1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * Common eBPF ELF object loading operations.
5 *
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.
11 */
12
13 #ifndef _GNU_SOURCE
14 #define _GNU_SOURCE
15 #endif
16 #include <stdlib.h>
17 #include <stdio.h>
18 #include <stdarg.h>
19 #include <libgen.h>
20 #include <inttypes.h>
21 #include <limits.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <endian.h>
25 #include <fcntl.h>
26 #include <errno.h>
27 #include <ctype.h>
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 <linux/version.h>
38 #include <sys/epoll.h>
39 #include <sys/ioctl.h>
40 #include <sys/mman.h>
41 #include <sys/stat.h>
42 #include <sys/types.h>
43 #include <sys/vfs.h>
44 #include <sys/utsname.h>
45 #include <sys/resource.h>
46 #ifdef HAVE_LIBELF
47 #include <libelf.h>
48 #include <gelf.h>
49 #endif
50
51 #include <zlib.h>
52
53 #include "libbpf.h"
54 #include "bpf.h"
55 #include "btf.h"
56 #include "str_error.h"
57 #include "libbpf_internal.h"
58 #include "hashmap.h"
59 #include "bpf_gen_internal.h"
60
61 #ifdef HAVE_ELFIO
62 #include "elfio_c_wrapper.h"
63 #include <linux/memfd.h>
64 #include <sys/syscall.h>
65 #include <limits.h>
66
67 typedef struct Elf64_Ehdr Elf64_Ehdr;
68 typedef struct Elf64_Shdr Elf64_Shdr;
69 typedef struct Elf64_Sym Elf64_Sym;
70 typedef struct Elf64_Rel Elf64_Rel;
71 typedef struct {
72 void *d_buf;
73 size_t d_size;
74 } Elf_Data;
75
76 #define ELF64_ST_TYPE(val) ELF_ST_TYPE (val)
77 #define ELF64_ST_BIND(val) ELF_ST_BIND (val)
78 #define elf_errmsg(val) "error"
79
80 #endif
81
82
83 #ifndef BPF_FS_MAGIC
84 #define BPF_FS_MAGIC 0xcafe4a11
85 #endif
86
87 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
88
89 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
90 * compilation if user enables corresponding warning. Disable it explicitly.
91 */
92 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
93
94 #define __printf(a, b) __attribute__((format(printf, a, b)))
95
96 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
97 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
98
99 static const char * const attach_type_name[] = {
100 [BPF_CGROUP_INET_INGRESS] = "cgroup_inet_ingress",
101 [BPF_CGROUP_INET_EGRESS] = "cgroup_inet_egress",
102 [BPF_CGROUP_INET_SOCK_CREATE] = "cgroup_inet_sock_create",
103 [BPF_CGROUP_INET_SOCK_RELEASE] = "cgroup_inet_sock_release",
104 [BPF_CGROUP_SOCK_OPS] = "cgroup_sock_ops",
105 [BPF_CGROUP_DEVICE] = "cgroup_device",
106 [BPF_CGROUP_INET4_BIND] = "cgroup_inet4_bind",
107 [BPF_CGROUP_INET6_BIND] = "cgroup_inet6_bind",
108 [BPF_CGROUP_INET4_CONNECT] = "cgroup_inet4_connect",
109 [BPF_CGROUP_INET6_CONNECT] = "cgroup_inet6_connect",
110 [BPF_CGROUP_INET4_POST_BIND] = "cgroup_inet4_post_bind",
111 [BPF_CGROUP_INET6_POST_BIND] = "cgroup_inet6_post_bind",
112 [BPF_CGROUP_INET4_GETPEERNAME] = "cgroup_inet4_getpeername",
113 [BPF_CGROUP_INET6_GETPEERNAME] = "cgroup_inet6_getpeername",
114 [BPF_CGROUP_INET4_GETSOCKNAME] = "cgroup_inet4_getsockname",
115 [BPF_CGROUP_INET6_GETSOCKNAME] = "cgroup_inet6_getsockname",
116 [BPF_CGROUP_UDP4_SENDMSG] = "cgroup_udp4_sendmsg",
117 [BPF_CGROUP_UDP6_SENDMSG] = "cgroup_udp6_sendmsg",
118 [BPF_CGROUP_SYSCTL] = "cgroup_sysctl",
119 [BPF_CGROUP_UDP4_RECVMSG] = "cgroup_udp4_recvmsg",
120 [BPF_CGROUP_UDP6_RECVMSG] = "cgroup_udp6_recvmsg",
121 [BPF_CGROUP_GETSOCKOPT] = "cgroup_getsockopt",
122 [BPF_CGROUP_SETSOCKOPT] = "cgroup_setsockopt",
123 [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser",
124 [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict",
125 [BPF_SK_SKB_VERDICT] = "sk_skb_verdict",
126 [BPF_SK_MSG_VERDICT] = "sk_msg_verdict",
127 [BPF_LIRC_MODE2] = "lirc_mode2",
128 [BPF_FLOW_DISSECTOR] = "flow_dissector",
129 [BPF_TRACE_RAW_TP] = "trace_raw_tp",
130 [BPF_TRACE_FENTRY] = "trace_fentry",
131 [BPF_TRACE_FEXIT] = "trace_fexit",
132 [BPF_MODIFY_RETURN] = "modify_return",
133 [BPF_LSM_MAC] = "lsm_mac",
134 [BPF_LSM_CGROUP] = "lsm_cgroup",
135 [BPF_SK_LOOKUP] = "sk_lookup",
136 [BPF_TRACE_ITER] = "trace_iter",
137 [BPF_XDP_DEVMAP] = "xdp_devmap",
138 [BPF_XDP_CPUMAP] = "xdp_cpumap",
139 [BPF_XDP] = "xdp",
140 [BPF_SK_REUSEPORT_SELECT] = "sk_reuseport_select",
141 [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_reuseport_select_or_migrate",
142 [BPF_PERF_EVENT] = "perf_event",
143 [BPF_TRACE_KPROBE_MULTI] = "trace_kprobe_multi",
144 };
145
146 static const char * const link_type_name[] = {
147 [BPF_LINK_TYPE_UNSPEC] = "unspec",
148 [BPF_LINK_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
149 [BPF_LINK_TYPE_TRACING] = "tracing",
150 [BPF_LINK_TYPE_CGROUP] = "cgroup",
151 [BPF_LINK_TYPE_ITER] = "iter",
152 [BPF_LINK_TYPE_NETNS] = "netns",
153 [BPF_LINK_TYPE_XDP] = "xdp",
154 [BPF_LINK_TYPE_PERF_EVENT] = "perf_event",
155 [BPF_LINK_TYPE_KPROBE_MULTI] = "kprobe_multi",
156 [BPF_LINK_TYPE_STRUCT_OPS] = "struct_ops",
157 };
158
159 static const char * const map_type_name[] = {
160 [BPF_MAP_TYPE_UNSPEC] = "unspec",
161 [BPF_MAP_TYPE_HASH] = "hash",
162 [BPF_MAP_TYPE_ARRAY] = "array",
163 [BPF_MAP_TYPE_PROG_ARRAY] = "prog_array",
164 [BPF_MAP_TYPE_PERF_EVENT_ARRAY] = "perf_event_array",
165 [BPF_MAP_TYPE_PERCPU_HASH] = "percpu_hash",
166 [BPF_MAP_TYPE_PERCPU_ARRAY] = "percpu_array",
167 [BPF_MAP_TYPE_STACK_TRACE] = "stack_trace",
168 [BPF_MAP_TYPE_CGROUP_ARRAY] = "cgroup_array",
169 [BPF_MAP_TYPE_LRU_HASH] = "lru_hash",
170 [BPF_MAP_TYPE_LRU_PERCPU_HASH] = "lru_percpu_hash",
171 [BPF_MAP_TYPE_LPM_TRIE] = "lpm_trie",
172 [BPF_MAP_TYPE_ARRAY_OF_MAPS] = "array_of_maps",
173 [BPF_MAP_TYPE_HASH_OF_MAPS] = "hash_of_maps",
174 [BPF_MAP_TYPE_DEVMAP] = "devmap",
175 [BPF_MAP_TYPE_DEVMAP_HASH] = "devmap_hash",
176 [BPF_MAP_TYPE_SOCKMAP] = "sockmap",
177 [BPF_MAP_TYPE_CPUMAP] = "cpumap",
178 [BPF_MAP_TYPE_XSKMAP] = "xskmap",
179 [BPF_MAP_TYPE_SOCKHASH] = "sockhash",
180 [BPF_MAP_TYPE_CGROUP_STORAGE] = "cgroup_storage",
181 [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY] = "reuseport_sockarray",
182 [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE] = "percpu_cgroup_storage",
183 [BPF_MAP_TYPE_QUEUE] = "queue",
184 [BPF_MAP_TYPE_STACK] = "stack",
185 [BPF_MAP_TYPE_SK_STORAGE] = "sk_storage",
186 [BPF_MAP_TYPE_STRUCT_OPS] = "struct_ops",
187 [BPF_MAP_TYPE_RINGBUF] = "ringbuf",
188 [BPF_MAP_TYPE_INODE_STORAGE] = "inode_storage",
189 [BPF_MAP_TYPE_TASK_STORAGE] = "task_storage",
190 [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter",
191 [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf",
192 [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage",
193 };
194
195 static const char * const prog_type_name[] = {
196 [BPF_PROG_TYPE_UNSPEC] = "unspec",
197 [BPF_PROG_TYPE_SOCKET_FILTER] = "socket_filter",
198 [BPF_PROG_TYPE_KPROBE] = "kprobe",
199 [BPF_PROG_TYPE_SCHED_CLS] = "sched_cls",
200 [BPF_PROG_TYPE_SCHED_ACT] = "sched_act",
201 [BPF_PROG_TYPE_TRACEPOINT] = "tracepoint",
202 [BPF_PROG_TYPE_XDP] = "xdp",
203 [BPF_PROG_TYPE_PERF_EVENT] = "perf_event",
204 [BPF_PROG_TYPE_CGROUP_SKB] = "cgroup_skb",
205 [BPF_PROG_TYPE_CGROUP_SOCK] = "cgroup_sock",
206 [BPF_PROG_TYPE_LWT_IN] = "lwt_in",
207 [BPF_PROG_TYPE_LWT_OUT] = "lwt_out",
208 [BPF_PROG_TYPE_LWT_XMIT] = "lwt_xmit",
209 [BPF_PROG_TYPE_SOCK_OPS] = "sock_ops",
210 [BPF_PROG_TYPE_SK_SKB] = "sk_skb",
211 [BPF_PROG_TYPE_CGROUP_DEVICE] = "cgroup_device",
212 [BPF_PROG_TYPE_SK_MSG] = "sk_msg",
213 [BPF_PROG_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
214 [BPF_PROG_TYPE_CGROUP_SOCK_ADDR] = "cgroup_sock_addr",
215 [BPF_PROG_TYPE_LWT_SEG6LOCAL] = "lwt_seg6local",
216 [BPF_PROG_TYPE_LIRC_MODE2] = "lirc_mode2",
217 [BPF_PROG_TYPE_SK_REUSEPORT] = "sk_reuseport",
218 [BPF_PROG_TYPE_FLOW_DISSECTOR] = "flow_dissector",
219 [BPF_PROG_TYPE_CGROUP_SYSCTL] = "cgroup_sysctl",
220 [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable",
221 [BPF_PROG_TYPE_CGROUP_SOCKOPT] = "cgroup_sockopt",
222 [BPF_PROG_TYPE_TRACING] = "tracing",
223 [BPF_PROG_TYPE_STRUCT_OPS] = "struct_ops",
224 [BPF_PROG_TYPE_EXT] = "ext",
225 [BPF_PROG_TYPE_LSM] = "lsm",
226 [BPF_PROG_TYPE_SK_LOOKUP] = "sk_lookup",
227 [BPF_PROG_TYPE_SYSCALL] = "syscall",
228 };
229
__base_pr(enum libbpf_print_level level,const char * format,va_list args)230 static int __base_pr(enum libbpf_print_level level, const char *format,
231 va_list args)
232 {
233 if (level == LIBBPF_DEBUG)
234 return 0;
235
236 return vfprintf(stderr, format, args);
237 }
238
239 static libbpf_print_fn_t __libbpf_pr = __base_pr;
240
libbpf_set_print(libbpf_print_fn_t fn)241 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
242 {
243 libbpf_print_fn_t old_print_fn = __libbpf_pr;
244
245 __libbpf_pr = fn;
246 return old_print_fn;
247 }
248
249 __printf(2, 3)
libbpf_print(enum libbpf_print_level level,const char * format,...)250 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
251 {
252 va_list args;
253 int old_errno;
254
255 if (!__libbpf_pr)
256 return;
257
258 old_errno = errno;
259
260 va_start(args, format);
261 __libbpf_pr(level, format, args);
262 va_end(args);
263
264 errno = old_errno;
265 }
266
pr_perm_msg(int err)267 static void pr_perm_msg(int err)
268 {
269 struct rlimit limit;
270 char buf[100];
271
272 if (err != -EPERM || geteuid() != 0)
273 return;
274
275 err = getrlimit(RLIMIT_MEMLOCK, &limit);
276 if (err)
277 return;
278
279 if (limit.rlim_cur == RLIM_INFINITY)
280 return;
281
282 if (limit.rlim_cur < 1024)
283 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
284 else if (limit.rlim_cur < 1024*1024)
285 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
286 else
287 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
288
289 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
290 buf);
291 }
292
293 #define STRERR_BUFSIZE 128
294
295 /* Copied from tools/perf/util/util.h */
296 #ifndef zfree
297 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
298 #endif
299
300 #ifndef zclose
301 # define zclose(fd) ({ \
302 int ___err = 0; \
303 if ((fd) >= 0) \
304 ___err = close((fd)); \
305 fd = -1; \
306 ___err; })
307 #endif
308
ptr_to_u64(const void * ptr)309 static inline __u64 ptr_to_u64(const void *ptr)
310 {
311 return (__u64) (unsigned long) ptr;
312 }
313
libbpf_set_strict_mode(enum libbpf_strict_mode mode)314 int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
315 {
316 /* as of v1.0 libbpf_set_strict_mode() is a no-op */
317 return 0;
318 }
319
libbpf_major_version(void)320 __u32 libbpf_major_version(void)
321 {
322 return LIBBPF_MAJOR_VERSION;
323 }
324
libbpf_minor_version(void)325 __u32 libbpf_minor_version(void)
326 {
327 return LIBBPF_MINOR_VERSION;
328 }
329
libbpf_version_string(void)330 const char *libbpf_version_string(void)
331 {
332 #define __S(X) #X
333 #define _S(X) __S(X)
334 return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
335 #undef _S
336 #undef __S
337 }
338
339 enum reloc_type {
340 RELO_LD64,
341 RELO_CALL,
342 RELO_DATA,
343 RELO_EXTERN_VAR,
344 RELO_EXTERN_FUNC,
345 RELO_SUBPROG_ADDR,
346 RELO_CORE,
347 };
348
349 struct reloc_desc {
350 enum reloc_type type;
351 int insn_idx;
352 union {
353 const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
354 struct {
355 int map_idx;
356 int sym_off;
357 };
358 };
359 };
360
361 /* stored as sec_def->cookie for all libbpf-supported SEC()s */
362 enum sec_def_flags {
363 SEC_NONE = 0,
364 /* expected_attach_type is optional, if kernel doesn't support that */
365 SEC_EXP_ATTACH_OPT = 1,
366 /* legacy, only used by libbpf_get_type_names() and
367 * libbpf_attach_type_by_name(), not used by libbpf itself at all.
368 * This used to be associated with cgroup (and few other) BPF programs
369 * that were attachable through BPF_PROG_ATTACH command. Pretty
370 * meaningless nowadays, though.
371 */
372 SEC_ATTACHABLE = 2,
373 SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
374 /* attachment target is specified through BTF ID in either kernel or
375 * other BPF program's BTF object
376 */
377 SEC_ATTACH_BTF = 4,
378 /* BPF program type allows sleeping/blocking in kernel */
379 SEC_SLEEPABLE = 8,
380 /* BPF program support non-linear XDP buffer */
381 SEC_XDP_FRAGS = 16,
382 };
383
384 struct bpf_sec_def {
385 char *sec;
386 enum bpf_prog_type prog_type;
387 enum bpf_attach_type expected_attach_type;
388 long cookie;
389 int handler_id;
390
391 libbpf_prog_setup_fn_t prog_setup_fn;
392 libbpf_prog_prepare_load_fn_t prog_prepare_load_fn;
393 libbpf_prog_attach_fn_t prog_attach_fn;
394 };
395
396 /*
397 * bpf_prog should be a better name but it has been used in
398 * linux/filter.h.
399 */
400 struct bpf_program {
401 char *name;
402 char *sec_name;
403 size_t sec_idx;
404 const struct bpf_sec_def *sec_def;
405 /* this program's instruction offset (in number of instructions)
406 * within its containing ELF section
407 */
408 size_t sec_insn_off;
409 /* number of original instructions in ELF section belonging to this
410 * program, not taking into account subprogram instructions possible
411 * appended later during relocation
412 */
413 size_t sec_insn_cnt;
414 /* Offset (in number of instructions) of the start of instruction
415 * belonging to this BPF program within its containing main BPF
416 * program. For the entry-point (main) BPF program, this is always
417 * zero. For a sub-program, this gets reset before each of main BPF
418 * programs are processed and relocated and is used to determined
419 * whether sub-program was already appended to the main program, and
420 * if yes, at which instruction offset.
421 */
422 size_t sub_insn_off;
423
424 /* instructions that belong to BPF program; insns[0] is located at
425 * sec_insn_off instruction within its ELF section in ELF file, so
426 * when mapping ELF file instruction index to the local instruction,
427 * one needs to subtract sec_insn_off; and vice versa.
428 */
429 struct bpf_insn *insns;
430 /* actual number of instruction in this BPF program's image; for
431 * entry-point BPF programs this includes the size of main program
432 * itself plus all the used sub-programs, appended at the end
433 */
434 size_t insns_cnt;
435
436 struct reloc_desc *reloc_desc;
437 int nr_reloc;
438
439 /* BPF verifier log settings */
440 char *log_buf;
441 size_t log_size;
442 __u32 log_level;
443
444 struct bpf_object *obj;
445
446 int fd;
447 bool autoload;
448 bool autoattach;
449 bool mark_btf_static;
450 enum bpf_prog_type type;
451 enum bpf_attach_type expected_attach_type;
452
453 int prog_ifindex;
454 __u32 attach_btf_obj_fd;
455 __u32 attach_btf_id;
456 __u32 attach_prog_fd;
457
458 void *func_info;
459 __u32 func_info_rec_size;
460 __u32 func_info_cnt;
461
462 void *line_info;
463 __u32 line_info_rec_size;
464 __u32 line_info_cnt;
465 __u32 prog_flags;
466 };
467
468 struct bpf_struct_ops {
469 const char *tname;
470 const struct btf_type *type;
471 struct bpf_program **progs;
472 __u32 *kern_func_off;
473 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
474 void *data;
475 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
476 * btf_vmlinux's format.
477 * struct bpf_struct_ops_tcp_congestion_ops {
478 * [... some other kernel fields ...]
479 * struct tcp_congestion_ops data;
480 * }
481 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
482 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
483 * from "data".
484 */
485 void *kern_vdata;
486 __u32 type_id;
487 };
488
489 #define DATA_SEC ".data"
490 #define BSS_SEC ".bss"
491 #define RODATA_SEC ".rodata"
492 #define KCONFIG_SEC ".kconfig"
493 #define KSYMS_SEC ".ksyms"
494 #define STRUCT_OPS_SEC ".struct_ops"
495
496 enum libbpf_map_type {
497 LIBBPF_MAP_UNSPEC,
498 LIBBPF_MAP_DATA,
499 LIBBPF_MAP_BSS,
500 LIBBPF_MAP_RODATA,
501 LIBBPF_MAP_KCONFIG,
502 };
503
504 struct bpf_map_def {
505 unsigned int type;
506 unsigned int key_size;
507 unsigned int value_size;
508 unsigned int max_entries;
509 unsigned int map_flags;
510 };
511
512 struct bpf_map {
513 struct bpf_object *obj;
514 char *name;
515 /* real_name is defined for special internal maps (.rodata*,
516 * .data*, .bss, .kconfig) and preserves their original ELF section
517 * name. This is important to be able to find corresponding BTF
518 * DATASEC information.
519 */
520 char *real_name;
521 int fd;
522 int sec_idx;
523 size_t sec_offset;
524 int map_ifindex;
525 int inner_map_fd;
526 struct bpf_map_def def;
527 __u32 numa_node;
528 __u32 btf_var_idx;
529 __u32 btf_key_type_id;
530 __u32 btf_value_type_id;
531 __u32 btf_vmlinux_value_type_id;
532 enum libbpf_map_type libbpf_type;
533 void *mmaped;
534 struct bpf_struct_ops *st_ops;
535 struct bpf_map *inner_map;
536 void **init_slots;
537 int init_slots_sz;
538 char *pin_path;
539 bool pinned;
540 bool reused;
541 bool autocreate;
542 __u64 map_extra;
543 };
544
545 enum extern_type {
546 EXT_UNKNOWN,
547 EXT_KCFG,
548 EXT_KSYM,
549 };
550
551 enum kcfg_type {
552 KCFG_UNKNOWN,
553 KCFG_CHAR,
554 KCFG_BOOL,
555 KCFG_INT,
556 KCFG_TRISTATE,
557 KCFG_CHAR_ARR,
558 };
559
560 struct extern_desc {
561 enum extern_type type;
562 int sym_idx;
563 int btf_id;
564 int sec_btf_id;
565 const char *name;
566 bool is_set;
567 bool is_weak;
568 union {
569 struct {
570 enum kcfg_type type;
571 int sz;
572 int align;
573 int data_off;
574 bool is_signed;
575 } kcfg;
576 struct {
577 unsigned long long addr;
578
579 /* target btf_id of the corresponding kernel var. */
580 int kernel_btf_obj_fd;
581 int kernel_btf_id;
582
583 /* local btf_id of the ksym extern's type. */
584 __u32 type_id;
585 /* BTF fd index to be patched in for insn->off, this is
586 * 0 for vmlinux BTF, index in obj->fd_array for module
587 * BTF
588 */
589 __s16 btf_fd_idx;
590 } ksym;
591 };
592 };
593
594 struct module_btf {
595 struct btf *btf;
596 char *name;
597 __u32 id;
598 int fd;
599 int fd_array_idx;
600 };
601
602 enum sec_type {
603 SEC_UNUSED = 0,
604 SEC_RELO,
605 SEC_BSS,
606 SEC_DATA,
607 SEC_RODATA,
608 };
609
610 struct elf_sec_desc {
611 enum sec_type sec_type;
612 #if defined HAVE_LIBELF
613 Elf64_Shdr *shdr;
614 #elif defined HAVE_ELFIO
615 psection_t psection;
616 Elf_Data realdata;
617 #endif
618 Elf_Data *data;
619 };
620
621 struct elf_state {
622 int fd;
623 const void *obj_buf;
624 size_t obj_buf_sz;
625 #if defined HAVE_LIBELF
626 Elf *elf;
627 #elif defined HAVE_ELFIO
628 pelfio_t elf;
629 Elf64_Ehdr eheader;
630 pstring_t shstring;
631 pstring_t strstring;
632 Elf_Data realsymbols;
633 Elf_Data realst_ops_data;
634 #endif
635 Elf64_Ehdr *ehdr;
636 Elf_Data *symbols;
637 Elf_Data *st_ops_data;
638 size_t shstrndx; /* section index for section name strings */
639 size_t strtabidx;
640 struct elf_sec_desc *secs;
641 size_t sec_cnt;
642 int btf_maps_shndx;
643 __u32 btf_maps_sec_btf_id;
644 int text_shndx;
645 int symbols_shndx;
646 int st_ops_shndx;
647 };
648
649 struct usdt_manager;
650
651 struct bpf_object {
652 char name[BPF_OBJ_NAME_LEN];
653 char license[64];
654 __u32 kern_version;
655
656 struct bpf_program *programs;
657 size_t nr_programs;
658 struct bpf_map *maps;
659 size_t nr_maps;
660 size_t maps_cap;
661
662 char *kconfig;
663 struct extern_desc *externs;
664 int nr_extern;
665 int kconfig_map_idx;
666
667 bool loaded;
668 bool has_subcalls;
669 bool has_rodata;
670
671 struct bpf_gen *gen_loader;
672
673 /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
674 struct elf_state efile;
675
676 struct btf *btf;
677 struct btf_ext *btf_ext;
678
679 /* Parse and load BTF vmlinux if any of the programs in the object need
680 * it at load time.
681 */
682 struct btf *btf_vmlinux;
683 /* Path to the custom BTF to be used for BPF CO-RE relocations as an
684 * override for vmlinux BTF.
685 */
686 char *btf_custom_path;
687 /* vmlinux BTF override for CO-RE relocations */
688 struct btf *btf_vmlinux_override;
689 /* Lazily initialized kernel module BTFs */
690 struct module_btf *btf_modules;
691 bool btf_modules_loaded;
692 size_t btf_module_cnt;
693 size_t btf_module_cap;
694
695 /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
696 char *log_buf;
697 size_t log_size;
698 __u32 log_level;
699
700 int *fd_array;
701 size_t fd_array_cap;
702 size_t fd_array_cnt;
703
704 struct usdt_manager *usdt_man;
705
706 char path[];
707 };
708
709 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
710 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
711 #ifdef HAVE_LIBELF
712 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
713 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
714 #endif
715 #if defined HAVE_LIBELF
716 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
717 #elif defined HAVE_ELFIO
718 static Elf64_Shdr *elf_sec_hdr_by_idx(const struct bpf_object *obj, size_t idx, Elf64_Shdr *sheader);
719 #endif
720 #if defined HAVE_LIBELF
721 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
722 #elif defined HAVE_ELFIO
723 static const char *elf_sec_name_by_idx(const struct bpf_object *obj, size_t idx);
724 #endif
725 #if defined HAVE_LIBELF
726 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
727 #elif defined HAVE_ELFIO
728 static Elf_Data *elf_sec_data_by_name(const struct bpf_object *obj, const char *name, Elf_Data *data);
729 static Elf_Data *elf_sec_data_by_idx(const struct bpf_object *obj, size_t idx, Elf_Data *data);
730 #endif
731 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
732 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
733
bpf_program__unload(struct bpf_program * prog)734 void bpf_program__unload(struct bpf_program *prog)
735 {
736 if (!prog)
737 return;
738
739 zclose(prog->fd);
740
741 zfree(&prog->func_info);
742 zfree(&prog->line_info);
743 }
744
bpf_program__exit(struct bpf_program * prog)745 static void bpf_program__exit(struct bpf_program *prog)
746 {
747 if (!prog)
748 return;
749
750 bpf_program__unload(prog);
751 zfree(&prog->name);
752 zfree(&prog->sec_name);
753 zfree(&prog->insns);
754 zfree(&prog->reloc_desc);
755
756 prog->nr_reloc = 0;
757 prog->insns_cnt = 0;
758 prog->sec_idx = -1;
759 }
760
insn_is_subprog_call(const struct bpf_insn * insn)761 static bool insn_is_subprog_call(const struct bpf_insn *insn)
762 {
763 return BPF_CLASS(insn->code) == BPF_JMP &&
764 BPF_OP(insn->code) == BPF_CALL &&
765 BPF_SRC(insn->code) == BPF_K &&
766 insn->src_reg == BPF_PSEUDO_CALL &&
767 insn->dst_reg == 0 &&
768 insn->off == 0;
769 }
770
is_call_insn(const struct bpf_insn * insn)771 static bool is_call_insn(const struct bpf_insn *insn)
772 {
773 return insn->code == (BPF_JMP | BPF_CALL);
774 }
775
insn_is_pseudo_func(struct bpf_insn * insn)776 static bool insn_is_pseudo_func(struct bpf_insn *insn)
777 {
778 return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
779 }
780
781 static int
bpf_object__init_prog(struct bpf_object * obj,struct bpf_program * prog,const char * name,size_t sec_idx,const char * sec_name,size_t sec_off,void * insn_data,size_t insn_data_sz)782 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
783 const char *name, size_t sec_idx, const char *sec_name,
784 size_t sec_off, void *insn_data, size_t insn_data_sz)
785 {
786 if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
787 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
788 sec_name, name, sec_off, insn_data_sz);
789 return -EINVAL;
790 }
791
792 memset(prog, 0, sizeof(*prog));
793 prog->obj = obj;
794
795 prog->sec_idx = sec_idx;
796 prog->sec_insn_off = sec_off / BPF_INSN_SZ;
797 prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
798 /* insns_cnt can later be increased by appending used subprograms */
799 prog->insns_cnt = prog->sec_insn_cnt;
800
801 prog->type = BPF_PROG_TYPE_UNSPEC;
802 prog->fd = -1;
803
804 /* libbpf's convention for SEC("?abc...") is that it's just like
805 * SEC("abc...") but the corresponding bpf_program starts out with
806 * autoload set to false.
807 */
808 if (sec_name[0] == '?') {
809 prog->autoload = false;
810 /* from now on forget there was ? in section name */
811 sec_name++;
812 } else {
813 prog->autoload = true;
814 }
815
816 prog->autoattach = true;
817
818 /* inherit object's log_level */
819 prog->log_level = obj->log_level;
820
821 prog->sec_name = strdup(sec_name);
822 if (!prog->sec_name)
823 goto errout;
824
825 prog->name = strdup(name);
826 if (!prog->name)
827 goto errout;
828
829 prog->insns = malloc(insn_data_sz);
830 if (!prog->insns)
831 goto errout;
832 memcpy(prog->insns, insn_data, insn_data_sz);
833
834 return 0;
835 errout:
836 pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
837 bpf_program__exit(prog);
838 return -ENOMEM;
839 }
840
841 static int
bpf_object__add_programs(struct bpf_object * obj,Elf_Data * sec_data,const char * sec_name,int sec_idx)842 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
843 const char *sec_name, int sec_idx)
844 {
845 Elf_Data *symbols = obj->efile.symbols;
846 struct bpf_program *prog, *progs;
847 void *data = sec_data->d_buf;
848 size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
849 int nr_progs, err, i;
850 const char *name;
851 Elf64_Sym *sym;
852
853 progs = obj->programs;
854 nr_progs = obj->nr_programs;
855 nr_syms = symbols->d_size / sizeof(Elf64_Sym);
856 sec_off = 0;
857
858 for (i = 0; i < nr_syms; i++) {
859 sym = elf_sym_by_idx(obj, i);
860
861 if (sym->st_shndx != sec_idx)
862 continue;
863 if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
864 continue;
865
866 prog_sz = sym->st_size;
867 sec_off = sym->st_value;
868
869 name = elf_sym_str(obj, sym->st_name);
870 if (!name) {
871 pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
872 sec_name, sec_off);
873 return -LIBBPF_ERRNO__FORMAT;
874 }
875
876 if (sec_off + prog_sz > sec_sz) {
877 pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
878 sec_name, sec_off);
879 return -LIBBPF_ERRNO__FORMAT;
880 }
881
882 if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
883 pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
884 return -ENOTSUP;
885 }
886
887 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
888 sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
889
890 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
891 if (!progs) {
892 /*
893 * In this case the original obj->programs
894 * is still valid, so don't need special treat for
895 * bpf_close_object().
896 */
897 pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
898 sec_name, name);
899 return -ENOMEM;
900 }
901 obj->programs = progs;
902
903 prog = &progs[nr_progs];
904
905 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
906 sec_off, data + sec_off, prog_sz);
907 if (err)
908 return err;
909
910 /* if function is a global/weak symbol, but has restricted
911 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
912 * as static to enable more permissive BPF verification mode
913 * with more outside context available to BPF verifier
914 */
915 if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL
916 && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
917 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
918 prog->mark_btf_static = true;
919
920 nr_progs++;
921 obj->nr_programs = nr_progs;
922 }
923
924 return 0;
925 }
926
get_kernel_version(void)927 __u32 get_kernel_version(void)
928 {
929 /* On Ubuntu LINUX_VERSION_CODE doesn't correspond to info.release,
930 * but Ubuntu provides /proc/version_signature file, as described at
931 * https://ubuntu.com/kernel, with an example contents below, which we
932 * can use to get a proper LINUX_VERSION_CODE.
933 *
934 * Ubuntu 5.4.0-12.15-generic 5.4.8
935 *
936 * In the above, 5.4.8 is what kernel is actually expecting, while
937 * uname() call will return 5.4.0 in info.release.
938 */
939 const char *ubuntu_kver_file = "/proc/version_signature";
940 __u32 major, minor, patch;
941 struct utsname info;
942
943 if (faccessat(AT_FDCWD, ubuntu_kver_file, R_OK, AT_EACCESS) == 0) {
944 FILE *f;
945
946 f = fopen(ubuntu_kver_file, "r");
947 if (f) {
948 if (fscanf(f, "%*s %*s %d.%d.%d\n", &major, &minor, &patch) == 3) {
949 fclose(f);
950 return KERNEL_VERSION(major, minor, patch);
951 }
952 fclose(f);
953 }
954 /* something went wrong, fall back to uname() approach */
955 }
956
957 uname(&info);
958 if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
959 return 0;
960 return KERNEL_VERSION(major, minor, patch);
961 }
962
963 static const struct btf_member *
find_member_by_offset(const struct btf_type * t,__u32 bit_offset)964 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
965 {
966 struct btf_member *m;
967 int i;
968
969 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
970 if (btf_member_bit_offset(t, i) == bit_offset)
971 return m;
972 }
973
974 return NULL;
975 }
976
977 static const struct btf_member *
find_member_by_name(const struct btf * btf,const struct btf_type * t,const char * name)978 find_member_by_name(const struct btf *btf, const struct btf_type *t,
979 const char *name)
980 {
981 struct btf_member *m;
982 int i;
983
984 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
985 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
986 return m;
987 }
988
989 return NULL;
990 }
991
992 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
993 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
994 const char *name, __u32 kind);
995
996 static int
find_struct_ops_kern_types(const struct btf * btf,const char * tname,const struct btf_type ** type,__u32 * type_id,const struct btf_type ** vtype,__u32 * vtype_id,const struct btf_member ** data_member)997 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
998 const struct btf_type **type, __u32 *type_id,
999 const struct btf_type **vtype, __u32 *vtype_id,
1000 const struct btf_member **data_member)
1001 {
1002 const struct btf_type *kern_type, *kern_vtype;
1003 const struct btf_member *kern_data_member;
1004 __s32 kern_vtype_id, kern_type_id;
1005 __u32 i;
1006
1007 kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
1008 if (kern_type_id < 0) {
1009 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
1010 tname);
1011 return kern_type_id;
1012 }
1013 kern_type = btf__type_by_id(btf, kern_type_id);
1014
1015 /* Find the corresponding "map_value" type that will be used
1016 * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example,
1017 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
1018 * btf_vmlinux.
1019 */
1020 kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
1021 tname, BTF_KIND_STRUCT);
1022 if (kern_vtype_id < 0) {
1023 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
1024 STRUCT_OPS_VALUE_PREFIX, tname);
1025 return kern_vtype_id;
1026 }
1027 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
1028
1029 /* Find "struct tcp_congestion_ops" from
1030 * struct bpf_struct_ops_tcp_congestion_ops {
1031 * [ ... ]
1032 * struct tcp_congestion_ops data;
1033 * }
1034 */
1035 kern_data_member = btf_members(kern_vtype);
1036 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
1037 if (kern_data_member->type == kern_type_id)
1038 break;
1039 }
1040 if (i == btf_vlen(kern_vtype)) {
1041 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
1042 tname, STRUCT_OPS_VALUE_PREFIX, tname);
1043 return -EINVAL;
1044 }
1045
1046 *type = kern_type;
1047 *type_id = kern_type_id;
1048 *vtype = kern_vtype;
1049 *vtype_id = kern_vtype_id;
1050 *data_member = kern_data_member;
1051
1052 return 0;
1053 }
1054
bpf_map__is_struct_ops(const struct bpf_map * map)1055 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
1056 {
1057 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
1058 }
1059
1060 /* Init the map's fields that depend on kern_btf */
bpf_map__init_kern_struct_ops(struct bpf_map * map,const struct btf * btf,const struct btf * kern_btf)1061 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
1062 const struct btf *btf,
1063 const struct btf *kern_btf)
1064 {
1065 const struct btf_member *member, *kern_member, *kern_data_member;
1066 const struct btf_type *type, *kern_type, *kern_vtype;
1067 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
1068 struct bpf_struct_ops *st_ops;
1069 void *data, *kern_data;
1070 const char *tname;
1071 int err;
1072
1073 st_ops = map->st_ops;
1074 type = st_ops->type;
1075 tname = st_ops->tname;
1076 err = find_struct_ops_kern_types(kern_btf, tname,
1077 &kern_type, &kern_type_id,
1078 &kern_vtype, &kern_vtype_id,
1079 &kern_data_member);
1080 if (err)
1081 return err;
1082
1083 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
1084 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
1085
1086 map->def.value_size = kern_vtype->size;
1087 map->btf_vmlinux_value_type_id = kern_vtype_id;
1088
1089 st_ops->kern_vdata = calloc(1, kern_vtype->size);
1090 if (!st_ops->kern_vdata)
1091 return -ENOMEM;
1092
1093 data = st_ops->data;
1094 kern_data_off = kern_data_member->offset / 8;
1095 kern_data = st_ops->kern_vdata + kern_data_off;
1096
1097 member = btf_members(type);
1098 for (i = 0; i < btf_vlen(type); i++, member++) {
1099 const struct btf_type *mtype, *kern_mtype;
1100 __u32 mtype_id, kern_mtype_id;
1101 void *mdata, *kern_mdata;
1102 __s64 msize, kern_msize;
1103 __u32 moff, kern_moff;
1104 __u32 kern_member_idx;
1105 const char *mname;
1106
1107 mname = btf__name_by_offset(btf, member->name_off);
1108 kern_member = find_member_by_name(kern_btf, kern_type, mname);
1109 if (!kern_member) {
1110 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
1111 map->name, mname);
1112 return -ENOTSUP;
1113 }
1114
1115 kern_member_idx = kern_member - btf_members(kern_type);
1116 if (btf_member_bitfield_size(type, i) ||
1117 btf_member_bitfield_size(kern_type, kern_member_idx)) {
1118 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
1119 map->name, mname);
1120 return -ENOTSUP;
1121 }
1122
1123 moff = member->offset / 8;
1124 kern_moff = kern_member->offset / 8;
1125
1126 mdata = data + moff;
1127 kern_mdata = kern_data + kern_moff;
1128
1129 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
1130 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
1131 &kern_mtype_id);
1132 if (BTF_INFO_KIND(mtype->info) !=
1133 BTF_INFO_KIND(kern_mtype->info)) {
1134 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
1135 map->name, mname, BTF_INFO_KIND(mtype->info),
1136 BTF_INFO_KIND(kern_mtype->info));
1137 return -ENOTSUP;
1138 }
1139
1140 if (btf_is_ptr(mtype)) {
1141 struct bpf_program *prog;
1142
1143 prog = st_ops->progs[i];
1144 if (!prog)
1145 continue;
1146
1147 kern_mtype = skip_mods_and_typedefs(kern_btf,
1148 kern_mtype->type,
1149 &kern_mtype_id);
1150
1151 /* mtype->type must be a func_proto which was
1152 * guaranteed in bpf_object__collect_st_ops_relos(),
1153 * so only check kern_mtype for func_proto here.
1154 */
1155 if (!btf_is_func_proto(kern_mtype)) {
1156 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1157 map->name, mname);
1158 return -ENOTSUP;
1159 }
1160
1161 prog->attach_btf_id = kern_type_id;
1162 prog->expected_attach_type = kern_member_idx;
1163
1164 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1165
1166 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1167 map->name, mname, prog->name, moff,
1168 kern_moff);
1169
1170 continue;
1171 }
1172
1173 msize = btf__resolve_size(btf, mtype_id);
1174 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1175 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
1176 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1177 map->name, mname, (ssize_t)msize,
1178 (ssize_t)kern_msize);
1179 return -ENOTSUP;
1180 }
1181
1182 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1183 map->name, mname, (unsigned int)msize,
1184 moff, kern_moff);
1185 memcpy(kern_mdata, mdata, msize);
1186 }
1187
1188 return 0;
1189 }
1190
bpf_object__init_kern_struct_ops_maps(struct bpf_object * obj)1191 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1192 {
1193 struct bpf_map *map;
1194 size_t i;
1195 int err;
1196
1197 for (i = 0; i < obj->nr_maps; i++) {
1198 map = &obj->maps[i];
1199
1200 if (!bpf_map__is_struct_ops(map))
1201 continue;
1202
1203 err = bpf_map__init_kern_struct_ops(map, obj->btf,
1204 obj->btf_vmlinux);
1205 if (err)
1206 return err;
1207 }
1208
1209 return 0;
1210 }
1211
bpf_object__init_struct_ops_maps(struct bpf_object * obj)1212 static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
1213 {
1214 const struct btf_type *type, *datasec;
1215 const struct btf_var_secinfo *vsi;
1216 struct bpf_struct_ops *st_ops;
1217 const char *tname, *var_name;
1218 __s32 type_id, datasec_id;
1219 const struct btf *btf;
1220 struct bpf_map *map;
1221 __u32 i;
1222
1223 if (obj->efile.st_ops_shndx == -1)
1224 return 0;
1225
1226 btf = obj->btf;
1227 datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
1228 BTF_KIND_DATASEC);
1229 if (datasec_id < 0) {
1230 pr_warn("struct_ops init: DATASEC %s not found\n",
1231 STRUCT_OPS_SEC);
1232 return -EINVAL;
1233 }
1234
1235 datasec = btf__type_by_id(btf, datasec_id);
1236 vsi = btf_var_secinfos(datasec);
1237 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1238 type = btf__type_by_id(obj->btf, vsi->type);
1239 var_name = btf__name_by_offset(obj->btf, type->name_off);
1240
1241 type_id = btf__resolve_type(obj->btf, vsi->type);
1242 if (type_id < 0) {
1243 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1244 vsi->type, STRUCT_OPS_SEC);
1245 return -EINVAL;
1246 }
1247
1248 type = btf__type_by_id(obj->btf, type_id);
1249 tname = btf__name_by_offset(obj->btf, type->name_off);
1250 if (!tname[0]) {
1251 pr_warn("struct_ops init: anonymous type is not supported\n");
1252 return -ENOTSUP;
1253 }
1254 if (!btf_is_struct(type)) {
1255 pr_warn("struct_ops init: %s is not a struct\n", tname);
1256 return -EINVAL;
1257 }
1258
1259 map = bpf_object__add_map(obj);
1260 if (IS_ERR(map))
1261 return PTR_ERR(map);
1262
1263 map->sec_idx = obj->efile.st_ops_shndx;
1264 map->sec_offset = vsi->offset;
1265 map->name = strdup(var_name);
1266 if (!map->name)
1267 return -ENOMEM;
1268
1269 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1270 map->def.key_size = sizeof(int);
1271 map->def.value_size = type->size;
1272 map->def.max_entries = 1;
1273
1274 map->st_ops = calloc(1, sizeof(*map->st_ops));
1275 if (!map->st_ops)
1276 return -ENOMEM;
1277 st_ops = map->st_ops;
1278 st_ops->data = malloc(type->size);
1279 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1280 st_ops->kern_func_off = malloc(btf_vlen(type) *
1281 sizeof(*st_ops->kern_func_off));
1282 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1283 return -ENOMEM;
1284
1285 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
1286 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1287 var_name, STRUCT_OPS_SEC);
1288 return -EINVAL;
1289 }
1290
1291 memcpy(st_ops->data,
1292 obj->efile.st_ops_data->d_buf + vsi->offset,
1293 type->size);
1294 st_ops->tname = tname;
1295 st_ops->type = type;
1296 st_ops->type_id = type_id;
1297
1298 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1299 tname, type_id, var_name, vsi->offset);
1300 }
1301
1302 return 0;
1303 }
1304
bpf_object__new(const char * path,const void * obj_buf,size_t obj_buf_sz,const char * obj_name)1305 static struct bpf_object *bpf_object__new(const char *path,
1306 const void *obj_buf,
1307 size_t obj_buf_sz,
1308 const char *obj_name)
1309 {
1310 struct bpf_object *obj;
1311 char *end;
1312
1313 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1314 if (!obj) {
1315 pr_warn("alloc memory failed for %s\n", path);
1316 return ERR_PTR(-ENOMEM);
1317 }
1318
1319 strcpy(obj->path, path);
1320 if (obj_name) {
1321 libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1322 } else {
1323 /* Using basename() GNU version which doesn't modify arg. */
1324 libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1325 end = strchr(obj->name, '.');
1326 if (end)
1327 *end = 0;
1328 }
1329
1330 obj->efile.fd = -1;
1331 /*
1332 * Caller of this function should also call
1333 * bpf_object__elf_finish() after data collection to return
1334 * obj_buf to user. If not, we should duplicate the buffer to
1335 * avoid user freeing them before elf finish.
