);
12516 for (i = 0; i < n; i++) {
12522 WRITE_ONCE(cpus, tmp_cpus);
12526 static int populate_skeleton_maps(const struct bpf_object *obj,
12527 struct bpf_map_skeleton *maps,
12532 for (i = 0; i < map_cnt; i++) {
12533 struct bpf_map **map = maps[i].map;
12534 const char *name = maps[i].name;
12535 void **mmaped = maps[i].mmaped;
12537 *map = bpf_object__find_map_by_name(obj, name);
12539 pr_warn("failed to find skeleton map '%s'\n", name);
12543 /* externs shouldn't be pre-setup from user code */
12544 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
12545 *mmaped = (*map)->mmaped;
12550 static int populate_skeleton_progs(const struct bpf_object *obj,
12551 struct bpf_prog_skeleton *progs,
12556 for (i = 0; i < prog_cnt; i++) {
12557 struct bpf_program **prog = progs[i].prog;
12558 const char *name = progs[i].name;
12560 *prog = bpf_object__find_program_by_name(obj, name);
12562 pr_warn("failed to find skeleton program '%s'\n", name);
12569 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
12570 const struct bpf_object_open_opts *opts)
12572 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
12573 .object_name = s->name,
12575 struct bpf_object *obj;
12578 /* Attempt to preserve opts->object_name, unless overriden by user
12579 * explicitly. Overwriting object name for skeletons is discouraged,
12580 * as it breaks global data maps, because they contain object name
12581 * prefix as their own map name prefix. When skeleton is generated,
12582 * bpftool is making an assumption that this name will stay the same.
12585 memcpy(&skel_opts, opts, sizeof(*opts));
12586 if (!opts->object_name)
12587 skel_opts.object_name = s->name;
12590 obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
12591 err = libbpf_get_error(obj);
12593 pr_warn("failed to initialize skeleton BPF object '%s': %d\n",
12595 return libbpf_err(err);
12599 err = populate_skeleton_maps(obj, s->maps, s->map_cnt);
12601 pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err);
12602 return libbpf_err(err);
12605 err = populate_skeleton_progs(obj, s->progs, s->prog_cnt);
12607 pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err);
12608 return libbpf_err(err);
12614 int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s)
12616 int err, len, var_idx, i;
12617 const char *var_name;
12618 const struct bpf_map *map;
12621 const struct btf_type *map_type, *var_type;
12622 const struct bpf_var_skeleton *var_skel;
12623 struct btf_var_secinfo *var;
12626 return libbpf_err(-EINVAL);
12628 btf = bpf_object__btf(s->obj);
12630 pr_warn("subskeletons require BTF at runtime (object %s)\n",
12631 bpf_object__name(s->obj));
12632 return libbpf_err(-errno);
12635 err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt);
12637 pr_warn("failed to populate subskeleton maps: %d\n", err);
12638 return libbpf_err(err);
12641 err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt);
12643 pr_warn("failed to populate subskeleton maps: %d\n", err);
12644 return libbpf_err(err);
12647 for (var_idx = 0; var_idx < s->var_cnt; var_idx++) {
12648 var_skel = &s->vars[var_idx];
12649 map = *var_skel->map;
12650 map_type_id = bpf_map__btf_value_type_id(map);
12651 map_type = btf__type_by_id(btf, map_type_id);
12653 if (!btf_is_datasec(map_type)) {
12654 pr_warn("type for map '%1$s' is not a datasec: %2$s",
12655 bpf_map__name(map),
12656 __btf_kind_str(btf_kind(map_type)));
12657 return libbpf_err(-EINVAL);
12660 len = btf_vlen(map_type);
12661 var = btf_var_secinfos(map_type);
12662 for (i = 0; i < len; i++, var++) {
12663 var_type = btf__type_by_id(btf, var->type);
12664 var_name = btf__name_by_offset(btf, var_type->name_off);
12665 if (strcmp(var_name, var_skel->name) == 0) {
12666 *var_skel->addr = map->mmaped + var->offset;
12674 void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s)
12684 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
12688 err = bpf_object__load(*s->obj);
12690 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
12691 return libbpf_err(err);
12694 for (i = 0; i < s->map_cnt; i++) {
12695 struct bpf_map *map = *s->maps[i].map;
12696 size_t mmap_sz = bpf_map_mmap_sz(map);
12697 int prot, map_fd = bpf_map__fd(map);
12698 void **mmaped = s->maps[i].mmaped;
12703 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
12708 if (map->def.map_flags & BPF_F_RDONLY_PROG)
12711 prot = PROT_READ | PROT_WRITE;
12713 /* Remap anonymous mmap()-ed "map initialization image" as
12714 * a BPF map-backed mmap()-ed memory, but preserving the same
12715 * memory address. This will cause kernel to change process'
12716 * page table to point to a different piece of kernel memory,
12717 * but from userspace point of view memory address (and its
12718 * contents, being identical at this point) will stay the
12719 * same. This mapping will be released by bpf_object__close()
12720 * as per normal clean up procedure, so we don't need to worry
12721 * about it from skeleton's clean up perspective.
12723 *mmaped = mmap(map->mmaped, mmap_sz, prot,
12724 MAP_SHARED | MAP_FIXED, map_fd, 0);
12725 if (*mmaped == MAP_FAILED) {
12728 pr_warn("failed to re-mmap() map '%s': %d\n",
12729 bpf_map__name(map), err);
12730 return libbpf_err(err);
12737 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
12741 for (i = 0; i < s->prog_cnt; i++) {
12742 struct bpf_program *prog = *s->progs[i].prog;
12743 struct bpf_link **link = s->progs[i].link;
12745 if (!prog->autoload || !prog->autoattach)
12748 /* auto-attaching not supported for this program */
12749 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
12752 /* if user already set the link manually, don't attempt auto-attach */
12756 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link);
12758 pr_warn("prog '%s': failed to auto-attach: %d\n",
12759 bpf_program__name(prog), err);
12760 return libbpf_err(err);
12763 /* It's possible that for some SEC() definitions auto-attach
12764 * is supported in some cases (e.g., if definition completely
12765 * specifies target information), but is not in other cases.
12766 * SEC("uprobe") is one such case. If user specified target
12767 * binary and function name, such BPF program can be
12768 * auto-attached. But if not, it shouldn't trigger skeleton's
12769 * attach to fail. It should just be skipped.
12770 * attach_fn signals such case with returning 0 (no error) and
12771 * setting link to NULL.
12778 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
12782 for (i = 0; i < s->prog_cnt; i++) {
12783 struct bpf_link **link = s->progs[i].link;
12785 bpf_link__destroy(*link);
12790 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
12796 bpf_object__detach_skeleton(s);
12798 bpf_object__close(*s->obj);