1336 */
1337 obj->efile.obj_buf = obj_buf;
1338 obj->efile.obj_buf_sz = obj_buf_sz;
1339 obj->efile.btf_maps_shndx = -1;
1340 obj->efile.st_ops_shndx = -1;
1341 obj->kconfig_map_idx = -1;
1342
1343 obj->kern_version = get_kernel_version();
1344 obj->loaded = false;
1345
1346 return obj;
1347 }
1348
bpf_object__elf_finish(struct bpf_object * obj)1349 static void bpf_object__elf_finish(struct bpf_object *obj)
1350 {
1351 if (!obj->efile.elf)
1352 return;
1353 #if defined HAVE_LIBELF
1354 elf_end(obj->efile.elf);
1355 #elif defined HAVE_ELFIO
1356 if (obj->efile.shstring) {
1357 elfio_string_section_accessor_delete(obj->efile.shstring);
1358 }
1359 if (obj->efile.strstring) {
1360 elfio_string_section_accessor_delete(obj->efile.strstring);
1361 }
1362 elfio_delete(obj->efile.elf);
1363 #endif
1364 obj->efile.elf = NULL;
1365 obj->efile.symbols = NULL;
1366 obj->efile.st_ops_data = NULL;
1367
1368 zfree(&obj->efile.secs);
1369 obj->efile.sec_cnt = 0;
1370 zclose(obj->efile.fd);
1371 obj->efile.obj_buf = NULL;
1372 obj->efile.obj_buf_sz = 0;
1373 }
1374
bpf_object__elf_init(struct bpf_object * obj)1375 static int bpf_object__elf_init(struct bpf_object *obj)
1376 {
1377 Elf64_Ehdr *ehdr;
1378 int err = 0;
1379 #ifdef HAVE_LIBELF
1380 Elf *elf;
1381 #elif defined HAVE_ELFIO
1382 pelfio_t elf;
1383 #endif
1384
1385 if (obj->efile.elf) {
1386 pr_warn("elf: init internal error\n");
1387 return -LIBBPF_ERRNO__LIBELF;
1388 }
1389
1390 if (obj->efile.obj_buf_sz > 0) {
1391 /* obj_buf should have been validated by bpf_object__open_mem(). */
1392 #ifdef HAVE_LIBELF
1393 elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1394 #elif defined HAVE_ELFIO
1395 char memfd_path[PATH_MAX] = {0};
1396 elf = elfio_new();
1397 int fdm = syscall(__NR_memfd_create, "bpfelf", MFD_CLOEXEC);
1398 ftruncate(fdm, obj->efile.obj_buf_sz);
1399 write(fdm, (char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1400 snprintf(memfd_path, PATH_MAX, "/proc/self/fd/%d", fdm);
1401 elfio_load(elf, memfd_path);
1402 #endif
1403 } else {
1404 obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1405 if (obj->efile.fd < 0) {
1406 char errmsg[STRERR_BUFSIZE], *cp;
1407
1408 err = -errno;
1409 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1410 pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1411 return err;
1412 }
1413 #ifdef HAVE_LIBELF
1414 elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1415 #endif
1416 }
1417
1418 if (!elf) {
1419 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1420 err = -LIBBPF_ERRNO__LIBELF;
1421 goto errout;
1422 }
1423
1424 obj->efile.elf = elf;
1425 #ifdef HAVE_LIBELF
1426 if (elf_kind(elf) != ELF_K_ELF) {
1427 err = -LIBBPF_ERRNO__FORMAT;
1428 pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1429 goto errout;
1430 }
1431
1432 if (gelf_getclass(elf) != ELFCLASS64) {
1433 #elif defined HAVE_ELFIO
1434 if (elfio_get_class(elf) != ELFCLASS64 ) {
1435 #endif
1436 err = -LIBBPF_ERRNO__FORMAT;
1437 pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1438 goto errout;
1439 }
1440 #ifdef HAVE_LIBELF
1441 obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1442 #elif defined HAVE_ELFIO
1443 obj->efile.ehdr = ehdr = (Elf64_Ehdr*)obj->efile.obj_buf;
1444 #endif
1445 if (!obj->efile.ehdr) {
1446 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1447 err = -LIBBPF_ERRNO__FORMAT;
1448 goto errout;
1449 }
1450
1451 #ifdef HAVE_LIBELF
1452 if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1453 pr_warn("elf: failed to get section names section index for %s: %s\n",
1454 obj->path, elf_errmsg(-1));
1455 err = -LIBBPF_ERRNO__FORMAT;
1456 goto errout;
1457 }
1458
1459 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1460 if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1461 pr_warn("elf: failed to get section names strings from %s: %s\n",
1462 obj->path, elf_errmsg(-1));
1463 err = -LIBBPF_ERRNO__FORMAT;
1464 goto errout;
1465 }
1466 #elif defined HAVE_ELFIO
1467 obj->efile.shstrndx = elfio_get_section_name_str_index(elf);
1468 #endif
1469 /* Old LLVM set e_machine to EM_NONE */
1470 if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1471 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1472 err = -LIBBPF_ERRNO__FORMAT;
1473 goto errout;
1474 }
1475
1476 return 0;
1477 errout:
1478 bpf_object__elf_finish(obj);
1479 return err;
1480 }
1481
1482 static int bpf_object__check_endianness(struct bpf_object *obj)
1483 {
1484 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1485 if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB)
1486 return 0;
1487 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1488 if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB)
1489 return 0;
1490 #else
1491 # error "Unrecognized __BYTE_ORDER__"
1492 #endif
1493 pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1494 return -LIBBPF_ERRNO__ENDIAN;
1495 }
1496
1497 static int
1498 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1499 {
1500 if (!data) {
1501 pr_warn("invalid license section in %s\n", obj->path);
1502 return -LIBBPF_ERRNO__FORMAT;
1503 }
1504 /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1505 * go over allowed ELF data section buffer
1506 */
1507 libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1508 pr_debug("license of %s is %s\n", obj->path, obj->license);
1509 return 0;
1510 }
1511
1512 static int
1513 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1514 {
1515 __u32 kver;
1516
1517 if (!data || size != sizeof(kver)) {
1518 pr_warn("invalid kver section in %s\n", obj->path);
1519 return -LIBBPF_ERRNO__FORMAT;
1520 }
1521 memcpy(&kver, data, sizeof(kver));
1522 obj->kern_version = kver;
1523 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1524 return 0;
1525 }
1526
1527 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1528 {
1529 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1530 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1531 return true;
1532 return false;
1533 }
1534
1535 static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1536 {
1537 Elf_Data *data;
1538 #ifdef HAVE_LIBELF
1539 Elf_Scn *scn;
1540 #endif
1541
1542 if (!name)
1543 return -EINVAL;
1544 #if defined HAVE_LIBELF
1545 scn = elf_sec_by_name(obj, name);
1546 data = elf_sec_data(obj, scn);
1547 #elif defined HAVE_ELFIO
1548 Elf_Data realdata;
1549 data = &realdata;
1550 data = elf_sec_data_by_name(obj, name, data);
1551 #endif
1552 if (data) {
1553 *size = data->d_size;
1554 return 0; /* found it */
1555 }
1556
1557 return -ENOENT;
1558 }
1559
1560 static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name)
1561 {
1562 Elf_Data *symbols = obj->efile.symbols;
1563 const char *sname;
1564 size_t si;
1565
1566 for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1567 Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1568
1569 if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1570 continue;
1571
1572 if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL &&
1573 ELF64_ST_BIND(sym->st_info) != STB_WEAK)
1574 continue;
1575
1576 sname = elf_sym_str(obj, sym->st_name);
1577 if (!sname) {
1578 pr_warn("failed to get sym name string for var %s\n", name);
1579 return ERR_PTR(-EIO);
1580 }
1581 if (strcmp(name, sname) == 0)
1582 return sym;
1583 }
1584
1585 return ERR_PTR(-ENOENT);
1586 }
1587
1588 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1589 {
1590 struct bpf_map *map;
1591 int err;
1592
1593 err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap,
1594 sizeof(*obj->maps), obj->nr_maps + 1);
1595 if (err)
1596 return ERR_PTR(err);
1597
1598 map = &obj->maps[obj->nr_maps++];
1599 map->obj = obj;
1600 map->fd = -1;
1601 map->inner_map_fd = -1;
1602 map->autocreate = true;
1603
1604 return map;
1605 }
1606
1607 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1608 {
1609 long page_sz = sysconf(_SC_PAGE_SIZE);
1610 size_t map_sz;
1611
1612 map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1613 map_sz = roundup(map_sz, page_sz);
1614 return map_sz;
1615 }
1616
1617 static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1618 {
1619 char map_name[BPF_OBJ_NAME_LEN], *p;
1620 int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1621
1622 /* This is one of the more confusing parts of libbpf for various
1623 * reasons, some of which are historical. The original idea for naming
1624 * internal names was to include as much of BPF object name prefix as
1625 * possible, so that it can be distinguished from similar internal
1626 * maps of a different BPF object.
1627 * As an example, let's say we have bpf_object named 'my_object_name'
1628 * and internal map corresponding to '.rodata' ELF section. The final
1629 * map name advertised to user and to the kernel will be
1630 * 'my_objec.rodata', taking first 8 characters of object name and
1631 * entire 7 characters of '.rodata'.
1632 * Somewhat confusingly, if internal map ELF section name is shorter
1633 * than 7 characters, e.g., '.bss', we still reserve 7 characters
1634 * for the suffix, even though we only have 4 actual characters, and
1635 * resulting map will be called 'my_objec.bss', not even using all 15
1636 * characters allowed by the kernel. Oh well, at least the truncated
1637 * object name is somewhat consistent in this case. But if the map
1638 * name is '.kconfig', we'll still have entirety of '.kconfig' added
1639 * (8 chars) and thus will be left with only first 7 characters of the
1640 * object name ('my_obje'). Happy guessing, user, that the final map
1641 * name will be "my_obje.kconfig".
1642 * Now, with libbpf starting to support arbitrarily named .rodata.*
1643 * and .data.* data sections, it's possible that ELF section name is
1644 * longer than allowed 15 chars, so we now need to be careful to take
1645 * only up to 15 first characters of ELF name, taking no BPF object
1646 * name characters at all. So '.rodata.abracadabra' will result in
1647 * '.rodata.abracad' kernel and user-visible name.
1648 * We need to keep this convoluted logic intact for .data, .bss and
1649 * .rodata maps, but for new custom .data.custom and .rodata.custom
1650 * maps we use their ELF names as is, not prepending bpf_object name
1651 * in front. We still need to truncate them to 15 characters for the
1652 * kernel. Full name can be recovered for such maps by using DATASEC
1653 * BTF type associated with such map's value type, though.
1654 */
1655 if (sfx_len >= BPF_OBJ_NAME_LEN)
1656 sfx_len = BPF_OBJ_NAME_LEN - 1;
1657
1658 /* if there are two or more dots in map name, it's a custom dot map */
1659 if (strchr(real_name + 1, '.') != NULL)
1660 pfx_len = 0;
1661 else
1662 pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1663
1664 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1665 sfx_len, real_name);
1666
1667 /* sanitise map name to characters allowed by kernel */
1668 for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1669 if (!isalnum(*p) && *p != '_' && *p != '.')
1670 *p = '_';
1671
1672 return strdup(map_name);
1673 }
1674
1675 static int
1676 map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map);
1677
1678 /* Internal BPF map is mmap()'able only if at least one of corresponding
1679 * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL
1680 * variable and it's not marked as __hidden (which turns it into, effectively,
1681 * a STATIC variable).
1682 */
1683 static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map)
1684 {
1685 const struct btf_type *t, *vt;
1686 struct btf_var_secinfo *vsi;
1687 int i, n;
1688
1689 if (!map->btf_value_type_id)
1690 return false;
1691
1692 t = btf__type_by_id(obj->btf, map->btf_value_type_id);
1693 if (!btf_is_datasec(t))
1694 return false;
1695
1696 vsi = btf_var_secinfos(t);
1697 for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) {
1698 vt = btf__type_by_id(obj->btf, vsi->type);
1699 if (!btf_is_var(vt))
1700 continue;
1701
1702 if (btf_var(vt)->linkage != BTF_VAR_STATIC)
1703 return true;
1704 }
1705
1706 return false;
1707 }
1708
1709 static int
1710 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1711 const char *real_name, int sec_idx, void *data, size_t data_sz)
1712 {
1713 struct bpf_map_def *def;
1714 struct bpf_map *map;
1715 int err;
1716
1717 map = bpf_object__add_map(obj);
1718 if (IS_ERR(map))
1719 return PTR_ERR(map);
1720
1721 map->libbpf_type = type;
1722 map->sec_idx = sec_idx;
1723 map->sec_offset = 0;
1724 map->real_name = strdup(real_name);
1725 map->name = internal_map_name(obj, real_name);
1726 if (!map->real_name || !map->name) {
1727 zfree(&map->real_name);
1728 zfree(&map->name);
1729 return -ENOMEM;
1730 }
1731
1732 def = &map->def;
1733 def->type = BPF_MAP_TYPE_ARRAY;
1734 def->key_size = sizeof(int);
1735 def->value_size = data_sz;
1736 def->max_entries = 1;
1737 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1738 ? BPF_F_RDONLY_PROG : 0;
1739
1740 /* failures are fine because of maps like .rodata.str1.1 */
1741 (void) map_fill_btf_type_info(obj, map);
1742
1743 if (map_is_mmapable(obj, map))
1744 def->map_flags |= BPF_F_MMAPABLE;
1745
1746 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1747 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1748
1749 map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1750 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1751 if (map->mmaped == MAP_FAILED) {
1752 err = -errno;
1753 map->mmaped = NULL;
1754 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1755 map->name, err);
1756 zfree(&map->real_name);
1757 zfree(&map->name);
1758 return err;
1759 }
1760
1761 if (data)
1762 memcpy(map->mmaped, data, data_sz);
1763
1764 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1765 return 0;
1766 }
1767
1768 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1769 {
1770 struct elf_sec_desc *sec_desc;
1771 const char *sec_name;
1772 int err = 0, sec_idx;
1773
1774 /*
1775 * Populate obj->maps with libbpf internal maps.
1776 */
1777 for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
1778 sec_desc = &obj->efile.secs[sec_idx];
1779
1780 /* Skip recognized sections with size 0. */
1781 if (!sec_desc->data || sec_desc->data->d_size == 0)
1782 continue;
1783
1784 switch (sec_desc->sec_type) {
1785 case SEC_DATA:
1786 #if defined HAVE_LIBELF
1787 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1788 #elif defined HAVE_ELFIO
1789 sec_name = elf_sec_name_by_idx(obj, sec_idx);
1790 #endif
1791 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1792 sec_name, sec_idx,
1793 sec_desc->data->d_buf,
1794 sec_desc->data->d_size);
1795 break;
1796 case SEC_RODATA:
1797 obj->has_rodata = true;
1798 #if defined HAVE_LIBELF
1799 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1800 #elif defined HAVE_ELFIO
1801 sec_name = elf_sec_name_by_idx(obj, sec_idx);
1802 #endif
1803 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1804 sec_name, sec_idx,
1805 sec_desc->data->d_buf,
1806 sec_desc->data->d_size);
1807 break;
1808 case SEC_BSS:
1809 #if defined HAVE_LIBELF
1810 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1811 #elif defined HAVE_ELFIO
1812 sec_name = elf_sec_name_by_idx(obj, sec_idx);
1813 #endif
1814 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1815 sec_name, sec_idx,
1816 NULL,
1817 sec_desc->data->d_size);
1818 break;
1819 default:
1820 /* skip */
1821 break;
1822 }
1823 if (err)
1824 return err;
1825 }
1826 return 0;
1827 }
1828
1829
1830 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1831 const void *name)
1832 {
1833 int i;
1834
1835 for (i = 0; i < obj->nr_extern; i++) {
1836 if (strcmp(obj->externs[i].name, name) == 0)
1837 return &obj->externs[i];
1838 }
1839 return NULL;
1840 }
1841
1842 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1843 char value)
1844 {
1845 switch (ext->kcfg.type) {
1846 case KCFG_BOOL:
1847 if (value == 'm') {
1848 pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n",
1849 ext->name, value);
1850 return -EINVAL;
1851 }
1852 *(bool *)ext_val = value == 'y' ? true : false;
1853 break;
1854 case KCFG_TRISTATE:
1855 if (value == 'y')
1856 *(enum libbpf_tristate *)ext_val = TRI_YES;
1857 else if (value == 'm')
1858 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1859 else /* value == 'n' */
1860 *(enum libbpf_tristate *)ext_val = TRI_NO;
1861 break;
1862 case KCFG_CHAR:
1863 *(char *)ext_val = value;
1864 break;
1865 case KCFG_UNKNOWN:
1866 case KCFG_INT:
1867 case KCFG_CHAR_ARR:
1868 default:
1869 pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n",
1870 ext->name, value);
1871 return -EINVAL;
1872 }
1873 ext->is_set = true;
1874 return 0;
1875 }
1876
1877 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1878 const char *value)
1879 {
1880 size_t len;
1881
1882 if (ext->kcfg.type != KCFG_CHAR_ARR) {
1883 pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n",
1884 ext->name, value);
1885 return -EINVAL;
1886 }
1887
1888 len = strlen(value);
1889 if (value[len - 1] != '"') {
1890 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1891 ext->name, value);
1892 return -EINVAL;
1893 }
1894
1895 /* strip quotes */
1896 len -= 2;
1897 if (len >= ext->kcfg.sz) {
1898 pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n",
1899 ext->name, value, len, ext->kcfg.sz - 1);
1900 len = ext->kcfg.sz - 1;
1901 }
1902 memcpy(ext_val, value + 1, len);
1903 ext_val[len] = '\0';
1904 ext->is_set = true;
1905 return 0;
1906 }
1907
1908 static int parse_u64(const char *value, __u64 *res)
1909 {
1910 char *value_end;
1911 int err;
1912
1913 errno = 0;
1914 *res = strtoull(value, &value_end, 0);
1915 if (errno) {
1916 err = -errno;
1917 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1918 return err;
1919 }
1920 if (*value_end) {
1921 pr_warn("failed to parse '%s' as integer completely\n", value);
1922 return -EINVAL;
1923 }
1924 return 0;
1925 }
1926
1927 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1928 {
1929 int bit_sz = ext->kcfg.sz * 8;
1930
1931 if (ext->kcfg.sz == 8)
1932 return true;
1933
1934 /* Validate that value stored in u64 fits in integer of `ext->sz`
1935 * bytes size without any loss of information. If the target integer
1936 * is signed, we rely on the following limits of integer type of
1937 * Y bits and subsequent transformation:
1938 *
1939 * -2^(Y-1) <= X <= 2^(Y-1) - 1
1940 * 0 <= X + 2^(Y-1) <= 2^Y - 1
1941 * 0 <= X + 2^(Y-1) < 2^Y
1942 *
1943 * For unsigned target integer, check that all the (64 - Y) bits are
1944 * zero.
1945 */
1946 if (ext->kcfg.is_signed)
1947 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1948 else
1949 return (v >> bit_sz) == 0;
1950 }
1951
1952 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1953 __u64 value)
1954 {
1955 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR &&
1956 ext->kcfg.type != KCFG_BOOL) {
1957 pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n",
1958 ext->name, (unsigned long long)value);
1959 return -EINVAL;
1960 }
1961 if (ext->kcfg.type == KCFG_BOOL && value > 1) {
1962 pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n",
1963 ext->name, (unsigned long long)value);
1964 return -EINVAL;
1965
1966 }
1967 if (!is_kcfg_value_in_range(ext, value)) {
1968 pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n",
1969 ext->name, (unsigned long long)value, ext->kcfg.sz);
1970 return -ERANGE;
1971 }
1972 switch (ext->kcfg.sz) {
1973 case 1:
1974 *(__u8 *)ext_val = value;
1975 break;
1976 case 2:
1977 *(__u16 *)ext_val = value;
1978 break;
1979 case 4:
1980 *(__u32 *)ext_val = value;
1981 break;
1982 case 8:
1983 *(__u64 *)ext_val = value;
1984 break;
1985 default:
1986 return -EINVAL;
1987 }
1988 ext->is_set = true;
1989 return 0;
1990 }
1991
1992 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1993 char *buf, void *data)
1994 {
1995 struct extern_desc *ext;
1996 char *sep, *value;
1997 int len, err = 0;
1998 void *ext_val;
1999 __u64 num;
2000
2001 if (!str_has_pfx(buf, "CONFIG_"))
2002 return 0;
2003
2004 sep = strchr(buf, '=');
2005 if (!sep) {
2006 pr_warn("failed to parse '%s': no separator\n", buf);
2007 return -EINVAL;
2008 }
2009
2010 /* Trim ending '\n' */
2011 len = strlen(buf);
2012 if (buf[len - 1] == '\n')
2013 buf[len - 1] = '\0';
2014 /* Split on '=' and ensure that a value is present. */
2015 *sep = '\0';
2016 if (!sep[1]) {
2017 *sep = '=';
2018 pr_warn("failed to parse '%s': no value\n", buf);
2019 return -EINVAL;
2020 }
2021
2022 ext = find_extern_by_name(obj, buf);
2023 if (!ext || ext->is_set)
2024 return 0;
2025
2026 ext_val = data + ext->kcfg.data_off;
2027 value = sep + 1;
2028
2029 switch (*value) {
2030 case 'y': case 'n': case 'm':
2031 err = set_kcfg_value_tri(ext, ext_val, *value);
2032 break;
2033 case '"':
2034 err = set_kcfg_value_str(ext, ext_val, value);
2035 break;
2036 default:
2037 /* assume integer */
2038 err = parse_u64(value, &num);
2039 if (err) {
2040 pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value);
2041 return err;
2042 }
2043 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
2044 pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value);
2045 return -EINVAL;
2046 }
2047 err = set_kcfg_value_num(ext, ext_val, num);
2048 break;
2049 }
2050 if (err)
2051 return err;
2052 pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value);
2053 return 0;
2054 }
2055
2056 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
2057 {
2058 char buf[PATH_MAX];
2059 struct utsname uts;
2060 int len, err = 0;
2061 gzFile file;
2062
2063 uname(&uts);
2064 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
2065 if (len < 0)
2066 return -EINVAL;
2067 else if (len >= PATH_MAX)
2068 return -ENAMETOOLONG;
2069
2070 /* gzopen also accepts uncompressed files. */
2071 file = gzopen(buf, "r");
2072 if (!file)
2073 file = gzopen("/proc/config.gz", "r");
2074
2075 if (!file) {
2076 pr_warn("failed to open system Kconfig\n");
2077 return -ENOENT;
2078 }
2079
2080 while (gzgets(file, buf, sizeof(buf))) {
2081 err = bpf_object__process_kconfig_line(obj, buf, data);
2082 if (err) {
2083 pr_warn("error parsing system Kconfig line '%s': %d\n",
2084 buf, err);
2085 goto out;
2086 }
2087 }
2088
2089 out:
2090 gzclose(file);
2091 return err;
2092 }
2093
2094 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
2095 const char *config, void *data)
2096 {
2097 char buf[PATH_MAX];
2098 int err = 0;
2099 FILE *file;
2100
2101 file = fmemopen((void *)config, strlen(config), "r");
2102 if (!file) {
2103 err = -errno;
2104 pr_warn("failed to open in-memory Kconfig: %d\n", err);
2105 return err;
2106 }
2107
2108 while (fgets(buf, sizeof(buf), file)) {
2109 err = bpf_object__process_kconfig_line(obj, buf, data);
2110 if (err) {
2111 pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
2112 buf, err);
2113 break;
2114 }
2115 }
2116
2117 fclose(file);
2118 return err;
2119 }
2120
2121 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
2122 {
2123 struct extern_desc *last_ext = NULL, *ext;
2124 size_t map_sz;
2125 int i, err;
2126
2127 for (i = 0; i < obj->nr_extern; i++) {
2128 ext = &obj->externs[i];
2129 if (ext->type == EXT_KCFG)
2130 last_ext = ext;
2131 }
2132
2133 if (!last_ext)
2134 return 0;
2135
2136 map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
2137 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
2138 ".kconfig", obj->efile.symbols_shndx,
2139 NULL, map_sz);
2140 if (err)
2141 return err;
2142
2143 obj->kconfig_map_idx = obj->nr_maps - 1;
2144
2145 return 0;
2146 }
2147
2148 const struct btf_type *
2149 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2150 {
2151 const struct btf_type *t = btf__type_by_id(btf, id);
2152
2153 if (res_id)
2154 *res_id = id;
2155
2156 while (btf_is_mod(t) || btf_is_typedef(t)) {
2157 if (res_id)
2158 *res_id = t->type;
2159 t = btf__type_by_id(btf, t->type);
2160 }
2161
2162 return t;
2163 }
2164
2165 static const struct btf_type *
2166 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2167 {
2168 const struct btf_type *t;
2169
2170 t = skip_mods_and_typedefs(btf, id, NULL);
2171 if (!btf_is_ptr(t))
2172 return NULL;
2173
2174 t = skip_mods_and_typedefs(btf, t->type, res_id);
2175
2176 return btf_is_func_proto(t) ? t : NULL;
2177 }
2178
2179 static const char *__btf_kind_str(__u16 kind)
2180 {
2181 switch (kind) {
2182 case BTF_KIND_UNKN: return "void";
2183 case BTF_KIND_INT: return "int";
2184 case BTF_KIND_PTR: return "ptr";
2185 case BTF_KIND_ARRAY: return "array";
2186 case BTF_KIND_STRUCT: return "struct";
2187 case BTF_KIND_UNION: return "union";
2188 case BTF_KIND_ENUM: return "enum";
2189 case BTF_KIND_FWD: return "fwd";
2190 case BTF_KIND_TYPEDEF: return "typedef";
2191 case BTF_KIND_VOLATILE: return "volatile";
2192 case BTF_KIND_CONST: return "const";
2193 case BTF_KIND_RESTRICT: return "restrict";
2194 case BTF_KIND_FUNC: return "func";
2195 case BTF_KIND_FUNC_PROTO: return "func_proto";
2196 case BTF_KIND_VAR: return "var";
2197 case BTF_KIND_DATASEC: return "datasec";
2198 case BTF_KIND_FLOAT: return "float";
2199 case BTF_KIND_DECL_TAG: return "decl_tag";
2200 case BTF_KIND_TYPE_TAG: return "type_tag";
2201 case BTF_KIND_ENUM64: return "enum64";
2202 default: return "unknown";
2203 }
2204 }
2205
2206 const char *btf_kind_str(const struct btf_type *t)
2207 {
2208 return __btf_kind_str(btf_kind(t));
2209 }
2210
2211 /*
2212 * Fetch integer attribute of BTF map definition. Such attributes are
2213 * represented using a pointer to an array, in which dimensionality of array
2214 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2215 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2216 * type definition, while using only sizeof(void *) space in ELF data section.
2217 */
2218 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2219 const struct btf_member *m, __u32 *res)
2220 {
2221 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2222 const char *name = btf__name_by_offset(btf, m->name_off);
2223 const struct btf_array *arr_info;
2224 const struct btf_type *arr_t;
2225
2226 if (!btf_is_ptr(t)) {
2227 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2228 map_name, name, btf_kind_str(t));
2229 return false;
2230 }
2231
2232 arr_t = btf__type_by_id(btf, t->type);
2233 if (!arr_t) {
2234 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2235 map_name, name, t->type);
2236 return false;
2237 }
2238 if (!btf_is_array(arr_t)) {
2239 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2240 map_name, name, btf_kind_str(arr_t));
2241 return false;
2242 }
2243 arr_info = btf_array(arr_t);
2244 *res = arr_info->nelems;
2245 return true;
2246 }
2247
2248 static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name)
2249 {
2250 int len;
2251
2252 len = snprintf(buf, buf_sz, "%s/%s", path, name);
2253 if (len < 0)
2254 return -EINVAL;
2255 if (len >= buf_sz)
2256 return -ENAMETOOLONG;
2257
2258 return 0;
2259 }
2260
2261 static int build_map_pin_path(struct bpf_map *map, const char *path)
2262 {
2263 char buf[PATH_MAX];
2264 int err;
2265
2266 if (!path)
2267 path = "/sys/fs/bpf";
2268
2269 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
2270 if (err)
2271 return err;
2272
2273 return bpf_map__set_pin_path(map, buf);
2274 }
2275
2276 /* should match definition in bpf_helpers.h */
2277 enum libbpf_pin_type {
2278 LIBBPF_PIN_NONE,
2279 /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
2280 LIBBPF_PIN_BY_NAME,
2281 };
2282
2283 int parse_btf_map_def(const char *map_name, struct btf *btf,
2284 const struct btf_type *def_t, bool strict,
2285 struct btf_map_def *map_def, struct btf_map_def *inner_def)
2286 {
2287 const struct btf_type *t;
2288 const struct btf_member *m;
2289 bool is_inner = inner_def == NULL;
2290 int vlen, i;
2291
2292 vlen = btf_vlen(def_t);
2293 m = btf_members(def_t);
2294 for (i = 0; i < vlen; i++, m++) {
2295 const char *name = btf__name_by_offset(btf, m->name_off);
2296
2297 if (!name) {
2298 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2299 return -EINVAL;
2300 }
2301 if (strcmp(name, "type") == 0) {
2302 if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2303 return -EINVAL;
2304 map_def->parts |= MAP_DEF_MAP_TYPE;
2305 } else if (strcmp(name, "max_entries") == 0) {
2306 if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2307 return -EINVAL;
2308 map_def->parts |= MAP_DEF_MAX_ENTRIES;
2309 } else if (strcmp(name, "map_flags") == 0) {
2310 if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2311 return -EINVAL;
2312 map_def->parts |= MAP_DEF_MAP_FLAGS;
2313 } else if (strcmp(name, "numa_node") == 0) {
2314 if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2315 return -EINVAL;
2316 map_def->parts |= MAP_DEF_NUMA_NODE;
2317 } else if (strcmp(name, "key_size") == 0) {
2318 __u32 sz;
2319
2320 if (!get_map_field_int(map_name, btf, m, &sz))
2321 return -EINVAL;
2322 if (map_def->key_size && map_def->key_size != sz) {
2323 pr_warn("map '%s': conflicting key size %u != %u.\n",
2324 map_name, map_def->key_size, sz);
2325 return -EINVAL;
2326 }
2327 map_def->key_size = sz;
2328 map_def->parts |= MAP_DEF_KEY_SIZE;
2329 } else if (strcmp(name, "key") == 0) {
2330 __s64 sz;
2331
2332 t = btf__type_by_id(btf, m->type);
2333 if (!t) {
2334 pr_warn("map '%s': key type [%d] not found.\n",
2335 map_name, m->type);
2336 return -EINVAL;
2337 }
2338 if (!btf_is_ptr(t)) {
2339 pr_warn("map '%s': key spec is not PTR: %s.\n",
2340 map_name, btf_kind_str(t));
2341 return -EINVAL;
2342 }
2343 sz = btf__resolve_size(btf, t->type);
2344 if (sz < 0) {
2345 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2346 map_name, t->type, (ssize_t)sz);
2347 return sz;
2348 }
2349 if (map_def->key_size && map_def->key_size != sz) {
2350 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2351 map_name, map_def->key_size, (ssize_t)sz);
2352 return -EINVAL;
2353 }
2354 map_def->key_size = sz;
2355 map_def->key_type_id = t->type;
2356 map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2357 } else if (strcmp(name, "value_size") == 0) {
2358 __u32 sz;
2359
2360 if (!get_map_field_int(map_name, btf, m, &sz))
2361 return -EINVAL;
2362 if (map_def->value_size && map_def->value_size != sz) {
2363 pr_warn("map '%s': conflicting value size %u != %u.\n",
2364 map_name, map_def->value_size, sz);
2365 return -EINVAL;
2366 }
2367 map_def->value_size = sz;
2368 map_def->parts |= MAP_DEF_VALUE_SIZE;
2369 } else if (strcmp(name, "value") == 0) {
2370 __s64 sz;
2371
2372 t = btf__type_by_id(btf, m->type);
2373 if (!t) {
2374 pr_warn("map '%s': value type [%d] not found.\n",
2375 map_name, m->type);
2376 return -EINVAL;
2377 }
2378 if (!btf_is_ptr(t)) {
2379 pr_warn("map '%s': value spec is not PTR: %s.\n",
2380 map_name, btf_kind_str(t));
2381 return -EINVAL;
2382 }
2383 sz = btf__resolve_size(btf, t->type);
2384 if (sz < 0) {
2385 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2386 map_name, t->type, (ssize_t)sz);
2387 return sz;
2388 }
2389 if (map_def->value_size && map_def->value_size != sz) {
2390 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2391 map_name, map_def->value_size, (ssize_t)sz);
2392 return -EINVAL;
2393 }
2394 map_def->value_size = sz;
2395 map_def->value_type_id = t->type;
2396 map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2397 }
2398 else if (strcmp(name, "values") == 0) {
2399 bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2400 bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2401 const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2402 char inner_map_name[128];
2403 int err;
2404
2405 if (is_inner) {
2406 pr_warn("map '%s': multi-level inner maps not supported.\n",
2407 map_name);
2408 return -ENOTSUP;
2409 }
2410 if (i != vlen - 1) {
2411 pr_warn("map '%s': '%s' member should be last.\n",
2412 map_name, name);
2413 return -EINVAL;
2414 }
2415 if (!is_map_in_map && !is_prog_array) {
2416 pr_warn("map '%s': should be map-in-map or prog-array.\n",
2417 map_name);
2418 return -ENOTSUP;
2419 }
2420 if (map_def->value_size && map_def->value_size != 4) {
2421 pr_warn("map '%s': conflicting value size %u != 4.\n",
2422 map_name, map_def->value_size);
2423 return -EINVAL;
2424 }
2425 map_def->value_size = 4;
2426 t = btf__type_by_id(btf, m->type);
2427 if (!t) {
2428 pr_warn("map '%s': %s type [%d] not found.\n",
2429 map_name, desc, m->type);
2430 return -EINVAL;
2431 }
2432 if (!btf_is_array(t) || btf_array(t)->nelems) {
2433 pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2434 map_name, desc);
2435 return -EINVAL;
2436 }
2437 t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2438 if (!btf_is_ptr(t)) {
2439 pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2440 map_name, desc, btf_kind_str(t));
2441 return -EINVAL;
2442 }
2443 t = skip_mods_and_typedefs(btf, t->type, NULL);
2444 if (is_prog_array) {
2445 if (!btf_is_func_proto(t)) {
2446 pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2447 map_name, btf_kind_str(t));
2448 return -EINVAL;
2449 }
2450 continue;
2451 }
2452 if (!btf_is_struct(t)) {
2453 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2454 map_name, btf_kind_str(t));
2455 return -EINVAL;
2456 }
2457
2458 snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2459 err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2460 if (err)
2461 return err;
2462
2463 map_def->parts |= MAP_DEF_INNER_MAP;
2464 } else if (strcmp(name, "pinning") == 0) {
2465 __u32 val;
2466
2467 if (is_inner) {
2468 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2469 return -EINVAL;
2470 }
2471 if (!get_map_field_int(map_name, btf, m, &val))
2472 return -EINVAL;
2473 if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2474 pr_warn("map '%s': invalid pinning value %u.\n",
2475 map_name, val);
2476 return -EINVAL;
2477 }
2478 map_def->pinning = val;
2479 map_def->parts |= MAP_DEF_PINNING;
2480 } else if (strcmp(name, "map_extra") == 0) {
2481 __u32 map_extra;
2482
2483 if (!get_map_field_int(map_name, btf, m, &map_extra))
2484 return -EINVAL;
2485 map_def->map_extra = map_extra;
2486 map_def->parts |= MAP_DEF_MAP_EXTRA;
2487 } else {
2488 if (strict) {
2489 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2490 return -ENOTSUP;
2491 }
2492 pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2493 }
2494 }
2495
2496 if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2497 pr_warn("map '%s': map type isn't specified.\n", map_name);
2498 return -EINVAL;
2499 }
2500
2501 return 0;
2502 }
2503
2504 static size_t adjust_ringbuf_sz(size_t sz)
2505 {
2506 __u32 page_sz = sysconf(_SC_PAGE_SIZE);
2507 __u32 mul;
2508
2509 /* if user forgot to set any size, make sure they see error */
2510 if (sz == 0)
2511 return 0;
2512 /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be
2513 * a power-of-2 multiple of kernel's page size. If user diligently
2514 * satisified these conditions, pass the size through.
2515 */
2516 if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz))
2517 return sz;
2518
2519 /* Otherwise find closest (page_sz * power_of_2) product bigger than
2520 * user-set size to satisfy both user size request and kernel
2521 * requirements and substitute correct max_entries for map creation.
2522 */
2523 for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) {
2524 if (mul * page_sz > sz)
2525 return mul * page_sz;
2526 }
2527
2528 /* if it's impossible to satisfy the conditions (i.e., user size is
2529 * very close to UINT_MAX but is not a power-of-2 multiple of
2530 * page_size) then just return original size and let kernel reject it
2531 */
2532 return sz;
2533 }
2534
2535 static bool map_is_ringbuf(const struct bpf_map *map)
2536 {
2537 return map->def.type == BPF_MAP_TYPE_RINGBUF ||
2538 map->def.type == BPF_MAP_TYPE_USER_RINGBUF;
2539 }
2540
2541 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2542 {
2543 map->def.type = def->map_type;
2544 map->def.key_size = def->key_size;
2545 map->def.value_size = def->value_size;
2546 map->def.max_entries = def->max_entries;
2547 map->def.map_flags = def->map_flags;
2548 map->map_extra = def->map_extra;
2549
2550 map->numa_node = def->numa_node;
2551 map->btf_key_type_id = def->key_type_id;
2552 map->btf_value_type_id = def->value_type_id;
2553
2554 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
2555 if (map_is_ringbuf(map))
2556 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
2557
2558 if (def->parts & MAP_DEF_MAP_TYPE)
2559 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2560
2561 if (def->parts & MAP_DEF_KEY_TYPE)
2562 pr_debug("map '%s': found key [%u], sz = %u.\n",
2563 map->name, def->key_type_id, def->key_size);
2564 else if (def->parts & MAP_DEF_KEY_SIZE)
2565 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2566
2567 if (def->parts & MAP_DEF_VALUE_TYPE)
2568 pr_debug("map '%s': found value [%u], sz = %u.\n",
2569 map->name, def->value_type_id, def->value_size);
2570 else if (def->parts & MAP_DEF_VALUE_SIZE)
2571 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2572
2573 if (def->parts & MAP_DEF_MAX_ENTRIES)
2574 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2575 if (def->parts & MAP_DEF_MAP_FLAGS)
2576 pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2577 if (def->parts & MAP_DEF_MAP_EXTRA)
2578 pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2579 (unsigned long long)def->map_extra);
2580 if (def->parts & MAP_DEF_PINNING)
2581 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2582 if (def->parts & MAP_DEF_NUMA_NODE)
2583 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2584
2585 if (def->parts & MAP_DEF_INNER_MAP)
2586 pr_debug("map '%s': found inner map definition.\n", map->name);
2587 }
2588
2589 static const char *btf_var_linkage_str(__u32 linkage)
2590 {
2591 switch (linkage) {
2592 case BTF_VAR_STATIC: return "static";
2593 case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2594 case BTF_VAR_GLOBAL_EXTERN: return "extern";
2595 default: return "unknown";
2596 }
2597 }
2598
2599 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2600 const struct btf_type *sec,
2601 int var_idx, int sec_idx,
2602 const Elf_Data *data, bool strict,
2603 const char *pin_root_path)
2604 {
2605 struct btf_map_def map_def = {}, inner_def = {};
2606 const struct btf_type *var, *def;
2607 const struct btf_var_secinfo *vi;
2608 const struct btf_var *var_extra;
2609 const char *map_name;
2610 struct bpf_map *map;
2611 int err;
2612
2613 vi = btf_var_secinfos(sec) + var_idx;
2614 var = btf__type_by_id(obj->btf, vi->type);
2615 var_extra = btf_var(var);
2616 map_name = btf__name_by_offset(obj->btf, var->name_off);
2617
2618 if (map_name == NULL || map_name[0] == '\0') {
2619 pr_warn("map #%d: empty name.\n", var_idx);
2620 return -EINVAL;
2621 }
2622 if ((__u64)vi->offset + vi->size > data->d_size) {
2623 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2624 return -EINVAL;
2625 }
2626 if (!btf_is_var(var)) {
2627 pr_warn("map '%s': unexpected var kind %s.\n",
2628 map_name, btf_kind_str(var));
2629 return -EINVAL;
2630 }
2631 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2632 pr_warn("map '%s': unsupported map linkage %s.\n",
2633 map_name, btf_var_linkage_str(var_extra->linkage));
2634 return -EOPNOTSUPP;
2635 }
2636
2637 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2638 if (!btf_is_struct(def)) {
2639 pr_warn("map '%s': unexpected def kind %s.\n",
2640 map_name, btf_kind_str(var));
2641 return -EINVAL;
2642 }
2643 if (def->size > vi->size) {
2644 pr_warn("map '%s': invalid def size.\n", map_name);
2645 return -EINVAL;
2646 }
2647
2648 map = bpf_object__add_map(obj);
2649 if (IS_ERR(map))
2650 return PTR_ERR(map);
2651 map->name = strdup(map_name);
2652 if (!map->name) {
2653 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2654 return -ENOMEM;
2655 }
2656 map->libbpf_type = LIBBPF_MAP_UNSPEC;
2657 map->def.type = BPF_MAP_TYPE_UNSPEC;
2658 map->sec_idx = sec_idx;
2659 map->sec_offset = vi->offset;
2660 map->btf_var_idx = var_idx;
2661 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2662 map_name, map->sec_idx, map->sec_offset);
2663
2664 err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2665 if (err)
2666 return err;
2667
2668 fill_map_from_def(map, &map_def);
2669
2670 if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2671 err = build_map_pin_path(map, pin_root_path);
2672 if (err) {
2673 pr_warn("map '%s': couldn't build pin path.\n", map->name);
2674 return err;
2675 }
2676 }
2677
2678 if (map_def.parts & MAP_DEF_INNER_MAP) {
2679 map->inner_map = calloc(1, sizeof(*map->inner_map));
2680 if (!map->inner_map)
2681 return -ENOMEM;
2682 map->inner_map->fd = -1;
2683 map->inner_map->sec_idx = sec_idx;
2684 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2685 if (!map->inner_map->name)
2686 return -ENOMEM;
2687 sprintf(map->inner_map->name, "%s.inner", map_name);
2688
2689 fill_map_from_def(map->inner_map, &inner_def);
2690 }
2691
2692 err = map_fill_btf_type_info(obj, map);
2693 if (err)
2694 return err;
2695
2696 return 0;
2697 }
2698
2699 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2700 const char *pin_root_path)
2701 {
2702 const struct btf_type *sec = NULL;
2703 int nr_types, i, vlen, err;
2704 const struct btf_type *t;
2705 const char *name;
2706 Elf_Data *data;
2707 #ifdef HAVE_LIBELF
2708 Elf_Scn *scn;
2709 #endif
2710
2711 if (obj->efile.btf_maps_shndx < 0)
2712 return 0;
2713 #if defined HAVE_LIBELF
2714 scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2715 data = elf_sec_data(obj, scn);
2716 if (!scn || !data) {
2717 #elif defined HAVE_ELFIO
2718 Elf_Data realdata;
2719 data = elf_sec_data_by_idx(obj, obj->efile.btf_maps_shndx, &realdata);
2720 if (!data) {
2721 #endif
2722 pr_warn("elf: failed to get %s map definitions for %s\n",
2723 MAPS_ELF_SEC, obj->path);
2724 return -EINVAL;
2725 }
2726
2727 nr_types = btf__type_cnt(obj->btf);
2728 for (i = 1; i < nr_types; i++) {
2729 t = btf__type_by_id(obj->btf, i);
2730 if (!btf_is_datasec(t))
2731 continue;
2732 name = btf__name_by_offset(obj->btf, t->name_off);
2733 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2734 sec = t;
2735 obj->efile.btf_maps_sec_btf_id = i;
2736 break;
2737 }
2738 }
2739
2740 if (!sec) {
2741 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2742 return -ENOENT;
2743 }
2744
2745 vlen = btf_vlen(sec);
2746 for (i = 0; i < vlen; i++) {
2747 err = bpf_object__init_user_btf_map(obj, sec, i,
2748 obj->efile.btf_maps_shndx,
2749 data, strict,
2750 pin_root_path);
2751 if (err)
2752 return err;
2753 }
2754
2755 return 0;
2756 }
2757
2758 static int bpf_object__init_maps(struct bpf_object *obj,
2759 const struct bpf_object_open_opts *opts)
2760 {
2761 const char *pin_root_path;
2762 bool strict;
2763 int err = 0;
2764
2765 strict = !OPTS_GET(opts, relaxed_maps, false);
2766 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2767
2768 err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2769 err = err ?: bpf_object__init_global_data_maps(obj);
2770 err = err ?: bpf_object__init_kconfig_map(obj);
2771 err = err ?: bpf_object__init_struct_ops_maps(obj);
2772
2773 return err;
2774 }
2775
2776 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2777 {
2778 Elf64_Shdr *sh;
2779 #if defined HAVE_LIBELF
2780 sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
2781 #elif defined HAVE_ELFIO
2782 Elf64_Shdr header;
2783 sh = elf_sec_hdr_by_idx(obj, idx, &header);
2784 #endif
2785 if (!sh)
2786 return false;
2787
2788 return sh->sh_flags & SHF_EXECINSTR;
2789 }
2790
2791 static bool btf_needs_sanitization(struct bpf_object *obj)
2792 {
2793 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2794 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2795 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2796 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2797 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2798 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2799 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
2800
2801 return !has_func || !has_datasec || !has_func_global || !has_float ||
2802 !has_decl_tag || !has_type_tag || !has_enum64;
2803 }
2804
2805 static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2806 {
2807 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2808 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2809 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2810 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2811 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2812 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2813 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
2814 int enum64_placeholder_id = 0;
2815 struct btf_type *t;
2816 int i, j, vlen;
2817
2818 for (i = 1; i < btf__type_cnt(btf); i++) {
2819 t = (struct btf_type *)btf__type_by_id(btf, i);
2820
2821 if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
2822 /* replace VAR/DECL_TAG with INT */
2823 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2824 /*
2825 * using size = 1 is the safest choice, 4 will be too
2826 * big and cause kernel BTF validation failure if
2827 * original variable took less than 4 bytes
2828 */
2829 t->size = 1;
2830 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2831 } else if (!has_datasec && btf_is_datasec(t)) {
2832 /* replace DATASEC with STRUCT */
2833 const struct btf_var_secinfo *v = btf_var_secinfos(t);
2834 struct btf_member *m = btf_members(t);
2835 struct btf_type *vt;
2836 char *name;
2837
2838 name = (char *)btf__name_by_offset(btf, t->name_off);
2839 while (*name) {
2840 if (*name == '.')
2841 *name = '_';
2842 name++;
2843 }
2844
2845 vlen = btf_vlen(t);
2846 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2847 for (j = 0; j < vlen; j++, v++, m++) {
2848 /* order of field assignments is important */
2849 m->offset = v->offset * 8;
2850 m->type = v->type;
2851 /* preserve variable name as member name */
2852 vt = (void *)btf__type_by_id(btf, v->type);
2853 m->name_off = vt->name_off;
2854 }
2855 } else if (!has_func && btf_is_func_proto(t)) {
2856 /* replace FUNC_PROTO with ENUM */
2857 vlen = btf_vlen(t);
2858 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2859 t->size = sizeof(__u32); /* kernel enforced */
2860 } else if (!has_func && btf_is_func(t)) {
2861 /* replace FUNC with TYPEDEF */
2862 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2863 } else if (!has_func_global && btf_is_func(t)) {
2864 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2865 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2866 } else if (!has_float && btf_is_float(t)) {
2867 /* replace FLOAT with an equally-sized empty STRUCT;
2868 * since C compilers do not accept e.g. "float" as a
2869 * valid struct name, make it anonymous
2870 */
2871 t->name_off = 0;
2872 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
2873 } else if (!has_type_tag && btf_is_type_tag(t)) {
2874 /* replace TYPE_TAG with a CONST */
2875 t->name_off = 0;
2876 t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
2877 } else if (!has_enum64 && btf_is_enum(t)) {
2878 /* clear the kflag */
2879 t->info = btf_type_info(btf_kind(t), btf_vlen(t), false);
2880 } else if (!has_enum64 && btf_is_enum64(t)) {
2881 /* replace ENUM64 with a union */
2882 struct btf_member *m;
2883
2884 if (enum64_placeholder_id == 0) {
2885 enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0);
2886 if (enum64_placeholder_id < 0)
2887 return enum64_placeholder_id;
2888
2889 t = (struct btf_type *)btf__type_by_id(btf, i);
2890 }
2891
2892 m = btf_members(t);
2893 vlen = btf_vlen(t);
2894 t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen);
2895 for (j = 0; j < vlen; j++, m++) {
2896 m->type = enum64_placeholder_id;
2897 m->offset = 0;
2898 }
2899 }
2900 }
2901
2902 return 0;
2903 }
2904
2905 static bool libbpf_needs_btf(const struct bpf_object *obj)
2906 {
2907 return obj->efile.btf_maps_shndx >= 0 ||
2908 obj->efile.st_ops_shndx >= 0 ||
2909 obj->nr_extern > 0;
2910 }
2911
2912 static bool kernel_needs_btf(const struct bpf_object *obj)
2913 {
2914 return obj->efile.st_ops_shndx >= 0;
2915 }
2916
2917 static int bpf_object__init_btf(struct bpf_object *obj,
2918 Elf_Data *btf_data,
2919 Elf_Data *btf_ext_data)
2920 {
2921 int err = -ENOENT;
2922
2923 if (btf_data) {
2924 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2925 err = libbpf_get_error(obj->btf);
2926 if (err) {
2927 obj->btf = NULL;
2928 pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err);
2929 goto out;
2930 }
2931 /* enforce 8-byte pointers for BPF-targeted BTFs */
2932 btf__set_pointer_size(obj->btf, 8);
2933 }
2934 if (btf_ext_data) {
2935 struct btf_ext_info *ext_segs[3];
2936 int seg_num, sec_num;
2937
2938 if (!obj->btf) {
2939 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2940 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2941 goto out;
2942 }
2943 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
2944 err = libbpf_get_error(obj->btf_ext);
2945 if (err) {
2946 pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n",
2947 BTF_EXT_ELF_SEC, err);
2948 obj->btf_ext = NULL;
2949 goto out;
2950 }
2951
2952 /* setup .BTF.ext to ELF section mapping */
2953 ext_segs[0] = &obj->btf_ext->func_info;
2954 ext_segs[1] = &obj->btf_ext->line_info;
2955 ext_segs[2] = &obj->btf_ext->core_relo_info;
2956 for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) {
2957 struct btf_ext_info *seg = ext_segs[seg_num];
2958 const struct btf_ext_info_sec *sec;
2959 const char *sec_name;
2960 #ifdef HAVE_LIBELF
2961 Elf_Scn *scn;
2962 #elif defined HAVE_ELFIO
2963 psection_t sec_obj;
2964 #endif
2965
2966 if (seg->sec_cnt == 0)
2967 continue;
2968
2969 seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs));
2970 if (!seg->sec_idxs) {
2971 err = -ENOMEM;
2972 goto out;
2973 }
2974
2975 sec_num = 0;
2976 for_each_btf_ext_sec(seg, sec) {
2977 /* preventively increment index to avoid doing
2978 * this before every continue below
2979 */
2980 sec_num++;
2981
2982 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
2983 if (str_is_empty(sec_name))
2984 continue;
2985 #ifdef HAVE_LIBELF
2986 scn = elf_sec_by_name(obj, sec_name);
2987 if (!scn)
2988 continue;
2989 #elif defined HAVE_ELFIO
2990 pelfio_t elf = obj->efile.elf;
2991 sec_obj = elfio_get_section_by_name(elf, sec_name);
2992 if (!sec_obj)
2993 continue;
2994 #endif
2995 #ifdef HAVE_LIBELF
2996 seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn);
2997 #elif defined HAVE_ELFIO
2998 seg->sec_idxs[sec_num - 1] = elfio_section_get_index(sec_obj);
2999 #endif
3000 }
3001 }
3002 }
3003 out:
3004 if (err && libbpf_needs_btf(obj)) {
3005 pr_warn("BTF is required, but is missing or corrupted.\n");
3006 return err;
3007 }
3008 return 0;
3009 }
3010
3011 static int compare_vsi_off(const void *_a, const void *_b)
3012 {
3013 const struct btf_var_secinfo *a = _a;
3014 const struct btf_var_secinfo *b = _b;
3015
3016 return a->offset - b->offset;
3017 }
3018
3019 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
3020 struct btf_type *t)
3021 {
3022 __u32 size = 0, i, vars = btf_vlen(t);
3023 const char *sec_name = btf__name_by_offset(btf, t->name_off);
3024 struct btf_var_secinfo *vsi;
3025 bool fixup_offsets = false;
3026 int err;
3027
3028 if (!sec_name) {
3029 pr_debug("No name found in string section for DATASEC kind.\n");
3030 return -ENOENT;
3031 }
3032
3033 /* Extern-backing datasecs (.ksyms, .kconfig) have their size and
3034 * variable offsets set at the previous step. Further, not every
3035 * extern BTF VAR has corresponding ELF symbol preserved, so we skip
3036 * all fixups altogether for such sections and go straight to sorting
3037 * VARs within their DATASEC.
3038 */
3039 if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0)
3040 goto sort_vars;
3041
3042 /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to
3043 * fix this up. But BPF static linker already fixes this up and fills
3044 * all the sizes and offsets during static linking. So this step has
3045 * to be optional. But the STV_HIDDEN handling is non-optional for any
3046 * non-extern DATASEC, so the variable fixup loop below handles both
3047 * functions at the same time, paying the cost of BTF VAR <-> ELF
3048 * symbol matching just once.
3049 */
3050 if (t->size == 0) {
3051 err = find_elf_sec_sz(obj, sec_name, &size);
3052 if (err || !size) {
3053 pr_debug("sec '%s': failed to determine size from ELF: size %u, err %d\n",
3054 sec_name, size, err);
3055 return -ENOENT;
3056 }
3057
3058 t->size = size;
3059 fixup_offsets = true;
3060 }
3061
3062 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
3063 const struct btf_type *t_var;
3064 struct btf_var *var;
3065 const char *var_name;
3066 Elf64_Sym *sym;
3067
3068 t_var = btf__type_by_id(btf, vsi->type);
3069 if (!t_var || !btf_is_var(t_var)) {
3070 pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name);
3071 return -EINVAL;
3072 }
3073
3074 var = btf_var(t_var);
3075 if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN)
3076 continue;
3077
3078 var_name = btf__name_by_offset(btf, t_var->name_off);
3079 if (!var_name) {
3080 pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n",
3081 sec_name, i);
3082 return -ENOENT;
3083 }
3084
3085 sym = find_elf_var_sym(obj, var_name);
3086 if (IS_ERR(sym)) {
3087 pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n",
3088 sec_name, var_name);
3089 return -ENOENT;
3090 }
3091
3092 if (fixup_offsets)
3093 vsi->offset = sym->st_value;
3094
3095 /* if variable is a global/weak symbol, but has restricted
3096 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR
3097 * as static. This follows similar logic for functions (BPF
3098 * subprogs) and influences libbpf's further decisions about
3099 * whether to make global data BPF array maps as
3100 * BPF_F_MMAPABLE.
3101 */
3102 if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
3103 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)
3104 var->linkage = BTF_VAR_STATIC;
3105 }
3106
3107 sort_vars:
3108 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
3109 return 0;
3110 }
3111
3112 static int bpf_object_fixup_btf(struct bpf_object *obj)
3113 {
3114 int i, n, err = 0;
3115
3116 if (!obj->btf)
3117 return 0;
3118
3119 n = btf__type_cnt(obj->btf);
3120 for (i = 1; i < n; i++) {
3121 struct btf_type *t = btf_type_by_id(obj->btf, i);
3122
3123 /* Loader needs to fix up some of the things compiler
3124 * couldn't get its hands on while emitting BTF. This
3125 * is section size and global variable offset. We use
3126 * the info from the ELF itself for this purpose.
3127 */
3128 if (btf_is_datasec(t)) {
3129 err = btf_fixup_datasec(obj, obj->btf, t);
3130 if (err)
3131 return err;
3132 }
3133 }
3134
3135 return 0;
3136 }
3137
3138 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
3139 {
3140 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
3141 prog->type == BPF_PROG_TYPE_LSM)
3142 return true;
3143
3144 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
3145 * also need vmlinux BTF
3146 */
3147 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
3148 return true;
3149
3150 return false;
3151 }
3152
3153 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
3154 {
3155 struct bpf_program *prog;
3156 int i;
3157
3158 /* CO-RE relocations need kernel BTF, only when btf_custom_path
3159 * is not specified
3160 */
3161 if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
3162 return true;
3163
3164 /* Support for typed ksyms needs kernel BTF */
3165 for (i = 0; i < obj->nr_extern; i++) {
3166 const struct extern_desc *ext;
3167
3168 ext = &obj->externs[i];
3169 if (ext->type == EXT_KSYM && ext->ksym.type_id)
3170 return true;
3171 }
3172
3173 bpf_object__for_each_program(prog, obj) {
3174 if (!prog->autoload)
3175 continue;
3176 if (prog_needs_vmlinux_btf(prog))
3177 return true;
3178 }
3179
3180 return false;
3181 }
3182
3183 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
3184 {
3185 int err;
3186
3187 /* btf_vmlinux could be loaded earlier */
3188 if (obj->btf_vmlinux || obj->gen_loader)
3189 return 0;
3190
3191 if (!force && !obj_needs_vmlinux_btf(obj))
3192 return 0;
3193
3194 obj->btf_vmlinux = btf__load_vmlinux_btf();
3195 err = libbpf_get_error(obj->btf_vmlinux);
3196 if (err) {
3197 pr_warn("Error loading vmlinux BTF: %d\n", err);
3198 obj->btf_vmlinux = NULL;
3199 return err;
3200 }
3201 return 0;
3202 }
3203
3204 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
3205 {
3206 struct btf *kern_btf = obj->btf;
3207 bool btf_mandatory, sanitize;
3208 int i, err = 0;
3209
3210 if (!obj->btf)
3211 return 0;
3212
3213 if (!kernel_supports(obj, FEAT_BTF)) {
3214 if (kernel_needs_btf(obj)) {
3215 err = -EOPNOTSUPP;
3216 goto report;
3217 }
3218 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
3219 return 0;
3220 }
3221
3222 /* Even though some subprogs are global/weak, user might prefer more
3223 * permissive BPF verification process that BPF verifier performs for
3224 * static functions, taking into account more context from the caller
3225 * functions. In such case, they need to mark such subprogs with
3226 * __attribute__((visibility("hidden"))) and libbpf will adjust
3227 * corresponding FUNC BTF type to be marked as static and trigger more
3228 * involved BPF verification process.
3229 */
3230 for (i = 0; i < obj->nr_programs; i++) {
3231 struct bpf_program *prog = &obj->programs[i];
3232 struct btf_type *t;
3233 const char *name;
3234 int j, n;
3235
3236 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
3237 continue;
3238
3239 n = btf__type_cnt(obj->btf);
3240 for (j = 1; j < n; j++) {
3241 t = btf_type_by_id(obj->btf, j);
3242 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
3243 continue;
3244
3245 name = btf__str_by_offset(obj->btf, t->name_off);
3246 if (strcmp(name, prog->name) != 0)
3247 continue;
3248
3249 t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
3250 break;
3251 }
3252 }
3253
3254 sanitize = btf_needs_sanitization(obj);
3255 if (sanitize) {
3256 const void *raw_data;
3257 __u32 sz;
3258
3259 /* clone BTF to sanitize a copy and leave the original intact */
3260 raw_data = btf__raw_data(obj->btf, &sz);
3261 kern_btf = btf__new(raw_data, sz);
3262 err = libbpf_get_error(kern_btf);
3263 if (err)
3264 return err;
3265
3266 /* enforce 8-byte pointers for BPF-targeted BTFs */
3267 btf__set_pointer_size(obj->btf, 8);
3268 err = bpf_object__sanitize_btf(obj, kern_btf);
3269 if (err)
3270 return err;
3271 }
3272
3273 if (obj->gen_loader) {
3274 __u32 raw_size = 0;
3275 const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3276
3277 if (!raw_data)
3278 return -ENOMEM;
3279 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3280 /* Pretend to have valid FD to pass various fd >= 0 checks.
3281 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3282 */
3283 btf__set_fd(kern_btf, 0);
3284 } else {
3285 /* currently BPF_BTF_LOAD only supports log_level 1 */
3286 err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3287 obj->log_level ? 1 : 0);
3288 }
3289 if (sanitize) {
3290 if (!err) {
3291 /* move fd to libbpf's BTF */
3292 btf__set_fd(obj->btf, btf__fd(kern_btf));
3293 btf__set_fd(kern_btf, -1);
3294 }
3295 btf__free(kern_btf);
3296 }
3297 report:
3298 if (err) {
3299 btf_mandatory = kernel_needs_btf(obj);
3300 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
3301 btf_mandatory ? "BTF is mandatory, can't proceed."
3302 : "BTF is optional, ignoring.");
3303 if (!btf_mandatory)
3304 err = 0;
3305 }
3306 return err;
3307 }
3308
3309 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3310 {
3311 const char *name;
3312 #if defined HAVE_LIBELF
3313 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3314 #elif defined HAVE_ELFIO
3315 name = elfio_string_get_string(obj->efile.strstring, off);
3316 #endif
3317 if (!name) {
3318 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3319 off, obj->path, elf_errmsg(-1));
3320 return NULL;
3321 }
3322
3323 return name;
3324 }
3325
3326 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3327 {
3328 const char *name;
3329 #if defined HAVE_LIBELF
3330 name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3331 #elif defined HAVE_ELFIO
3332 name = elfio_string_get_string(obj->efile.shstring, off);
3333 #endif
3334
3335 if (!name) {
3336 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3337 off, obj->path, elf_errmsg(-1));
3338 return NULL;
3339 }
3340
3341 return name;
3342 }
3343
3344 #ifdef HAVE_LIBELF
3345 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3346 {
3347 Elf_Scn *scn;
3348
3349 scn = elf_getscn(obj->efile.elf, idx);
3350 if (!scn) {
3351 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3352 idx, obj->path, elf_errmsg(-1));
3353 return NULL;
3354 }
3355 return scn;
3356 }
3357
3358 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3359 {
3360 Elf_Scn *scn = NULL;
3361 Elf *elf = obj->efile.elf;
3362 const char *sec_name;
3363
3364 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3365 sec_name = elf_sec_name(obj, scn);
3366 if (!sec_name)
3367 return NULL;
3368
3369 if (strcmp(sec_name, name) != 0)
3370 continue;
3371
3372 return scn;
3373 }
3374 return NULL;
3375 }
3376
3377 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3378 {
3379 Elf64_Shdr *shdr;
3380
3381 if (!scn)
3382 return NULL;
3383
3384 shdr = elf64_getshdr(scn);
3385 if (!shdr) {
3386 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3387 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3388 return NULL;
3389 }
3390
3391 return shdr;
3392 }
3393
3394 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3395 {
3396 const char *name;
3397 Elf64_Shdr *sh;
3398
3399 if (!scn)
3400 return NULL;
3401
3402 sh = elf_sec_hdr(obj, scn);
3403 if (!sh)
3404 return NULL;
3405
3406 name = elf_sec_str(obj, sh->sh_name);
3407 if (!name) {
3408 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3409 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3410 return NULL;
3411 }
3412
3413 return name;
3414 }
3415 #elif defined HAVE_ELFIO
3416 static Elf64_Shdr *elf_sec_hdr_by_idx(const struct bpf_object *obj, size_t idx, Elf64_Shdr *sheader)
3417 {
3418 psection_t psection = elfio_get_section_by_index(obj->efile.elf, idx);
3419
3420 sheader->sh_name = elfio_section_get_name_string_offset(psection);
3421 sheader->sh_type = elfio_section_get_type(psection);
3422 sheader->sh_flags = elfio_section_get_flags(psection);
3423 sheader->sh_addr = elfio_section_get_address(psection);
3424 sheader->sh_offset = elfio_section_get_offset(psection);
3425 sheader->sh_size = elfio_section_get_size(psection);
3426 sheader->sh_link = elfio_section_get_link(psection);
3427 sheader->sh_info = elfio_section_get_info(psection);
3428 sheader->sh_addralign = elfio_section_get_addr_align(psection);
3429 sheader->sh_entsize = elfio_section_get_entry_size(psection);
3430
3431 return sheader;
3432 }
3433
3434 static const char *elf_sec_name_by_idx(const struct bpf_object *obj, size_t idx)
3435 {
3436 const char *name;
3437 Elf64_Shdr sh;
3438
3439 elf_sec_hdr_by_idx(obj, idx, &sh);
3440
3441 name = elf_sec_str(obj, sh.sh_name);
3442 if (!name) {
3443 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3444 idx, obj->path, elf_errmsg(-1));
3445 return NULL;
3446 }
3447
3448 return name;
3449 }
3450 #endif
3451
3452 #if defined HAVE_LIBELF
3453 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3454 {
3455 Elf_Data *data;
3456
3457 if (!scn)
3458 return NULL;
3459
3460 data = elf_getdata(scn, 0);
3461 if (!data) {
3462 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3463 elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3464 obj->path, elf_errmsg(-1));
3465 return NULL;
3466 }
3467
3468 return data;
3469 }
3470 #elif defined HAVE_ELFIO
3471 static Elf_Data *elf_sec_data_by_name(const struct bpf_object *obj, const char *name, Elf_Data *data)
3472 {
3473 pelfio_t elf = obj->efile.elf;
3474 psection_t psection_name = elfio_get_section_by_name(elf, name);
3475 data->d_buf = (void*)elfio_section_get_data(psection_name);
3476 data->d_size = elfio_section_get_size(psection_name);
3477
3478 return data;
3479 }
3480
3481 static Elf_Data *elf_sec_data_by_idx(const struct bpf_object *obj, size_t idx, Elf_Data *data)
3482 {
3483 pelfio_t elf = obj->efile.elf;
3484 psection_t psection_index = elfio_get_section_by_index(elf, idx);
3485 data->d_buf = (void*)elfio_section_get_data(psection_index);
3486 data->d_size = elfio_section_get_size(psection_index);
3487
3488 return data;
3489 }
3490 #endif
3491
3492 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3493 {
3494 if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3495 return NULL;
3496
3497 return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3498 }
3499
3500 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3501 {
3502 if (idx >= data->d_size / sizeof(Elf64_Rel))
3503 return NULL;
3504
3505 return (Elf64_Rel *)data->d_buf + idx;
3506 }
3507
3508 static bool is_sec_name_dwarf(const char *name)
3509 {
3510 /* approximation, but the actual list is too long */
3511 return str_has_pfx(name, ".debug_");
3512 }
3513
3514 static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3515 {
3516 /* no special handling of .strtab */
3517 if (hdr->sh_type == SHT_STRTAB)
3518 return true;
3519
3520 /* ignore .llvm_addrsig section as well */
3521 if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3522 return true;
3523
3524 /* no subprograms will lead to an empty .text section, ignore it */
3525 if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3526 strcmp(name, ".text") == 0)
3527 return true;
3528
3529 /* DWARF sections */
3530 if (is_sec_name_dwarf(name))
3531 return true;
3532
3533 if (str_has_pfx(name, ".rel")) {
3534 name += sizeof(".rel") - 1;
3535 /* DWARF section relocations */
3536 if (is_sec_name_dwarf(name))
3537 return true;
3538
3539 /* .BTF and .BTF.ext don't need relocations */
3540 if (strcmp(name, BTF_ELF_SEC) == 0 ||
3541 strcmp(name, BTF_EXT_ELF_SEC) == 0)
3542 return true;
3543 }
3544
3545 return false;
3546 }
3547
3548 static int cmp_progs(const void *_a, const void *_b)
3549 {
3550 const struct bpf_program *a = _a;
3551 const struct bpf_program *b = _b;
3552
3553 if (a->sec_idx != b->sec_idx)
3554 return a->sec_idx < b->sec_idx ? -1 : 1;
3555
3556 /* sec_insn_off can't be the same within the section */
3557 return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3558 }
3559
3560 static int bpf_object__elf_collect(struct bpf_object *obj)
3561 {
3562 struct elf_sec_desc *sec_desc;
3563 #if defined HAVE_LIBELF
3564 Elf *elf = obj->efile.elf;
3565 #elif defined HAVE_ELFIO
3566 pelfio_t elf = obj->efile.elf;
3567 #endif
3568 Elf_Data *btf_ext_data = NULL;
3569 Elf_Data *btf_data = NULL;
3570 int idx = 0, err = 0;
3571 const char *name;
3572 Elf_Data *data;
3573 #ifdef HAVE_LIBELF
3574 Elf_Scn *scn;
3575 #endif
3576 Elf64_Shdr *sh;
3577 #ifdef HAVE_ELFIO
3578 Elf64_Shdr secHeader = {0};
3579 sh = &secHeader;
3580 #endif
3581
3582 /* ELF section indices are 0-based, but sec #0 is special "invalid"
3583 * section. Since section count retrieved by elf_getshdrnum() does
3584 * include sec #0, it is already the necessary size of an array to keep
3585 * all the sections.
3586 */
3587 #ifdef HAVE_LIBELF
3588 if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) {
3589 pr_warn("elf: failed to get the number of sections for %s: %s\n",
3590 obj->path, elf_errmsg(-1));
3591 return -LIBBPF_ERRNO__FORMAT;
3592 }
3593 #elif defined HAVE_ELFIO
3594 obj->efile.sec_cnt = elfio_get_sections_num(elf);
3595 #endif
3596 obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3597 if (!obj->efile.secs)
3598 return -ENOMEM;
3599
3600 /* a bunch of ELF parsing functionality depends on processing symbols,
3601 * so do the first pass and find the symbol table
3602 */
3603 #if defined HAVE_LIBELF
3604 scn = NULL;
3605 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3606 sh = elf_sec_hdr(obj, scn);
3607 #elif defined HAVE_ELFIO
3608 int secno = elfio_get_sections_num(elf);
3609 for ( int i = 0; i < secno; i++ ) {
3610 Elf_Data realdata;
3611 sh = elf_sec_hdr_by_idx(obj, i, sh);
3612 #endif
3613 if (!sh)
3614 return -LIBBPF_ERRNO__FORMAT;
3615
3616 if (sh->sh_type == SHT_SYMTAB) {
3617 if (obj->efile.symbols) {
3618 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3619 return -LIBBPF_ERRNO__FORMAT;
3620 }
3621 #if defined HAVE_LIBELF
3622 data = elf_sec_data(obj, scn);
3623 #elif defined HAVE_ELFIO
3624 data = elf_sec_data_by_idx(obj, i, &realdata);
3625 #endif
3626 if (!data)
3627 return -LIBBPF_ERRNO__FORMAT;
3628 #ifdef HAVE_LIBELF
3629 idx = elf_ndxscn(scn);
3630 #endif
3631
3632 #if defined HAVE_LIBELF
3633 obj->efile.symbols = data;
3634 #elif defined HAVE_ELFIO
3635 obj->efile.realsymbols.d_buf = data->d_buf;
3636 obj->efile.realsymbols.d_size = data->d_size;
3637 obj->efile.symbols = &(obj->efile.realsymbols);
3638 #endif
3639
3640 #if defined HAVE_LIBELF
3641 obj->efile.symbols_shndx = idx;
3642 #elif defined HAVE_ELFIO
3643 obj->efile.symbols_shndx = i;
3644 #endif
3645 obj->efile.strtabidx = sh->sh_link;
3646 }
3647 }
3648
3649 #ifdef HAVE_ELFIO
3650 pstring_t shstring;
3651 pstring_t strstring;
3652
3653 psection_t psection = elfio_get_section_by_index(elf, obj->efile.strtabidx);
3654 if (!psection)
3655 return -LIBBPF_ERRNO__FORMAT;
3656 strstring = elfio_string_section_accessor_new(psection);
3657
3658 psection = elfio_get_section_by_index(elf, obj->efile.shstrndx);
3659 if (!psection)
3660 return -LIBBPF_ERRNO__FORMAT;
3661 shstring = elfio_string_section_accessor_new(psection);
3662
3663 if (!strstring || !shstring)
3664 return -LIBBPF_ERRNO__FORMAT;
3665 obj->efile.strstring = strstring;
3666 obj->efile.shstring = shstring;
3667 #endif
3668
3669 if (!obj->efile.symbols) {
3670 pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3671 obj->path);
3672 return -ENOENT;
3673 }
3674
3675 #ifdef HAVE_LIBELF
3676 scn = NULL;
3677 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3678 #elif defined HAVE_ELFIO
3679 for ( int i = 0; i < secno; i++ ) {
3680 psection_t ptmpsection = elfio_get_section_by_index(elf, i);
3681 elf_sec_hdr_by_idx(obj, i, sh);
3682 #endif
3683
3684 #if defined HAVE_LIBELF
3685 idx = elf_ndxscn(scn);
3686 #elif defined HAVE_ELFIO
3687 idx = i;
3688 #endif
3689 sec_desc = &obj->efile.secs[idx];
3690
3691 #if defined HAVE_LIBELF
3692 sh = elf_sec_hdr(obj, scn);
3693 #elif defined HAVE_ELFIO
3694 sh = elf_sec_hdr_by_idx(obj, i, sh);
3695 #endif
3696
3697 if (!sh)
3698 return -LIBBPF_ERRNO__FORMAT;
3699
3700 name = elf_sec_str(obj, sh->sh_name);
3701 if (!name)
3702 return -LIBBPF_ERRNO__FORMAT;
3703
3704 if (ignore_elf_section(sh, name))
3705 continue;
3706
3707 #if defined HAVE_LIBELF
3708 data = elf_sec_data(obj, scn);
3709 #elif defined HAVE_ELFIO
3710 data = elf_sec_data_by_idx(obj, i, &sec_desc->realdata);
3711 #endif
3712 if (!data)
3713 return -LIBBPF_ERRNO__FORMAT;
3714
3715 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3716 idx, name, (unsigned long)data->d_size,
3717 (int)sh->sh_link, (unsigned long)sh->sh_flags,
3718 (int)sh->sh_type);
3719
3720 if (strcmp(name, "license") == 0) {
3721 err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3722 if (err)
3723 return err;
3724 } else if (strcmp(name, "version") == 0) {
3725 err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3726 if (err)
3727 return err;
3728 } else if (strcmp(name, "maps") == 0) {
3729 pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n");
3730 return -ENOTSUP;
3731 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3732 obj->efile.btf_maps_shndx = idx;
3733 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3734 if (sh->sh_type != SHT_PROGBITS)
3735 return -LIBBPF_ERRNO__FORMAT;
3736 btf_data = data;
3737 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3738 if (sh->sh_type != SHT_PROGBITS)
3739 return -LIBBPF_ERRNO__FORMAT;
3740 btf_ext_data = data;
3741 } else if (sh->sh_type == SHT_SYMTAB) {
3742 /* already processed during the first pass above */
3743 } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3744 if (sh->sh_flags & SHF_EXECINSTR) {
3745 if (strcmp(name, ".text") == 0)
3746 obj->efile.text_shndx = idx;
3747 err = bpf_object__add_programs(obj, data, name, idx);
3748 if (err)
3749 return err;
3750 } else if (strcmp(name, DATA_SEC) == 0 ||
3751 str_has_pfx(name, DATA_SEC ".")) {
3752 sec_desc->sec_type = SEC_DATA;
3753 #if defined HAVE_LIBELF
3754 sec_desc->shdr = sh;
3755 sec_desc->data = data;
3756 #elif defined HAVE_ELFIO
3757 sec_desc->psection = ptmpsection;
3758 sec_desc->realdata.d_buf = data->d_buf;
3759 sec_desc->realdata.d_size = data->d_size;
3760 sec_desc->data = &(sec_desc->realdata);
3761 #endif
3762 } else if (strcmp(name, RODATA_SEC) == 0 ||
3763 str_has_pfx(name, RODATA_SEC ".")) {
3764 sec_desc->sec_type = SEC_RODATA;
3765 #if defined HAVE_LIBELF
3766 sec_desc->shdr = sh;
3767 sec_desc->data = data;
3768 #elif defined HAVE_ELFIO
3769 sec_desc->psection = ptmpsection;
3770 sec_desc->realdata.d_buf = data->d_buf;
3771 sec_desc->realdata.d_size = data->d_size;
3772 sec_desc->data = &(sec_desc->realdata);
3773 #endif
3774
3775 } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
3776 #if defined HAVE_LIBELF
3777 obj->efile.st_ops_data = data;
3778 #elif defined HAVE_ELFIO
3779 obj->efile.realst_ops_data.d_buf = data->d_buf;
3780 obj->efile.realst_ops_data.d_size = data->d_size;
3781 obj->efile.st_ops_data = &(obj->efile.realst_ops_data);
3782 #endif
3783 obj->efile.st_ops_shndx = idx;
3784 } else {
3785 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3786 idx, name);
3787 }
3788 } else if (sh->sh_type == SHT_REL) {
3789 int targ_sec_idx = sh->sh_info; /* points to other section */
3790
3791 if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3792 targ_sec_idx >= obj->efile.sec_cnt)
3793 return -LIBBPF_ERRNO__FORMAT;
3794
3795 /* Only do relo for section with exec instructions */
3796 if (!section_have_execinstr(obj, targ_sec_idx) &&
3797 strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3798 strcmp(name, ".rel" MAPS_ELF_SEC)) {
3799 #if defined HAVE_LIBELF
3800 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3801 idx, name, targ_sec_idx,
3802 elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
3803 #elif defined HAVE_ELFIO
3804 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3805 idx, name, targ_sec_idx,
3806 elf_sec_name_by_idx(obj, targ_sec_idx) ?: "<?>");
3807 #endif
3808 continue;
3809 }
3810
3811 sec_desc->sec_type = SEC_RELO;
3812 #if defined HAVE_LIBELF
3813 sec_desc->shdr = sh;
3814 #elif defined HAVE_ELFIO
3815 sec_desc->psection = ptmpsection;
3816 #endif
3817 sec_desc->data = data;
3818 } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 ||
3819 str_has_pfx(name, BSS_SEC "."))) {
3820 sec_desc->sec_type = SEC_BSS;
3821 #if defined HAVE_LIBELF
3822 sec_desc->shdr = sh;
3823 #elif defined HAVE_ELFIO
3824 sec_desc->psection = ptmpsection;
3825 #endif
3826 sec_desc->data = data;
3827 } else {
3828 pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3829 (size_t)sh->sh_size);
3830 }
3831 }
3832
3833 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3834 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3835 return -LIBBPF_ERRNO__FORMAT;
3836 }
3837
3838 /* sort BPF programs by section name and in-section instruction offset
3839 * for faster search
3840 */
3841 if (obj->nr_programs)
3842 qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
3843
3844 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
3845 }
3846
3847 static bool sym_is_extern(const Elf64_Sym *sym)
3848 {
3849 int bind = ELF64_ST_BIND(sym->st_info);
3850 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
3851 return sym->st_shndx == SHN_UNDEF &&
3852 (bind == STB_GLOBAL || bind == STB_WEAK) &&
3853 ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
3854 }
3855
3856 static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
3857 {
3858 int bind = ELF64_ST_BIND(sym->st_info);
3859 int type = ELF64_ST_TYPE(sym->st_info);
3860
3861 /* in .text section */
3862 if (sym->st_shndx != text_shndx)
3863 return false;
3864
3865 /* local function */
3866 if (bind == STB_LOCAL && type == STT_SECTION)
3867 return true;
3868
3869 /* global function */
3870 return bind == STB_GLOBAL && type == STT_FUNC;
3871 }
3872
3873 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
3874 {
3875 const struct btf_type *t;
3876 const char *tname;
3877 int i, n;
3878
3879 if (!btf)
3880 return -ESRCH;
3881
3882 n = btf__type_cnt(btf);
3883 for (i = 1; i < n; i++) {
3884 t = btf__type_by_id(btf, i);
3885
3886 if (!btf_is_var(t) && !btf_is_func(t))
3887 continue;
3888
3889 tname = btf__name_by_offset(btf, t->name_off);
3890 if (strcmp(tname, ext_name))
3891 continue;
3892
3893 if (btf_is_var(t) &&
3894 btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
3895 return -EINVAL;
3896
3897 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
3898 return -EINVAL;
3899
3900 return i;
3901 }
3902
3903 return -ENOENT;
3904 }
3905
3906 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3907 const struct btf_var_secinfo *vs;
3908 const struct btf_type *t;
3909 int i, j, n;
3910
3911 if (!btf)
3912 return -ESRCH;
3913
3914 n = btf__type_cnt(btf);
3915 for (i = 1; i < n; i++) {
3916 t = btf__type_by_id(btf, i);
3917
3918 if (!btf_is_datasec(t))
3919 continue;
3920
3921 vs = btf_var_secinfos(t);
3922 for (j = 0; j < btf_vlen(t); j++, vs++) {
3923 if (vs->type == ext_btf_id)
3924 return i;
3925 }
3926 }
3927
3928 return -ENOENT;
3929 }
3930
3931 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3932 bool *is_signed)
3933 {
3934 const struct btf_type *t;
3935 const char *name;
3936
3937 t = skip_mods_and_typedefs(btf, id, NULL);
3938 name = btf__name_by_offset(btf, t->name_off);
3939
3940 if (is_signed)
3941 *is_signed = false;
3942 switch (btf_kind(t)) {
3943 case BTF_KIND_INT: {
3944 int enc = btf_int_encoding(t);
3945
3946 if (enc & BTF_INT_BOOL)
3947 return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3948 if (is_signed)
3949 *is_signed = enc & BTF_INT_SIGNED;
3950 if (t->size == 1)
3951 return KCFG_CHAR;
3952 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3953 return KCFG_UNKNOWN;
3954 return KCFG_INT;
3955 }
3956 case BTF_KIND_ENUM:
3957 if (t->size != 4)
3958 return KCFG_UNKNOWN;
3959 if (strcmp(name, "libbpf_tristate"))
3960 return KCFG_UNKNOWN;
3961 return KCFG_TRISTATE;
3962 case BTF_KIND_ENUM64:
3963 if (strcmp(name, "libbpf_tristate"))
3964 return KCFG_UNKNOWN;
3965 return KCFG_TRISTATE;
3966 case BTF_KIND_ARRAY:
3967 if (btf_array(t)->nelems == 0)
3968 return KCFG_UNKNOWN;
3969 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3970 return KCFG_UNKNOWN;
3971 return KCFG_CHAR_ARR;
3972 default:
3973 return KCFG_UNKNOWN;
3974 }
3975 }
3976
3977 static int cmp_externs(const void *_a, const void *_b)
3978 {
3979 const struct extern_desc *a = _a;
3980 const struct extern_desc *b = _b;
3981
3982 if (a->type != b->type)
3983 return a->type < b->type ? -1 : 1;
3984
3985 if (a->type == EXT_KCFG) {
3986 /* descending order by alignment requirements */
3987 if (a->kcfg.align != b->kcfg.align)
3988 return a->kcfg.align > b->kcfg.align ? -1 : 1;
3989 /* ascending order by size, within same alignment class */
3990 if (a->kcfg.sz != b->kcfg.sz)
3991 return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3992 }
3993
3994 /* resolve ties by name */
3995 return strcmp(a->name, b->name);
3996 }
3997
3998 static int find_int_btf_id(const struct btf *btf)
3999 {
4000 const struct btf_type *t;
4001 int i, n;
4002
4003 n = btf__type_cnt(btf);
4004 for (i = 1; i < n; i++) {
4005 t = btf__type_by_id(btf, i);
4006
4007 if (btf_is_int(t) && btf_int_bits(t) == 32)
4008 return i;
4009 }
4010
4011 return 0;
4012 }
4013
4014 static int add_dummy_ksym_var(struct btf *btf)
4015 {
4016 int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
4017 const struct btf_var_secinfo *vs;
4018 const struct btf_type *sec;
4019
4020 if (!btf)
4021 return 0;
4022
4023 sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
4024 BTF_KIND_DATASEC);
4025 if (sec_btf_id < 0)
4026 return 0;
4027
4028 sec = btf__type_by_id(btf, sec_btf_id);
4029 vs = btf_var_secinfos(sec);
4030 for (i = 0; i < btf_vlen(sec); i++, vs++) {
4031 const struct btf_type *vt;
4032
4033 vt = btf__type_by_id(btf, vs->type);
4034 if (btf_is_func(vt))
4035 break;
4036 }
4037
4038 /* No func in ksyms sec. No need to add dummy var. */
4039 if (i == btf_vlen(sec))
4040 return 0;
4041
4042 int_btf_id = find_int_btf_id(btf);
4043 dummy_var_btf_id = btf__add_var(btf,
4044 "dummy_ksym",
4045 BTF_VAR_GLOBAL_ALLOCATED,
4046 int_btf_id);
4047 if (dummy_var_btf_id < 0)
4048 pr_warn("cannot create a dummy_ksym var\n");
4049
4050 return dummy_var_btf_id;
4051 }
4052
4053 static int bpf_object__collect_externs(struct bpf_object *obj)
4054 {
4055 struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
4056 const struct btf_type *t;
4057 struct extern_desc *ext;
4058 int i, n, off, dummy_var_btf_id;
4059 const char *ext_name, *sec_name;
4060 #ifdef HAVE_LIBELF
4061 Elf_Scn *scn;
4062 #endif
4063 Elf64_Shdr *sh;
4064 Elf64_Shdr shheader;
4065
4066 if (!obj->efile.symbols)
4067 return 0;
4068
4069 #if defined HAVE_LIBELF
4070 scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
4071 sh = elf_sec_hdr(obj, scn);
4072 #elif defined HAVE_ELFIO
4073 sh = &shheader;
4074 sh = elf_sec_hdr_by_idx(obj, obj->efile.symbols_shndx, sh);
4075 #endif
4076
4077 if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
4078 return -LIBBPF_ERRNO__FORMAT;
4079
4080 dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
4081 if (dummy_var_btf_id < 0)
4082 return dummy_var_btf_id;
4083
4084 n = sh->sh_size / sh->sh_entsize;
4085 pr_debug("looking for externs among %d symbols...\n", n);
4086
4087 for (i = 0; i < n; i++) {
4088 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
4089
4090 if (!sym)
4091 return -LIBBPF_ERRNO__FORMAT;
4092 if (!sym_is_extern(sym))
4093 continue;
4094 ext_name = elf_sym_str(obj, sym->st_name);
4095 if (!ext_name || !ext_name[0])
4096 continue;
4097
4098 ext = obj->externs;
4099 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
4100 if (!ext)
4101 return -ENOMEM;
4102 obj->externs = ext;
4103 ext = &ext[obj->nr_extern];
4104 memset(ext, 0, sizeof(*ext));
4105 obj->nr_extern++;
4106
4107 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
4108 if (ext->btf_id <= 0) {
4109 pr_warn("failed to find BTF for extern '%s': %d\n",
4110 ext_name, ext->btf_id);
4111 return ext->btf_id;
4112 }
4113 t = btf__type_by_id(obj->btf, ext->btf_id);
4114 ext->name = btf__name_by_offset(obj->btf, t->name_off);
4115 ext->sym_idx = i;
4116 ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
4117
4118 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
4119 if (ext->sec_btf_id <= 0) {
4120 pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
4121 ext_name, ext->btf_id, ext->sec_btf_id);
4122 return ext->sec_btf_id;
4123 }
4124 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
4125 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
4126
4127 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
4128 if (btf_is_func(t)) {
4129 pr_warn("extern function %s is unsupported under %s section\n",
4130 ext->name, KCONFIG_SEC);
4131 return -ENOTSUP;
4132 }
4133 kcfg_sec = sec;
4134 ext->type = EXT_KCFG;
4135 ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
4136 if (ext->kcfg.sz <= 0) {
4137 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
4138 ext_name, ext->kcfg.sz);
4139 return ext->kcfg.sz;
4140 }
4141 ext->kcfg.align = btf__align_of(obj->btf, t->type);
4142 if (ext->kcfg.align <= 0) {
4143 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
4144 ext_name, ext->kcfg.align);
4145 return -EINVAL;
4146 }
4147 ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
4148 &ext->kcfg.is_signed);
4149 if (ext->kcfg.type == KCFG_UNKNOWN) {
4150 pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name);
4151 return -ENOTSUP;
4152 }
4153 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
4154 ksym_sec = sec;
4155 ext->type = EXT_KSYM;
4156 skip_mods_and_typedefs(obj->btf, t->type,
4157 &ext->ksym.type_id);
4158 } else {
4159 pr_warn("unrecognized extern section '%s'\n", sec_name);
4160 return -ENOTSUP;
4161 }
4162 }
4163 pr_debug("collected %d externs total\n", obj->nr_extern);
4164
4165 if (!obj->nr_extern)
4166 return 0;
4167
4168 /* sort externs by type, for kcfg ones also by (align, size, name) */
4169 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
4170
4171 /* for .ksyms section, we need to turn all externs into allocated
4172 * variables in BTF to pass kernel verification; we do this by
4173 * pretending that each extern is a 8-byte variable
4174 */
4175 if (ksym_sec) {
4176 /* find existing 4-byte integer type in BTF to use for fake
4177 * extern variables in DATASEC
4178 */
4179 int int_btf_id = find_int_btf_id(obj->btf);
4180 /* For extern function, a dummy_var added earlier
4181 * will be used to replace the vs->type and
4182 * its name string will be used to refill
4183 * the missing param's name.
4184 */
4185 const struct btf_type *dummy_var;
4186
4187 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
4188 for (i = 0; i < obj->nr_extern; i++) {
4189 ext = &obj->externs[i];
4190 if (ext->type != EXT_KSYM)
4191 continue;
4192 pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
4193 i, ext->sym_idx, ext->name);
4194 }
4195
4196 sec = ksym_sec;
4197 n = btf_vlen(sec);
4198 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
4199 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4200 struct btf_type *vt;
4201
4202 vt = (void *)btf__type_by_id(obj->btf, vs->type);
4203 ext_name = btf__name_by_offset(obj->btf, vt->name_off);
4204 ext = find_extern_by_name(obj, ext_name);
4205 if (!ext) {
4206 pr_warn("failed to find extern definition for BTF %s '%s'\n",
4207 btf_kind_str(vt), ext_name);
4208 return -ESRCH;
4209 }
4210 if (btf_is_func(vt)) {
4211 const struct btf_type *func_proto;
4212 struct btf_param *param;
4213 int j;
4214
4215 func_proto = btf__type_by_id(obj->btf,
4216 vt->type);
4217 param = btf_params(func_proto);
4218 /* Reuse the dummy_var string if the
4219 * func proto does not have param name.
4220 */
4221 for (j = 0; j < btf_vlen(func_proto); j++)
4222 if (param[j].type && !param[j].name_off)
4223 param[j].name_off =
4224 dummy_var->name_off;
4225 vs->type = dummy_var_btf_id;
4226 vt->info &= ~0xffff;
4227 vt->info |= BTF_FUNC_GLOBAL;
4228 } else {
4229 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4230 vt->type = int_btf_id;
4231 }
4232 vs->offset = off;
4233 vs->size = sizeof(int);
4234 }
4235 sec->size = off;
4236 }
4237
4238 if (kcfg_sec) {
4239 sec = kcfg_sec;
4240 /* for kcfg externs calculate their offsets within a .kconfig map */
4241 off = 0;
4242 for (i = 0; i < obj->nr_extern; i++) {
4243 ext = &obj->externs[i];
4244 if (ext->type != EXT_KCFG)
4245 continue;
4246
4247 ext->kcfg.data_off = roundup(off, ext->kcfg.align);
4248 off = ext->kcfg.data_off + ext->kcfg.sz;
4249 pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
4250 i, ext->sym_idx, ext->kcfg.data_off, ext->name);
4251 }
4252 sec->size = off;
4253 n = btf_vlen(sec);
4254 for (i = 0; i < n; i++) {
4255 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4256
4257 t = btf__type_by_id(obj->btf, vs->type);
4258 ext_name = btf__name_by_offset(obj->btf, t->name_off);
4259 ext = find_extern_by_name(obj, ext_name);
4260 if (!ext) {
4261 pr_warn("failed to find extern definition for BTF var '%s'\n",
4262 ext_name);
4263 return -ESRCH;
4264 }
4265 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4266 vs->offset = ext->kcfg.data_off;
4267 }
4268 }
4269 return 0;
4270 }
4271
4272 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog)
4273 {
4274 return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
4275 }
4276
4277 struct bpf_program *
4278 bpf_object__find_program_by_name(const struct bpf_object *obj,
4279 const char *name)
4280 {
4281 struct bpf_program *prog;
4282
4283 bpf_object__for_each_program(prog, obj) {
4284 if (prog_is_subprog(obj, prog))
4285 continue;
4286 if (!strcmp(prog->name, name))
4287 return prog;
4288 }
4289 return errno = ENOENT, NULL;
4290 }
4291
4292 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
4293 int shndx)
4294 {
4295 switch (obj->efile.secs[shndx].sec_type) {
4296 case SEC_BSS:
4297 case SEC_DATA:
4298 case SEC_RODATA:
4299 return true;
4300 default:
4301 return false;
4302 }
4303 }
4304
4305 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
4306 int shndx)
4307 {
4308 return shndx == obj->efile.btf_maps_shndx;
4309 }
4310
4311 static enum libbpf_map_type
4312 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
4313 {
4314 if (shndx == obj->efile.symbols_shndx)
4315 return LIBBPF_MAP_KCONFIG;
4316
4317 switch (obj->efile.secs[shndx].sec_type) {
4318 case SEC_BSS:
4319 return LIBBPF_MAP_BSS;
4320 case SEC_DATA:
4321 return LIBBPF_MAP_DATA;
4322 case SEC_RODATA:
4323 return LIBBPF_MAP_RODATA;
4324 default:
4325 return LIBBPF_MAP_UNSPEC;
4326 }
4327 }
4328
4329 static int bpf_program__record_reloc(struct bpf_program *prog,
4330 struct reloc_desc *reloc_desc,
4331 __u32 insn_idx, const char *sym_name,
4332 const Elf64_Sym *sym, const Elf64_Rel *rel)
4333 {
4334 struct bpf_insn *insn = &prog->insns[insn_idx];
4335 size_t map_idx, nr_maps = prog->obj->nr_maps;
4336 struct bpf_object *obj = prog->obj;
4337 __u32 shdr_idx = sym->st_shndx;
4338 enum libbpf_map_type type;
4339 const char *sym_sec_name;
4340 struct bpf_map *map;
4341
4342 if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
4343 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
4344 prog->name, sym_name, insn_idx, insn->code);
4345 return -LIBBPF_ERRNO__RELOC;
4346 }
4347
4348 if (sym_is_extern(sym)) {
4349 int sym_idx = ELF64_R_SYM(rel->r_info);
4350 int i, n = obj->nr_extern;
4351 struct extern_desc *ext;
4352
4353 for (i = 0; i < n; i++) {
4354 ext = &obj->externs[i];
4355 if (ext->sym_idx == sym_idx)
4356 break;
4357 }
4358 if (i >= n) {
4359 pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
4360 prog->name, sym_name, sym_idx);
4361 return -LIBBPF_ERRNO__RELOC;
4362 }
4363 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
4364 prog->name, i, ext->name, ext->sym_idx, insn_idx);
4365 if (insn->code == (BPF_JMP | BPF_CALL))
4366 reloc_desc->type = RELO_EXTERN_FUNC;
4367 else
4368 reloc_desc->type = RELO_EXTERN_VAR;
4369 reloc_desc->insn_idx = insn_idx;
4370 reloc_desc->sym_off = i; /* sym_off stores extern index */
4371 return 0;
4372 }
4373
4374 /* sub-program call relocation */
4375 if (is_call_insn(insn)) {
4376 if (insn->src_reg != BPF_PSEUDO_CALL) {
4377 pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
4378 return -LIBBPF_ERRNO__RELOC;
4379 }
4380 /* text_shndx can be 0, if no default "main" program exists */
4381 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
4382 #if defined HAVE_LIBELF
4383 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4384 #elif defined HAVE_ELFIO
4385 sym_sec_name = elf_sec_name_by_idx(obj, shdr_idx);
4386 #endif
4387 pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
4388 prog->name, sym_name, sym_sec_name);
4389 return -LIBBPF_ERRNO__RELOC;
4390 }
4391 if (sym->st_value % BPF_INSN_SZ) {
4392 pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
4393 prog->name, sym_name, (size_t)sym->st_value);
4394 return -LIBBPF_ERRNO__RELOC;
4395 }
4396 reloc_desc->type = RELO_CALL;
4397 reloc_desc->insn_idx = insn_idx;
4398 reloc_desc->sym_off = sym->st_value;
4399 return 0;
4400 }
4401
4402 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
4403 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
4404 prog->name, sym_name, shdr_idx);
4405 return -LIBBPF_ERRNO__RELOC;
4406 }
4407
4408 /* loading subprog addresses */
4409 if (sym_is_subprog(sym, obj->efile.text_shndx)) {
4410 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
4411 * local_func: sym->st_value = 0, insn->imm = offset in the section.
4412 */
4413 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
4414 pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
4415 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
4416 return -LIBBPF_ERRNO__RELOC;
4417 }
4418
4419 reloc_desc->type = RELO_SUBPROG_ADDR;
4420 reloc_desc->insn_idx = insn_idx;
4421 reloc_desc->sym_off = sym->st_value;
4422 return 0;
4423 }
4424
4425 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
4426 #if defined HAVE_LIBELF
4427 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4428 #elif defined HAVE_ELFIO
4429 sym_sec_name = elf_sec_name_by_idx(obj, shdr_idx);
4430 #endif
4431 /* generic map reference relocation */
4432 if (type == LIBBPF_MAP_UNSPEC) {
4433 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
4434 pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
4435 prog->name, sym_name, sym_sec_name);
4436 return -LIBBPF_ERRNO__RELOC;
4437 }
4438 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4439 map = &obj->maps[map_idx];
4440 if (map->libbpf_type != type ||
4441 map->sec_idx != sym->st_shndx ||
4442 map->sec_offset != sym->st_value)
4443 continue;
4444 pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4445 prog->name, map_idx, map->name, map->sec_idx,
4446 map->sec_offset, insn_idx);
4447 break;
4448 }
4449 if (map_idx >= nr_maps) {
4450 pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4451 prog->name, sym_sec_name, (size_t)sym->st_value);
4452 return -LIBBPF_ERRNO__RELOC;
4453 }
4454 reloc_desc->type = RELO_LD64;
4455 reloc_desc->insn_idx = insn_idx;
4456 reloc_desc->map_idx = map_idx;
4457 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4458 return 0;
4459 }
4460
4461 /* global data map relocation */
4462 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4463 pr_warn("prog '%s': bad data relo against section '%s'\n",
4464 prog->name, sym_sec_name);
4465 return -LIBBPF_ERRNO__RELOC;
4466 }
4467 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4468 map = &obj->maps[map_idx];
4469 if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4470 continue;
4471 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4472 prog->name, map_idx, map->name, map->sec_idx,
4473 map->sec_offset, insn_idx);
4474 break;
4475 }
4476 if (map_idx >= nr_maps) {
4477 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4478 prog->name, sym_sec_name);
4479 return -LIBBPF_ERRNO__RELOC;
4480 }
4481
4482 reloc_desc->type = RELO_DATA;
4483 reloc_desc->insn_idx = insn_idx;
4484 reloc_desc->map_idx = map_idx;
4485 reloc_desc->sym_off = sym->st_value;
4486 return 0;
4487 }
4488
4489 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4490 {
4491 return insn_idx >= prog->sec_insn_off &&
4492 insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4493 }
4494
4495 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4496 size_t sec_idx, size_t insn_idx)
4497 {
4498 int l = 0, r = obj->nr_programs - 1, m;
4499 struct bpf_program *prog;
4500
4501 if (!obj->nr_programs)
4502 return NULL;
4503
4504 while (l < r) {
4505 m = l + (r - l + 1) / 2;
4506 prog = &obj->programs[m];
4507
4508 if (prog->sec_idx < sec_idx ||
4509 (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4510 l = m;
4511 else
4512 r = m - 1;
4513 }
4514 /* matching program could be at index l, but it still might be the
4515 * wrong one, so we need to double check conditions for the last time
4516 */
4517 prog = &obj->programs[l];
4518 if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4519 return prog;
4520 return NULL;
4521 }
4522
4523 static int
4524 bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4525 {
4526 const char *relo_sec_name, *sec_name;
4527 size_t sec_idx = shdr->sh_info, sym_idx;
4528 struct bpf_program *prog;
4529 struct reloc_desc *relos;
4530 int err, i, nrels;
4531 const char *sym_name;
4532 __u32 insn_idx;
4533 #ifdef HAVE_LIBELF
4534 Elf_Scn *scn;
4535 #endif
4536 Elf_Data *scn_data;
4537 Elf64_Sym *sym;
4538 Elf64_Rel *rel;
4539
4540 if (sec_idx >= obj->efile.sec_cnt)
4541 return -EINVAL;
4542
4543 #if defined HAVE_LIBELF
4544 scn = elf_sec_by_idx(obj, sec_idx);
4545 scn_data = elf_sec_data(obj, scn);
4546
4547 relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4548 sec_name = elf_sec_name(obj, scn);
4549 if (!relo_sec_name || !sec_name)
4550 return -EINVAL;
4551 #elif defined HAVE_ELFIO
4552 Elf_Data realdata;
4553 scn_data = elf_sec_data_by_idx(obj, sec_idx, &realdata);
4554
4555 relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4556 sec_name = elf_sec_name_by_idx(obj, sec_idx);
4557 if (!relo_sec_name || !sec_name)
4558 return -EINVAL;
4559 #endif
4560
4561 pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4562 relo_sec_name, sec_idx, sec_name);
4563 nrels = shdr->sh_size / shdr->sh_entsize;
4564
4565 for (i = 0; i < nrels; i++) {
4566 rel = elf_rel_by_idx(data, i);
4567 if (!rel) {
4568 pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4569 return -LIBBPF_ERRNO__FORMAT;
4570 }
4571
4572 sym_idx = ELF64_R_SYM(rel->r_info);
4573 sym = elf_sym_by_idx(obj, sym_idx);
4574 if (!sym) {
4575 pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4576 relo_sec_name, sym_idx, i);
4577 return -LIBBPF_ERRNO__FORMAT;
4578 }
4579
4580 if (sym->st_shndx >= obj->efile.sec_cnt) {
4581 pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4582 relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4583 return -LIBBPF_ERRNO__FORMAT;
4584 }
4585
4586 if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4587 pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4588 relo_sec_name, (size_t)rel->r_offset, i);
4589 return -LIBBPF_ERRNO__FORMAT;
4590 }
4591
4592 insn_idx = rel->r_offset / BPF_INSN_SZ;
4593 /* relocations against static functions are recorded as
4594 * relocations against the section that contains a function;
4595 * in such case, symbol will be STT_SECTION and sym.st_name
4596 * will point to empty string (0), so fetch section name
4597 * instead
4598 */
4599 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4600 #if defined HAVE_LIBELF
4601 sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4602 #elif defined HAVE_ELFIO
4603 sym_name = elf_sec_name_by_idx(obj, sym->st_shndx);
4604 #endif
4605 else
4606 sym_name = elf_sym_str(obj, sym->st_name);
4607 sym_name = sym_name ?: "<?";
4608
4609 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4610 relo_sec_name, i, insn_idx, sym_name);
4611
4612 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4613 if (!prog) {
4614 pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4615 relo_sec_name, i, sec_name, insn_idx);
4616 continue;
4617 }
4618
4619 relos = libbpf_reallocarray(prog->reloc_desc,
4620 prog->nr_reloc + 1, sizeof(*relos));
4621 if (!relos)
4622 return -ENOMEM;
4623 prog->reloc_desc = relos;
4624
4625 /* adjust insn_idx to local BPF program frame of reference */
4626 insn_idx -= prog->sec_insn_off;
4627 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4628 insn_idx, sym_name, sym, rel);
4629 if (err)
4630 return err;
4631
4632 prog->nr_reloc++;
4633 }
4634 return 0;
4635 }
4636
4637 static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map)
4638 {
4639 int id;
4640
4641 if (!obj->btf)
4642 return -ENOENT;
4643
4644 /* if it's BTF-defined map, we don't need to search for type IDs.
4645 * For struct_ops map, it does not need btf_key_type_id and
4646 * btf_value_type_id.
4647 */
4648 if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map))
4649 return 0;
4650
4651 /*
4652 * LLVM annotates global data differently in BTF, that is,
4653 * only as '.data', '.bss' or '.rodata'.
4654 */
4655 if (!bpf_map__is_internal(map))
4656 return -ENOENT;
4657
4658 id = btf__find_by_name(obj->btf, map->real_name);
4659 if (id < 0)
4660 return id;
4661
4662 map->btf_key_type_id = 0;
4663 map->btf_value_type_id = id;
4664 return 0;
4665 }
4666
4667 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4668 {
4669 char file[PATH_MAX], buff[4096];
4670 FILE *fp;
4671 __u32 val;
4672 int err;
4673
4674 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4675 memset(info, 0, sizeof(*info));
4676
4677 fp = fopen(file, "r");
4678 if (!fp) {
4679 err = -errno;
4680 pr_warn("failed to open %s: %d. No procfs support?\n", file,
4681 err);
4682 return err;
4683 }
4684
4685 while (fgets(buff, sizeof(buff), fp)) {
4686 if (sscanf(buff, "map_type:\t%u", &val) == 1)
4687 info->type = val;
4688 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4689 info->key_size = val;
4690 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4691 info->value_size = val;
4692 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4693 info->max_entries = val;
4694 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4695 info->map_flags = val;
4696 }
4697
4698 fclose(fp);
4699
4700 return 0;
4701 }
4702
4703 bool bpf_map__autocreate(const struct bpf_map *map)
4704 {
4705 return map->autocreate;
4706 }
4707
4708 int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate)
4709 {
4710 if (map->obj->loaded)
4711 return libbpf_err(-EBUSY);
4712
4713 map->autocreate = autocreate;
4714 return 0;
4715 }
4716
4717 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4718 {
4719 struct bpf_map_info info;
4720 __u32 len = sizeof(info), name_len;
4721 int new_fd, err;
4722 char *new_name;
4723
4724 memset(&info, 0, len);
4725 err = bpf_obj_get_info_by_fd(fd, &info, &len);
4726 if (err && errno == EINVAL)
4727 err = bpf_get_map_info_from_fdinfo(fd, &info);
4728 if (err)
4729 return libbpf_err(err);
4730
4731 name_len = strlen(info.name);
4732 if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0)
4733 new_name = strdup(map->name);
4734 else
4735 new_name = strdup(info.name);
4736
4737 if (!new_name)
4738 return libbpf_err(-errno);
4739
4740 new_fd = open("/", O_RDONLY | O_CLOEXEC);
4741 if (new_fd < 0) {
4742 err = -errno;
4743 goto err_free_new_name;
4744 }
4745
4746 new_fd = dup3(fd, new_fd, O_CLOEXEC);
4747 if (new_fd < 0) {
4748 err = -errno;
4749 goto err_close_new_fd;
4750 }
4751
4752 err = zclose(map->fd);
4753 if (err) {
4754 err = -errno;
4755 goto err_close_new_fd;
4756 }
4757 free(map->name);
4758
4759 map->fd = new_fd;
4760 map->name = new_name;
4761 map->def.type = info.type;
4762 map->def.key_size = info.key_size;
4763 map->def.value_size = info.value_size;
4764 map->def.max_entries = info.max_entries;
4765 map->def.map_flags = info.map_flags;
4766 map->btf_key_type_id = info.btf_key_type_id;
4767 map->btf_value_type_id = info.btf_value_type_id;
4768 map->reused = true;
4769 map->map_extra = info.map_extra;
4770
4771 return 0;
4772
4773 err_close_new_fd:
4774 close(new_fd);
4775 err_free_new_name:
4776 free(new_name);
4777 return libbpf_err(err);
4778 }
4779
4780 __u32 bpf_map__max_entries(const struct bpf_map *map)
4781 {
4782 return map->def.max_entries;
4783 }
4784
4785 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
4786 {
4787 if (!bpf_map_type__is_map_in_map(map->def.type))
4788 return errno = EINVAL, NULL;
4789
4790 return map->inner_map;
4791 }
4792
4793 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4794 {
4795 if (map->obj->loaded)
4796 return libbpf_err(-EBUSY);
4797
4798 map->def.max_entries = max_entries;
4799
4800 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
4801 if (map_is_ringbuf(map))
4802 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
4803
4804 return 0;
4805 }
4806
4807 static int
4808 bpf_object__probe_loading(struct bpf_object *obj)
4809 {
4810 char *cp, errmsg[STRERR_BUFSIZE];
4811 struct bpf_insn insns[] = {
4812 BPF_MOV64_IMM(BPF_REG_0, 0),
4813 BPF_EXIT_INSN(),
4814 };
4815 int ret, insn_cnt = ARRAY_SIZE(insns);
4816
4817 if (obj->gen_loader)
4818 return 0;
4819
4820 ret = bump_rlimit_memlock();
4821 if (ret)
4822 pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret);
4823
4824 /* make sure basic loading works */
4825 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4826 if (ret < 0)
4827 ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4828 if (ret < 0) {
4829 ret = errno;
4830 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4831 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
4832 "program. Make sure your kernel supports BPF "
4833 "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
4834 "set to big enough value.\n", __func__, cp, ret);
4835 return -ret;
4836 }
4837 close(ret);
4838
4839 return 0;
4840 }
4841
4842 static int probe_fd(int fd)
4843 {
4844 if (fd >= 0)
4845 close(fd);
4846 return fd >= 0;
4847 }
4848
4849 static int probe_kern_prog_name(void)
4850 {
4851 const size_t attr_sz = offsetofend(union bpf_attr, prog_name);
4852 struct bpf_insn insns[] = {
4853 BPF_MOV64_IMM(BPF_REG_0, 0),
4854 BPF_EXIT_INSN(),
4855 };
4856 union bpf_attr attr;
4857 int ret;
4858
4859 memset(&attr, 0, attr_sz);
4860 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4861 attr.license = ptr_to_u64("GPL");
4862 attr.insns = ptr_to_u64(insns);
4863 attr.insn_cnt = (__u32)ARRAY_SIZE(insns);
4864 libbpf_strlcpy(attr.prog_name, "libbpf_nametest", sizeof(attr.prog_name));
4865
4866 /* make sure loading with name works */
4867 ret = sys_bpf_prog_load(&attr, attr_sz, PROG_LOAD_ATTEMPTS);
4868 return probe_fd(ret);
4869 }
4870
4871 static int probe_kern_global_data(void)
4872 {
4873 char *cp, errmsg[STRERR_BUFSIZE];
4874 struct bpf_insn insns[] = {
4875 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
4876 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
4877 BPF_MOV64_IMM(BPF_REG_0, 0),
4878 BPF_EXIT_INSN(),
4879 };
4880 int ret, map, insn_cnt = ARRAY_SIZE(insns);
4881
4882 map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_global", sizeof(int), 32, 1, NULL);
4883 if (map < 0) {
4884 ret = -errno;
4885 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4886 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4887 __func__, cp, -ret);
4888 return ret;
4889 }
4890
4891 insns[0].imm = map;
4892
4893 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4894 close(map);
4895 return probe_fd(ret);
4896 }
4897
4898 static int probe_kern_btf(void)
4899 {
4900 static const char strs[] = "\0int";
4901 __u32 types[] = {
4902 /* int */
4903 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4904 };
4905
4906 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4907 strs, sizeof(strs)));
4908 }
4909
4910 static int probe_kern_btf_func(void)
4911 {
4912 static const char strs[] = "\0int\0x\0a";
4913 /* void x(int a) {} */
4914 __u32 types[] = {
4915 /* int */
4916 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4917 /* FUNC_PROTO */ /* [2] */
4918 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4919 BTF_PARAM_ENC(7, 1),
4920 /* FUNC x */ /* [3] */
4921 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
4922 };
4923
4924 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4925 strs, sizeof(strs)));
4926 }
4927
4928 static int probe_kern_btf_func_global(void)
4929 {
4930 static const char strs[] = "\0int\0x\0a";
4931 /* static void x(int a) {} */
4932 __u32 types[] = {
4933 /* int */
4934 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4935 /* FUNC_PROTO */ /* [2] */
4936 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4937 BTF_PARAM_ENC(7, 1),
4938 /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */
4939 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
4940 };
4941
4942 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4943 strs, sizeof(strs)));
4944 }
4945
4946 static int probe_kern_btf_datasec(void)
4947 {
4948 static const char strs[] = "\0x\0.data";
4949 /* static int a; */
4950 __u32 types[] = {
4951 /* int */
4952 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4953 /* VAR x */ /* [2] */
4954 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4955 BTF_VAR_STATIC,
4956 /* DATASEC val */ /* [3] */
4957 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
4958 BTF_VAR_SECINFO_ENC(2, 0, 4),
4959 };
4960
4961 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4962 strs, sizeof(strs)));
4963 }
4964
4965 static int probe_kern_btf_float(void)
4966 {
4967 static const char strs[] = "\0float";
4968 __u32 types[] = {
4969 /* float */
4970 BTF_TYPE_FLOAT_ENC(1, 4),
4971 };
4972
4973 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4974 strs, sizeof(strs)));
4975 }
4976
4977 static int probe_kern_btf_decl_tag(void)
4978 {
4979 static const char strs[] = "\0tag";
4980 __u32 types[] = {
4981 /* int */
4982 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4983 /* VAR x */ /* [2] */
4984 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4985 BTF_VAR_STATIC,
4986 /* attr */
4987 BTF_TYPE_DECL_TAG_ENC(1, 2, -1),
4988 };
4989
4990 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4991 strs, sizeof(strs)));
4992 }
4993
4994 static int probe_kern_btf_type_tag(void)
4995 {
4996 static const char strs[] = "\0tag";
4997 __u32 types[] = {
4998 /* int */
4999 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
5000 /* attr */
5001 BTF_TYPE_TYPE_TAG_ENC(1, 1), /* [2] */
5002 /* ptr */
5003 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2), /* [3] */
5004 };
5005
5006 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
5007 strs, sizeof(strs)));
5008 }
5009
5010 static int probe_kern_array_mmap(void)
5011 {
5012 LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
5013 int fd;
5014
5015 fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_mmap", sizeof(int), sizeof(int), 1, &opts);
5016 return probe_fd(fd);
5017 }
5018
5019 static int probe_kern_exp_attach_type(void)
5020 {
5021 LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE);
5022 struct bpf_insn insns[] = {
5023 BPF_MOV64_IMM(BPF_REG_0, 0),
5024 BPF_EXIT_INSN(),
5025 };
5026 int fd, insn_cnt = ARRAY_SIZE(insns);
5027
5028 /* use any valid combination of program type and (optional)
5029 * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
5030 * to see if kernel supports expected_attach_type field for
5031 * BPF_PROG_LOAD command
5032 */
5033 fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts);
5034 return probe_fd(fd);
5035 }
5036
5037 static int probe_kern_probe_read_kernel(void)
5038 {
5039 struct bpf_insn insns[] = {
5040 BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */
5041 BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */
5042 BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */
5043 BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */
5044 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
5045 BPF_EXIT_INSN(),
5046 };
5047 int fd, insn_cnt = ARRAY_SIZE(insns);
5048
5049 fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
5050 return probe_fd(fd);
5051 }
5052
5053 static int probe_prog_bind_map(void)
5054 {
5055 char *cp, errmsg[STRERR_BUFSIZE];
5056 struct bpf_insn insns[] = {
5057 BPF_MOV64_IMM(BPF_REG_0, 0),
5058 BPF_EXIT_INSN(),
5059 };
5060 int ret, map, prog, insn_cnt = ARRAY_SIZE(insns);
5061
5062 map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_det_bind", sizeof(int), 32, 1, NULL);
5063 if (map < 0) {
5064 ret = -errno;
5065 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
5066 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
5067 __func__, cp, -ret);
5068 return ret;
5069 }
5070
5071 prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
5072 if (prog < 0) {
5073 close(map);
5074 return 0;
5075 }
5076
5077 ret = bpf_prog_bind_map(prog, map, NULL);
5078
5079 close(map);
5080 close(prog);
5081
5082 return ret >= 0;
5083 }
5084
5085 static int probe_module_btf(void)
5086 {
5087 static const char strs[] = "\0int";
5088 __u32 types[] = {
5089 /* int */
5090 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
5091 };
5092 struct bpf_btf_info info;
5093 __u32 len = sizeof(info);
5094 char name[16];
5095 int fd, err;
5096
5097 fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
5098 if (fd < 0)
5099 return 0; /* BTF not supported at all */
5100
5101 memset(&info, 0, sizeof(info));
5102 info.name = ptr_to_u64(name);
5103 info.name_len = sizeof(name);
5104
5105 /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
5106 * kernel's module BTF support coincides with support for
5107 * name/name_len fields in struct bpf_btf_info.
5108 */
5109 err = bpf_obj_get_info_by_fd(fd, &info, &len);
5110 close(fd);
5111 return !err;
5112 }
5113
5114 static int probe_perf_link(void)
5115 {
5116 struct bpf_insn insns[] = {
5117 BPF_MOV64_IMM(BPF_REG_0, 0),
5118 BPF_EXIT_INSN(),
5119 };
5120 int prog_fd, link_fd, err;
5121
5122 prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL",
5123 insns, ARRAY_SIZE(insns), NULL);
5124 if (prog_fd < 0)
5125 return -errno;
5126
5127 /* use invalid perf_event FD to get EBADF, if link is supported;
5128 * otherwise EINVAL should be returned
5129 */
5130 link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL);
5131 err = -errno; /* close() can clobber errno */
5132
5133 if (link_fd >= 0)
5134 close(link_fd);
5135 close(prog_fd);
5136
5137 return link_fd < 0 && err == -EBADF;
5138 }
5139
5140 static int probe_kern_bpf_cookie(void)
5141 {
5142 struct bpf_insn insns[] = {
5143 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_attach_cookie),
5144 BPF_EXIT_INSN(),
5145 };
5146 int ret, insn_cnt = ARRAY_SIZE(insns);
5147
5148 ret = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL);
5149 return probe_fd(ret);
5150 }
5151
5152 static int probe_kern_btf_enum64(void)
5153 {
5154 static const char strs[] = "\0enum64";
5155 __u32 types[] = {
5156 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_ENUM64, 0, 0), 8),
5157 };
5158
5159 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
5160 strs, sizeof(strs)));
5161 }
5162
5163 static int probe_kern_syscall_wrapper(void);
5164
5165 enum kern_feature_result {
5166 FEAT_UNKNOWN = 0,
5167 FEAT_SUPPORTED = 1,
5168 FEAT_MISSING = 2,
5169 };
5170
5171 typedef int (*feature_probe_fn)(void);
5172
5173 static struct kern_feature_desc {
5174 const char *desc;
5175 feature_probe_fn probe;
5176 enum kern_feature_result res;
5177 } feature_probes[__FEAT_CNT] = {
5178 [FEAT_PROG_NAME] = {
5179 "BPF program name", probe_kern_prog_name,
5180 },
5181 [FEAT_GLOBAL_DATA] = {
5182 "global variables", probe_kern_global_data,
5183 },
5184 [FEAT_BTF] = {
5185 "minimal BTF", probe_kern_btf,
5186 },
5187 [FEAT_BTF_FUNC] = {
5188 "BTF functions", probe_kern_btf_func,
5189 },
5190 [FEAT_BTF_GLOBAL_FUNC] = {
5191 "BTF global function", probe_kern_btf_func_global,
5192 },
5193 [FEAT_BTF_DATASEC] = {
5194 "BTF data section and variable", probe_kern_btf_datasec,
5195 },
5196 [FEAT_ARRAY_MMAP] = {
5197 "ARRAY map mmap()", probe_kern_array_mmap,
5198 },
5199 [FEAT_EXP_ATTACH_TYPE] = {
5200 "BPF_PROG_LOAD expected_attach_type attribute",
5201 probe_kern_exp_attach_type,
5202 },
5203 [FEAT_PROBE_READ_KERN] = {
5204 "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
5205 },
5206 [FEAT_PROG_BIND_MAP] = {
5207 "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
5208 },
5209 [FEAT_MODULE_BTF] = {
5210 "module BTF support", probe_module_btf,
5211 },
5212 [FEAT_BTF_FLOAT] = {
5213 "BTF_KIND_FLOAT support", probe_kern_btf_float,
5214 },
5215 [FEAT_PERF_LINK] = {
5216 "BPF perf link support", probe_perf_link,
5217 },
5218 [FEAT_BTF_DECL_TAG] = {
5219 "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag,
5220 },
5221 [FEAT_BTF_TYPE_TAG] = {
5222 "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag,
5223 },
5224 [FEAT_MEMCG_ACCOUNT] = {
5225 "memcg-based memory accounting", probe_memcg_account,
5226 },
5227 [FEAT_BPF_COOKIE] = {
5228 "BPF cookie support", probe_kern_bpf_cookie,
5229 },
5230 [FEAT_BTF_ENUM64] = {
5231 "BTF_KIND_ENUM64 support", probe_kern_btf_enum64,
5232 },
5233 [FEAT_SYSCALL_WRAPPER] = {
5234 "Kernel using syscall wrapper", probe_kern_syscall_wrapper,
5235 },
5236 };
5237
5238 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
5239 {
5240 struct kern_feature_desc *feat = &feature_probes[feat_id];
5241 int ret;
5242
5243 if (obj && obj->gen_loader)
5244 /* To generate loader program assume the latest kernel
5245 * to avoid doing extra prog_load, map_create syscalls.
5246 */
5247 return true;
5248
5249 if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
5250 ret = feat->probe();
5251 if (ret > 0) {
5252 WRITE_ONCE(feat->res, FEAT_SUPPORTED);
5253 } else if (ret == 0) {
5254 WRITE_ONCE(feat->res, FEAT_MISSING);
5255 } else {
5256 pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
5257 WRITE_ONCE(feat->res, FEAT_MISSING);
5258 }
5259 }
5260
5261 return READ_ONCE(feat->res) == FEAT_SUPPORTED;
5262 }
5263
5264 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
5265 {
5266 struct bpf_map_info map_info;
5267 char msg[STRERR_BUFSIZE];
5268 __u32 map_info_len = sizeof(map_info);
5269 int err;
5270
5271 memset(&map_info, 0, map_info_len);
5272 err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len);
5273 if (err && errno == EINVAL)
5274 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
5275 if (err) {
5276 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
5277 libbpf_strerror_r(errno, msg, sizeof(msg)));
5278 return false;
5279 }
5280
5281 return (map_info.type == map->def.type &&
5282 map_info.key_size == map->def.key_size &&
5283 map_info.value_size == map->def.value_size &&
5284 map_info.max_entries == map->def.max_entries &&
5285 map_info.map_flags == map->def.map_flags &&
5286 map_info.map_extra == map->map_extra);
5287 }
5288
5289 static int
5290 bpf_object__reuse_map(struct bpf_map *map)
5291 {
5292 char *cp, errmsg[STRERR_BUFSIZE];
5293 int err, pin_fd;
5294
5295 pin_fd = bpf_obj_get(map->pin_path);
5296 if (pin_fd < 0) {
5297 err = -errno;
5298 if (err == -ENOENT) {
5299 pr_debug("found no pinned map to reuse at '%s'\n",
5300 map->pin_path);
5301 return 0;
5302 }
5303
5304 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5305 pr_warn("couldn't retrieve pinned map '%s': %s\n",
5306 map->pin_path, cp);
5307 return err;
5308 }
5309
5310 if (!map_is_reuse_compat(map, pin_fd)) {
5311 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
5312 map->pin_path);
5313 close(pin_fd);
5314 return -EINVAL;
5315 }
5316
5317 err = bpf_map__reuse_fd(map, pin_fd);
5318 close(pin_fd);
5319 if (err)
5320 return err;
5321
5322 map->pinned = true;
5323 pr_debug("reused pinned map at '%s'\n", map->pin_path);
5324
5325 return 0;
5326 }
5327
5328 static int
5329 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
5330 {
5331 enum libbpf_map_type map_type = map->libbpf_type;
5332 char *cp, errmsg[STRERR_BUFSIZE];
5333 int err, zero = 0;
5334
5335 if (obj->gen_loader) {
5336 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
5337 map->mmaped, map->def.value_size);
5338 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
5339 bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
5340 return 0;
5341 }
5342 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
5343 if (err) {
5344 err = -errno;
5345 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5346 pr_warn("Error setting initial map(%s) contents: %s\n",
5347 map->name, cp);
5348 return err;
5349 }
5350
5351 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
5352 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
5353 err = bpf_map_freeze(map->fd);
5354 if (err) {
5355 err = -errno;
5356 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5357 pr_warn("Error freezing map(%s) as read-only: %s\n",
5358 map->name, cp);
5359 return err;
5360 }
5361 }
5362 return 0;
5363 }
5364
5365 static void bpf_map__destroy(struct bpf_map *map);
5366
5367 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
5368 {
5369 LIBBPF_OPTS(bpf_map_create_opts, create_attr);
5370 struct bpf_map_def *def = &map->def;
5371 const char *map_name = NULL;
5372 int err = 0;
5373
5374 if (kernel_supports(obj, FEAT_PROG_NAME))
5375 map_name = map->name;
5376 create_attr.map_ifindex = map->map_ifindex;
5377 create_attr.map_flags = def->map_flags;
5378 create_attr.numa_node = map->numa_node;
5379 create_attr.map_extra = map->map_extra;
5380
5381 if (bpf_map__is_struct_ops(map))
5382 create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
5383
5384 if (obj->btf && btf__fd(obj->btf) >= 0) {
5385 create_attr.btf_fd = btf__fd(obj->btf);
5386 create_attr.btf_key_type_id = map->btf_key_type_id;
5387 create_attr.btf_value_type_id = map->btf_value_type_id;
5388 }
5389
5390 if (bpf_map_type__is_map_in_map(def->type)) {
5391 if (map->inner_map) {
5392 err = bpf_object__create_map(obj, map->inner_map, true);
5393 if (err) {
5394 pr_warn("map '%s': failed to create inner map: %d\n",
5395 map->name, err);
5396 return err;
5397 }
5398 map->inner_map_fd = bpf_map__fd(map->inner_map);
5399 }
5400 if (map->inner_map_fd >= 0)
5401 create_attr.inner_map_fd = map->inner_map_fd;
5402 }
5403
5404 switch (def->type) {
5405 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
5406 case BPF_MAP_TYPE_CGROUP_ARRAY:
5407 case BPF_MAP_TYPE_STACK_TRACE:
5408 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
5409 case BPF_MAP_TYPE_HASH_OF_MAPS:
5410 case BPF_MAP_TYPE_DEVMAP:
5411 case BPF_MAP_TYPE_DEVMAP_HASH:
5412 case BPF_MAP_TYPE_CPUMAP:
5413 case BPF_MAP_TYPE_XSKMAP:
5414 case BPF_MAP_TYPE_SOCKMAP:
5415 case BPF_MAP_TYPE_SOCKHASH:
5416 case BPF_MAP_TYPE_QUEUE:
5417 case BPF_MAP_TYPE_STACK:
5418 create_attr.btf_fd = 0;
5419 create_attr.btf_key_type_id = 0;
5420 create_attr.btf_value_type_id = 0;
5421 map->btf_key_type_id = 0;
5422 map->btf_value_type_id = 0;
5423 break;
5424 default:
5425 break;
5426 }
5427
5428 if (obj->gen_loader) {
5429 bpf_gen__map_create(obj->gen_loader, def->type, map_name,
5430 def->key_size, def->value_size, def->max_entries,
5431 &create_attr, is_inner ? -1 : map - obj->maps);
5432 /* Pretend to have valid FD to pass various fd >= 0 checks.
5433 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
5434 */
5435 map->fd = 0;
5436 } else {
5437 map->fd = bpf_map_create(def->type, map_name,
5438 def->key_size, def->value_size,
5439 def->max_entries, &create_attr);
5440 }
5441 if (map->fd < 0 && (create_attr.btf_key_type_id ||
5442 create_attr.btf_value_type_id)) {
5443 char *cp, errmsg[STRERR_BUFSIZE];
5444
5445 err = -errno;
5446 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5447 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
5448 map->name, cp, err);
5449 create_attr.btf_fd = 0;
5450 create_attr.btf_key_type_id = 0;
5451 create_attr.btf_value_type_id = 0;
5452 map->btf_key_type_id = 0;
5453 map->btf_value_type_id = 0;
5454 map->fd = bpf_map_create(def->type, map_name,
5455 def->key_size, def->value_size,
5456 def->max_entries, &create_attr);
5457 }
5458
5459 err = map->fd < 0 ? -errno : 0;
5460
5461 if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
5462 if (obj->gen_loader)
5463 map->inner_map->fd = -1;
5464 bpf_map__destroy(map->inner_map);
5465 zfree(&map->inner_map);
5466 }
5467
5468 return err;
5469 }
5470
5471 static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
5472 {
5473 const struct bpf_map *targ_map;
5474 unsigned int i;
5475 int fd, err = 0;
5476
5477 for (i = 0; i < map->init_slots_sz; i++) {
5478 if (!map->init_slots[i])
5479 continue;
5480
5481 targ_map = map->init_slots[i];
5482 fd = bpf_map__fd(targ_map);
5483
5484 if (obj->gen_loader) {
5485 bpf_gen__populate_outer_map(obj->gen_loader,
5486 map - obj->maps, i,
5487 targ_map - obj->maps);
5488 } else {
5489 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5490 }
5491 if (err) {
5492 err = -errno;
5493 pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
5494 map->name, i, targ_map->name, fd, err);
5495 return err;
5496 }
5497 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5498 map->name, i, targ_map->name, fd);
5499 }
5500
5501 zfree(&map->init_slots);
5502 map->init_slots_sz = 0;
5503
5504 return 0;
5505 }
5506
5507 static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5508 {
5509 const struct bpf_program *targ_prog;
5510 unsigned int i;
5511 int fd, err;
5512
5513 if (obj->gen_loader)
5514 return -ENOTSUP;
5515
5516 for (i = 0; i < map->init_slots_sz; i++) {
5517 if (!map->init_slots[i])
5518 continue;
5519
5520 targ_prog = map->init_slots[i];
5521 fd = bpf_program__fd(targ_prog);
5522
5523 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5524 if (err) {
5525 err = -errno;
5526 pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n",
5527 map->name, i, targ_prog->name, fd, err);
5528 return err;
5529 }
5530 pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5531 map->name, i, targ_prog->name, fd);
5532 }
5533
5534 zfree(&map->init_slots);
5535 map->init_slots_sz = 0;
5536
5537 return 0;
5538 }
5539
5540 static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5541 {
5542 struct bpf_map *map;
5543 int i, err;
5544
5545 for (i = 0; i < obj->nr_maps; i++) {
5546 map = &obj->maps[i];
5547
5548 if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5549 continue;
5550
5551 err = init_prog_array_slots(obj, map);
5552 if (err < 0) {
5553 zclose(map->fd);
5554 return err;
5555 }
5556 }
5557 return 0;
5558 }
5559
5560 static int map_set_def_max_entries(struct bpf_map *map)
5561 {
5562 if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) {
5563 int nr_cpus;
5564
5565 nr_cpus = libbpf_num_possible_cpus();
5566 if (nr_cpus < 0) {
5567 pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
5568 map->name, nr_cpus);
5569 return nr_cpus;
5570 }
5571 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
5572 map->def.max_entries = nr_cpus;
5573 }
5574
5575 return 0;
5576 }
5577
5578 static int
5579 bpf_object__create_maps(struct bpf_object *obj)
5580 {
5581 struct bpf_map *map;
5582 char *cp, errmsg[STRERR_BUFSIZE];
5583 unsigned int i, j;
5584 int err;
5585 bool retried;
5586
5587 for (i = 0; i < obj->nr_maps; i++) {
5588 map = &obj->maps[i];
5589
5590 /* To support old kernels, we skip creating global data maps
5591 * (.rodata, .data, .kconfig, etc); later on, during program
5592 * loading, if we detect that at least one of the to-be-loaded
5593 * programs is referencing any global data map, we'll error
5594 * out with program name and relocation index logged.
5595 * This approach allows to accommodate Clang emitting
5596 * unnecessary .rodata.str1.1 sections for string literals,
5597 * but also it allows to have CO-RE applications that use
5598 * global variables in some of BPF programs, but not others.
5599 * If those global variable-using programs are not loaded at
5600 * runtime due to bpf_program__set_autoload(prog, false),
5601 * bpf_object loading will succeed just fine even on old
5602 * kernels.
5603 */
5604 if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA))
5605 map->autocreate = false;
5606
5607 if (!map->autocreate) {
5608 pr_debug("map '%s': skipped auto-creating...\n", map->name);
5609 continue;
5610 }
5611
5612 err = map_set_def_max_entries(map);
5613 if (err)
5614 goto err_out;
5615
5616 retried = false;
5617 retry:
5618 if (map->pin_path) {
5619 err = bpf_object__reuse_map(map);
5620 if (err) {
5621 pr_warn("map '%s': error reusing pinned map\n",
5622 map->name);
5623 goto err_out;
5624 }
5625 if (retried && map->fd < 0) {
5626 pr_warn("map '%s': cannot find pinned map\n",
5627 map->name);
5628 err = -ENOENT;
5629 goto err_out;
5630 }
5631 }
5632
5633 if (map->fd >= 0) {
5634 pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5635 map->name, map->fd);
5636 } else {
5637 err = bpf_object__create_map(obj, map, false);
5638 if (err)
5639 goto err_out;
5640
5641 pr_debug("map '%s': created successfully, fd=%d\n",
5642 map->name, map->fd);
5643
5644 if (bpf_map__is_internal(map)) {
5645 err = bpf_object__populate_internal_map(obj, map);
5646 if (err < 0) {
5647 zclose(map->fd);
5648 goto err_out;
5649 }
5650 }
5651
5652 if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5653 err = init_map_in_map_slots(obj, map);
5654 if (err < 0) {
5655 zclose(map->fd);
5656 goto err_out;
5657 }
5658 }
5659 }
5660
5661 if (map->pin_path && !map->pinned) {
5662 err = bpf_map__pin(map, NULL);
5663 if (err) {
5664 zclose(map->fd);
5665 if (!retried && err == -EEXIST) {
5666 retried = true;
5667 goto retry;
5668 }
5669 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
5670 map->name, map->pin_path, err);
5671 goto err_out;
5672 }
5673 }
5674 }
5675
5676 return 0;
5677
5678 err_out:
5679 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5680 pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
5681 pr_perm_msg(err);
5682 for (j = 0; j < i; j++)
5683 zclose(obj->maps[j].fd);
5684 return err;
5685 }
5686
5687 static bool bpf_core_is_flavor_sep(const char *s)
5688 {
5689 /* check X___Y name pattern, where X and Y are not underscores */
5690 return s[0] != '_' && /* X */
5691 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
5692 s[4] != '_'; /* Y */
5693 }
5694
5695 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
5696 * before last triple underscore. Struct name part after last triple
5697 * underscore is ignored by BPF CO-RE relocation during relocation matching.
5698 */
5699 size_t bpf_core_essential_name_len(const char *name)
5700 {
5701 size_t n = strlen(name);
5702 int i;
5703
5704 for (i = n - 5; i >= 0; i--) {
5705 if (bpf_core_is_flavor_sep(name + i))
5706 return i + 1;
5707 }
5708 return n;
5709 }
5710
5711 void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5712 {
5713 if (!cands)
5714 return;
5715
5716 free(cands->cands);
5717 free(cands);
5718 }
5719
5720 int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5721 size_t local_essent_len,
5722 const struct btf *targ_btf,
5723 const char *targ_btf_name,
5724 int targ_start_id,
5725 struct bpf_core_cand_list *cands)
5726 {
5727 struct bpf_core_cand *new_cands, *cand;
5728 const struct btf_type *t, *local_t;
5729 const char *targ_name, *local_name;
5730 size_t targ_essent_len;
5731 int n, i;
5732
5733 local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5734 local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5735
5736 n = btf__type_cnt(targ_btf);
5737 for (i = targ_start_id; i < n; i++) {
5738 t = btf__type_by_id(targ_btf, i);
5739 if (!btf_kind_core_compat(t, local_t))
5740 continue;
5741
5742 targ_name = btf__name_by_offset(targ_btf, t->name_off);
5743 if (str_is_empty(targ_name))
5744 continue;
5745
5746 targ_essent_len = bpf_core_essential_name_len(targ_name);
5747 if (targ_essent_len != local_essent_len)
5748 continue;
5749
5750 if (strncmp(local_name, targ_name, local_essent_len) != 0)
5751 continue;
5752
5753 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5754 local_cand->id, btf_kind_str(local_t),
5755 local_name, i, btf_kind_str(t), targ_name,
5756 targ_btf_name);
5757 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5758 sizeof(*cands->cands));
5759 if (!new_cands)
5760 return -ENOMEM;
5761
5762 cand = &new_cands[cands->len];
5763 cand->btf = targ_btf;
5764 cand->id = i;
5765
5766 cands->cands = new_cands;
5767 cands->len++;
5768 }
5769 return 0;
5770 }
5771
5772 static int load_module_btfs(struct bpf_object *obj)
5773 {
5774 struct bpf_btf_info info;
5775 struct module_btf *mod_btf;
5776 struct btf *btf;
5777 char name[64];
5778 __u32 id = 0, len;
5779 int err, fd;
5780
5781 if (obj->btf_modules_loaded)
5782 return 0;
5783
5784 if (obj->gen_loader)
5785 return 0;
5786
5787 /* don't do this again, even if we find no module BTFs */
5788 obj->btf_modules_loaded = true;
5789
5790 /* kernel too old to support module BTFs */
5791 if (!kernel_supports(obj, FEAT_MODULE_BTF))
5792 return 0;
5793
5794 while (true) {
5795 err = bpf_btf_get_next_id(id, &id);
5796 if (err && errno == ENOENT)
5797 return 0;
5798 if (err) {
5799 err = -errno;
5800 pr_warn("failed to iterate BTF objects: %d\n", err);
5801 return err;
5802 }
5803
5804 fd = bpf_btf_get_fd_by_id(id);
5805 if (fd < 0) {
5806 if (errno == ENOENT)
5807 continue; /* expected race: BTF was unloaded */
5808 err = -errno;
5809 pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
5810 return err;
5811 }
5812
5813 len = sizeof(info);
5814 memset(&info, 0, sizeof(info));
5815 info.name = ptr_to_u64(name);
5816 info.name_len = sizeof(name);
5817
5818 err = bpf_obj_get_info_by_fd(fd, &info, &len);
5819 if (err) {
5820 err = -errno;
5821 pr_warn("failed to get BTF object #%d info: %d\n", id, err);
5822 goto err_out;
5823 }
5824
5825 /* ignore non-module BTFs */
5826 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5827 close(fd);
5828 continue;
5829 }
5830
5831 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5832 err = libbpf_get_error(btf);
5833 if (err) {
5834 pr_warn("failed to load module [%s]'s BTF object #%d: %d\n",
5835 name, id, err);
5836 goto err_out;
5837 }
5838
5839 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5840 sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5841 if (err)
5842 goto err_out;
5843
5844 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
5845
5846 mod_btf->btf = btf;
5847 mod_btf->id = id;
5848 mod_btf->fd = fd;
5849 mod_btf->name = strdup(name);
5850 if (!mod_btf->name) {
5851 err = -ENOMEM;
5852 goto err_out;
5853 }
5854 continue;
5855
5856 err_out:
5857 close(fd);
5858 return err;
5859 }
5860
5861 return 0;
5862 }
5863
5864 static struct bpf_core_cand_list *
5865 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5866 {
5867 struct bpf_core_cand local_cand = {};
5868 struct bpf_core_cand_list *cands;
5869 const struct btf *main_btf;
5870 const struct btf_type *local_t;
5871 const char *local_name;
5872 size_t local_essent_len;
5873 int err, i;
5874
5875 local_cand.btf = local_btf;
5876 local_cand.id = local_type_id;
5877 local_t = btf__type_by_id(local_btf, local_type_id);
5878 if (!local_t)
5879 return ERR_PTR(-EINVAL);
5880
5881 local_name = btf__name_by_offset(local_btf, local_t->name_off);
5882 if (str_is_empty(local_name))
5883 return ERR_PTR(-EINVAL);
5884 local_essent_len = bpf_core_essential_name_len(local_name);
5885
5886 cands = calloc(1, sizeof(*cands));
5887 if (!cands)
5888 return ERR_PTR(-ENOMEM);
5889
5890 /* Attempt to find target candidates in vmlinux BTF first */
5891 main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5892 err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5893 if (err)
5894 goto err_out;
5895
5896 /* if vmlinux BTF has any candidate, don't got for module BTFs */
5897 if (cands->len)
5898 return cands;
5899
5900 /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5901 if (obj->btf_vmlinux_override)
5902 return cands;
5903
5904 /* now look through module BTFs, trying to still find candidates */
5905 err = load_module_btfs(obj);
5906 if (err)
5907 goto err_out;
5908
5909 for (i = 0; i < obj->btf_module_cnt; i++) {
5910 err = bpf_core_add_cands(&local_cand, local_essent_len,
5911 obj->btf_modules[i].btf,
5912 obj->btf_modules[i].name,
5913 btf__type_cnt(obj->btf_vmlinux),
5914 cands);
5915 if (err)
5916 goto err_out;
5917 }
5918
5919 return cands;
5920 err_out:
5921 bpf_core_free_cands(cands);
5922 return ERR_PTR(err);
5923 }
5924
5925 /* Check local and target types for compatibility. This check is used for
5926 * type-based CO-RE relocations and follow slightly different rules than
5927 * field-based relocations. This function assumes that root types were already
5928 * checked for name match. Beyond that initial root-level name check, names
5929 * are completely ignored. Compatibility rules are as follows:
5930 * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5931 * kind should match for local and target types (i.e., STRUCT is not
5932 * compatible with UNION);
5933 * - for ENUMs, the size is ignored;
5934 * - for INT, size and signedness are ignored;
5935 * - for ARRAY, dimensionality is ignored, element types are checked for
5936 * compatibility recursively;
5937 * - CONST/VOLATILE/RESTRICT modifiers are ignored;
5938 * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5939 * - FUNC_PROTOs are compatible if they have compatible signature: same
5940 * number of input args and compatible return and argument types.
5941 * These rules are not set in stone and probably will be adjusted as we get
5942 * more experience with using BPF CO-RE relocations.
5943 */
5944 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5945 const struct btf *targ_btf, __u32 targ_id)
5946 {
5947 return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32);
5948 }
5949
5950 int bpf_core_types_match(const struct btf *local_btf, __u32 local_id,
5951 const struct btf *targ_btf, __u32 targ_id)
5952 {
5953 return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32);
5954 }
5955
5956 static size_t bpf_core_hash_fn(const long key, void *ctx)
5957 {
5958 return key;
5959 }
5960
5961 static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx)
5962 {
5963 return k1 == k2;
5964 }
5965
5966 static int record_relo_core(struct bpf_program *prog,
5967 const struct bpf_core_relo *core_relo, int insn_idx)
5968 {
5969 struct reloc_desc *relos, *relo;
5970
5971 relos = libbpf_reallocarray(prog->reloc_desc,
5972 prog->nr_reloc + 1, sizeof(*relos));
5973 if (!relos)
5974 return -ENOMEM;
5975 relo = &relos[prog->nr_reloc];
5976 relo->type = RELO_CORE;
5977 relo->insn_idx = insn_idx;
5978 relo->core_relo = core_relo;
5979 prog->reloc_desc = relos;
5980 prog->nr_reloc++;
5981 return 0;
5982 }
5983
5984 static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx)
5985 {
5986 struct reloc_desc *relo;
5987 int i;
5988
5989 for (i = 0; i < prog->nr_reloc; i++) {
5990 relo = &prog->reloc_desc[i];
5991 if (relo->type != RELO_CORE || relo->insn_idx != insn_idx)
5992 continue;
5993
5994 return relo->core_relo;
5995 }
5996
5997 return NULL;
5998 }
5999
6000 static int bpf_core_resolve_relo(struct bpf_program *prog,
6001 const struct bpf_core_relo *relo,
6002 int relo_idx,
6003 const struct btf *local_btf,
6004 struct hashmap *cand_cache,
6005 struct bpf_core_relo_res *targ_res)
6006 {
6007 struct bpf_core_spec specs_scratch[3] = {};
6008 struct bpf_core_cand_list *cands = NULL;
6009 const char *prog_name = prog->name;
6010 const struct btf_type *local_type;
6011 const char *local_name;
6012 __u32 local_id = relo->type_id;
6013 int err;
6014
6015 local_type = btf__type_by_id(local_btf, local_id);
6016 if (!local_type)
6017 return -EINVAL;
6018
6019 local_name = btf__name_by_offset(local_btf, local_type->name_off);
6020 if (!local_name)
6021 return -EINVAL;
6022
6023 if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
6024 !hashmap__find(cand_cache, local_id, &cands)) {
6025 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
6026 if (IS_ERR(cands)) {
6027 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
6028 prog_name, relo_idx, local_id, btf_kind_str(local_type),
6029 local_name, PTR_ERR(cands));
6030 return PTR_ERR(cands);
6031 }
6032 err = hashmap__set(cand_cache, local_id, cands, NULL, NULL);
6033 if (err) {
6034 bpf_core_free_cands(cands);
6035 return err;
6036 }
6037 }
6038
6039 return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch,
6040 targ_res);
6041 }
6042
6043 static int
6044 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
6045 {
6046 const struct btf_ext_info_sec *sec;
6047 struct bpf_core_relo_res targ_res;
6048 const struct bpf_core_relo *rec;
6049 const struct btf_ext_info *seg;
6050 struct hashmap_entry *entry;
6051 struct hashmap *cand_cache = NULL;
6052 struct bpf_program *prog;
6053 struct bpf_insn *insn;
6054 const char *sec_name;
6055 int i, err = 0, insn_idx, sec_idx, sec_num;
6056
6057 if (obj->btf_ext->core_relo_info.len == 0)
6058 return 0;
6059
6060 if (targ_btf_path) {
6061 #ifdef HAVE_LIBELF
6062 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
6063 #endif
6064 err = libbpf_get_error(obj->btf_vmlinux_override);
6065 if (err) {
6066 pr_warn("failed to parse target BTF: %d\n", err);
6067 return err;
6068 }
6069 }
6070
6071 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
6072 if (IS_ERR(cand_cache)) {
6073 err = PTR_ERR(cand_cache);
6074 goto out;
6075 }
6076
6077 seg = &obj->btf_ext->core_relo_info;
6078 sec_num = 0;
6079 for_each_btf_ext_sec(seg, sec) {
6080 sec_idx = seg->sec_idxs[sec_num];
6081 sec_num++;
6082
6083 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
6084 if (str_is_empty(sec_name)) {
6085 err = -EINVAL;
6086 goto out;
6087 }
6088
6089 pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info);
6090
6091 for_each_btf_ext_rec(seg, sec, i, rec) {
6092 if (rec->insn_off % BPF_INSN_SZ)
6093 return -EINVAL;
6094 insn_idx = rec->insn_off / BPF_INSN_SZ;
6095 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
6096 if (!prog) {
6097 /* When __weak subprog is "overridden" by another instance
6098 * of the subprog from a different object file, linker still
6099 * appends all the .BTF.ext info that used to belong to that
6100 * eliminated subprogram.
6101 * This is similar to what x86-64 linker does for relocations.
6102 * So just ignore such relocations just like we ignore
6103 * subprog instructions when discovering subprograms.
6104 */
6105 pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n",
6106 sec_name, i, insn_idx);
6107 continue;
6108 }
6109 /* no need to apply CO-RE relocation if the program is
6110 * not going to be loaded
6111 */
6112 if (!prog->autoload)
6113 continue;
6114
6115 /* adjust insn_idx from section frame of reference to the local
6116 * program's frame of reference; (sub-)program code is not yet
6117 * relocated, so it's enough to just subtract in-section offset
6118 */
6119 insn_idx = insn_idx - prog->sec_insn_off;
6120 if (insn_idx >= prog->insns_cnt)
6121 return -EINVAL;
6122 insn = &prog->insns[insn_idx];
6123
6124 err = record_relo_core(prog, rec, insn_idx);
6125 if (err) {
6126 pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n",
6127 prog->name, i, err);
6128 goto out;
6129 }
6130
6131 if (prog->obj->gen_loader)
6132 continue;
6133
6134 err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res);
6135 if (err) {
6136 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
6137 prog->name, i, err);
6138 goto out;
6139 }
6140
6141 err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res);
6142 if (err) {
6143 pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n",
6144 prog->name, i, insn_idx, err);
6145 goto out;
6146 }
6147 }
6148 }
6149
6150 out:
6151 /* obj->btf_vmlinux and module BTFs are freed after object load */
6152 btf__free(obj->btf_vmlinux_override);
6153 obj->btf_vmlinux_override = NULL;
6154
6155 if (!IS_ERR_OR_NULL(cand_cache)) {
6156 hashmap__for_each_entry(cand_cache, entry, i) {
6157 bpf_core_free_cands(entry->pvalue);
6158 }
6159 hashmap__free(cand_cache);
6160 }
6161 return err;
6162 }
6163
6164 /* base map load ldimm64 special constant, used also for log fixup logic */
6165 #define MAP_LDIMM64_POISON_BASE 2001000000
6166 #define MAP_LDIMM64_POISON_PFX "200100"
6167
6168 static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx,
6169 int insn_idx, struct bpf_insn *insn,
6170 int map_idx, const struct bpf_map *map)
6171 {
6172 int i;
6173
6174 pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n",
6175 prog->name, relo_idx, insn_idx, map_idx, map->name);
6176
6177 /* we turn single ldimm64 into two identical invalid calls */
6178 for (i = 0; i < 2; i++) {
6179 insn->code = BPF_JMP | BPF_CALL;
6180 insn->dst_reg = 0;
6181 insn->src_reg = 0;
6182 insn->off = 0;
6183 /* if this instruction is reachable (not a dead code),
6184 * verifier will complain with something like:
6185 * invalid func unknown#2001000123
6186 * where lower 123 is map index into obj->maps[] array
6187 */
6188 insn->imm = MAP_LDIMM64_POISON_BASE + map_idx;
6189
6190 insn++;
6191 }
6192 }
6193
6194 /* Relocate data references within program code:
6195 * - map references;
6196 * - global variable references;
6197 * - extern references.
6198 */
6199 static int
6200 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
6201 {
6202 int i;
6203
6204 for (i = 0; i < prog->nr_reloc; i++) {
6205 struct reloc_desc *relo = &prog->reloc_desc[i];
6206 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6207 const struct bpf_map *map;
6208 struct extern_desc *ext;
6209
6210 switch (relo->type) {
6211 case RELO_LD64:
6212 map = &obj->maps[relo->map_idx];
6213 if (obj->gen_loader) {
6214 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
6215 insn[0].imm = relo->map_idx;
6216 } else if (map->autocreate) {
6217 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
6218 insn[0].imm = map->fd;
6219 } else {
6220 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6221 relo->map_idx, map);
6222 }
6223 break;
6224 case RELO_DATA:
6225 map = &obj->maps[relo->map_idx];
6226 insn[1].imm = insn[0].imm + relo->sym_off;
6227 if (obj->gen_loader) {
6228 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6229 insn[0].imm = relo->map_idx;
6230 } else if (map->autocreate) {
6231 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6232 insn[0].imm = map->fd;
6233 } else {
6234 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6235 relo->map_idx, map);
6236 }
6237 break;
6238 case RELO_EXTERN_VAR:
6239 ext = &obj->externs[relo->sym_off];
6240 if (ext->type == EXT_KCFG) {
6241 if (obj->gen_loader) {
6242 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6243 insn[0].imm = obj->kconfig_map_idx;
6244 } else {
6245 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6246 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
6247 }
6248 insn[1].imm = ext->kcfg.data_off;
6249 } else /* EXT_KSYM */ {
6250 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
6251 insn[0].src_reg = BPF_PSEUDO_BTF_ID;
6252 insn[0].imm = ext->ksym.kernel_btf_id;
6253 insn[1].imm = ext->ksym.kernel_btf_obj_fd;
6254 } else { /* typeless ksyms or unresolved typed ksyms */
6255 insn[0].imm = (__u32)ext->ksym.addr;
6256 insn[1].imm = ext->ksym.addr >> 32;
6257 }
6258 }
6259 break;
6260 case RELO_EXTERN_FUNC:
6261 ext = &obj->externs[relo->sym_off];
6262 insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
6263 if (ext->is_set) {
6264 insn[0].imm = ext->ksym.kernel_btf_id;
6265 insn[0].off = ext->ksym.btf_fd_idx;
6266 } else { /* unresolved weak kfunc */
6267 insn[0].imm = 0;
6268 insn[0].off = 0;
6269 }
6270 break;
6271 case RELO_SUBPROG_ADDR:
6272 if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
6273 pr_warn("prog '%s': relo #%d: bad insn\n",
6274 prog->name, i);
6275 return -EINVAL;
6276 }
6277 /* handled already */
6278 break;
6279 case RELO_CALL:
6280 /* handled already */
6281 break;
6282 case RELO_CORE:
6283 /* will be handled by bpf_program_record_relos() */
6284 break;
6285 default:
6286 pr_warn("prog '%s': relo #%d: bad relo type %d\n",
6287 prog->name, i, relo->type);
6288 return -EINVAL;
6289 }
6290 }
6291
6292 return 0;
6293 }
6294
6295 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
6296 const struct bpf_program *prog,
6297 const struct btf_ext_info *ext_info,
6298 void **prog_info, __u32 *prog_rec_cnt,
6299 __u32 *prog_rec_sz)
6300 {
6301 void *copy_start = NULL, *copy_end = NULL;
6302 void *rec, *rec_end, *new_prog_info;
6303 const struct btf_ext_info_sec *sec;
6304 size_t old_sz, new_sz;
6305 int i, sec_num, sec_idx, off_adj;
6306
6307 sec_num = 0;
6308 for_each_btf_ext_sec(ext_info, sec) {
6309 sec_idx = ext_info->sec_idxs[sec_num];
6310 sec_num++;
6311 if (prog->sec_idx != sec_idx)
6312 continue;
6313
6314 for_each_btf_ext_rec(ext_info, sec, i, rec) {
6315 __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
6316
6317 if (insn_off < prog->sec_insn_off)
6318 continue;
6319 if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
6320 break;
6321
6322 if (!copy_start)
6323 copy_start = rec;
6324 copy_end = rec + ext_info->rec_size;
6325 }
6326
6327 if (!copy_start)
6328 return -ENOENT;
6329
6330 /* append func/line info of a given (sub-)program to the main
6331 * program func/line info
6332 */
6333 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
6334 new_sz = old_sz + (copy_end - copy_start);
6335 new_prog_info = realloc(*prog_info, new_sz);
6336 if (!new_prog_info)
6337 return -ENOMEM;
6338 *prog_info = new_prog_info;
6339 *prog_rec_cnt = new_sz / ext_info->rec_size;
6340 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
6341
6342 /* Kernel instruction offsets are in units of 8-byte
6343 * instructions, while .BTF.ext instruction offsets generated
6344 * by Clang are in units of bytes. So convert Clang offsets
6345 * into kernel offsets and adjust offset according to program
6346 * relocated position.
6347 */
6348 off_adj = prog->sub_insn_off - prog->sec_insn_off;
6349 rec = new_prog_info + old_sz;
6350 rec_end = new_prog_info + new_sz;
6351 for (; rec < rec_end; rec += ext_info->rec_size) {
6352 __u32 *insn_off = rec;
6353
6354 *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
6355 }
6356 *prog_rec_sz = ext_info->rec_size;
6357 return 0;
6358 }
6359
6360 return -ENOENT;
6361 }
6362
6363 static int
6364 reloc_prog_func_and_line_info(const struct bpf_object *obj,
6365 struct bpf_program *main_prog,
6366 const struct bpf_program *prog)
6367 {
6368 int err;
6369
6370 /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
6371 * supprot func/line info
6372 */
6373 if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
6374 return 0;
6375
6376 /* only attempt func info relocation if main program's func_info
6377 * relocation was successful
6378 */
6379 if (main_prog != prog && !main_prog->func_info)
6380 goto line_info;
6381
6382 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
6383 &main_prog->func_info,
6384 &main_prog->func_info_cnt,
6385 &main_prog->func_info_rec_size);
6386 if (err) {
6387 if (err != -ENOENT) {
6388 pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
6389 prog->name, err);
6390 return err;
6391 }
6392 if (main_prog->func_info) {
6393 /*
6394 * Some info has already been found but has problem
6395 * in the last btf_ext reloc. Must have to error out.
6396 */
6397 pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
6398 return err;
6399 }
6400 /* Have problem loading the very first info. Ignore the rest. */
6401 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
6402 prog->name);
6403 }
6404
6405 line_info:
6406 /* don't relocate line info if main program's relocation failed */
6407 if (main_prog != prog && !main_prog->line_info)
6408 return 0;
6409
6410 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
6411 &main_prog->line_info,
6412 &main_prog->line_info_cnt,
6413 &main_prog->line_info_rec_size);
6414 if (err) {
6415 if (err != -ENOENT) {
6416 pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
6417 prog->name, err);
6418 return err;
6419 }
6420 if (main_prog->line_info) {
6421 /*
6422 * Some info has already been found but has problem
6423 * in the last btf_ext reloc. Must have to error out.
6424 */
6425 pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
6426 return err;
6427 }
6428 /* Have problem loading the very first info. Ignore the rest. */
6429 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
6430 prog->name);
6431 }
6432 return 0;
6433 }
6434
6435 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
6436 {
6437 size_t insn_idx = *(const size_t *)key;
6438 const struct reloc_desc *relo = elem;
6439
6440 if (insn_idx == relo->insn_idx)
6441 return 0;
6442 return insn_idx < relo->insn_idx ? -1 : 1;
6443 }
6444
6445 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
6446 {
6447 if (!prog->nr_reloc)
6448 return NULL;
6449 return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
6450 sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
6451 }
6452
6453 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
6454 {
6455 int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
6456 struct reloc_desc *relos;
6457 int i;
6458
6459 if (main_prog == subprog)
6460 return 0;
6461 relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
6462 if (!relos)
6463 return -ENOMEM;
6464 if (subprog->nr_reloc)
6465 memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
6466 sizeof(*relos) * subprog->nr_reloc);
6467
6468 for (i = main_prog->nr_reloc; i < new_cnt; i++)
6469 relos[i].insn_idx += subprog->sub_insn_off;
6470 /* After insn_idx adjustment the 'relos' array is still sorted
6471 * by insn_idx and doesn't break bsearch.
6472 */
6473 main_prog->reloc_desc = relos;
6474 main_prog->nr_reloc = new_cnt;
6475 return 0;
6476 }
6477
6478 static int
6479 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6480 struct bpf_program *prog)
6481 {
6482 size_t sub_insn_idx, insn_idx, new_cnt;
6483 struct bpf_program *subprog;
6484 struct bpf_insn *insns, *insn;
6485 struct reloc_desc *relo;
6486 int err;
6487
6488 err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6489 if (err)
6490 return err;
6491
6492 for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6493 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6494 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
6495 continue;
6496
6497 relo = find_prog_insn_relo(prog, insn_idx);
6498 if (relo && relo->type == RELO_EXTERN_FUNC)
6499 /* kfunc relocations will be handled later
6500 * in bpf_object__relocate_data()
6501 */
6502 continue;
6503 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6504 pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6505 prog->name, insn_idx, relo->type);
6506 return -LIBBPF_ERRNO__RELOC;
6507 }
6508 if (relo) {
6509 /* sub-program instruction index is a combination of
6510 * an offset of a symbol pointed to by relocation and
6511 * call instruction's imm field; for global functions,
6512 * call always has imm = -1, but for static functions
6513 * relocation is against STT_SECTION and insn->imm
6514 * points to a start of a static function
6515 *
6516 * for subprog addr relocation, the relo->sym_off + insn->imm is
6517 * the byte offset in the corresponding section.
6518 */
6519 if (relo->type == RELO_CALL)
6520 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6521 else
6522 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6523 } else if (insn_is_pseudo_func(insn)) {
6524 /*
6525 * RELO_SUBPROG_ADDR relo is always emitted even if both
6526 * functions are in the same section, so it shouldn't reach here.
6527 */
6528 pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6529 prog->name, insn_idx);
6530 return -LIBBPF_ERRNO__RELOC;
6531 } else {
6532 /* if subprogram call is to a static function within
6533 * the same ELF section, there won't be any relocation
6534 * emitted, but it also means there is no additional
6535 * offset necessary, insns->imm is relative to
6536 * instruction's original position within the section
6537 */
6538 sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6539 }
6540
6541 /* we enforce that sub-programs should be in .text section */
6542 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6543 if (!subprog) {
6544 pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6545 prog->name);
6546 return -LIBBPF_ERRNO__RELOC;
6547 }
6548
6549 /* if it's the first call instruction calling into this
6550 * subprogram (meaning this subprog hasn't been processed
6551 * yet) within the context of current main program:
6552 * - append it at the end of main program's instructions blog;
6553 * - process is recursively, while current program is put on hold;
6554 * - if that subprogram calls some other not yet processes
6555 * subprogram, same thing will happen recursively until
6556 * there are no more unprocesses subprograms left to append
6557 * and relocate.
6558 */
6559 if (subprog->sub_insn_off == 0) {
6560 subprog->sub_insn_off = main_prog->insns_cnt;
6561
6562 new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6563 insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6564 if (!insns) {
6565 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6566 return -ENOMEM;
6567 }
6568 main_prog->insns = insns;
6569 main_prog->insns_cnt = new_cnt;
6570
6571 memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6572 subprog->insns_cnt * sizeof(*insns));
6573
6574 pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6575 main_prog->name, subprog->insns_cnt, subprog->name);
6576
6577 /* The subprog insns are now appended. Append its relos too. */
6578 err = append_subprog_relos(main_prog, subprog);
6579 if (err)
6580 return err;
6581 err = bpf_object__reloc_code(obj, main_prog, subprog);
6582 if (err)
6583 return err;
6584 }
6585
6586 /* main_prog->insns memory could have been re-allocated, so
6587 * calculate pointer again
6588 */
6589 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6590 /* calculate correct instruction position within current main
6591 * prog; each main prog can have a different set of
6592 * subprograms appended (potentially in different order as
6593 * well), so position of any subprog can be different for
6594 * different main programs
6595 */
6596 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6597
6598 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6599 prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6600 }
6601
6602 return 0;
6603 }
6604
6605 /*
6606 * Relocate sub-program calls.
6607 *
6608 * Algorithm operates as follows. Each entry-point BPF program (referred to as
6609 * main prog) is processed separately. For each subprog (non-entry functions,
6610 * that can be called from either entry progs or other subprogs) gets their
6611 * sub_insn_off reset to zero. This serves as indicator that this subprogram
6612 * hasn't been yet appended and relocated within current main prog. Once its
6613 * relocated, sub_insn_off will point at the position within current main prog
6614 * where given subprog was appended. This will further be used to relocate all
6615 * the call instructions jumping into this subprog.
6616 *
6617 * We start with main program and process all call instructions. If the call
6618 * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6619 * is zero), subprog instructions are appended at the end of main program's
6620 * instruction array. Then main program is "put on hold" while we recursively
6621 * process newly appended subprogram. If that subprogram calls into another
6622 * subprogram that hasn't been appended, new subprogram is appended again to
6623 * the *main* prog's instructions (subprog's instructions are always left
6624 * untouched, as they need to be in unmodified state for subsequent main progs
6625 * and subprog instructions are always sent only as part of a main prog) and
6626 * the process continues recursively. Once all the subprogs called from a main
6627 * prog or any of its subprogs are appended (and relocated), all their
6628 * positions within finalized instructions array are known, so it's easy to
6629 * rewrite call instructions with correct relative offsets, corresponding to
6630 * desired target subprog.
6631 *
6632 * Its important to realize that some subprogs might not be called from some
6633 * main prog and any of its called/used subprogs. Those will keep their
6634 * subprog->sub_insn_off as zero at all times and won't be appended to current
6635 * main prog and won't be relocated within the context of current main prog.
6636 * They might still be used from other main progs later.
6637 *
6638 * Visually this process can be shown as below. Suppose we have two main
6639 * programs mainA and mainB and BPF object contains three subprogs: subA,
6640 * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6641 * subC both call subB:
6642 *
6643 * +--------+ +-------+
6644 * | v v |
6645 * +--+---+ +--+-+-+ +---+--+
6646 * | subA | | subB | | subC |
6647 * +--+---+ +------+ +---+--+
6648 * ^ ^
6649 * | |
6650 * +---+-------+ +------+----+
6651 * | mainA | | mainB |
6652 * +-----------+ +-----------+
6653 *
6654 * We'll start relocating mainA, will find subA, append it and start
6655 * processing sub A recursively:
6656 *
6657 * +-----------+------+
6658 * | mainA | subA |
6659 * +-----------+------+
6660 *
6661 * At this point we notice that subB is used from subA, so we append it and
6662 * relocate (there are no further subcalls from subB):
6663 *
6664 * +-----------+------+------+
6665 * | mainA | subA | subB |
6666 * +-----------+------+------+
6667 *
6668 * At this point, we relocate subA calls, then go one level up and finish with
6669 * relocatin mainA calls. mainA is done.
6670 *
6671 * For mainB process is similar but results in different order. We start with
6672 * mainB and skip subA and subB, as mainB never calls them (at least
6673 * directly), but we see subC is needed, so we append and start processing it:
6674 *
6675 * +-----------+------+
6676 * | mainB | subC |
6677 * +-----------+------+
6678 * Now we see subC needs subB, so we go back to it, append and relocate it:
6679 *
6680 * +-----------+------+------+
6681 * | mainB | subC | subB |
6682 * +-----------+------+------+
6683 *
6684 * At this point we unwind recursion, relocate calls in subC, then in mainB.
6685 */
6686 static int
6687 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6688 {
6689 struct bpf_program *subprog;
6690 int i, err;
6691
6692 /* mark all subprogs as not relocated (yet) within the context of
6693 * current main program
6694 */
6695 for (i = 0; i < obj->nr_programs; i++) {
6696 subprog = &obj->programs[i];
6697 if (!prog_is_subprog(obj, subprog))
6698 continue;
6699
6700 subprog->sub_insn_off = 0;
6701 }
6702
6703 err = bpf_object__reloc_code(obj, prog, prog);
6704 if (err)
6705 return err;
6706
6707 return 0;
6708 }
6709
6710 static void
6711 bpf_object__free_relocs(struct bpf_object *obj)
6712 {
6713 struct bpf_program *prog;
6714 int i;
6715
6716 /* free up relocation descriptors */
6717 for (i = 0; i < obj->nr_programs; i++) {
6718 prog = &obj->programs[i];
6719 zfree(&prog->reloc_desc);
6720 prog->nr_reloc = 0;
6721 }
6722 }
6723
6724 static int cmp_relocs(const void *_a, const void *_b)
6725 {
6726 const struct reloc_desc *a = _a;
6727 const struct reloc_desc *b = _b;
6728
6729 if (a->insn_idx != b->insn_idx)
6730 return a->insn_idx < b->insn_idx ? -1 : 1;
6731
6732 /* no two relocations should have the same insn_idx, but ... */
6733 if (a->type != b->type)
6734 return a->type < b->type ? -1 : 1;
6735
6736 return 0;
6737 }
6738
6739 static void bpf_object__sort_relos(struct bpf_object *obj)
6740 {
6741 int i;
6742
6743 for (i = 0; i < obj->nr_programs; i++) {
6744 struct bpf_program *p = &obj->programs[i];
6745
6746 if (!p->nr_reloc)
6747 continue;
6748
6749 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6750 }
6751 }
6752
6753 static int
6754 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
6755 {
6756 struct bpf_program *prog;
6757 size_t i, j;
6758 int err;
6759
6760 if (obj->btf_ext) {
6761 err = bpf_object__relocate_core(obj, targ_btf_path);
6762 if (err) {
6763 pr_warn("failed to perform CO-RE relocations: %d\n",
6764 err);
6765 return err;
6766 }
6767 bpf_object__sort_relos(obj);
6768 }
6769
6770 /* Before relocating calls pre-process relocations and mark
6771 * few ld_imm64 instructions that points to subprogs.
6772 * Otherwise bpf_object__reloc_code() later would have to consider
6773 * all ld_imm64 insns as relocation candidates. That would
6774 * reduce relocation speed, since amount of find_prog_insn_relo()
6775 * would increase and most of them will fail to find a relo.
6776 */
6777 for (i = 0; i < obj->nr_programs; i++) {
6778 prog = &obj->programs[i];
6779 for (j = 0; j < prog->nr_reloc; j++) {
6780 struct reloc_desc *relo = &prog->reloc_desc[j];
6781 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6782
6783 /* mark the insn, so it's recognized by insn_is_pseudo_func() */
6784 if (relo->type == RELO_SUBPROG_ADDR)
6785 insn[0].src_reg = BPF_PSEUDO_FUNC;
6786 }
6787 }
6788
6789 /* relocate subprogram calls and append used subprograms to main
6790 * programs; each copy of subprogram code needs to be relocated
6791 * differently for each main program, because its code location might
6792 * have changed.
6793 * Append subprog relos to main programs to allow data relos to be
6794 * processed after text is completely relocated.
6795 */
6796 for (i = 0; i < obj->nr_programs; i++) {
6797 prog = &obj->programs[i];
6798 /* sub-program's sub-calls are relocated within the context of
6799 * its main program only
6800 */
6801 if (prog_is_subprog(obj, prog))
6802 continue;
6803 if (!prog->autoload)
6804 continue;
6805
6806 err = bpf_object__relocate_calls(obj, prog);
6807 if (err) {
6808 pr_warn("prog '%s': failed to relocate calls: %d\n",
6809 prog->name, err);
6810 return err;
6811 }
6812 }
6813 /* Process data relos for main programs */
6814 for (i = 0; i < obj->nr_programs; i++) {
6815 prog = &obj->programs[i];
6816 if (prog_is_subprog(obj, prog))
6817 continue;
6818 if (!prog->autoload)
6819 continue;
6820 err = bpf_object__relocate_data(obj, prog);
6821 if (err) {
6822 pr_warn("prog '%s': failed to relocate data references: %d\n",
6823 prog->name, err);
6824 return err;
6825 }
6826 }
6827
6828 return 0;
6829 }
6830
6831 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
6832 Elf64_Shdr *shdr, Elf_Data *data);
6833
6834 static int bpf_object__collect_map_relos(struct bpf_object *obj,
6835 Elf64_Shdr *shdr, Elf_Data *data)
6836 {
6837 const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
6838 int i, j, nrels, new_sz;
6839 const struct btf_var_secinfo *vi = NULL;
6840 const struct btf_type *sec, *var, *def;
6841 struct bpf_map *map = NULL, *targ_map = NULL;
6842 struct bpf_program *targ_prog = NULL;
6843 bool is_prog_array, is_map_in_map;
6844 const struct btf_member *member;
6845 const char *name, *mname, *type;
6846 unsigned int moff;
6847 Elf64_Sym *sym;
6848 Elf64_Rel *rel;
6849 void *tmp;
6850
6851 if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
6852 return -EINVAL;
6853 sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
6854 if (!sec)
6855 return -EINVAL;
6856
6857 nrels = shdr->sh_size / shdr->sh_entsize;
6858 for (i = 0; i < nrels; i++) {
6859 rel = elf_rel_by_idx(data, i);
6860 if (!rel) {
6861 pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
6862 return -LIBBPF_ERRNO__FORMAT;
6863 }
6864
6865 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
6866 if (!sym) {
6867 pr_warn(".maps relo #%d: symbol %zx not found\n",
6868 i, (size_t)ELF64_R_SYM(rel->r_info));
6869 return -LIBBPF_ERRNO__FORMAT;
6870 }
6871 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
6872
6873 pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
6874 i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
6875 (size_t)rel->r_offset, sym->st_name, name);
6876
6877 for (j = 0; j < obj->nr_maps; j++) {
6878 map = &obj->maps[j];
6879 if (map->sec_idx != obj->efile.btf_maps_shndx)
6880 continue;
6881
6882 vi = btf_var_secinfos(sec) + map->btf_var_idx;
6883 if (vi->offset <= rel->r_offset &&
6884 rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
6885 break;
6886 }
6887 if (j == obj->nr_maps) {
6888 pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
6889 i, name, (size_t)rel->r_offset);
6890 return -EINVAL;
6891 }
6892
6893 is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
6894 is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
6895 type = is_map_in_map ? "map" : "prog";
6896 if (is_map_in_map) {
6897 if (sym->st_shndx != obj->efile.btf_maps_shndx) {
6898 pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
6899 i, name);
6900 return -LIBBPF_ERRNO__RELOC;
6901 }
6902 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
6903 map->def.key_size != sizeof(int)) {
6904 pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
6905 i, map->name, sizeof(int));
6906 return -EINVAL;
6907 }
6908 targ_map = bpf_object__find_map_by_name(obj, name);
6909 if (!targ_map) {
6910 pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
6911 i, name);
6912 return -ESRCH;
6913 }
6914 } else if (is_prog_array) {
6915 targ_prog = bpf_object__find_program_by_name(obj, name);
6916 if (!targ_prog) {
6917 pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
6918 i, name);
6919 return -ESRCH;
6920 }
6921 if (targ_prog->sec_idx != sym->st_shndx ||
6922 targ_prog->sec_insn_off * 8 != sym->st_value ||
6923 prog_is_subprog(obj, targ_prog)) {
6924 pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
6925 i, name);
6926 return -LIBBPF_ERRNO__RELOC;
6927 }
6928 } else {
6929 return -EINVAL;
6930 }
6931
6932 var = btf__type_by_id(obj->btf, vi->type);
6933 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
6934 if (btf_vlen(def) == 0)
6935 return -EINVAL;
6936 member = btf_members(def) + btf_vlen(def) - 1;
6937 mname = btf__name_by_offset(obj->btf, member->name_off);
6938 if (strcmp(mname, "values"))
6939 return -EINVAL;
6940
6941 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
6942 if (rel->r_offset - vi->offset < moff)
6943 return -EINVAL;
6944
6945 moff = rel->r_offset - vi->offset - moff;
6946 /* here we use BPF pointer size, which is always 64 bit, as we
6947 * are parsing ELF that was built for BPF target
6948 */
6949 if (moff % bpf_ptr_sz)
6950 return -EINVAL;
6951 moff /= bpf_ptr_sz;
6952 if (moff >= map->init_slots_sz) {
6953 new_sz = moff + 1;
6954 tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
6955 if (!tmp)
6956 return -ENOMEM;
6957 map->init_slots = tmp;
6958 memset(map->init_slots + map->init_slots_sz, 0,
6959 (new_sz - map->init_slots_sz) * host_ptr_sz);
6960 map->init_slots_sz = new_sz;
6961 }
6962 map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
6963
6964 pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
6965 i, map->name, moff, type, name);
6966 }
6967
6968 return 0;
6969 }
6970
6971 static int bpf_object__collect_relos(struct bpf_object *obj)
6972 {
6973 int i, err;
6974
6975 for (i = 0; i < obj->efile.sec_cnt; i++) {
6976 struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
6977 Elf64_Shdr *shdr;
6978 Elf_Data *data;
6979 int idx;
6980 Elf64_Shdr shdrelf;
6981
6982 if (sec_desc->sec_type != SEC_RELO)
6983 continue;
6984
6985 #if defined HAVE_LIBELF
6986 shdr = sec_desc->shdr;
6987 #elif defined HAVE_ELFIO
6988 shdr = elf_sec_hdr_by_idx(obj, i, &shdrelf);
6989 #endif
6990 data = sec_desc->data;
6991 idx = shdr->sh_info;
6992
6993 if (shdr->sh_type != SHT_REL) {
6994 pr_warn("internal error at %d\n", __LINE__);
6995 return -LIBBPF_ERRNO__INTERNAL;
6996 }
6997
6998 if (idx == obj->efile.st_ops_shndx)
6999 err = bpf_object__collect_st_ops_relos(obj, shdr, data);
7000 else if (idx == obj->efile.btf_maps_shndx)
7001 err = bpf_object__collect_map_relos(obj, shdr, data);
7002 else
7003 err = bpf_object__collect_prog_relos(obj, shdr, data);
7004 if (err)
7005 return err;
7006 }
7007
7008 bpf_object__sort_relos(obj);
7009 return 0;
7010 }
7011
7012 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
7013 {
7014 if (BPF_CLASS(insn->code) == BPF_JMP &&
7015 BPF_OP(insn->code) == BPF_CALL &&
7016 BPF_SRC(insn->code) == BPF_K &&
7017 insn->src_reg == 0 &&
7018 insn->dst_reg == 0) {
7019 *func_id = insn->imm;
7020 return true;
7021 }
7022 return false;
7023 }
7024
7025 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
7026 {
7027 struct bpf_insn *insn = prog->insns;
7028 enum bpf_func_id func_id;
7029 int i;
7030
7031 if (obj->gen_loader)
7032 return 0;
7033
7034 for (i = 0; i < prog->insns_cnt; i++, insn++) {
7035 if (!insn_is_helper_call(insn, &func_id))
7036 continue;
7037
7038 /* on kernels that don't yet support
7039 * bpf_probe_read_{kernel,user}[_str] helpers, fall back
7040 * to bpf_probe_read() which works well for old kernels
7041 */
7042 switch (func_id) {
7043 case BPF_FUNC_probe_read_kernel:
7044 case BPF_FUNC_probe_read_user:
7045 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
7046 insn->imm = BPF_FUNC_probe_read;
7047 break;
7048 case BPF_FUNC_probe_read_kernel_str:
7049 case BPF_FUNC_probe_read_user_str:
7050 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
7051 insn->imm = BPF_FUNC_probe_read_str;
7052 break;
7053 default:
7054 break;
7055 }
7056 }
7057 return 0;
7058 }
7059
7060 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
7061 int *btf_obj_fd, int *btf_type_id);
7062
7063 /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */
7064 static int libbpf_prepare_prog_load(struct bpf_program *prog,
7065 struct bpf_prog_load_opts *opts, long cookie)
7066 {
7067 enum sec_def_flags def = cookie;
7068
7069 /* old kernels might not support specifying expected_attach_type */
7070 if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
7071 opts->expected_attach_type = 0;
7072
7073 if (def & SEC_SLEEPABLE)
7074 opts->prog_flags |= BPF_F_SLEEPABLE;
7075
7076 if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
7077 opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
7078
7079 if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) {
7080 int btf_obj_fd = 0, btf_type_id = 0, err;
7081 const char *attach_name;
7082
7083 attach_name = strchr(prog->sec_name, '/');
7084 if (!attach_name) {
7085 /* if BPF program is annotated with just SEC("fentry")
7086 * (or similar) without declaratively specifying
7087 * target, then it is expected that target will be
7088 * specified with bpf_program__set_attach_target() at
7089 * runtime before BPF object load step. If not, then
7090 * there is nothing to load into the kernel as BPF
7091 * verifier won't be able to validate BPF program
7092 * correctness anyways.
7093 */
7094 pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n",
7095 prog->name);
7096 return -EINVAL;
7097 }
7098 attach_name++; /* skip over / */
7099
7100 err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
7101 if (err)
7102 return err;
7103
7104 /* cache resolved BTF FD and BTF type ID in the prog */
7105 prog->attach_btf_obj_fd = btf_obj_fd;
7106 prog->attach_btf_id = btf_type_id;
7107
7108 /* but by now libbpf common logic is not utilizing
7109 * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
7110 * this callback is called after opts were populated by
7111 * libbpf, so this callback has to update opts explicitly here
7112 */
7113 opts->attach_btf_obj_fd = btf_obj_fd;
7114 opts->attach_btf_id = btf_type_id;
7115 }
7116 return 0;
7117 }
7118
7119 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz);
7120
7121 static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
7122 struct bpf_insn *insns, int insns_cnt,
7123 const char *license, __u32 kern_version, int *prog_fd)
7124 {
7125 LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
7126 const char *prog_name = NULL;
7127 char *cp, errmsg[STRERR_BUFSIZE];
7128 size_t log_buf_size = 0;
7129 char *log_buf = NULL, *tmp;
7130 int btf_fd, ret, err;
7131 bool own_log_buf = true;
7132 __u32 log_level = prog->log_level;
7133
7134 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
7135 /*
7136 * The program type must be set. Most likely we couldn't find a proper
7137 * section definition at load time, and thus we didn't infer the type.
7138 */
7139 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
7140 prog->name, prog->sec_name);
7141 return -EINVAL;
7142 }
7143
7144 if (!insns || !insns_cnt)
7145 return -EINVAL;
7146
7147 load_attr.expected_attach_type = prog->expected_attach_type;
7148 if (kernel_supports(obj, FEAT_PROG_NAME))
7149 prog_name = prog->name;
7150 load_attr.attach_prog_fd = prog->attach_prog_fd;
7151 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
7152 load_attr.attach_btf_id = prog->attach_btf_id;
7153 load_attr.kern_version = kern_version;
7154 load_attr.prog_ifindex = prog->prog_ifindex;
7155
7156 /* specify func_info/line_info only if kernel supports them */
7157 btf_fd = bpf_object__btf_fd(obj);
7158 if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
7159 load_attr.prog_btf_fd = btf_fd;
7160 load_attr.func_info = prog->func_info;
7161 load_attr.func_info_rec_size = prog->func_info_rec_size;
7162 load_attr.func_info_cnt = prog->func_info_cnt;
7163 load_attr.line_info = prog->line_info;
7164 load_attr.line_info_rec_size = prog->line_info_rec_size;
7165 load_attr.line_info_cnt = prog->line_info_cnt;
7166 }
7167 load_attr.log_level = log_level;
7168 load_attr.prog_flags = prog->prog_flags;
7169 load_attr.fd_array = obj->fd_array;
7170
7171 /* adjust load_attr if sec_def provides custom preload callback */
7172 if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) {
7173 err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie);
7174 if (err < 0) {
7175 pr_warn("prog '%s': failed to prepare load attributes: %d\n",
7176 prog->name, err);
7177 return err;
7178 }
7179 insns = prog->insns;
7180 insns_cnt = prog->insns_cnt;
7181 }
7182
7183 if (obj->gen_loader) {
7184 bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
7185 license, insns, insns_cnt, &load_attr,
7186 prog - obj->programs);
7187 *prog_fd = -1;
7188 return 0;
7189 }
7190
7191 retry_load:
7192 /* if log_level is zero, we don't request logs initially even if
7193 * custom log_buf is specified; if the program load fails, then we'll
7194 * bump log_level to 1 and use either custom log_buf or we'll allocate
7195 * our own and retry the load to get details on what failed
7196 */
7197 if (log_level) {
7198 if (prog->log_buf) {
7199 log_buf = prog->log_buf;
7200 log_buf_size = prog->log_size;
7201 own_log_buf = false;
7202 } else if (obj->log_buf) {
7203 log_buf = obj->log_buf;
7204 log_buf_size = obj->log_size;
7205 own_log_buf = false;
7206 } else {
7207 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
7208 tmp = realloc(log_buf, log_buf_size);
7209 if (!tmp) {
7210 ret = -ENOMEM;
7211 goto out;
7212 }
7213 log_buf = tmp;
7214 log_buf[0] = '\0';
7215 own_log_buf = true;
7216 }
7217 }
7218
7219 load_attr.log_buf = log_buf;
7220 load_attr.log_size = log_buf_size;
7221 load_attr.log_level = log_level;
7222
7223 ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
7224 if (ret >= 0) {
7225 if (log_level && own_log_buf) {
7226 pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7227 prog->name, log_buf);
7228 }
7229
7230 if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
7231 struct bpf_map *map;
7232 int i;
7233
7234 for (i = 0; i < obj->nr_maps; i++) {
7235 map = &prog->obj->maps[i];
7236 if (map->libbpf_type != LIBBPF_MAP_RODATA)
7237 continue;
7238
7239 if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) {
7240 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7241 pr_warn("prog '%s': failed to bind map '%s': %s\n",
7242 prog->name, map->real_name, cp);
7243 /* Don't fail hard if can't bind rodata. */
7244 }
7245 }
7246 }
7247
7248 *prog_fd = ret;
7249 ret = 0;
7250 goto out;
7251 }
7252
7253 if (log_level == 0) {
7254 log_level = 1;
7255 goto retry_load;
7256 }
7257 /* On ENOSPC, increase log buffer size and retry, unless custom
7258 * log_buf is specified.
7259 * Be careful to not overflow u32, though. Kernel's log buf size limit
7260 * isn't part of UAPI so it can always be bumped to full 4GB. So don't
7261 * multiply by 2 unless we are sure we'll fit within 32 bits.
7262 * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
7263 */
7264 if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
7265 goto retry_load;
7266
7267 ret = -errno;
7268
7269 /* post-process verifier log to improve error descriptions */
7270 fixup_verifier_log(prog, log_buf, log_buf_size);
7271
7272 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7273 pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp);
7274 pr_perm_msg(ret);
7275
7276 if (own_log_buf && log_buf && log_buf[0] != '\0') {
7277 pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7278 prog->name, log_buf);
7279 }
7280
7281 out:
7282 if (own_log_buf)
7283 free(log_buf);
7284 return ret;
7285 }
7286
7287 static char *find_prev_line(char *buf, char *cur)
7288 {
7289 char *p;
7290
7291 if (cur == buf) /* end of a log buf */
7292 return NULL;
7293
7294 p = cur - 1;
7295 while (p - 1 >= buf && *(p - 1) != '\n')
7296 p--;
7297
7298 return p;
7299 }
7300
7301 static void patch_log(char *buf, size_t buf_sz, size_t log_sz,
7302 char *orig, size_t orig_sz, const char *patch)
7303 {
7304 /* size of the remaining log content to the right from the to-be-replaced part */
7305 size_t rem_sz = (buf + log_sz) - (orig + orig_sz);
7306 size_t patch_sz = strlen(patch);
7307
7308 if (patch_sz != orig_sz) {
7309 /* If patch line(s) are longer than original piece of verifier log,
7310 * shift log contents by (patch_sz - orig_sz) bytes to the right
7311 * starting from after to-be-replaced part of the log.
7312 *
7313 * If patch line(s) are shorter than original piece of verifier log,
7314 * shift log contents by (orig_sz - patch_sz) bytes to the left
7315 * starting from after to-be-replaced part of the log
7316 *
7317 * We need to be careful about not overflowing available
7318 * buf_sz capacity. If that's the case, we'll truncate the end
7319 * of the original log, as necessary.
7320 */
7321 if (patch_sz > orig_sz) {
7322 if (orig + patch_sz >= buf + buf_sz) {
7323 /* patch is big enough to cover remaining space completely */
7324 patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1;
7325 rem_sz = 0;
7326 } else if (patch_sz - orig_sz > buf_sz - log_sz) {
7327 /* patch causes part of remaining log to be truncated */
7328 rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz);
7329 }
7330 }
7331 /* shift remaining log to the right by calculated amount */
7332 memmove(orig + patch_sz, orig + orig_sz, rem_sz);
7333 }
7334
7335 memcpy(orig, patch, patch_sz);
7336 }
7337
7338 static void fixup_log_failed_core_relo(struct bpf_program *prog,
7339 char *buf, size_t buf_sz, size_t log_sz,
7340 char *line1, char *line2, char *line3)
7341 {
7342 /* Expected log for failed and not properly guarded CO-RE relocation:
7343 * line1 -> 123: (85) call unknown#195896080
7344 * line2 -> invalid func unknown#195896080
7345 * line3 -> <anything else or end of buffer>
7346 *
7347 * "123" is the index of the instruction that was poisoned. We extract
7348 * instruction index to find corresponding CO-RE relocation and
7349 * replace this part of the log with more relevant information about
7350 * failed CO-RE relocation.
7351 */
7352 const struct bpf_core_relo *relo;
7353 struct bpf_core_spec spec;
7354 char patch[512], spec_buf[256];
7355 int insn_idx, err, spec_len;
7356
7357 if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1)
7358 return;
7359
7360 relo = find_relo_core(prog, insn_idx);
7361 if (!relo)
7362 return;
7363
7364 err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec);
7365 if (err)
7366 return;
7367
7368 spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec);
7369 snprintf(patch, sizeof(patch),
7370 "%d: <invalid CO-RE relocation>\n"
7371 "failed to resolve CO-RE relocation %s%s\n",
7372 insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : "");
7373
7374 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7375 }
7376
7377 static void fixup_log_missing_map_load(struct bpf_program *prog,
7378 char *buf, size_t buf_sz, size_t log_sz,
7379 char *line1, char *line2, char *line3)
7380 {
7381 /* Expected log for failed and not properly guarded CO-RE relocation:
7382 * line1 -> 123: (85) call unknown#2001000345
7383 * line2 -> invalid func unknown#2001000345
7384 * line3 -> <anything else or end of buffer>
7385 *
7386 * "123" is the index of the instruction that was poisoned.
7387 * "345" in "2001000345" are map index in obj->maps to fetch map name.
7388 */
7389 struct bpf_object *obj = prog->obj;
7390 const struct bpf_map *map;
7391 int insn_idx, map_idx;
7392 char patch[128];
7393
7394 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2)
7395 return;
7396
7397 map_idx -= MAP_LDIMM64_POISON_BASE;
7398 if (map_idx < 0 || map_idx >= obj->nr_maps)
7399 return;
7400 map = &obj->maps[map_idx];
7401
7402 snprintf(patch, sizeof(patch),
7403 "%d: <invalid BPF map reference>\n"
7404 "BPF map '%s' is referenced but wasn't created\n",
7405 insn_idx, map->name);
7406
7407 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7408 }
7409
7410 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz)
7411 {
7412 /* look for familiar error patterns in last N lines of the log */
7413 const size_t max_last_line_cnt = 10;
7414 char *prev_line, *cur_line, *next_line;
7415 size_t log_sz;
7416 int i;
7417
7418 if (!buf)
7419 return;
7420
7421 log_sz = strlen(buf) + 1;
7422 next_line = buf + log_sz - 1;
7423
7424 for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) {
7425 cur_line = find_prev_line(buf, next_line);
7426 if (!cur_line)
7427 return;
7428
7429 /* failed CO-RE relocation case */
7430 if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) {
7431 prev_line = find_prev_line(buf, cur_line);
7432 if (!prev_line)
7433 continue;
7434
7435 fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz,
7436 prev_line, cur_line, next_line);
7437 return;
7438 } else if (str_has_pfx(cur_line, "invalid func unknown#"MAP_LDIMM64_POISON_PFX)) {
7439 prev_line = find_prev_line(buf, cur_line);
7440 if (!prev_line)
7441 continue;
7442
7443 fixup_log_missing_map_load(prog, buf, buf_sz, log_sz,
7444 prev_line, cur_line, next_line);
7445 return;
7446 }
7447 }
7448 }
7449
7450 static int bpf_program_record_relos(struct bpf_program *prog)
7451 {
7452 struct bpf_object *obj = prog->obj;
7453 int i;
7454
7455 for (i = 0; i < prog->nr_reloc; i++) {
7456 struct reloc_desc *relo = &prog->reloc_desc[i];
7457 struct extern_desc *ext = &obj->externs[relo->sym_off];
7458
7459 switch (relo->type) {
7460 case RELO_EXTERN_VAR:
7461 if (ext->type != EXT_KSYM)
7462 continue;
7463 bpf_gen__record_extern(obj->gen_loader, ext->name,
7464 ext->is_weak, !ext->ksym.type_id,
7465 BTF_KIND_VAR, relo->insn_idx);
7466 break;
7467 case RELO_EXTERN_FUNC:
7468 bpf_gen__record_extern(obj->gen_loader, ext->name,
7469 ext->is_weak, false, BTF_KIND_FUNC,
7470 relo->insn_idx);
7471 break;
7472 case RELO_CORE: {
7473 struct bpf_core_relo cr = {
7474 .insn_off = relo->insn_idx * 8,
7475 .type_id = relo->core_relo->type_id,
7476 .access_str_off = relo->core_relo->access_str_off,
7477 .kind = relo->core_relo->kind,
7478 };
7479
7480 bpf_gen__record_relo_core(obj->gen_loader, &cr);
7481 break;
7482 }
7483 default:
7484 continue;
7485 }
7486 }
7487 return 0;
7488 }
7489
7490 static int
7491 bpf_object__load_progs(struct bpf_object *obj, int log_level)
7492 {
7493 struct bpf_program *prog;
7494 size_t i;
7495 int err;
7496
7497 for (i = 0; i < obj->nr_programs; i++) {
7498 prog = &obj->programs[i];
7499 err = bpf_object__sanitize_prog(obj, prog);
7500 if (err)
7501 return err;
7502 }
7503
7504 for (i = 0; i < obj->nr_programs; i++) {
7505 prog = &obj->programs[i];
7506 if (prog_is_subprog(obj, prog))
7507 continue;
7508 if (!prog->autoload) {
7509 pr_debug("prog '%s': skipped loading\n", prog->name);
7510 continue;
7511 }
7512 prog->log_level |= log_level;
7513
7514 if (obj->gen_loader)
7515 bpf_program_record_relos(prog);
7516
7517 err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt,
7518 obj->license, obj->kern_version, &prog->fd);
7519 if (err) {
7520 pr_warn("prog '%s': failed to load: %d\n", prog->name, err);
7521 return err;
7522 }
7523 }
7524
7525 bpf_object__free_relocs(obj);
7526 return 0;
7527 }
7528
7529 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
7530
7531 static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
7532 {
7533 struct bpf_program *prog;
7534 int err;
7535
7536 bpf_object__for_each_program(prog, obj) {
7537 prog->sec_def = find_sec_def(prog->sec_name);
7538 if (!prog->sec_def) {
7539 /* couldn't guess, but user might manually specify */
7540 pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
7541 prog->name, prog->sec_name);
7542 continue;
7543 }
7544
7545 prog->type = prog->sec_def->prog_type;
7546 prog->expected_attach_type = prog->sec_def->expected_attach_type;
7547
7548 /* sec_def can have custom callback which should be called
7549 * after bpf_program is initialized to adjust its properties
7550 */
7551 if (prog->sec_def->prog_setup_fn) {
7552 err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie);
7553 if (err < 0) {
7554 pr_warn("prog '%s': failed to initialize: %d\n",
7555 prog->name, err);
7556 return err;
7557 }
7558 }
7559 }
7560
7561 return 0;
7562 }
7563
7564 static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
7565 const struct bpf_object_open_opts *opts)
7566 {
7567 const char *obj_name, *kconfig, *btf_tmp_path;
7568 struct bpf_object *obj;
7569 char tmp_name[64];
7570 int err;
7571 char *log_buf;
7572 size_t log_size;
7573 __u32 log_level;
7574
7575 #ifdef HAVE_LIBELF
7576 if (elf_version(EV_CURRENT) == EV_NONE) {
7577 pr_warn("failed to init libelf for %s\n",
7578 path ? : "(mem buf)");
7579 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
7580 }
7581 #endif
7582
7583 if (!OPTS_VALID(opts, bpf_object_open_opts))
7584 return ERR_PTR(-EINVAL);
7585
7586 obj_name = OPTS_GET(opts, object_name, NULL);
7587 if (obj_buf) {
7588 if (!obj_name) {
7589 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
7590 (unsigned long)obj_buf,
7591 (unsigned long)obj_buf_sz);
7592 obj_name = tmp_name;
7593 }
7594 path = obj_name;
7595 pr_debug("loading object '%s' from buffer\n", obj_name);
7596 }
7597
7598 log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
7599 log_size = OPTS_GET(opts, kernel_log_size, 0);
7600 log_level = OPTS_GET(opts, kernel_log_level, 0);
7601 if (log_size > UINT_MAX)
7602 return ERR_PTR(-EINVAL);
7603 if (log_size && !log_buf)
7604 return ERR_PTR(-EINVAL);
7605
7606 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
7607 if (IS_ERR(obj))
7608 return obj;
7609
7610 obj->log_buf = log_buf;
7611 obj->log_size = log_size;
7612 obj->log_level = log_level;
7613
7614 btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
7615 if (btf_tmp_path) {
7616 if (strlen(btf_tmp_path) >= PATH_MAX) {
7617 err = -ENAMETOOLONG;
7618 goto out;
7619 }
7620 obj->btf_custom_path = strdup(btf_tmp_path);
7621 if (!obj->btf_custom_path) {
7622 err = -ENOMEM;
7623 goto out;
7624 }
7625 }
7626
7627 kconfig = OPTS_GET(opts, kconfig, NULL);
7628 if (kconfig) {
7629 obj->kconfig = strdup(kconfig);
7630 if (!obj->kconfig) {
7631 err = -ENOMEM;
7632 goto out;
7633 }
7634 }
7635
7636 err = bpf_object__elf_init(obj);
7637 err = err ? : bpf_object__check_endianness(obj);
7638 err = err ? : bpf_object__elf_collect(obj);
7639 err = err ? : bpf_object__collect_externs(obj);
7640 err = err ? : bpf_object_fixup_btf(obj);
7641 err = err ? : bpf_object__init_maps(obj, opts);
7642 err = err ? : bpf_object_init_progs(obj, opts);
7643 err = err ? : bpf_object__collect_relos(obj);
7644 if (err)
7645 goto out;
7646
7647 bpf_object__elf_finish(obj);
7648
7649 return obj;
7650 out:
7651 bpf_object__close(obj);
7652 return ERR_PTR(err);
7653 }
7654
7655 struct bpf_object *
7656 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
7657 {
7658 if (!path)
7659 return libbpf_err_ptr(-EINVAL);
7660
7661 pr_debug("loading %s\n", path);
7662
7663 return libbpf_ptr(bpf_object_open(path, NULL, 0, opts));
7664 }
7665
7666 struct bpf_object *bpf_object__open(const char *path)
7667 {
7668 return bpf_object__open_file(path, NULL);
7669 }
7670
7671 struct bpf_object *
7672 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
7673 const struct bpf_object_open_opts *opts)
7674 {
7675 if (!obj_buf || obj_buf_sz == 0)
7676 return libbpf_err_ptr(-EINVAL);
7677
7678 return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts));
7679 }
7680
7681 static int bpf_object_unload(struct bpf_object *obj)
7682 {
7683 size_t i;
7684
7685 if (!obj)
7686 return libbpf_err(-EINVAL);
7687
7688 for (i = 0; i < obj->nr_maps; i++) {
7689 zclose(obj->maps[i].fd);
7690 if (obj->maps[i].st_ops)
7691 zfree(&obj->maps[i].st_ops->kern_vdata);
7692 }
7693
7694 for (i = 0; i < obj->nr_programs; i++)
7695 bpf_program__unload(&obj->programs[i]);
7696
7697 return 0;
7698 }
7699
7700 static int bpf_object__sanitize_maps(struct bpf_object *obj)
7701 {
7702 struct bpf_map *m;
7703
7704 bpf_object__for_each_map(m, obj) {
7705 if (!bpf_map__is_internal(m))
7706 continue;
7707 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
7708 m->def.map_flags ^= BPF_F_MMAPABLE;
7709 }
7710
7711 return 0;
7712 }
7713
7714 int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx)
7715 {
7716 char sym_type, sym_name[500];
7717 unsigned long long sym_addr;
7718 int ret, err = 0;
7719 FILE *f;
7720
7721 f = fopen("/proc/kallsyms", "r");
7722 if (!f) {
7723 err = -errno;
7724 pr_warn("failed to open /proc/kallsyms: %d\n", err);
7725 return err;
7726 }
7727
7728 while (true) {
7729 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
7730 &sym_addr, &sym_type, sym_name);
7731 if (ret == EOF && feof(f))
7732 break;
7733 if (ret != 3) {
7734 pr_warn("failed to read kallsyms entry: %d\n", ret);
7735 err = -EINVAL;
7736 break;
7737 }
7738
7739 err = cb(sym_addr, sym_type, sym_name, ctx);
7740 if (err)
7741 break;
7742 }
7743
7744 fclose(f);
7745 return err;
7746 }
7747
7748 static int kallsyms_cb(unsigned long long sym_addr, char sym_type,
7749 const char *sym_name, void *ctx)
7750 {
7751 struct bpf_object *obj = ctx;
7752 const struct btf_type *t;
7753 struct extern_desc *ext;
7754
7755 ext = find_extern_by_name(obj, sym_name);
7756 if (!ext || ext->type != EXT_KSYM)
7757 return 0;
7758
7759 t = btf__type_by_id(obj->btf, ext->btf_id);
7760 if (!btf_is_var(t))
7761 return 0;
7762
7763 if (ext->is_set && ext->ksym.addr != sym_addr) {
7764 pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n",
7765 sym_name, ext->ksym.addr, sym_addr);
7766 return -EINVAL;
7767 }
7768 if (!ext->is_set) {
7769 ext->is_set = true;
7770 ext->ksym.addr = sym_addr;
7771 pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr);
7772 }
7773 return 0;
7774 }
7775
7776 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
7777 {
7778 return libbpf_kallsyms_parse(kallsyms_cb, obj);
7779 }
7780
7781 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
7782 __u16 kind, struct btf **res_btf,
7783 struct module_btf **res_mod_btf)
7784 {
7785 struct module_btf *mod_btf;
7786 struct btf *btf;
7787 int i, id, err;
7788
7789 btf = obj->btf_vmlinux;
7790 mod_btf = NULL;
7791 id = btf__find_by_name_kind(btf, ksym_name, kind);
7792
7793 if (id == -ENOENT) {
7794 err = load_module_btfs(obj);
7795 if (err)
7796 return err;
7797
7798 for (i = 0; i < obj->btf_module_cnt; i++) {
7799 /* we assume module_btf's BTF FD is always >0 */
7800 mod_btf = &obj->btf_modules[i];
7801 btf = mod_btf->btf;
7802 id = btf__find_by_name_kind_own(btf, ksym_name, kind);
7803 if (id != -ENOENT)
7804 break;
7805 }
7806 }
7807 if (id <= 0)
7808 return -ESRCH;
7809
7810 *res_btf = btf;
7811 *res_mod_btf = mod_btf;
7812 return id;
7813 }
7814
7815 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
7816 struct extern_desc *ext)
7817 {
7818 const struct btf_type *targ_var, *targ_type;
7819 __u32 targ_type_id, local_type_id;
7820 struct module_btf *mod_btf = NULL;
7821 const char *targ_var_name;
7822 struct btf *btf = NULL;
7823 int id, err;
7824
7825 id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
7826 if (id < 0) {
7827 if (id == -ESRCH && ext->is_weak)
7828 return 0;
7829 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
7830 ext->name);
7831 return id;
7832 }
7833
7834 /* find local type_id */
7835 local_type_id = ext->ksym.type_id;
7836
7837 /* find target type_id */
7838 targ_var = btf__type_by_id(btf, id);
7839 targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
7840 targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
7841
7842 err = bpf_core_types_are_compat(obj->btf, local_type_id,
7843 btf, targ_type_id);
7844 if (err <= 0) {
7845 const struct btf_type *local_type;
7846 const char *targ_name, *local_name;
7847
7848 local_type = btf__type_by_id(obj->btf, local_type_id);
7849 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
7850 targ_name = btf__name_by_offset(btf, targ_type->name_off);
7851
7852 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
7853 ext->name, local_type_id,
7854 btf_kind_str(local_type), local_name, targ_type_id,
7855 btf_kind_str(targ_type), targ_name);
7856 return -EINVAL;
7857 }
7858
7859 ext->is_set = true;
7860 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7861 ext->ksym.kernel_btf_id = id;
7862 pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
7863 ext->name, id, btf_kind_str(targ_var), targ_var_name);
7864
7865 return 0;
7866 }
7867
7868 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
7869 struct extern_desc *ext)
7870 {
7871 int local_func_proto_id, kfunc_proto_id, kfunc_id;
7872 struct module_btf *mod_btf = NULL;
7873 const struct btf_type *kern_func;
7874 struct btf *kern_btf = NULL;
7875 int ret;
7876
7877 local_func_proto_id = ext->ksym.type_id;
7878
7879 kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf);
7880 if (kfunc_id < 0) {
7881 if (kfunc_id == -ESRCH && ext->is_weak)
7882 return 0;
7883 pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
7884 ext->name);
7885 return kfunc_id;
7886 }
7887
7888 kern_func = btf__type_by_id(kern_btf, kfunc_id);
7889 kfunc_proto_id = kern_func->type;
7890
7891 ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
7892 kern_btf, kfunc_proto_id);
7893 if (ret <= 0) {
7894 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n",
7895 ext->name, local_func_proto_id, kfunc_proto_id);
7896 return -EINVAL;
7897 }
7898
7899 /* set index for module BTF fd in fd_array, if unset */
7900 if (mod_btf && !mod_btf->fd_array_idx) {
7901 /* insn->off is s16 */
7902 if (obj->fd_array_cnt == INT16_MAX) {
7903 pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
7904 ext->name, mod_btf->fd_array_idx);
7905 return -E2BIG;
7906 }
7907 /* Cannot use index 0 for module BTF fd */
7908 if (!obj->fd_array_cnt)
7909 obj->fd_array_cnt = 1;
7910
7911 ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
7912 obj->fd_array_cnt + 1);
7913 if (ret)
7914 return ret;
7915 mod_btf->fd_array_idx = obj->fd_array_cnt;
7916 /* we assume module BTF FD is always >0 */
7917 obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
7918 }
7919
7920 ext->is_set = true;
7921 ext->ksym.kernel_btf_id = kfunc_id;
7922 ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
7923 pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n",
7924 ext->name, kfunc_id);
7925
7926 return 0;
7927 }
7928
7929 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
7930 {
7931 const struct btf_type *t;
7932 struct extern_desc *ext;
7933 int i, err;
7934
7935 for (i = 0; i < obj->nr_extern; i++) {
7936 ext = &obj->externs[i];
7937 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
7938 continue;
7939
7940 if (obj->gen_loader) {
7941 ext->is_set = true;
7942 ext->ksym.kernel_btf_obj_fd = 0;
7943 ext->ksym.kernel_btf_id = 0;
7944 continue;
7945 }
7946 t = btf__type_by_id(obj->btf, ext->btf_id);
7947 if (btf_is_var(t))
7948 err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
7949 else
7950 err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
7951 if (err)
7952 return err;
7953 }
7954 return 0;
7955 }
7956
7957 static int bpf_object__resolve_externs(struct bpf_object *obj,
7958 const char *extra_kconfig)
7959 {
7960 bool need_config = false, need_kallsyms = false;
7961 bool need_vmlinux_btf = false;
7962 struct extern_desc *ext;
7963 void *kcfg_data = NULL;
7964 int err, i;
7965
7966 if (obj->nr_extern == 0)
7967 return 0;
7968
7969 if (obj->kconfig_map_idx >= 0)
7970 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
7971
7972 for (i = 0; i < obj->nr_extern; i++) {
7973 ext = &obj->externs[i];
7974
7975 if (ext->type == EXT_KSYM) {
7976 if (ext->ksym.type_id)
7977 need_vmlinux_btf = true;
7978 else
7979 need_kallsyms = true;
7980 continue;
7981 } else if (ext->type == EXT_KCFG) {
7982 void *ext_ptr = kcfg_data + ext->kcfg.data_off;
7983 __u64 value = 0;
7984
7985 /* Kconfig externs need actual /proc/config.gz */
7986 if (str_has_pfx(ext->name, "CONFIG_")) {
7987 need_config = true;
7988 continue;
7989 }
7990
7991 /* Virtual kcfg externs are customly handled by libbpf */
7992 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
7993 value = get_kernel_version();
7994 if (!value) {
7995 pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name);
7996 return -EINVAL;
7997 }
7998 } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) {
7999 value = kernel_supports(obj, FEAT_BPF_COOKIE);
8000 } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) {
8001 value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER);
8002 } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) {
8003 /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed
8004 * __kconfig externs, where LINUX_ ones are virtual and filled out
8005 * customly by libbpf (their values don't come from Kconfig).
8006 * If LINUX_xxx variable is not recognized by libbpf, but is marked
8007 * __weak, it defaults to zero value, just like for CONFIG_xxx
8008 * externs.
8009 */
8010 pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name);
8011 return -EINVAL;
8012 }
8013
8014 err = set_kcfg_value_num(ext, ext_ptr, value);
8015 if (err)
8016 return err;
8017 pr_debug("extern (kcfg) '%s': set to 0x%llx\n",
8018 ext->name, (long long)value);
8019 } else {
8020 pr_warn("extern '%s': unrecognized extern kind\n", ext->name);
8021 return -EINVAL;
8022 }
8023 }
8024 if (need_config && extra_kconfig) {
8025 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
8026 if (err)
8027 return -EINVAL;
8028 need_config = false;
8029 for (i = 0; i < obj->nr_extern; i++) {
8030 ext = &obj->externs[i];
8031 if (ext->type == EXT_KCFG && !ext->is_set) {
8032 need_config = true;
8033 break;
8034 }
8035 }
8036 }
8037 if (need_config) {
8038 err = bpf_object__read_kconfig_file(obj, kcfg_data);
8039 if (err)
8040 return -EINVAL;
8041 }
8042 if (need_kallsyms) {
8043 err = bpf_object__read_kallsyms_file(obj);
8044 if (err)
8045 return -EINVAL;
8046 }
8047 if (need_vmlinux_btf) {
8048 err = bpf_object__resolve_ksyms_btf_id(obj);
8049 if (err)
8050 return -EINVAL;
8051 }
8052 for (i = 0; i < obj->nr_extern; i++) {
8053 ext = &obj->externs[i];
8054
8055 if (!ext->is_set && !ext->is_weak) {
8056 pr_warn("extern '%s' (strong): not resolved\n", ext->name);
8057 return -ESRCH;
8058 } else if (!ext->is_set) {
8059 pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n",
8060 ext->name);
8061 }
8062 }
8063
8064 return 0;
8065 }
8066
8067 static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
8068 {
8069 int err, i;
8070
8071 if (!obj)
8072 return libbpf_err(-EINVAL);
8073
8074 if (obj->loaded) {
8075 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
8076 return libbpf_err(-EINVAL);
8077 }
8078
8079 if (obj->gen_loader)
8080 bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
8081
8082 err = bpf_object__probe_loading(obj);
8083 err = err ? : bpf_object__load_vmlinux_btf(obj, false);
8084 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
8085 err = err ? : bpf_object__sanitize_and_load_btf(obj);
8086 err = err ? : bpf_object__sanitize_maps(obj);
8087 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
8088 err = err ? : bpf_object__create_maps(obj);
8089 err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
8090 err = err ? : bpf_object__load_progs(obj, extra_log_level);
8091 err = err ? : bpf_object_init_prog_arrays(obj);
8092
8093 if (obj->gen_loader) {
8094 /* reset FDs */
8095 if (obj->btf)
8096 btf__set_fd(obj->btf, -1);
8097 for (i = 0; i < obj->nr_maps; i++)
8098 obj->maps[i].fd = -1;
8099 if (!err)
8100 err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
8101 }
8102
8103 /* clean up fd_array */
8104 zfree(&obj->fd_array);
8105
8106 /* clean up module BTFs */
8107 for (i = 0; i < obj->btf_module_cnt; i++) {
8108 close(obj->btf_modules[i].fd);
8109 btf__free(obj->btf_modules[i].btf);
8110 free(obj->btf_modules[i].name);
8111 }
8112 free(obj->btf_modules);
8113
8114 /* clean up vmlinux BTF */
8115 btf__free(obj->btf_vmlinux);
8116 obj->btf_vmlinux = NULL;
8117
8118 obj->loaded = true; /* doesn't matter if successfully or not */
8119
8120 if (err)
8121 goto out;
8122
8123 return 0;
8124 out:
8125 /* unpin any maps that were auto-pinned during load */
8126 for (i = 0; i < obj->nr_maps; i++)
8127 if (obj->maps[i].pinned && !obj->maps[i].reused)
8128 bpf_map__unpin(&obj->maps[i], NULL);
8129
8130 bpf_object_unload(obj);
8131 pr_warn("failed to load object '%s'\n", obj->path);
8132 return libbpf_err(err);
8133 }
8134
8135 int bpf_object__load(struct bpf_object *obj)
8136 {
8137 return bpf_object_load(obj, 0, NULL);
8138 }
8139
8140 static int make_parent_dir(const char *path)
8141 {
8142 char *cp, errmsg[STRERR_BUFSIZE];
8143 char *dname, *dir;
8144 int err = 0;
8145
8146 dname = strdup(path);
8147 if (dname == NULL)
8148 return -ENOMEM;
8149
8150 dir = dirname(dname);
8151 if (mkdir(dir, 0700) && errno != EEXIST)
8152 err = -errno;
8153
8154 free(dname);
8155 if (err) {
8156 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
8157 pr_warn("failed to mkdir %s: %s\n", path, cp);
8158 }
8159 return err;
8160 }
8161
8162 static int check_path(const char *path)
8163 {
8164 char *cp, errmsg[STRERR_BUFSIZE];
8165 struct statfs st_fs;
8166 char *dname, *dir;
8167 int err = 0;
8168
8169 if (path == NULL)
8170 return -EINVAL;
8171
8172 dname = strdup(path);
8173 if (dname == NULL)
8174 return -ENOMEM;
8175
8176 dir = dirname(dname);
8177 if (statfs(dir, &st_fs)) {
8178 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
8179 pr_warn("failed to statfs %s: %s\n", dir, cp);
8180 err = -errno;
8181 }
8182 free(dname);
8183
8184 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
8185 pr_warn("specified path %s is not on BPF FS\n", path);
8186 err = -EINVAL;
8187 }
8188
8189 return err;
8190 }
8191
8192 int bpf_program__pin(struct bpf_program *prog, const char *path)
8193 {
8194 char *cp, errmsg[STRERR_BUFSIZE];
8195 int err;
8196
8197 if (prog->fd < 0) {
8198 pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name);
8199 return libbpf_err(-EINVAL);
8200 }
8201
8202 err = make_parent_dir(path);
8203 if (err)
8204 return libbpf_err(err);
8205
8206 err = check_path(path);
8207 if (err)
8208 return libbpf_err(err);
8209
8210 if (bpf_obj_pin(prog->fd, path)) {
8211 err = -errno;
8212 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
8213 pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, cp);
8214 return libbpf_err(err);
8215 }
8216
8217 pr_debug("prog '%s': pinned at '%s'\n", prog->name, path);
8218 return 0;
8219 }
8220
8221 int bpf_program__unpin(struct bpf_program *prog, const char *path)
8222 {
8223 int err;
8224
8225 if (prog->fd < 0) {
8226 pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name);
8227 return libbpf_err(-EINVAL);
8228 }
8229
8230 err = check_path(path);
8231 if (err)
8232 return libbpf_err(err);
8233
8234 err = unlink(path);
8235 if (err)
8236 return libbpf_err(-errno);
8237
8238 pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path);
8239 return 0;
8240 }
8241
8242 int bpf_map__pin(struct bpf_map *map, const char *path)
8243 {
8244 char *cp, errmsg[STRERR_BUFSIZE];
8245 int err;
8246
8247 if (map == NULL) {
8248 pr_warn("invalid map pointer\n");
8249 return libbpf_err(-EINVAL);
8250 }
8251
8252 if (map->pin_path) {
8253 if (path && strcmp(path, map->pin_path)) {
8254 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8255 bpf_map__name(map), map->pin_path, path);
8256 return libbpf_err(-EINVAL);
8257 } else if (map->pinned) {
8258 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
8259 bpf_map__name(map), map->pin_path);
8260 return 0;
8261 }
8262 } else {
8263 if (!path) {
8264 pr_warn("missing a path to pin map '%s' at\n",
8265 bpf_map__name(map));
8266 return libbpf_err(-EINVAL);
8267 } else if (map->pinned) {
8268 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
8269 return libbpf_err(-EEXIST);
8270 }
8271
8272 map->pin_path = strdup(path);
8273 if (!map->pin_path) {
8274 err = -errno;
8275 goto out_err;
8276 }
8277 }
8278
8279 err = make_parent_dir(map->pin_path);
8280 if (err)
8281 return libbpf_err(err);
8282
8283 err = check_path(map->pin_path);
8284 if (err)
8285 return libbpf_err(err);
8286
8287 if (bpf_obj_pin(map->fd, map->pin_path)) {
8288 err = -errno;
8289 goto out_err;
8290 }
8291
8292 map->pinned = true;
8293 pr_debug("pinned map '%s'\n", map->pin_path);
8294
8295 return 0;
8296
8297 out_err:
8298 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
8299 pr_warn("failed to pin map: %s\n", cp);
8300 return libbpf_err(err);
8301 }
8302
8303 int bpf_map__unpin(struct bpf_map *map, const char *path)
8304 {
8305 int err;
8306
8307 if (map == NULL) {
8308 pr_warn("invalid map pointer\n");
8309 return libbpf_err(-EINVAL);
8310 }
8311
8312 if (map->pin_path) {
8313 if (path && strcmp(path, map->pin_path)) {
8314 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8315 bpf_map__name(map), map->pin_path, path);
8316 return libbpf_err(-EINVAL);
8317 }
8318 path = map->pin_path;
8319 } else if (!path) {
8320 pr_warn("no path to unpin map '%s' from\n",
8321 bpf_map__name(map));
8322 return libbpf_err(-EINVAL);
8323 }
8324
8325 err = check_path(path);
8326 if (err)
8327 return libbpf_err(err);
8328
8329 err = unlink(path);
8330 if (err != 0)
8331 return libbpf_err(-errno);
8332
8333 map->pinned = false;
8334 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
8335
8336 return 0;
8337 }
8338
8339 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
8340 {
8341 char *new = NULL;
8342
8343 if (path) {
8344 new = strdup(path);
8345 if (!new)
8346 return libbpf_err(-errno);
8347 }
8348
8349 free(map->pin_path);
8350 map->pin_path = new;
8351 return 0;
8352 }
8353
8354 __alias(bpf_map__pin_path)
8355 const char *bpf_map__get_pin_path(const struct bpf_map *map);
8356
8357 const char *bpf_map__pin_path(const struct bpf_map *map)
8358 {
8359 return map->pin_path;
8360 }
8361
8362 bool bpf_map__is_pinned(const struct bpf_map *map)
8363 {
8364 return map->pinned;
8365 }
8366
8367 static void sanitize_pin_path(char *s)
8368 {
8369 /* bpffs disallows periods in path names */
8370 while (*s) {
8371 if (*s == '.')
8372 *s = '_';
8373 s++;
8374 }
8375 }
8376
8377 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
8378 {
8379 struct bpf_map *map;
8380 int err;
8381
8382 if (!obj)
8383 return libbpf_err(-ENOENT);
8384
8385 if (!obj->loaded) {
8386 pr_warn("object not yet loaded; load it first\n");
8387 return libbpf_err(-ENOENT);
8388 }
8389
8390 bpf_object__for_each_map(map, obj) {
8391 char *pin_path = NULL;
8392 char buf[PATH_MAX];
8393
8394 if (!map->autocreate)
8395 continue;
8396
8397 if (path) {
8398 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8399 if (err)
8400 goto err_unpin_maps;
8401 sanitize_pin_path(buf);
8402 pin_path = buf;
8403 } else if (!map->pin_path) {
8404 continue;
8405 }
8406
8407 err = bpf_map__pin(map, pin_path);
8408 if (err)
8409 goto err_unpin_maps;
8410 }
8411
8412 return 0;
8413
8414 err_unpin_maps:
8415 while ((map = bpf_object__prev_map(obj, map))) {
8416 if (!map->pin_path)
8417 continue;
8418
8419 bpf_map__unpin(map, NULL);
8420 }
8421
8422 return libbpf_err(err);
8423 }
8424
8425 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
8426 {
8427 struct bpf_map *map;
8428 int err;
8429
8430 if (!obj)
8431 return libbpf_err(-ENOENT);
8432
8433 bpf_object__for_each_map(map, obj) {
8434 char *pin_path = NULL;
8435 char buf[PATH_MAX];
8436
8437 if (path) {
8438 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8439 if (err)
8440 return libbpf_err(err);
8441 sanitize_pin_path(buf);
8442 pin_path = buf;
8443 } else if (!map->pin_path) {
8444 continue;
8445 }
8446
8447 err = bpf_map__unpin(map, pin_path);
8448 if (err)
8449 return libbpf_err(err);
8450 }
8451
8452 return 0;
8453 }
8454
8455 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
8456 {
8457 struct bpf_program *prog;
8458 char buf[PATH_MAX];
8459 int err;
8460
8461 if (!obj)
8462 return libbpf_err(-ENOENT);
8463
8464 if (!obj->loaded) {
8465 pr_warn("object not yet loaded; load it first\n");
8466 return libbpf_err(-ENOENT);
8467 }
8468
8469 bpf_object__for_each_program(prog, obj) {
8470 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8471 if (err)
8472 goto err_unpin_programs;
8473
8474 err = bpf_program__pin(prog, buf);
8475 if (err)
8476 goto err_unpin_programs;
8477 }
8478
8479 return 0;
8480
8481 err_unpin_programs:
8482 while ((prog = bpf_object__prev_program(obj, prog))) {
8483 if (pathname_concat(buf, sizeof(buf), path, prog->name))
8484 continue;
8485
8486 bpf_program__unpin(prog, buf);
8487 }
8488
8489 return libbpf_err(err);
8490 }
8491
8492 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8493 {
8494 struct bpf_program *prog;
8495 int err;
8496
8497 if (!obj)
8498 return libbpf_err(-ENOENT);
8499
8500 bpf_object__for_each_program(prog, obj) {
8501 char buf[PATH_MAX];
8502
8503 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8504 if (err)
8505 return libbpf_err(err);
8506
8507 err = bpf_program__unpin(prog, buf);
8508 if (err)
8509 return libbpf_err(err);
8510 }
8511
8512 return 0;
8513 }
8514
8515 int bpf_object__pin(struct bpf_object *obj, const char *path)
8516 {
8517 int err;
8518
8519 err = bpf_object__pin_maps(obj, path);
8520 if (err)
8521 return libbpf_err(err);
8522
8523 err = bpf_object__pin_programs(obj, path);
8524 if (err) {
8525 bpf_object__unpin_maps(obj, path);
8526 return libbpf_err(err);
8527 }
8528
8529 return 0;
8530 }
8531
8532 static void bpf_map__destroy(struct bpf_map *map)
8533 {
8534 if (map->inner_map) {
8535 bpf_map__destroy(map->inner_map);
8536 zfree(&map->inner_map);
8537 }
8538
8539 zfree(&map->init_slots);
8540 map->init_slots_sz = 0;
8541
8542 if (map->mmaped) {
8543 munmap(map->mmaped, bpf_map_mmap_sz(map));
8544 map->mmaped = NULL;
8545 }
8546
8547 if (map->st_ops) {
8548 zfree(&map->st_ops->data);
8549 zfree(&map->st_ops->progs);
8550 zfree(&map->st_ops->kern_func_off);
8551 zfree(&map->st_ops);
8552 }
8553
8554 zfree(&map->name);
8555 zfree(&map->real_name);
8556 zfree(&map->pin_path);
8557
8558 if (map->fd >= 0)
8559 zclose(map->fd);
8560 }
8561
8562 void bpf_object__close(struct bpf_object *obj)
8563 {
8564 size_t i;
8565
8566 if (IS_ERR_OR_NULL(obj))
8567 return;
8568 #ifdef HAVE_LIBELF
8569 usdt_manager_free(obj->usdt_man);
8570 obj->usdt_man = NULL;
8571 #endif //HAVE_LIBELF
8572 bpf_gen__free(obj->gen_loader);
8573 bpf_object__elf_finish(obj);
8574 bpf_object_unload(obj);
8575 btf__free(obj->btf);
8576 btf_ext__free(obj->btf_ext);
8577
8578 for (i = 0; i < obj->nr_maps; i++)
8579 bpf_map__destroy(&obj->maps[i]);
8580
8581 zfree(&obj->btf_custom_path);
8582 zfree(&obj->kconfig);
8583 zfree(&obj->externs);
8584 obj->nr_extern = 0;
8585
8586 zfree(&obj->maps);
8587 obj->nr_maps = 0;
8588
8589 if (obj->programs && obj->nr_programs) {
8590 for (i = 0; i < obj->nr_programs; i++)
8591 bpf_program__exit(&obj->programs[i]);
8592 }
8593 zfree(&obj->programs);
8594
8595 free(obj);
8596 }
8597
8598 const char *bpf_object__name(const struct bpf_object *obj)
8599 {
8600 return obj ? obj->name : libbpf_err_ptr(-EINVAL);
8601 }
8602
8603 unsigned int bpf_object__kversion(const struct bpf_object *obj)
8604 {
8605 return obj ? obj->kern_version : 0;
8606 }
8607
8608 struct btf *bpf_object__btf(const struct bpf_object *obj)
8609 {
8610 return obj ? obj->btf : NULL;
8611 }
8612
8613 int bpf_object__btf_fd(const struct bpf_object *obj)
8614 {
8615 return obj->btf ? btf__fd(obj->btf) : -1;
8616 }
8617
8618 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
8619 {
8620 if (obj->loaded)
8621 return libbpf_err(-EINVAL);
8622
8623 obj->kern_version = kern_version;
8624
8625 return 0;
8626 }
8627
8628 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
8629 {
8630 struct bpf_gen *gen;
8631
8632 if (!opts)
8633 return -EFAULT;
8634 if (!OPTS_VALID(opts, gen_loader_opts))
8635 return -EINVAL;
8636 gen = calloc(sizeof(*gen), 1);
8637 if (!gen)
8638 return -ENOMEM;
8639 gen->opts = opts;
8640 obj->gen_loader = gen;
8641 return 0;
8642 }
8643
8644 static struct bpf_program *
8645 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
8646 bool forward)
8647 {
8648 size_t nr_programs = obj->nr_programs;
8649 ssize_t idx;
8650
8651 if (!nr_programs)
8652 return NULL;
8653
8654 if (!p)
8655 /* Iter from the beginning */
8656 return forward ? &obj->programs[0] :
8657 &obj->programs[nr_programs - 1];
8658
8659 if (p->obj != obj) {
8660 pr_warn("error: program handler doesn't match object\n");
8661 return errno = EINVAL, NULL;
8662 }
8663
8664 idx = (p - obj->programs) + (forward ? 1 : -1);
8665 if (idx >= obj->nr_programs || idx < 0)
8666 return NULL;
8667 return &obj->programs[idx];
8668 }
8669
8670 struct bpf_program *
8671 bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
8672 {
8673 struct bpf_program *prog = prev;
8674
8675 do {
8676 prog = __bpf_program__iter(prog, obj, true);
8677 } while (prog && prog_is_subprog(obj, prog));
8678
8679 return prog;
8680 }
8681
8682 struct bpf_program *
8683 bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
8684 {
8685 struct bpf_program *prog = next;
8686
8687 do {
8688 prog = __bpf_program__iter(prog, obj, false);
8689 } while (prog && prog_is_subprog(obj, prog));
8690
8691 return prog;
8692 }
8693
8694 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
8695 {
8696 prog->prog_ifindex = ifindex;
8697 }
8698
8699 const char *bpf_program__name(const struct bpf_program *prog)
8700 {
8701 return prog->name;
8702 }
8703
8704 const char *bpf_program__section_name(const struct bpf_program *prog)
8705 {
8706 return prog->sec_name;
8707 }
8708
8709 bool bpf_program__autoload(const struct bpf_program *prog)
8710 {
8711 return prog->autoload;
8712 }
8713
8714 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
8715 {
8716 if (prog->obj->loaded)
8717 return libbpf_err(-EINVAL);
8718
8719 prog->autoload = autoload;
8720 return 0;
8721 }
8722
8723 bool bpf_program__autoattach(const struct bpf_program *prog)
8724 {
8725 return prog->autoattach;
8726 }
8727
8728 void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach)
8729 {
8730 prog->autoattach = autoattach;
8731 }
8732
8733 const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
8734 {
8735 return prog->insns;
8736 }
8737
8738 size_t bpf_program__insn_cnt(const struct bpf_program *prog)
8739 {
8740 return prog->insns_cnt;
8741 }
8742
8743 int bpf_program__set_insns(struct bpf_program *prog,
8744 struct bpf_insn *new_insns, size_t new_insn_cnt)
8745 {
8746 struct bpf_insn *insns;
8747
8748 if (prog->obj->loaded)
8749 return -EBUSY;
8750
8751 insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns));
8752 if (!insns) {
8753 pr_warn("prog '%s': failed to realloc prog code\n", prog->name);
8754 return -ENOMEM;
8755 }
8756 memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns));
8757
8758 prog->insns = insns;
8759 prog->insns_cnt = new_insn_cnt;
8760 return 0;
8761 }
8762
8763 int bpf_program__fd(const struct bpf_program *prog)
8764 {
8765 if (!prog)
8766 return libbpf_err(-EINVAL);
8767
8768 if (prog->fd < 0)
8769 return libbpf_err(-ENOENT);
8770
8771 return prog->fd;
8772 }
8773
8774 __alias(bpf_program__type)
8775 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
8776
8777 enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
8778 {
8779 return prog->type;
8780 }
8781
8782 int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
8783 {
8784 if (prog->obj->loaded)
8785 return libbpf_err(-EBUSY);
8786
8787 prog->type = type;
8788 return 0;
8789 }
8790
8791 __alias(bpf_program__expected_attach_type)
8792 enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
8793
8794 enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
8795 {
8796 return prog->expected_attach_type;
8797 }
8798
8799 int bpf_program__set_expected_attach_type(struct bpf_program *prog,
8800 enum bpf_attach_type type)
8801 {
8802 if (prog->obj->loaded)
8803 return libbpf_err(-EBUSY);
8804
8805 prog->expected_attach_type = type;
8806 return 0;
8807 }
8808
8809 __u32 bpf_program__flags(const struct bpf_program *prog)
8810 {
8811 return prog->prog_flags;
8812 }
8813
8814 int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
8815 {
8816 if (prog->obj->loaded)
8817 return libbpf_err(-EBUSY);
8818
8819 prog->prog_flags = flags;
8820 return 0;
8821 }
8822
8823 __u32 bpf_program__log_level(const struct bpf_program *prog)
8824 {
8825 return prog->log_level;
8826 }
8827
8828 int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
8829 {
8830 if (prog->obj->loaded)
8831 return libbpf_err(-EBUSY);
8832
8833 prog->log_level = log_level;
8834 return 0;
8835 }
8836
8837 const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
8838 {
8839 *log_size = prog->log_size;
8840 return prog->log_buf;
8841 }
8842
8843 int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
8844 {
8845 if (log_size && !log_buf)
8846 return -EINVAL;
8847 if (prog->log_size > UINT_MAX)
8848 return -EINVAL;
8849 if (prog->obj->loaded)
8850 return -EBUSY;
8851
8852 prog->log_buf = log_buf;
8853 prog->log_size = log_size;
8854 return 0;
8855 }
8856
8857 #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
8858 .sec = (char *)sec_pfx, \
8859 .prog_type = BPF_PROG_TYPE_##ptype, \
8860 .expected_attach_type = atype, \
8861 .cookie = (long)(flags), \
8862 .prog_prepare_load_fn = libbpf_prepare_prog_load, \
8863 __VA_ARGS__ \
8864 }
8865
8866 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8867 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8868 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8869 #ifdef HAVE_LIBELF
8870 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8871 #endif //HAVE_LIBELF
8872 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8873 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8874 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8875 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8876 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8877 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8878
8879 static const struct bpf_sec_def section_defs[] = {
8880 SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE),
8881 SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE),
8882 SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE),
8883 SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
8884 SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
8885 SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
8886 SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
8887 SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
8888 SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
8889 SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8890 SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8891 SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
8892 SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
8893 #ifdef HAVE_LIBELF
8894 SEC_DEF("usdt+", KPROBE, 0, SEC_NONE, attach_usdt),
8895 #endif //HAVE_LIBELF
8896 SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE),
8897 SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE),
8898 SEC_DEF("action", SCHED_ACT, 0, SEC_NONE),
8899 SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp),
8900 SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp),
8901 SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8902 SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8903 SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8904 SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8905 SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
8906 SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
8907 SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
8908 SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
8909 SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8910 SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8911 SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8912 SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace),
8913 SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
8914 SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
8915 SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF),
8916 SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
8917 SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
8918 SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE),
8919 SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
8920 SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
8921 SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
8922 SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
8923 SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS),
8924 SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT),
8925 SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE),
8926 SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE),
8927 SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE),
8928 SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE),
8929 SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE),
8930 SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT),
8931 SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT),
8932 SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT),
8933 SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE),
8934 SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT),
8935 SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT),
8936 SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT),
8937 SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT),
8938 SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT),
8939 SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE),
8940 SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE),
8941 SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE),
8942 SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT),
8943 SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE),
8944 SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE),
8945 SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE),
8946 SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE),
8947 SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE),
8948 SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE),
8949 SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE),
8950 SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE),
8951 SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE),
8952 SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE),
8953 SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE),
8954 SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE),
8955 SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE),
8956 SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE),
8957 SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE),
8958 SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE),
8959 SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE),
8960 SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT),
8961 SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE),
8962 SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE),
8963 };
8964
8965 static size_t custom_sec_def_cnt;
8966 static struct bpf_sec_def *custom_sec_defs;
8967 static struct bpf_sec_def custom_fallback_def;
8968 static bool has_custom_fallback_def;
8969
8970 static int last_custom_sec_def_handler_id;
8971
8972 int libbpf_register_prog_handler(const char *sec,
8973 enum bpf_prog_type prog_type,
8974 enum bpf_attach_type exp_attach_type,
8975 const struct libbpf_prog_handler_opts *opts)
8976 {
8977 struct bpf_sec_def *sec_def;
8978
8979 if (!OPTS_VALID(opts, libbpf_prog_handler_opts))
8980 return libbpf_err(-EINVAL);
8981
8982 if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */
8983 return libbpf_err(-E2BIG);
8984
8985 if (sec) {
8986 sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1,
8987 sizeof(*sec_def));
8988 if (!sec_def)
8989 return libbpf_err(-ENOMEM);
8990
8991 custom_sec_defs = sec_def;
8992 sec_def = &custom_sec_defs[custom_sec_def_cnt];
8993 } else {
8994 if (has_custom_fallback_def)
8995 return libbpf_err(-EBUSY);
8996
8997 sec_def = &custom_fallback_def;
8998 }
8999
9000 sec_def->sec = sec ? strdup(sec) : NULL;
9001 if (sec && !sec_def->sec)
9002 return libbpf_err(-ENOMEM);
9003
9004 sec_def->prog_type = prog_type;
9005 sec_def->expected_attach_type = exp_attach_type;
9006 sec_def->cookie = OPTS_GET(opts, cookie, 0);
9007
9008 sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL);
9009 sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL);
9010 sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL);
9011
9012 sec_def->handler_id = ++last_custom_sec_def_handler_id;
9013
9014 if (sec)
9015 custom_sec_def_cnt++;
9016 else
9017 has_custom_fallback_def = true;
9018
9019 return sec_def->handler_id;
9020 }
9021
9022 int libbpf_unregister_prog_handler(int handler_id)
9023 {
9024 struct bpf_sec_def *sec_defs;
9025 int i;
9026
9027 if (handler_id <= 0)
9028 return libbpf_err(-EINVAL);
9029
9030 if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) {
9031 memset(&custom_fallback_def, 0, sizeof(custom_fallback_def));
9032 has_custom_fallback_def = false;
9033 return 0;
9034 }
9035
9036 for (i = 0; i < custom_sec_def_cnt; i++) {
9037 if (custom_sec_defs[i].handler_id == handler_id)
9038 break;
9039 }
9040
9041 if (i == custom_sec_def_cnt)
9042 return libbpf_err(-ENOENT);
9043
9044 free(custom_sec_defs[i].sec);
9045 for (i = i + 1; i < custom_sec_def_cnt; i++)
9046 custom_sec_defs[i - 1] = custom_sec_defs[i];
9047 custom_sec_def_cnt--;
9048
9049 /* try to shrink the array, but it's ok if we couldn't */
9050 sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs));
9051 if (sec_defs)
9052 custom_sec_defs = sec_defs;
9053
9054 return 0;
9055 }
9056
9057 static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name)
9058 {
9059 size_t len = strlen(sec_def->sec);
9060
9061 /* "type/" always has to have proper SEC("type/extras") form */
9062 if (sec_def->sec[len - 1] == '/') {
9063 if (str_has_pfx(sec_name, sec_def->sec))
9064 return true;
9065 return false;
9066 }
9067
9068 /* "type+" means it can be either exact SEC("type") or
9069 * well-formed SEC("type/extras") with proper '/' separator
9070 */
9071 if (sec_def->sec[len - 1] == '+') {
9072 len--;
9073 /* not even a prefix */
9074 if (strncmp(sec_name, sec_def->sec, len) != 0)
9075 return false;
9076 /* exact match or has '/' separator */
9077 if (sec_name[len] == '\0' || sec_name[len] == '/')
9078 return true;
9079 return false;
9080 }
9081
9082 return strcmp(sec_name, sec_def->sec) == 0;
9083 }
9084
9085 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
9086 {
9087 const struct bpf_sec_def *sec_def;
9088 int i, n;
9089
9090 n = custom_sec_def_cnt;
9091 for (i = 0; i < n; i++) {
9092 sec_def = &custom_sec_defs[i];
9093 if (sec_def_matches(sec_def, sec_name))
9094 return sec_def;
9095 }
9096
9097 n = ARRAY_SIZE(section_defs);
9098 for (i = 0; i < n; i++) {
9099 sec_def = §ion_defs[i];
9100 if (sec_def_matches(sec_def, sec_name))
9101 return sec_def;
9102 }
9103
9104 if (has_custom_fallback_def)
9105 return &custom_fallback_def;
9106
9107 return NULL;
9108 }
9109
9110 #define MAX_TYPE_NAME_SIZE 32
9111
9112 static char *libbpf_get_type_names(bool attach_type)
9113 {
9114 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
9115 char *buf;
9116
9117 buf = malloc(len);
9118 if (!buf)
9119 return NULL;
9120
9121 buf[0] = '\0';
9122 /* Forge string buf with all available names */
9123 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
9124 const struct bpf_sec_def *sec_def = §ion_defs[i];
9125
9126 if (attach_type) {
9127 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9128 continue;
9129
9130 if (!(sec_def->cookie & SEC_ATTACHABLE))
9131 continue;
9132 }
9133
9134 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
9135 free(buf);
9136 return NULL;
9137 }
9138 strcat(buf, " ");
9139 strcat(buf, section_defs[i].sec);
9140 }
9141
9142 return buf;
9143 }
9144
9145 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
9146 enum bpf_attach_type *expected_attach_type)
9147 {
9148 const struct bpf_sec_def *sec_def;
9149 char *type_names;
9150
9151 if (!name)
9152 return libbpf_err(-EINVAL);
9153
9154 sec_def = find_sec_def(name);
9155 if (sec_def) {
9156 *prog_type = sec_def->prog_type;
9157 *expected_attach_type = sec_def->expected_attach_type;
9158 return 0;
9159 }
9160
9161 pr_debug("failed to guess program type from ELF section '%s'\n", name);
9162 type_names = libbpf_get_type_names(false);
9163 if (type_names != NULL) {
9164 pr_debug("supported section(type) names are:%s\n", type_names);
9165 free(type_names);
9166 }
9167
9168 return libbpf_err(-ESRCH);
9169 }
9170
9171 const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t)
9172 {
9173 if (t < 0 || t >= ARRAY_SIZE(attach_type_name))
9174 return NULL;
9175
9176 return attach_type_name[t];
9177 }
9178
9179 const char *libbpf_bpf_link_type_str(enum bpf_link_type t)
9180 {
9181 if (t < 0 || t >= ARRAY_SIZE(link_type_name))
9182 return NULL;
9183
9184 return link_type_name[t];
9185 }
9186
9187 const char *libbpf_bpf_map_type_str(enum bpf_map_type t)
9188 {
9189 if (t < 0 || t >= ARRAY_SIZE(map_type_name))
9190 return NULL;
9191
9192 return map_type_name[t];
9193 }
9194
9195 const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t)
9196 {
9197 if (t < 0 || t >= ARRAY_SIZE(prog_type_name))
9198 return NULL;
9199
9200 return prog_type_name[t];
9201 }
9202
9203 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
9204 size_t offset)
9205 {
9206 struct bpf_map *map;
9207 size_t i;
9208
9209 for (i = 0; i < obj->nr_maps; i++) {
9210 map = &obj->maps[i];
9211 if (!bpf_map__is_struct_ops(map))
9212 continue;
9213 if (map->sec_offset <= offset &&
9214 offset - map->sec_offset < map->def.value_size)
9215 return map;
9216 }
9217
9218 return NULL;
9219 }
9220
9221 /* Collect the reloc from ELF and populate the st_ops->progs[] */
9222 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
9223 Elf64_Shdr *shdr, Elf_Data *data)
9224 {
9225 const struct btf_member *member;
9226 struct bpf_struct_ops *st_ops;
9227 struct bpf_program *prog;
9228 unsigned int shdr_idx;
9229 const struct btf *btf;
9230 struct bpf_map *map;
9231 unsigned int moff, insn_idx;
9232 const char *name;
9233 __u32 member_idx;
9234 Elf64_Sym *sym;
9235 Elf64_Rel *rel;
9236 int i, nrels;
9237
9238 btf = obj->btf;
9239 nrels = shdr->sh_size / shdr->sh_entsize;
9240 for (i = 0; i < nrels; i++) {
9241 rel = elf_rel_by_idx(data, i);
9242 if (!rel) {
9243 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
9244 return -LIBBPF_ERRNO__FORMAT;
9245 }
9246
9247 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
9248 if (!sym) {
9249 pr_warn("struct_ops reloc: symbol %zx not found\n",
9250 (size_t)ELF64_R_SYM(rel->r_info));
9251 return -LIBBPF_ERRNO__FORMAT;
9252 }
9253
9254 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
9255 map = find_struct_ops_map_by_offset(obj, rel->r_offset);
9256 if (!map) {
9257 pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
9258 (size_t)rel->r_offset);
9259 return -EINVAL;
9260 }
9261
9262 moff = rel->r_offset - map->sec_offset;
9263 shdr_idx = sym->st_shndx;
9264 st_ops = map->st_ops;
9265 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",
9266 map->name,
9267 (long long)(rel->r_info >> 32),
9268 (long long)sym->st_value,
9269 shdr_idx, (size_t)rel->r_offset,
9270 map->sec_offset, sym->st_name, name);
9271
9272 if (shdr_idx >= SHN_LORESERVE) {
9273 pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
9274 map->name, (size_t)rel->r_offset, shdr_idx);
9275 return -LIBBPF_ERRNO__RELOC;
9276 }
9277 if (sym->st_value % BPF_INSN_SZ) {
9278 pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
9279 map->name, (unsigned long long)sym->st_value);
9280 return -LIBBPF_ERRNO__FORMAT;
9281 }
9282 insn_idx = sym->st_value / BPF_INSN_SZ;
9283
9284 member = find_member_by_offset(st_ops->type, moff * 8);
9285 if (!member) {
9286 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
9287 map->name, moff);
9288 return -EINVAL;
9289 }
9290 member_idx = member - btf_members(st_ops->type);
9291 name = btf__name_by_offset(btf, member->name_off);
9292
9293 if (!resolve_func_ptr(btf, member->type, NULL)) {
9294 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
9295 map->name, name);
9296 return -EINVAL;
9297 }
9298
9299 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
9300 if (!prog) {
9301 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
9302 map->name, shdr_idx, name);
9303 return -EINVAL;
9304 }
9305
9306 /* prevent the use of BPF prog with invalid type */
9307 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
9308 pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
9309 map->name, prog->name);
9310 return -EINVAL;
9311 }
9312
9313 /* if we haven't yet processed this BPF program, record proper
9314 * attach_btf_id and member_idx
9315 */
9316 if (!prog->attach_btf_id) {
9317 prog->attach_btf_id = st_ops->type_id;
9318 prog->expected_attach_type = member_idx;
9319 }
9320
9321 /* struct_ops BPF prog can be re-used between multiple
9322 * .struct_ops as long as it's the same struct_ops struct
9323 * definition and the same function pointer field
9324 */
9325 if (prog->attach_btf_id != st_ops->type_id ||
9326 prog->expected_attach_type != member_idx) {
9327 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",
9328 map->name, prog->name, prog->sec_name, prog->type,
9329 prog->attach_btf_id, prog->expected_attach_type, name);
9330 return -EINVAL;
9331 }
9332
9333 st_ops->progs[member_idx] = prog;
9334 }
9335
9336 return 0;
9337 }
9338
9339 #define BTF_TRACE_PREFIX "btf_trace_"
9340 #define BTF_LSM_PREFIX "bpf_lsm_"
9341 #define BTF_ITER_PREFIX "bpf_iter_"
9342 #define BTF_MAX_NAME_SIZE 128
9343
9344 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
9345 const char **prefix, int *kind)
9346 {
9347 switch (attach_type) {
9348 case BPF_TRACE_RAW_TP:
9349 *prefix = BTF_TRACE_PREFIX;
9350 *kind = BTF_KIND_TYPEDEF;
9351 break;
9352 case BPF_LSM_MAC:
9353 case BPF_LSM_CGROUP:
9354 *prefix = BTF_LSM_PREFIX;
9355 *kind = BTF_KIND_FUNC;
9356 break;
9357 case BPF_TRACE_ITER:
9358 *prefix = BTF_ITER_PREFIX;
9359 *kind = BTF_KIND_FUNC;
9360 break;
9361 default:
9362 *prefix = "";
9363 *kind = BTF_KIND_FUNC;
9364 }
9365 }
9366
9367 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
9368 const char *name, __u32 kind)
9369 {
9370 char btf_type_name[BTF_MAX_NAME_SIZE];
9371 int ret;
9372
9373 ret = snprintf(btf_type_name, sizeof(btf_type_name),
9374 "%s%s", prefix, name);
9375 /* snprintf returns the number of characters written excluding the
9376 * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
9377 * indicates truncation.
9378 */
9379 if (ret < 0 || ret >= sizeof(btf_type_name))
9380 return -ENAMETOOLONG;
9381 return btf__find_by_name_kind(btf, btf_type_name, kind);
9382 }
9383
9384 static inline int find_attach_btf_id(struct btf *btf, const char *name,
9385 enum bpf_attach_type attach_type)
9386 {
9387 const char *prefix;
9388 int kind;
9389
9390 btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
9391 return find_btf_by_prefix_kind(btf, prefix, name, kind);
9392 }
9393
9394 int libbpf_find_vmlinux_btf_id(const char *name,
9395 enum bpf_attach_type attach_type)
9396 {
9397 struct btf *btf;
9398 int err;
9399
9400 btf = btf__load_vmlinux_btf();
9401 err = libbpf_get_error(btf);
9402 if (err) {
9403 pr_warn("vmlinux BTF is not found\n");
9404 return libbpf_err(err);
9405 }
9406
9407 err = find_attach_btf_id(btf, name, attach_type);
9408 if (err <= 0)
9409 pr_warn("%s is not found in vmlinux BTF\n", name);
9410
9411 btf__free(btf);
9412 return libbpf_err(err);
9413 }
9414
9415 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
9416 {
9417 struct bpf_prog_info info;
9418 __u32 info_len = sizeof(info);
9419 struct btf *btf;
9420 int err;
9421
9422 memset(&info, 0, info_len);
9423 err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len);
9424 if (err) {
9425 pr_warn("failed bpf_obj_get_info_by_fd for FD %d: %d\n",
9426 attach_prog_fd, err);
9427 return err;
9428 }
9429
9430 err = -EINVAL;
9431 if (!info.btf_id) {
9432 pr_warn("The target program doesn't have BTF\n");
9433 goto out;
9434 }
9435 btf = btf__load_from_kernel_by_id(info.btf_id);
9436 err = libbpf_get_error(btf);
9437 if (err) {
9438 pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err);
9439 goto out;
9440 }
9441 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
9442 btf__free(btf);
9443 if (err <= 0) {
9444 pr_warn("%s is not found in prog's BTF\n", name);
9445 goto out;
9446 }
9447 out:
9448 return err;
9449 }
9450
9451 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
9452 enum bpf_attach_type attach_type,
9453 int *btf_obj_fd, int *btf_type_id)
9454 {
9455 int ret, i;
9456
9457 ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
9458 if (ret > 0) {
9459 *btf_obj_fd = 0; /* vmlinux BTF */
9460 *btf_type_id = ret;
9461 return 0;
9462 }
9463 if (ret != -ENOENT)
9464 return ret;
9465
9466 ret = load_module_btfs(obj);
9467 if (ret)
9468 return ret;
9469
9470 for (i = 0; i < obj->btf_module_cnt; i++) {
9471 const struct module_btf *mod = &obj->btf_modules[i];
9472
9473 ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
9474 if (ret > 0) {
9475 *btf_obj_fd = mod->fd;
9476 *btf_type_id = ret;
9477 return 0;
9478 }
9479 if (ret == -ENOENT)
9480 continue;
9481
9482 return ret;
9483 }
9484
9485 return -ESRCH;
9486 }
9487
9488 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
9489 int *btf_obj_fd, int *btf_type_id)
9490 {
9491 enum bpf_attach_type attach_type = prog->expected_attach_type;
9492 __u32 attach_prog_fd = prog->attach_prog_fd;
9493 int err = 0;
9494
9495 /* BPF program's BTF ID */
9496 if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) {
9497 if (!attach_prog_fd) {
9498 pr_warn("prog '%s': attach program FD is not set\n", prog->name);
9499 return -EINVAL;
9500 }
9501 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
9502 if (err < 0) {
9503 pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
9504 prog->name, attach_prog_fd, attach_name, err);
9505 return err;
9506 }
9507 *btf_obj_fd = 0;
9508 *btf_type_id = err;
9509 return 0;
9510 }
9511
9512 /* kernel/module BTF ID */
9513 if (prog->obj->gen_loader) {
9514 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
9515 *btf_obj_fd = 0;
9516 *btf_type_id = 1;
9517 } else {
9518 err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
9519 }
9520 if (err) {
9521 pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %d\n",
9522 prog->name, attach_name, err);
9523 return err;
9524 }
9525 return 0;
9526 }
9527
9528 int libbpf_attach_type_by_name(const char *name,
9529 enum bpf_attach_type *attach_type)
9530 {
9531 char *type_names;
9532 const struct bpf_sec_def *sec_def;
9533
9534 if (!name)
9535 return libbpf_err(-EINVAL);
9536
9537 sec_def = find_sec_def(name);
9538 if (!sec_def) {
9539 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
9540 type_names = libbpf_get_type_names(true);
9541 if (type_names != NULL) {
9542 pr_debug("attachable section(type) names are:%s\n", type_names);
9543 free(type_names);
9544 }
9545
9546 return libbpf_err(-EINVAL);
9547 }
9548
9549 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9550 return libbpf_err(-EINVAL);
9551 if (!(sec_def->cookie & SEC_ATTACHABLE))
9552 return libbpf_err(-EINVAL);
9553
9554 *attach_type = sec_def->expected_attach_type;
9555 return 0;
9556 }
9557
9558 int bpf_map__fd(const struct bpf_map *map)
9559 {
9560 return map ? map->fd : libbpf_err(-EINVAL);
9561 }
9562
9563 static bool map_uses_real_name(const struct bpf_map *map)
9564 {
9565 /* Since libbpf started to support custom .data.* and .rodata.* maps,
9566 * their user-visible name differs from kernel-visible name. Users see
9567 * such map's corresponding ELF section name as a map name.
9568 * This check distinguishes .data/.rodata from .data.* and .rodata.*
9569 * maps to know which name has to be returned to the user.
9570 */
9571 if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
9572 return true;
9573 if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
9574 return true;
9575 return false;
9576 }
9577
9578 const char *bpf_map__name(const struct bpf_map *map)
9579 {
9580 if (!map)
9581 return NULL;
9582
9583 if (map_uses_real_name(map))
9584 return map->real_name;
9585
9586 return map->name;
9587 }
9588
9589 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
9590 {
9591 return map->def.type;
9592 }
9593
9594 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
9595 {
9596 if (map->fd >= 0)
9597 return libbpf_err(-EBUSY);
9598 map->def.type = type;
9599 return 0;
9600 }
9601
9602 __u32 bpf_map__map_flags(const struct bpf_map *map)
9603 {
9604 return map->def.map_flags;
9605 }
9606
9607 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
9608 {
9609 if (map->fd >= 0)
9610 return libbpf_err(-EBUSY);
9611 map->def.map_flags = flags;
9612 return 0;
9613 }
9614
9615 __u64 bpf_map__map_extra(const struct bpf_map *map)
9616 {
9617 return map->map_extra;
9618 }
9619
9620 int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
9621 {
9622 if (map->fd >= 0)
9623 return libbpf_err(-EBUSY);
9624 map->map_extra = map_extra;
9625 return 0;
9626 }
9627
9628 __u32 bpf_map__numa_node(const struct bpf_map *map)
9629 {
9630 return map->numa_node;
9631 }
9632
9633 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
9634 {
9635 if (map->fd >= 0)
9636 return libbpf_err(-EBUSY);
9637 map->numa_node = numa_node;
9638 return 0;
9639 }
9640
9641 __u32 bpf_map__key_size(const struct bpf_map *map)
9642 {
9643 return map->def.key_size;
9644 }
9645
9646 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
9647 {
9648 if (map->fd >= 0)
9649 return libbpf_err(-EBUSY);
9650 map->def.key_size = size;
9651 return 0;
9652 }
9653
9654 __u32 bpf_map__value_size(const struct bpf_map *map)
9655 {
9656 return map->def.value_size;
9657 }
9658
9659 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
9660 {
9661 if (map->fd >= 0)
9662 return libbpf_err(-EBUSY);
9663 map->def.value_size = size;
9664 return 0;
9665 }
9666
9667 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
9668 {
9669 return map ? map->btf_key_type_id : 0;
9670 }
9671
9672 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
9673 {
9674 return map ? map->btf_value_type_id : 0;
9675 }
9676
9677 int bpf_map__set_initial_value(struct bpf_map *map,
9678 const void *data, size_t size)
9679 {
9680 if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
9681 size != map->def.value_size || map->fd >= 0)
9682 return libbpf_err(-EINVAL);
9683
9684 memcpy(map->mmaped, data, size);
9685 return 0;
9686 }
9687
9688 const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize)
9689 {
9690 if (!map->mmaped)
9691 return NULL;
9692 *psize = map->def.value_size;
9693 return map->mmaped;
9694 }
9695
9696 bool bpf_map__is_internal(const struct bpf_map *map)
9697 {
9698 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
9699 }
9700
9701 __u32 bpf_map__ifindex(const struct bpf_map *map)
9702 {
9703 return map->map_ifindex;
9704 }
9705
9706 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
9707 {
9708 if (map->fd >= 0)
9709 return libbpf_err(-EBUSY);
9710 map->map_ifindex = ifindex;
9711 return 0;
9712 }
9713
9714 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
9715 {
9716 if (!bpf_map_type__is_map_in_map(map->def.type)) {
9717 pr_warn("error: unsupported map type\n");
9718 return libbpf_err(-EINVAL);
9719 }
9720 if (map->inner_map_fd != -1) {
9721 pr_warn("error: inner_map_fd already specified\n");
9722 return libbpf_err(-EINVAL);
9723 }
9724 if (map->inner_map) {
9725 bpf_map__destroy(map->inner_map);
9726 zfree(&map->inner_map);
9727 }
9728 map->inner_map_fd = fd;
9729 return 0;
9730 }
9731
9732 static struct bpf_map *
9733 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9734 {
9735 ssize_t idx;
9736 struct bpf_map *s, *e;
9737
9738 if (!obj || !obj->maps)
9739 return errno = EINVAL, NULL;
9740
9741 s = obj->maps;
9742 e = obj->maps + obj->nr_maps;
9743
9744 if ((m < s) || (m >= e)) {
9745 pr_warn("error in %s: map handler doesn't belong to object\n",
9746 __func__);
9747 return errno = EINVAL, NULL;
9748 }
9749
9750 idx = (m - obj->maps) + i;
9751 if (idx >= obj->nr_maps || idx < 0)
9752 return NULL;
9753 return &obj->maps[idx];
9754 }
9755
9756 struct bpf_map *
9757 bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
9758 {
9759 if (prev == NULL)
9760 return obj->maps;
9761
9762 return __bpf_map__iter(prev, obj, 1);
9763 }
9764
9765 struct bpf_map *
9766 bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
9767 {
9768 if (next == NULL) {
9769 if (!obj->nr_maps)
9770 return NULL;
9771 return obj->maps + obj->nr_maps - 1;
9772 }
9773
9774 return __bpf_map__iter(next, obj, -1);
9775 }
9776
9777 struct bpf_map *
9778 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
9779 {
9780 struct bpf_map *pos;
9781
9782 bpf_object__for_each_map(pos, obj) {
9783 /* if it's a special internal map name (which always starts
9784 * with dot) then check if that special name matches the
9785 * real map name (ELF section name)
9786 */
9787 if (name[0] == '.') {
9788 if (pos->real_name && strcmp(pos->real_name, name) == 0)
9789 return pos;
9790 continue;
9791 }
9792 /* otherwise map name has to be an exact match */
9793 if (map_uses_real_name(pos)) {
9794 if (strcmp(pos->real_name, name) == 0)
9795 return pos;
9796 continue;
9797 }
9798 if (strcmp(pos->name, name) == 0)
9799 return pos;
9800 }
9801 return errno = ENOENT, NULL;
9802 }
9803
9804 int
9805 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
9806 {
9807 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
9808 }
9809
9810 static int validate_map_op(const struct bpf_map *map, size_t key_sz,
9811 size_t value_sz, bool check_value_sz)
9812 {
9813 if (map->fd <= 0)
9814 return -ENOENT;
9815
9816 if (map->def.key_size != key_sz) {
9817 pr_warn("map '%s': unexpected key size %zu provided, expected %u\n",
9818 map->name, key_sz, map->def.key_size);
9819 return -EINVAL;
9820 }
9821
9822 if (!check_value_sz)
9823 return 0;
9824
9825 switch (map->def.type) {
9826 case BPF_MAP_TYPE_PERCPU_ARRAY:
9827 case BPF_MAP_TYPE_PERCPU_HASH:
9828 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
9829 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: {
9830 int num_cpu = libbpf_num_possible_cpus();
9831 size_t elem_sz = roundup(map->def.value_size, 8);
9832
9833 if (value_sz != num_cpu * elem_sz) {
9834 pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n",
9835 map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz);
9836 return -EINVAL;
9837 }
9838 break;
9839 }
9840 default:
9841 if (map->def.value_size != value_sz) {
9842 pr_warn("map '%s': unexpected value size %zu provided, expected %u\n",
9843 map->name, value_sz, map->def.value_size);
9844 return -EINVAL;
9845 }
9846 break;
9847 }
9848 return 0;
9849 }
9850
9851 int bpf_map__lookup_elem(const struct bpf_map *map,
9852 const void *key, size_t key_sz,
9853 void *value, size_t value_sz, __u64 flags)
9854 {
9855 int err;
9856
9857 err = validate_map_op(map, key_sz, value_sz, true);
9858 if (err)
9859 return libbpf_err(err);
9860
9861 return bpf_map_lookup_elem_flags(map->fd, key, value, flags);
9862 }
9863
9864 int bpf_map__update_elem(const struct bpf_map *map,
9865 const void *key, size_t key_sz,
9866 const void *value, size_t value_sz, __u64 flags)
9867 {
9868 int err;
9869
9870 err = validate_map_op(map, key_sz, value_sz, true);
9871 if (err)
9872 return libbpf_err(err);
9873
9874 return bpf_map_update_elem(map->fd, key, value, flags);
9875 }
9876
9877 int bpf_map__delete_elem(const struct bpf_map *map,
9878 const void *key, size_t key_sz, __u64 flags)
9879 {
9880 int err;
9881
9882 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
9883 if (err)
9884 return libbpf_err(err);
9885
9886 return bpf_map_delete_elem_flags(map->fd, key, flags);
9887 }
9888
9889 int bpf_map__lookup_and_delete_elem(const struct bpf_map *map,
9890 const void *key, size_t key_sz,
9891 void *value, size_t value_sz, __u64 flags)
9892 {
9893 int err;
9894
9895 err = validate_map_op(map, key_sz, value_sz, true);
9896 if (err)
9897 return libbpf_err(err);
9898
9899 return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags);
9900 }
9901
9902 int bpf_map__get_next_key(const struct bpf_map *map,
9903 const void *cur_key, void *next_key, size_t key_sz)
9904 {
9905 int err;
9906
9907 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
9908 if (err)
9909 return libbpf_err(err);
9910
9911 return bpf_map_get_next_key(map->fd, cur_key, next_key);
9912 }
9913
9914 long libbpf_get_error(const void *ptr)
9915 {
9916 if (!IS_ERR_OR_NULL(ptr))
9917 return 0;
9918
9919 if (IS_ERR(ptr))
9920 errno = -PTR_ERR(ptr);
9921
9922 /* If ptr == NULL, then errno should be already set by the failing
9923 * API, because libbpf never returns NULL on success and it now always
9924 * sets errno on error. So no extra errno handling for ptr == NULL
9925 * case.
9926 */
9927 return -errno;
9928 }
9929
9930 /* Replace link's underlying BPF program with the new one */
9931 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
9932 {
9933 int ret;
9934
9935 ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
9936 return libbpf_err_errno(ret);
9937 }
9938
9939 /* Release "ownership" of underlying BPF resource (typically, BPF program
9940 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
9941 * link, when destructed through bpf_link__destroy() call won't attempt to
9942 * detach/unregisted that BPF resource. This is useful in situations where,
9943 * say, attached BPF program has to outlive userspace program that attached it
9944 * in the system. Depending on type of BPF program, though, there might be
9945 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
9946 * exit of userspace program doesn't trigger automatic detachment and clean up
9947 * inside the kernel.
9948 */
9949 void bpf_link__disconnect(struct bpf_link *link)
9950 {
9951 link->disconnected = true;
9952 }
9953
9954 int bpf_link__destroy(struct bpf_link *link)
9955 {
9956 int err = 0;
9957
9958 if (IS_ERR_OR_NULL(link))
9959 return 0;
9960
9961 if (!link->disconnected && link->detach)
9962 err = link->detach(link);
9963 if (link->pin_path)
9964 free(link->pin_path);
9965 if (link->dealloc)
9966 link->dealloc(link);
9967 else
9968 free(link);
9969
9970 return libbpf_err(err);
9971 }
9972
9973 int bpf_link__fd(const struct bpf_link *link)
9974 {
9975 return link->fd;
9976 }
9977
9978 const char *bpf_link__pin_path(const struct bpf_link *link)
9979 {
9980 return link->pin_path;
9981 }
9982
9983 static int bpf_link__detach_fd(struct bpf_link *link)
9984 {
9985 return libbpf_err_errno(close(link->fd));
9986 }
9987
9988 struct bpf_link *bpf_link__open(const char *path)
9989 {
9990 struct bpf_link *link;
9991 int fd;
9992
9993 fd = bpf_obj_get(path);
9994 if (fd < 0) {
9995 fd = -errno;
9996 pr_warn("failed to open link at %s: %d\n", path, fd);
9997 return libbpf_err_ptr(fd);
9998 }
9999
10000 link = calloc(1, sizeof(*link));
10001 if (!link) {
10002 close(fd);
10003 return libbpf_err_ptr(-ENOMEM);
10004 }
10005 link->detach = &bpf_link__detach_fd;
10006 link->fd = fd;
10007
10008 link->pin_path = strdup(path);
10009 if (!link->pin_path) {
10010 bpf_link__destroy(link);
10011 return libbpf_err_ptr(-ENOMEM);
10012 }
10013
10014 return link;
10015 }
10016
10017 int bpf_link__detach(struct bpf_link *link)
10018 {
10019 return bpf_link_detach(link->fd) ? -errno : 0;
10020 }
10021
10022 int bpf_link__pin(struct bpf_link *link, const char *path)
10023 {
10024 int err;
10025
10026 if (link->pin_path)
10027 return libbpf_err(-EBUSY);
10028 err = make_parent_dir(path);
10029 if (err)
10030 return libbpf_err(err);
10031 err = check_path(path);
10032 if (err)
10033 return libbpf_err(err);
10034
10035 link->pin_path = strdup(path);
10036 if (!link->pin_path)
10037 return libbpf_err(-ENOMEM);
10038
10039 if (bpf_obj_pin(link->fd, link->pin_path)) {
10040 err = -errno;
10041 zfree(&link->pin_path);
10042 return libbpf_err(err);
10043 }
10044
10045 pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
10046 return 0;
10047 }
10048
10049 int bpf_link__unpin(struct bpf_link *link)
10050 {
10051 int err;
10052
10053 if (!link->pin_path)
10054 return libbpf_err(-EINVAL);
10055
10056 err = unlink(link->pin_path);
10057 if (err != 0)
10058 return -errno;
10059
10060 pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
10061 zfree(&link->pin_path);
10062 return 0;
10063 }
10064
10065 struct bpf_link_perf {
10066 struct bpf_link link;
10067 int perf_event_fd;
10068 /* legacy kprobe support: keep track of probe identifier and type */
10069 char *legacy_probe_name;
10070 bool legacy_is_kprobe;
10071 bool legacy_is_retprobe;
10072 };
10073
10074 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
10075 static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
10076
10077 static int bpf_link_perf_detach(struct bpf_link *link)
10078 {
10079 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10080 int err = 0;
10081
10082 if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
10083 err = -errno;
10084
10085 if (perf_link->perf_event_fd != link->fd)
10086 close(perf_link->perf_event_fd);
10087 close(link->fd);
10088
10089 /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
10090 if (perf_link->legacy_probe_name) {
10091 if (perf_link->legacy_is_kprobe) {
10092 err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
10093 perf_link->legacy_is_retprobe);
10094 } else {
10095 err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
10096 perf_link->legacy_is_retprobe);
10097 }
10098 }
10099
10100 return err;
10101 }
10102
10103 static void bpf_link_perf_dealloc(struct bpf_link *link)
10104 {
10105 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10106
10107 free(perf_link->legacy_probe_name);
10108 free(perf_link);
10109 }
10110
10111 struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
10112 const struct bpf_perf_event_opts *opts)
10113 {
10114 char errmsg[STRERR_BUFSIZE];
10115 struct bpf_link_perf *link;
10116 int prog_fd, link_fd = -1, err;
10117
10118 if (!OPTS_VALID(opts, bpf_perf_event_opts))
10119 return libbpf_err_ptr(-EINVAL);
10120
10121 if (pfd < 0) {
10122 pr_warn("prog '%s': invalid perf event FD %d\n",
10123 prog->name, pfd);
10124 return libbpf_err_ptr(-EINVAL);
10125 }
10126 prog_fd = bpf_program__fd(prog);
10127 if (prog_fd < 0) {
10128 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
10129 prog->name);
10130 return libbpf_err_ptr(-EINVAL);
10131 }
10132
10133 link = calloc(1, sizeof(*link));
10134 if (!link)
10135 return libbpf_err_ptr(-ENOMEM);
10136 link->link.detach = &bpf_link_perf_detach;
10137 link->link.dealloc = &bpf_link_perf_dealloc;
10138 link->perf_event_fd = pfd;
10139
10140 if (kernel_supports(prog->obj, FEAT_PERF_LINK)) {
10141 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
10142 .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
10143
10144 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
10145 if (link_fd < 0) {
10146 err = -errno;
10147 pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n",
10148 prog->name, pfd,
10149 err, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10150 goto err_out;
10151 }
10152 link->link.fd = link_fd;
10153 } else {
10154 if (OPTS_GET(opts, bpf_cookie, 0)) {
10155 pr_warn("prog '%s': user context value is not supported\n", prog->name);
10156 err = -EOPNOTSUPP;
10157 goto err_out;
10158 }
10159
10160 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
10161 err = -errno;
10162 pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
10163 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10164 if (err == -EPROTO)
10165 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
10166 prog->name, pfd);
10167 goto err_out;
10168 }
10169 link->link.fd = pfd;
10170 }
10171 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
10172 err = -errno;
10173 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
10174 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10175 goto err_out;
10176 }
10177
10178 return &link->link;
10179 err_out:
10180 if (link_fd >= 0)
10181 close(link_fd);
10182 free(link);
10183 return libbpf_err_ptr(err);
10184 }
10185
10186 struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
10187 {
10188 return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
10189 }
10190
10191 /*
10192 * this function is expected to parse integer in the range of [0, 2^31-1] from
10193 * given file using scanf format string fmt. If actual parsed value is
10194 * negative, the result might be indistinguishable from error
10195 */
10196 static int parse_uint_from_file(const char *file, const char *fmt)
10197 {
10198 char buf[STRERR_BUFSIZE];
10199 int err, ret;
10200 FILE *f;
10201
10202 f = fopen(file, "r");
10203 if (!f) {
10204 err = -errno;
10205 pr_debug("failed to open '%s': %s\n", file,
10206 libbpf_strerror_r(err, buf, sizeof(buf)));
10207 return err;
10208 }
10209 err = fscanf(f, fmt, &ret);
10210 if (err != 1) {
10211 err = err == EOF ? -EIO : -errno;
10212 pr_debug("failed to parse '%s': %s\n", file,
10213 libbpf_strerror_r(err, buf, sizeof(buf)));
10214 fclose(f);
10215 return err;
10216 }
10217 fclose(f);
10218 return ret;
10219 }
10220
10221 static int determine_kprobe_perf_type(void)
10222 {
10223 const char *file = "/sys/bus/event_source/devices/kprobe/type";
10224
10225 return parse_uint_from_file(file, "%d\n");
10226 }
10227
10228 static int determine_uprobe_perf_type(void)
10229 {
10230 const char *file = "/sys/bus/event_source/devices/uprobe/type";
10231
10232 return parse_uint_from_file(file, "%d\n");
10233 }
10234
10235 static int determine_kprobe_retprobe_bit(void)
10236 {
10237 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
10238
10239 return parse_uint_from_file(file, "config:%d\n");
10240 }
10241
10242 static int determine_uprobe_retprobe_bit(void)
10243 {
10244 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
10245
10246 return parse_uint_from_file(file, "config:%d\n");
10247 }
10248
10249 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
10250 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
10251
10252 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
10253 uint64_t offset, int pid, size_t ref_ctr_off)
10254 {
10255 const size_t attr_sz = sizeof(struct perf_event_attr);
10256 struct perf_event_attr attr;
10257 char errmsg[STRERR_BUFSIZE];
10258 int type, pfd;
10259
10260 if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
10261 return -EINVAL;
10262
10263 memset(&attr, 0, attr_sz);
10264
10265 type = uprobe ? determine_uprobe_perf_type()
10266 : determine_kprobe_perf_type();
10267 if (type < 0) {
10268 pr_warn("failed to determine %s perf type: %s\n",
10269 uprobe ? "uprobe" : "kprobe",
10270 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
10271 return type;
10272 }
10273 if (retprobe) {
10274 int bit = uprobe ? determine_uprobe_retprobe_bit()
10275 : determine_kprobe_retprobe_bit();
10276
10277 if (bit < 0) {
10278 pr_warn("failed to determine %s retprobe bit: %s\n",
10279 uprobe ? "uprobe" : "kprobe",
10280 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
10281 return bit;
10282 }
10283 attr.config |= 1 << bit;
10284 }
10285 attr.size = attr_sz;
10286 attr.type = type;
10287 attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
10288 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
10289 attr.config2 = offset; /* kprobe_addr or probe_offset */
10290
10291 /* pid filter is meaningful only for uprobes */
10292 pfd = syscall(__NR_perf_event_open, &attr,
10293 pid < 0 ? -1 : pid /* pid */,
10294 pid == -1 ? 0 : -1 /* cpu */,
10295 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10296 return pfd >= 0 ? pfd : -errno;
10297 }
10298
10299 static int append_to_file(const char *file, const char *fmt, ...)
10300 {
10301 int fd, n, err = 0;
10302 va_list ap;
10303
10304 fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
10305 if (fd < 0)
10306 return -errno;
10307
10308 va_start(ap, fmt);
10309 n = vdprintf(fd, fmt, ap);
10310 va_end(ap);
10311
10312 if (n < 0)
10313 err = -errno;
10314
10315 close(fd);
10316 return err;
10317 }
10318
10319 #define DEBUGFS "/sys/kernel/debug/tracing"
10320 #define TRACEFS "/sys/kernel/tracing"
10321
10322 static bool use_debugfs(void)
10323 {
10324 static int has_debugfs = -1;
10325
10326 if (has_debugfs < 0)
10327 has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0;
10328
10329 return has_debugfs == 1;
10330 }
10331
10332 static const char *tracefs_path(void)
10333 {
10334 return use_debugfs() ? DEBUGFS : TRACEFS;
10335 }
10336
10337 static const char *tracefs_kprobe_events(void)
10338 {
10339 return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events";
10340 }
10341
10342 static const char *tracefs_uprobe_events(void)
10343 {
10344 return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events";
10345 }
10346
10347 static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz,
10348 const char *kfunc_name, size_t offset)
10349 {
10350 static int index = 0;
10351
10352 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset,
10353 __sync_fetch_and_add(&index, 1));
10354 }
10355
10356 static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
10357 const char *kfunc_name, size_t offset)
10358 {
10359 return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx",
10360 retprobe ? 'r' : 'p',
10361 retprobe ? "kretprobes" : "kprobes",
10362 probe_name, kfunc_name, offset);
10363 }
10364
10365 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
10366 {
10367 return append_to_file(tracefs_kprobe_events(), "-:%s/%s",
10368 retprobe ? "kretprobes" : "kprobes", probe_name);
10369 }
10370
10371 static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10372 {
10373 char file[256];
10374
10375 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
10376 tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name);
10377
10378 return parse_uint_from_file(file, "%d\n");
10379 }
10380
10381 static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
10382 const char *kfunc_name, size_t offset, int pid)
10383 {
10384 const size_t attr_sz = sizeof(struct perf_event_attr);
10385 struct perf_event_attr attr;
10386 char errmsg[STRERR_BUFSIZE];
10387 int type, pfd, err;
10388
10389 err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
10390 if (err < 0) {
10391 pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
10392 kfunc_name, offset,
10393 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10394 return err;
10395 }
10396 type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
10397 if (type < 0) {
10398 err = type;
10399 pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
10400 kfunc_name, offset,
10401 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10402 goto err_clean_legacy;
10403 }
10404
10405 memset(&attr, 0, attr_sz);
10406 attr.size = attr_sz;
10407 attr.config = type;
10408 attr.type = PERF_TYPE_TRACEPOINT;
10409
10410 pfd = syscall(__NR_perf_event_open, &attr,
10411 pid < 0 ? -1 : pid, /* pid */
10412 pid == -1 ? 0 : -1, /* cpu */
10413 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10414 if (pfd < 0) {
10415 err = -errno;
10416 pr_warn("legacy kprobe perf_event_open() failed: %s\n",
10417 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10418 goto err_clean_legacy;
10419 }
10420 return pfd;
10421
10422 err_clean_legacy:
10423 /* Clear the newly added legacy kprobe_event */
10424 remove_kprobe_event_legacy(probe_name, retprobe);
10425 return err;
10426 }
10427
10428 static const char *arch_specific_syscall_pfx(void)
10429 {
10430 #if defined(__x86_64__)
10431 return "x64";
10432 #elif defined(__i386__)
10433 return "ia32";
10434 #elif defined(__s390x__)
10435 return "s390x";
10436 #elif defined(__s390__)
10437 return "s390";
10438 #elif defined(__arm__)
10439 return "arm";
10440 #elif defined(__aarch64__)
10441 return "arm64";
10442 #elif defined(__mips__)
10443 return "mips";
10444 #elif defined(__riscv)
10445 return "riscv";
10446 #elif defined(__powerpc__)
10447 return "powerpc";
10448 #elif defined(__powerpc64__)
10449 return "powerpc64";
10450 #else
10451 return NULL;
10452 #endif
10453 }
10454
10455 static int probe_kern_syscall_wrapper(void)
10456 {
10457 char syscall_name[64];
10458 const char *ksys_pfx;
10459
10460 ksys_pfx = arch_specific_syscall_pfx();
10461 if (!ksys_pfx)
10462 return 0;
10463
10464 snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx);
10465
10466 if (determine_kprobe_perf_type() >= 0) {
10467 int pfd;
10468
10469 pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0);
10470 if (pfd >= 0)
10471 close(pfd);
10472
10473 return pfd >= 0 ? 1 : 0;
10474 } else { /* legacy mode */
10475 char probe_name[128];
10476
10477 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0);
10478 if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0)
10479 return 0;
10480
10481 (void)remove_kprobe_event_legacy(probe_name, false);
10482 return 1;
10483 }
10484 }
10485
10486 struct bpf_link *
10487 bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
10488 const char *func_name,
10489 const struct bpf_kprobe_opts *opts)
10490 {
10491 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10492 char errmsg[STRERR_BUFSIZE];
10493 char *legacy_probe = NULL;
10494 struct bpf_link *link;
10495 size_t offset;
10496 bool retprobe, legacy;
10497 int pfd, err;
10498
10499 if (!OPTS_VALID(opts, bpf_kprobe_opts))
10500 return libbpf_err_ptr(-EINVAL);
10501
10502 retprobe = OPTS_GET(opts, retprobe, false);
10503 offset = OPTS_GET(opts, offset, 0);
10504 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10505
10506 legacy = determine_kprobe_perf_type() < 0;
10507 if (!legacy) {
10508 pfd = perf_event_open_probe(false /* uprobe */, retprobe,
10509 func_name, offset,
10510 -1 /* pid */, 0 /* ref_ctr_off */);
10511 } else {
10512 char probe_name[256];
10513
10514 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name),
10515 func_name, offset);
10516
10517 legacy_probe = strdup(probe_name);
10518 if (!legacy_probe)
10519 return libbpf_err_ptr(-ENOMEM);
10520
10521 pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
10522 offset, -1 /* pid */);
10523 }
10524 if (pfd < 0) {
10525 err = -errno;
10526 pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
10527 prog->name, retprobe ? "kretprobe" : "kprobe",
10528 func_name, offset,
10529 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10530 goto err_out;
10531 }
10532 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10533 err = libbpf_get_error(link);
10534 if (err) {
10535 close(pfd);
10536 pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
10537 prog->name, retprobe ? "kretprobe" : "kprobe",
10538 func_name, offset,
10539 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10540 goto err_clean_legacy;
10541 }
10542 if (legacy) {
10543 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10544
10545 perf_link->legacy_probe_name = legacy_probe;
10546 perf_link->legacy_is_kprobe = true;
10547 perf_link->legacy_is_retprobe = retprobe;
10548 }
10549
10550 return link;
10551
10552 err_clean_legacy:
10553 if (legacy)
10554 remove_kprobe_event_legacy(legacy_probe, retprobe);
10555 err_out:
10556 free(legacy_probe);
10557 return libbpf_err_ptr(err);
10558 }
10559
10560 struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
10561 bool retprobe,
10562 const char *func_name)
10563 {
10564 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
10565 .retprobe = retprobe,
10566 );
10567
10568 return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
10569 }
10570
10571 struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog,
10572 const char *syscall_name,
10573 const struct bpf_ksyscall_opts *opts)
10574 {
10575 LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts);
10576 char func_name[128];
10577
10578 if (!OPTS_VALID(opts, bpf_ksyscall_opts))
10579 return libbpf_err_ptr(-EINVAL);
10580
10581 if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) {
10582 /* arch_specific_syscall_pfx() should never return NULL here
10583 * because it is guarded by kernel_supports(). However, since
10584 * compiler does not know that we have an explicit conditional
10585 * as well.
10586 */
10587 snprintf(func_name, sizeof(func_name), "__%s_sys_%s",
10588 arch_specific_syscall_pfx() ? : "", syscall_name);
10589 } else {
10590 snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name);
10591 }
10592
10593 kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false);
10594 kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10595
10596 return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts);
10597 }
10598
10599 /* Adapted from perf/util/string.c */
10600 static bool glob_match(const char *str, const char *pat)
10601 {
10602 while (*str && *pat && *pat != '*') {
10603 if (*pat == '?') { /* Matches any single character */
10604 str++;
10605 pat++;
10606 continue;
10607 }
10608 if (*str != *pat)
10609 return false;
10610 str++;
10611 pat++;
10612 }
10613 /* Check wild card */
10614 if (*pat == '*') {
10615 while (*pat == '*')
10616 pat++;
10617 if (!*pat) /* Tail wild card matches all */
10618 return true;
10619 while (*str)
10620 if (glob_match(str++, pat))
10621 return true;
10622 }
10623 return !*str && !*pat;
10624 }
10625
10626 struct kprobe_multi_resolve {
10627 const char *pattern;
10628 unsigned long *addrs;
10629 size_t cap;
10630 size_t cnt;
10631 };
10632
10633 static int
10634 resolve_kprobe_multi_cb(unsigned long long sym_addr, char sym_type,
10635 const char *sym_name, void *ctx)
10636 {
10637 struct kprobe_multi_resolve *res = ctx;
10638 int err;
10639
10640 if (!glob_match(sym_name, res->pattern))
10641 return 0;
10642
10643 err = libbpf_ensure_mem((void **) &res->addrs, &res->cap, sizeof(unsigned long),
10644 res->cnt + 1);
10645 if (err)
10646 return err;
10647
10648 res->addrs[res->cnt++] = (unsigned long) sym_addr;
10649 return 0;
10650 }
10651
10652 struct bpf_link *
10653 bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog,
10654 const char *pattern,
10655 const struct bpf_kprobe_multi_opts *opts)
10656 {
10657 LIBBPF_OPTS(bpf_link_create_opts, lopts);
10658 struct kprobe_multi_resolve res = {
10659 .pattern = pattern,
10660 };
10661 struct bpf_link *link = NULL;
10662 char errmsg[STRERR_BUFSIZE];
10663 const unsigned long *addrs;
10664 int err, link_fd, prog_fd;
10665 const __u64 *cookies;
10666 const char **syms;
10667 bool retprobe;
10668 size_t cnt;
10669
10670 if (!OPTS_VALID(opts, bpf_kprobe_multi_opts))
10671 return libbpf_err_ptr(-EINVAL);
10672
10673 syms = OPTS_GET(opts, syms, false);
10674 addrs = OPTS_GET(opts, addrs, false);
10675 cnt = OPTS_GET(opts, cnt, false);
10676 cookies = OPTS_GET(opts, cookies, false);
10677
10678 if (!pattern && !addrs && !syms)
10679 return libbpf_err_ptr(-EINVAL);
10680 if (pattern && (addrs || syms || cookies || cnt))
10681 return libbpf_err_ptr(-EINVAL);
10682 if (!pattern && !cnt)
10683 return libbpf_err_ptr(-EINVAL);
10684 if (addrs && syms)
10685 return libbpf_err_ptr(-EINVAL);
10686
10687 if (pattern) {
10688 err = libbpf_kallsyms_parse(resolve_kprobe_multi_cb, &res);
10689 if (err)
10690 goto error;
10691 if (!res.cnt) {
10692 err = -ENOENT;
10693 goto error;
10694 }
10695 addrs = res.addrs;
10696 cnt = res.cnt;
10697 }
10698
10699 retprobe = OPTS_GET(opts, retprobe, false);
10700
10701 lopts.kprobe_multi.syms = syms;
10702 lopts.kprobe_multi.addrs = addrs;
10703 lopts.kprobe_multi.cookies = cookies;
10704 lopts.kprobe_multi.cnt = cnt;
10705 lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0;
10706
10707 link = calloc(1, sizeof(*link));
10708 if (!link) {
10709 err = -ENOMEM;
10710 goto error;
10711 }
10712 link->detach = &bpf_link__detach_fd;
10713
10714 prog_fd = bpf_program__fd(prog);
10715 link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_KPROBE_MULTI, &lopts);
10716 if (link_fd < 0) {
10717 err = -errno;
10718 pr_warn("prog '%s': failed to attach: %s\n",
10719 prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10720 goto error;
10721 }
10722 link->fd = link_fd;
10723 free(res.addrs);
10724 return link;
10725
10726 error:
10727 free(link);
10728 free(res.addrs);
10729 return libbpf_err_ptr(err);
10730 }
10731
10732 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10733 {
10734 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
10735 unsigned long offset = 0;
10736 const char *func_name;
10737 char *func;
10738 int n;
10739
10740 *link = NULL;
10741
10742 /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */
10743 if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0)
10744 return 0;
10745
10746 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
10747 if (opts.retprobe)
10748 func_name = prog->sec_name + sizeof("kretprobe/") - 1;
10749 else
10750 func_name = prog->sec_name + sizeof("kprobe/") - 1;
10751
10752 n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
10753 if (n < 1) {
10754 pr_warn("kprobe name is invalid: %s\n", func_name);
10755 return -EINVAL;
10756 }
10757 if (opts.retprobe && offset != 0) {
10758 free(func);
10759 pr_warn("kretprobes do not support offset specification\n");
10760 return -EINVAL;
10761 }
10762
10763 opts.offset = offset;
10764 *link = bpf_program__attach_kprobe_opts(prog, func, &opts);
10765 free(func);
10766 return libbpf_get_error(*link);
10767 }
10768
10769 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10770 {
10771 LIBBPF_OPTS(bpf_ksyscall_opts, opts);
10772 const char *syscall_name;
10773
10774 *link = NULL;
10775
10776 /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */
10777 if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0)
10778 return 0;
10779
10780 opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/");
10781 if (opts.retprobe)
10782 syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1;
10783 else
10784 syscall_name = prog->sec_name + sizeof("ksyscall/") - 1;
10785
10786 *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts);
10787 return *link ? 0 : -errno;
10788 }
10789
10790 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10791 {
10792 LIBBPF_OPTS(bpf_kprobe_multi_opts, opts);
10793 const char *spec;
10794 char *pattern;
10795 int n;
10796
10797 *link = NULL;
10798
10799 /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */
10800 if (strcmp(prog->sec_name, "kprobe.multi") == 0 ||
10801 strcmp(prog->sec_name, "kretprobe.multi") == 0)
10802 return 0;
10803
10804 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/");
10805 if (opts.retprobe)
10806 spec = prog->sec_name + sizeof("kretprobe.multi/") - 1;
10807 else
10808 spec = prog->sec_name + sizeof("kprobe.multi/") - 1;
10809
10810 n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
10811 if (n < 1) {
10812 pr_warn("kprobe multi pattern is invalid: %s\n", pattern);
10813 return -EINVAL;
10814 }
10815
10816 *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
10817 free(pattern);
10818 return libbpf_get_error(*link);
10819 }
10820
10821 static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz,
10822 const char *binary_path, uint64_t offset)
10823 {
10824 int i;
10825
10826 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset);
10827
10828 /* sanitize binary_path in the probe name */
10829 for (i = 0; buf[i]; i++) {
10830 if (!isalnum(buf[i]))
10831 buf[i] = '_';
10832 }
10833 }
10834
10835 static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
10836 const char *binary_path, size_t offset)
10837 {
10838 return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx",
10839 retprobe ? 'r' : 'p',
10840 retprobe ? "uretprobes" : "uprobes",
10841 probe_name, binary_path, offset);
10842 }
10843
10844 static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
10845 {
10846 return append_to_file(tracefs_uprobe_events(), "-:%s/%s",
10847 retprobe ? "uretprobes" : "uprobes", probe_name);
10848 }
10849
10850 static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10851 {
10852 char file[512];
10853
10854 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
10855 tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name);
10856
10857 return parse_uint_from_file(file, "%d\n");
10858 }
10859
10860 static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
10861 const char *binary_path, size_t offset, int pid)
10862 {
10863 const size_t attr_sz = sizeof(struct perf_event_attr);
10864 struct perf_event_attr attr;
10865 int type, pfd, err;
10866
10867 err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
10868 if (err < 0) {
10869 pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n",
10870 binary_path, (size_t)offset, err);
10871 return err;
10872 }
10873 type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
10874 if (type < 0) {
10875 err = type;
10876 pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n",
10877 binary_path, offset, err);
10878 goto err_clean_legacy;
10879 }
10880
10881 memset(&attr, 0, attr_sz);
10882 attr.size = attr_sz;
10883 attr.config = type;
10884 attr.type = PERF_TYPE_TRACEPOINT;
10885
10886 pfd = syscall(__NR_perf_event_open, &attr,
10887 pid < 0 ? -1 : pid, /* pid */
10888 pid == -1 ? 0 : -1, /* cpu */
10889 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10890 if (pfd < 0) {
10891 err = -errno;
10892 pr_warn("legacy uprobe perf_event_open() failed: %d\n", err);
10893 goto err_clean_legacy;
10894 }
10895 return pfd;
10896
10897 err_clean_legacy:
10898 /* Clear the newly added legacy uprobe_event */
10899 remove_uprobe_event_legacy(probe_name, retprobe);
10900 return err;
10901 }
10902
10903 /* Return next ELF section of sh_type after scn, or first of that type if scn is NULL. */
10904 #ifdef HAVE_LIBELF
10905 static Elf_Scn *elf_find_next_scn_by_type(Elf *elf, int sh_type, Elf_Scn *scn)
10906 {
10907 while ((scn = elf_nextscn(elf, scn)) != NULL) {
10908 GElf_Shdr sh;
10909
10910 if (!gelf_getshdr(scn, &sh))
10911 continue;
10912 if (sh.sh_type == sh_type)
10913 return scn;
10914 }
10915 return NULL;
10916 }
10917 #elif HAVE_ELFIO
10918 static psection_t elf_find_next_scn_by_type(pelfio_t pelfio, int sh_type, psection_t pscn)
10919 {
10920 int secno = elfio_get_sections_num(pelfio);
10921 int j = 0;
10922 if (pscn != NULL) {
10923 for (int i = 0; i < secno; i++) {
10924 psection_t psection = elfio_get_section_by_index(pelfio, i);
10925 if (psection == pscn) {
10926 j = i;
10927 }
10928 }
10929 }
10930 for (; j < secno; j++) {
10931 psection_t psection = elfio_get_section_by_index(pelfio, j);
10932 Elf_Word sec_type = elfio_section_get_type(psection);
10933 if (sec_type == sh_type) {
10934 return psection;
10935 }
10936 }
10937 return NULL;
10938 }
10939 #endif
10940
10941 /* Find offset of function name in object specified by path. "name" matches
10942 * symbol name or name@@LIB for library functions.
10943 */
10944 static long elf_find_func_offset(const char *binary_path, const char *name)
10945 {
10946 int fd, i, sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB };
10947 bool is_shared_lib, is_name_qualified;
10948 char errmsg[STRERR_BUFSIZE];
10949 long ret = -ENOENT;
10950 size_t name_len;
10951 #ifdef HAVE_LIBELF
10952 GElf_Ehdr ehdr;
10953 Elf *elf;
10954 #elif HAVE_ELFIO
10955 pelfio_t pelfio = elfio_new();
10956 bool is_load = elfio_load(pelfio, binary_path);
10957 #endif //
10958 #ifdef HAVE_LIBELF
10959 fd = open(binary_path, O_RDONLY | O_CLOEXEC);
10960 if (fd < 0) {
10961 #elif HAVE_ELFIO
10962 if (!is_load) {
10963 #endif //
10964 ret = -errno;
10965 pr_warn("failed to open %s: %s\n", binary_path,
10966 libbpf_strerror_r(ret, errmsg, sizeof(errmsg)));
10967 return ret;
10968 }
10969 #ifdef HAVE_LIBELF
10970 elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
10971 if (!elf) {
10972 pr_warn("elf: could not read elf from %s: %s\n", binary_path, elf_errmsg(-1));
10973 close(fd);
10974 return -LIBBPF_ERRNO__FORMAT;
10975 }
10976 if (!gelf_getehdr(elf, &ehdr)) {
10977 pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1));
10978 ret = -LIBBPF_ERRNO__FORMAT;
10979 goto out;
10980 }
10981 /* for shared lib case, we do not need to calculate relative offset */
10982 is_shared_lib = ehdr.e_type == ET_DYN;
10983 #elif HAVA_ELFIO
10984 is_shared_lib = (ET_DYN == elfio_get_type(pelfio));
10985 #endif //HAVE_LIBELF
10986
10987 name_len = strlen(name);
10988 /* Does name specify "@@LIB"? */
10989 is_name_qualified = strstr(name, "@@") != NULL;
10990
10991 /* Search SHT_DYNSYM, SHT_SYMTAB for symbol. This search order is used because if
10992 * a binary is stripped, it may only have SHT_DYNSYM, and a fully-statically
10993 * linked binary may not have SHT_DYMSYM, so absence of a section should not be
10994 * reported as a warning/error.
10995 */
10996 for (i = 0; i < ARRAY_SIZE(sh_types); i++) {
10997 size_t nr_syms, strtabidx, idx;
10998 #ifdef HAVE_LIBELF
10999 Elf_Data *symbols = NULL;
11000 Elf_Scn *scn = NULL;
11001 GElf_Shdr sh;
11002 #elif HAVE_ELFIO
11003 Elf_Data realData = {};
11004 psection_t pSec = NULL;
11005 #endif //HAVE_LIBELF
11006 int last_bind = -1;
11007 const char *sname;
11008 #ifdef HAVE_LIBELF
11009 scn = elf_find_next_scn_by_type(elf, sh_types[i], NULL);
11010 #elif HAVE_ELFIO
11011 pSec = elf_find_next_scn_by_type(pelfio, sh_types[i], NULL);
11012 #endif //HAVE_LIBELF
11013 #ifdef HAVE_LIBELF
11014 if (!scn) {
11015 #elif HAVE_ELFIO
11016 if (!pSec) {
11017 #endif //HAVE_LIBELF
11018 pr_debug("elf: failed to find symbol table ELF sections in '%s'\n",
11019 binary_path);
11020 continue;
11021 }
11022 #ifdef HAVE_LIBELF
11023 if (!gelf_getshdr(scn, &sh))
11024 continue;
11025 strtabidx = sh.sh_link;
11026 #elif HAVE_ELFIO
11027 strtabidx = elfio_section_get_link(pSec);
11028 #endif //
11029 #ifdef HAVE_LIBELF
11030 symbols = elf_getdata(scn, 0);
11031 if (!symbols) {
11032 pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n",
11033 binary_path, elf_errmsg(-1));
11034 ret = -LIBBPF_ERRNO__FORMAT;
11035 goto out;
11036 }
11037 nr_syms = symbols->d_size / sh.sh_entsize;
11038 #elif HAVE_ELFIO
11039 realData.d_buf = (void*)elfio_section_get_data(pSec);
11040 realData.d_size = elfio_section_get_size(pSec);
11041 if (!realData.d_buf) {
11042 pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n",
11043 binary_path, elf_errmsg(-1));
11044 ret = -LIBBPF_ERRNO__FORMAT;
11045 goto out;
11046 }
11047 nr_syms = realData.d_size / elfio_section_get_entry_size(pSec);
11048 psymbol_t psymbols;
11049 psymbols = elfio_symbol_section_accessor_new(pelfio, pSec);
11050 if (!psymbols) {
11051 pr_warn("elf: failed to get symbols in '%s': %s\n",
11052 binary_path, elf_errmsg(-1));
11053 ret = -LIBBPF_ERRNO__FORMAT;
11054 goto out;
11055 }
11056 #endif //HAVE_LIBELF
11057 for (idx = 0; idx < nr_syms; idx++) {
11058 int curr_bind;
11059 #ifdef HAVE_LIBELF
11060 GElf_Sym sym;
11061 Elf_Scn *sym_scn;
11062 GElf_Shdr sym_sh;
11063
11064 if (!gelf_getsym(symbols, idx, &sym))
11065 continue;
11066
11067 if (GELF_ST_TYPE(sym.st_info) != STT_FUNC)
11068 continue;
11069
11070 sname = elf_strptr(elf, strtabidx, sym.st_name);
11071 if (!sname)
11072 continue;
11073
11074 curr_bind = GELF_ST_BIND(sym.st_info);
11075 #elif HAVE_ELFIO
11076 char sname[128] = {0};
11077 Elf64_Addr st_value;
11078 Elf_Xword size;
11079 unsigned char bind;
11080 unsigned char type;
11081 Elf_Half section_index;
11082 unsigned char other;
11083 elfio_symbol_get_symbol( psymbols, idx, sname, 128, &st_value, &size, &bind,
11084 &type, §ion_index, &other );
11085 if (type == STT_FUNC)
11086 continue;
11087 if (sname[0] == '\0')
11088 continue;
11089 curr_bind = bind;
11090 #endif //HAVE_LIBELF
11091
11092
11093 /* User can specify func, func@@LIB or func@@LIB_VERSION. */
11094 if (strncmp(sname, name, name_len) != 0)
11095 continue;
11096 /* ...but we don't want a search for "foo" to match 'foo2" also, so any
11097 * additional characters in sname should be of the form "@@LIB".
11098 */
11099 if (!is_name_qualified && sname[name_len] != '\0' && sname[name_len] != '@')
11100 continue;
11101
11102 if (ret >= 0) {
11103 /* handle multiple matches */
11104 if (last_bind != STB_WEAK && curr_bind != STB_WEAK) {
11105 /* Only accept one non-weak bind. */
11106 pr_warn("elf: ambiguous match for '%s', '%s' in '%s'\n",
11107 sname, name, binary_path);
11108 ret = -LIBBPF_ERRNO__FORMAT;
11109 goto out;
11110 } else if (curr_bind == STB_WEAK) {
11111 /* already have a non-weak bind, and
11112 * this is a weak bind, so ignore.
11113 */
11114 continue;
11115 }
11116 }
11117
11118 /* Transform symbol's virtual address (absolute for
11119 * binaries and relative for shared libs) into file
11120 * offset, which is what kernel is expecting for
11121 * uprobe/uretprobe attachment.
11122 * See Documentation/trace/uprobetracer.rst for more
11123 * details.
11124 * This is done by looking up symbol's containing
11125 * section's header and using it's virtual address
11126 * (sh_addr) and corresponding file offset (sh_offset)
11127 * to transform sym.st_value (virtual address) into
11128 * desired final file offset.
11129 */
11130 #ifdef HAVE_LIBELF
11131 sym_scn = elf_getscn(elf, sym.st_shndx);
11132 if (!sym_scn)
11133 continue;
11134 if (!gelf_getshdr(sym_scn, &sym_sh))
11135 continue;
11136 ret = sym.st_value - sym_sh.sh_addr + sym_sh.sh_offset;
11137 #elif HAVE_ELFIO
11138 psection_t pSymSection= NULL;
11139 pSymSection = elfio_get_section_by_index(pelfio,section_index);
11140 if (!pSymSection)
11141 continue;
11142 Elf64_Addr sh_addr = elfio_section_get_address(pSymSection);
11143 Elf64_Off sh_offset = elfio_section_get_offset(pSymSection);
11144 ret = st_value - sh_addr + sh_offset;
11145 #endif //HAVE_LIBELF
11146 last_bind = curr_bind;
11147 }
11148 #ifdef HAVE_ELFIO
11149 elfio_symbol_section_accessor_delete(psymbols);
11150 #endif //HAVE_ELFIO
11151 if (ret > 0)
11152 break;
11153 }
11154
11155 if (ret > 0) {
11156 pr_debug("elf: symbol address match for '%s' in '%s': 0x%lx\n", name, binary_path,
11157 ret);
11158 } else {
11159 if (ret == 0) {
11160 pr_warn("elf: '%s' is 0 in symtab for '%s': %s\n", name, binary_path,
11161 is_shared_lib ? "should not be 0 in a shared library" :
11162 "try using shared library path instead");
11163 ret = -ENOENT;
11164 } else {
11165 pr_warn("elf: failed to find symbol '%s' in '%s'\n", name, binary_path);
11166 }
11167 }
11168 out:
11169 #ifdef HAVE_LIBELF
11170 elf_end(elf);
11171 close(fd);
11172 #elif HAVE_ELFIO
11173 elfio_delete(pelfio);
11174 #endif //HAVE_LIBELF
11175 return ret;
11176 }
11177
11178
11179
11180 static const char *arch_specific_lib_paths(void)
11181 {
11182 /*
11183 * Based on https://packages.debian.org/sid/libc6.
11184 *
11185 * Assume that the traced program is built for the same architecture
11186 * as libbpf, which should cover the vast majority of cases.
11187 */
11188 #if defined(__x86_64__)
11189 return "/lib/x86_64-linux-gnu";
11190 #elif defined(__i386__)
11191 return "/lib/i386-linux-gnu";
11192 #elif defined(__s390x__)
11193 return "/lib/s390x-linux-gnu";
11194 #elif defined(__s390__)
11195 return "/lib/s390-linux-gnu";
11196 #elif defined(__arm__) && defined(__SOFTFP__)
11197 return "/lib/arm-linux-gnueabi";
11198 #elif defined(__arm__) && !defined(__SOFTFP__)
11199 return "/lib/arm-linux-gnueabihf";
11200 #elif defined(__aarch64__)
11201 return "/lib/aarch64-linux-gnu";
11202 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64
11203 return "/lib/mips64el-linux-gnuabi64";
11204 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32
11205 return "/lib/mipsel-linux-gnu";
11206 #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
11207 return "/lib/powerpc64le-linux-gnu";
11208 #elif defined(__sparc__) && defined(__arch64__)
11209 return "/lib/sparc64-linux-gnu";
11210 #elif defined(__riscv) && __riscv_xlen == 64
11211 return "/lib/riscv64-linux-gnu";
11212 #else
11213 return NULL;
11214 #endif
11215 }
11216
11217 /* Get full path to program/shared library. */
11218 static int resolve_full_path(const char *file, char *result, size_t result_sz)
11219 {
11220 const char *search_paths[3] = {};
11221 int i, perm;
11222
11223 if (str_has_sfx(file, ".so") || strstr(file, ".so.")) {
11224 search_paths[0] = getenv("LD_LIBRARY_PATH");
11225 search_paths[1] = "/usr/lib64:/usr/lib";
11226 search_paths[2] = arch_specific_lib_paths();
11227 perm = R_OK;
11228 } else {
11229 search_paths[0] = getenv("PATH");
11230 search_paths[1] = "/usr/bin:/usr/sbin";
11231 perm = R_OK | X_OK;
11232 }
11233
11234 for (i = 0; i < ARRAY_SIZE(search_paths); i++) {
11235 const char *s;
11236
11237 if (!search_paths[i])
11238 continue;
11239 for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) {
11240 char *next_path;
11241 int seg_len;
11242
11243 if (s[0] == ':')
11244 s++;
11245 next_path = strchr(s, ':');
11246 seg_len = next_path ? next_path - s : strlen(s);
11247 if (!seg_len)
11248 continue;
11249 snprintf(result, result_sz, "%.*s/%s", seg_len, s, file);
11250 /* ensure it has required permissions */
11251 if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0)
11252 continue;
11253 pr_debug("resolved '%s' to '%s'\n", file, result);
11254 return 0;
11255 }
11256 }
11257 return -ENOENT;
11258 }
11259
11260 LIBBPF_API struct bpf_link *
11261 bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
11262 const char *binary_path, size_t func_offset,
11263 const struct bpf_uprobe_opts *opts)
11264 {
11265 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11266 char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL;
11267 char full_binary_path[PATH_MAX];
11268 struct bpf_link *link;
11269 size_t ref_ctr_off;
11270 int pfd, err;
11271 bool retprobe, legacy;
11272 const char *func_name;
11273
11274 if (!OPTS_VALID(opts, bpf_uprobe_opts))
11275 return libbpf_err_ptr(-EINVAL);
11276
11277 retprobe = OPTS_GET(opts, retprobe, false);
11278 ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
11279 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11280
11281 if (!binary_path)
11282 return libbpf_err_ptr(-EINVAL);
11283
11284 if (!strchr(binary_path, '/')) {
11285 err = resolve_full_path(binary_path, full_binary_path,
11286 sizeof(full_binary_path));
11287 if (err) {
11288 pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11289 prog->name, binary_path, err);
11290 return libbpf_err_ptr(err);
11291 }
11292 binary_path = full_binary_path;
11293 }
11294 func_name = OPTS_GET(opts, func_name, NULL);
11295 if (func_name) {
11296 long sym_off;
11297
11298 sym_off = elf_find_func_offset(binary_path, func_name);
11299 if (sym_off < 0)
11300 return libbpf_err_ptr(sym_off);
11301 func_offset += sym_off;
11302 }
11303
11304 legacy = determine_uprobe_perf_type() < 0;
11305 if (!legacy) {
11306 pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
11307 func_offset, pid, ref_ctr_off);
11308 } else {
11309 char probe_name[PATH_MAX + 64];
11310
11311 if (ref_ctr_off)
11312 return libbpf_err_ptr(-EINVAL);
11313
11314 gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name),
11315 binary_path, func_offset);
11316
11317 legacy_probe = strdup(probe_name);
11318 if (!legacy_probe)
11319 return libbpf_err_ptr(-ENOMEM);
11320
11321 pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
11322 binary_path, func_offset, pid);
11323 }
11324 if (pfd < 0) {
11325 err = -errno;
11326 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
11327 prog->name, retprobe ? "uretprobe" : "uprobe",
11328 binary_path, func_offset,
11329 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11330 goto err_out;
11331 }
11332
11333 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11334 err = libbpf_get_error(link);
11335 if (err) {
11336 close(pfd);
11337 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
11338 prog->name, retprobe ? "uretprobe" : "uprobe",
11339 binary_path, func_offset,
11340 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11341 goto err_clean_legacy;
11342 }
11343 if (legacy) {
11344 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11345
11346 perf_link->legacy_probe_name = legacy_probe;
11347 perf_link->legacy_is_kprobe = false;
11348 perf_link->legacy_is_retprobe = retprobe;
11349 }
11350 return link;
11351
11352 err_clean_legacy:
11353 if (legacy)
11354 remove_uprobe_event_legacy(legacy_probe, retprobe);
11355 err_out:
11356 free(legacy_probe);
11357 return libbpf_err_ptr(err);
11358 }
11359
11360 /* Format of u[ret]probe section definition supporting auto-attach:
11361 * u[ret]probe/binary:function[+offset]
11362 *
11363 * binary can be an absolute/relative path or a filename; the latter is resolved to a
11364 * full binary path via bpf_program__attach_uprobe_opts.
11365 *
11366 * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be
11367 * specified (and auto-attach is not possible) or the above format is specified for
11368 * auto-attach.
11369 */
11370 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11371 {
11372 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts);
11373 char *probe_type = NULL, *binary_path = NULL, *func_name = NULL;
11374 int n, ret = -EINVAL;
11375 long offset = 0;
11376
11377 *link = NULL;
11378
11379 n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[a-zA-Z0-9_.]+%li",
11380 &probe_type, &binary_path, &func_name, &offset);
11381 switch (n) {
11382 case 1:
11383 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
11384 ret = 0;
11385 break;
11386 case 2:
11387 pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n",
11388 prog->name, prog->sec_name);
11389 break;
11390 case 3:
11391 case 4:
11392 opts.retprobe = strcmp(probe_type, "uretprobe") == 0 ||
11393 strcmp(probe_type, "uretprobe.s") == 0;
11394 if (opts.retprobe && offset != 0) {
11395 pr_warn("prog '%s': uretprobes do not support offset specification\n",
11396 prog->name);
11397 break;
11398 }
11399 opts.func_name = func_name;
11400 *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts);
11401 ret = libbpf_get_error(*link);
11402 break;
11403 default:
11404 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
11405 prog->sec_name);
11406 break;
11407 }
11408 free(probe_type);
11409 free(binary_path);
11410 free(func_name);
11411
11412 return ret;
11413 }
11414
11415 struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
11416 bool retprobe, pid_t pid,
11417 const char *binary_path,
11418 size_t func_offset)
11419 {
11420 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
11421
11422 return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
11423 }
11424
11425 #ifdef HAVE_LIBELF
11426 struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog,
11427 pid_t pid, const char *binary_path,
11428 const char *usdt_provider, const char *usdt_name,
11429 const struct bpf_usdt_opts *opts)
11430 {
11431 char resolved_path[512];
11432 struct bpf_object *obj = prog->obj;
11433 struct bpf_link *link;
11434 __u64 usdt_cookie;
11435 int err;
11436
11437 if (!OPTS_VALID(opts, bpf_uprobe_opts))
11438 return libbpf_err_ptr(-EINVAL);
11439
11440 if (bpf_program__fd(prog) < 0) {
11441 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
11442 prog->name);
11443 return libbpf_err_ptr(-EINVAL);
11444 }
11445
11446 if (!binary_path)
11447 return libbpf_err_ptr(-EINVAL);
11448
11449 if (!strchr(binary_path, '/')) {
11450 err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path));
11451 if (err) {
11452 pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11453 prog->name, binary_path, err);
11454 return libbpf_err_ptr(err);
11455 }
11456 binary_path = resolved_path;
11457 }
11458
11459 /* USDT manager is instantiated lazily on first USDT attach. It will
11460 * be destroyed together with BPF object in bpf_object__close().
11461 */
11462 if (IS_ERR(obj->usdt_man))
11463 return libbpf_ptr(obj->usdt_man);
11464 if (!obj->usdt_man) {
11465 obj->usdt_man = usdt_manager_new(obj);
11466 if (IS_ERR(obj->usdt_man))
11467 return libbpf_ptr(obj->usdt_man);
11468 }
11469
11470 usdt_cookie = OPTS_GET(opts, usdt_cookie, 0);
11471 link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path,
11472 usdt_provider, usdt_name, usdt_cookie);
11473 err = libbpf_get_error(link);
11474 if (err)
11475 return libbpf_err_ptr(err);
11476 return link;
11477 }
11478 #endif //HAVE_LIBELF
11479
11480 #ifdef HAVE_LIBELF
11481 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11482 {
11483 char *path = NULL, *provider = NULL, *name = NULL;
11484 const char *sec_name;
11485 int n, err;
11486
11487 sec_name = bpf_program__section_name(prog);
11488 if (strcmp(sec_name, "usdt") == 0) {
11489 /* no auto-attach for just SEC("usdt") */
11490 *link = NULL;
11491 return 0;
11492 }
11493
11494 n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name);
11495 if (n != 3) {
11496 pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n",
11497 sec_name);
11498 err = -EINVAL;
11499 } else {
11500 *link = bpf_program__attach_usdt(prog, -1 /* any process */, path,
11501 provider, name, NULL);
11502 err = libbpf_get_error(*link);
11503 }
11504 free(path);
11505 free(provider);
11506 free(name);
11507 return err;
11508 }
11509 #endif //HAVE_LIBELF
11510
11511 static int determine_tracepoint_id(const char *tp_category,
11512 const char *tp_name)
11513 {
11514 char file[PATH_MAX];
11515 int ret;
11516
11517 ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11518 tracefs_path(), tp_category, tp_name);
11519 if (ret < 0)
11520 return -errno;
11521 if (ret >= sizeof(file)) {
11522 pr_debug("tracepoint %s/%s path is too long\n",
11523 tp_category, tp_name);
11524 return -E2BIG;
11525 }
11526 return parse_uint_from_file(file, "%d\n");
11527 }
11528
11529 static int perf_event_open_tracepoint(const char *tp_category,
11530 const char *tp_name)
11531 {
11532 const size_t attr_sz = sizeof(struct perf_event_attr);
11533 struct perf_event_attr attr;
11534 char errmsg[STRERR_BUFSIZE];
11535 int tp_id, pfd, err;
11536
11537 tp_id = determine_tracepoint_id(tp_category, tp_name);
11538 if (tp_id < 0) {
11539 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
11540 tp_category, tp_name,
11541 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
11542 return tp_id;
11543 }
11544
11545 memset(&attr, 0, attr_sz);
11546 attr.type = PERF_TYPE_TRACEPOINT;
11547 attr.size = attr_sz;
11548 attr.config = tp_id;
11549
11550 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
11551 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11552 if (pfd < 0) {
11553 err = -errno;
11554 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
11555 tp_category, tp_name,
11556 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11557 return err;
11558 }
11559 return pfd;
11560 }
11561
11562 struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
11563 const char *tp_category,
11564 const char *tp_name,
11565 const struct bpf_tracepoint_opts *opts)
11566 {
11567 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11568 char errmsg[STRERR_BUFSIZE];
11569 struct bpf_link *link;
11570 int pfd, err;
11571
11572 if (!OPTS_VALID(opts, bpf_tracepoint_opts))
11573 return libbpf_err_ptr(-EINVAL);
11574
11575 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11576
11577 pfd = perf_event_open_tracepoint(tp_category, tp_name);
11578 if (pfd < 0) {
11579 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
11580 prog->name, tp_category, tp_name,
11581 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11582 return libbpf_err_ptr(pfd);
11583 }
11584 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11585 err = libbpf_get_error(link);
11586 if (err) {
11587 close(pfd);
11588 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
11589 prog->name, tp_category, tp_name,
11590 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11591 return libbpf_err_ptr(err);
11592 }
11593 return link;
11594 }
11595
11596 struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
11597 const char *tp_category,
11598 const char *tp_name)
11599 {
11600 return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
11601 }
11602
11603 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11604 {
11605 char *sec_name, *tp_cat, *tp_name;
11606
11607 *link = NULL;
11608
11609 /* no auto-attach for SEC("tp") or SEC("tracepoint") */
11610 if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0)
11611 return 0;
11612
11613 sec_name = strdup(prog->sec_name);
11614 if (!sec_name)
11615 return -ENOMEM;
11616
11617 /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
11618 if (str_has_pfx(prog->sec_name, "tp/"))
11619 tp_cat = sec_name + sizeof("tp/") - 1;
11620 else
11621 tp_cat = sec_name + sizeof("tracepoint/") - 1;
11622 tp_name = strchr(tp_cat, '/');
11623 if (!tp_name) {
11624 free(sec_name);
11625 return -EINVAL;
11626 }
11627 *tp_name = '\0';
11628 tp_name++;
11629
11630 *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
11631 free(sec_name);
11632 return libbpf_get_error(*link);
11633 }
11634
11635 struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
11636 const char *tp_name)
11637 {
11638 char errmsg[STRERR_BUFSIZE];
11639 struct bpf_link *link;
11640 int prog_fd, pfd;
11641
11642 prog_fd = bpf_program__fd(prog);
11643 if (prog_fd < 0) {
11644 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11645 return libbpf_err_ptr(-EINVAL);
11646 }
11647
11648 link = calloc(1, sizeof(*link));
11649 if (!link)
11650 return libbpf_err_ptr(-ENOMEM);
11651 link->detach = &bpf_link__detach_fd;
11652
11653 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
11654 if (pfd < 0) {
11655 pfd = -errno;
11656 free(link);
11657 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
11658 prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11659 return libbpf_err_ptr(pfd);
11660 }
11661 link->fd = pfd;
11662 return link;
11663 }
11664
11665 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11666 {
11667 static const char *const prefixes[] = {
11668 "raw_tp",
11669 "raw_tracepoint",
11670 "raw_tp.w",
11671 "raw_tracepoint.w",
11672 };
11673 size_t i;
11674 const char *tp_name = NULL;
11675
11676 *link = NULL;
11677
11678 for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
11679 size_t pfx_len;
11680
11681 if (!str_has_pfx(prog->sec_name, prefixes[i]))
11682 continue;
11683
11684 pfx_len = strlen(prefixes[i]);
11685 /* no auto-attach case of, e.g., SEC("raw_tp") */
11686 if (prog->sec_name[pfx_len] == '\0')
11687 return 0;
11688
11689 if (prog->sec_name[pfx_len] != '/')
11690 continue;
11691
11692 tp_name = prog->sec_name + pfx_len + 1;
11693 break;
11694 }
11695
11696 if (!tp_name) {
11697 pr_warn("prog '%s': invalid section name '%s'\n",
11698 prog->name, prog->sec_name);
11699 return -EINVAL;
11700 }
11701
11702 *link = bpf_program__attach_raw_tracepoint(prog, tp_name);
11703 return libbpf_get_error(*link);
11704 }
11705
11706 /* Common logic for all BPF program types that attach to a btf_id */
11707 static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog,
11708 const struct bpf_trace_opts *opts)
11709 {
11710 LIBBPF_OPTS(bpf_link_create_opts, link_opts);
11711 char errmsg[STRERR_BUFSIZE];
11712 struct bpf_link *link;
11713 int prog_fd, pfd;
11714
11715 if (!OPTS_VALID(opts, bpf_trace_opts))
11716 return libbpf_err_ptr(-EINVAL);
11717
11718 prog_fd = bpf_program__fd(prog);
11719 if (prog_fd < 0) {
11720 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11721 return libbpf_err_ptr(-EINVAL);
11722 }
11723
11724 link = calloc(1, sizeof(*link));
11725 if (!link)
11726 return libbpf_err_ptr(-ENOMEM);
11727 link->detach = &bpf_link__detach_fd;
11728
11729 /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */
11730 link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0);
11731 pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts);
11732 if (pfd < 0) {
11733 pfd = -errno;
11734 free(link);
11735 pr_warn("prog '%s': failed to attach: %s\n",
11736 prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11737 return libbpf_err_ptr(pfd);
11738 }
11739 link->fd = pfd;
11740 return link;
11741 }
11742
11743 struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
11744 {
11745 return bpf_program__attach_btf_id(prog, NULL);
11746 }
11747
11748 struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog,
11749 const struct bpf_trace_opts *opts)
11750 {
11751 return bpf_program__attach_btf_id(prog, opts);
11752 }
11753
11754 struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
11755 {
11756 return bpf_program__attach_btf_id(prog, NULL);
11757 }
11758
11759 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11760 {
11761 *link = bpf_program__attach_trace(prog);
11762 return libbpf_get_error(*link);
11763 }
11764
11765 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11766 {
11767 *link = bpf_program__attach_lsm(prog);
11768 return libbpf_get_error(*link);
11769 }
11770
11771 static struct bpf_link *
11772 bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id,
11773 const char *target_name)
11774 {
11775 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts,
11776 .target_btf_id = btf_id);
11777 enum bpf_attach_type attach_type;
11778 char errmsg[STRERR_BUFSIZE];
11779 struct bpf_link *link;
11780 int prog_fd, link_fd;
11781
11782 prog_fd = bpf_program__fd(prog);
11783 if (prog_fd < 0) {
11784 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11785 return libbpf_err_ptr(-EINVAL);
11786 }
11787
11788 link = calloc(1, sizeof(*link));
11789 if (!link)
11790 return libbpf_err_ptr(-ENOMEM);
11791 link->detach = &bpf_link__detach_fd;
11792
11793 attach_type = bpf_program__expected_attach_type(prog);
11794 link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts);
11795 if (link_fd < 0) {
11796 link_fd = -errno;
11797 free(link);
11798 pr_warn("prog '%s': failed to attach to %s: %s\n",
11799 prog->name, target_name,
11800 libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
11801 return libbpf_err_ptr(link_fd);
11802 }
11803 link->fd = link_fd;
11804 return link;
11805 }
11806
11807 struct bpf_link *
11808 bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
11809 {
11810 return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup");
11811 }
11812
11813 struct bpf_link *
11814 bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
11815 {
11816 return bpf_program__attach_fd(prog, netns_fd, 0, "netns");
11817 }
11818
11819 struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
11820 {
11821 /* target_fd/target_ifindex use the same field in LINK_CREATE */
11822 return bpf_program__attach_fd(prog, ifindex, 0, "xdp");
11823 }
11824
11825 struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
11826 int target_fd,
11827 const char *attach_func_name)
11828 {
11829 int btf_id;
11830
11831 if (!!target_fd != !!attach_func_name) {
11832 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
11833 prog->name);
11834 return libbpf_err_ptr(-EINVAL);
11835 }
11836
11837 if (prog->type != BPF_PROG_TYPE_EXT) {
11838 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
11839 prog->name);
11840 return libbpf_err_ptr(-EINVAL);
11841 }
11842
11843 if (target_fd) {
11844 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
11845 if (btf_id < 0)
11846 return libbpf_err_ptr(btf_id);
11847
11848 return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace");
11849 } else {
11850 /* no target, so use raw_tracepoint_open for compatibility
11851 * with old kernels
11852 */
11853 return bpf_program__attach_trace(prog);
11854 }
11855 }
11856
11857 struct bpf_link *
11858 bpf_program__attach_iter(const struct bpf_program *prog,
11859 const struct bpf_iter_attach_opts *opts)
11860 {
11861 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
11862 char errmsg[STRERR_BUFSIZE];
11863 struct bpf_link *link;
11864 int prog_fd, link_fd;
11865 __u32 target_fd = 0;
11866
11867 if (!OPTS_VALID(opts, bpf_iter_attach_opts))
11868 return libbpf_err_ptr(-EINVAL);
11869
11870 link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
11871 link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
11872
11873 prog_fd = bpf_program__fd(prog);
11874 if (prog_fd < 0) {
11875 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11876 return libbpf_err_ptr(-EINVAL);
11877 }
11878
11879 link = calloc(1, sizeof(*link));
11880 if (!link)
11881 return libbpf_err_ptr(-ENOMEM);
11882 link->detach = &bpf_link__detach_fd;
11883
11884 link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
11885 &link_create_opts);
11886 if (link_fd < 0) {
11887 link_fd = -errno;
11888 free(link);
11889 pr_warn("prog '%s': failed to attach to iterator: %s\n",
11890 prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
11891 return libbpf_err_ptr(link_fd);
11892 }
11893 link->fd = link_fd;
11894 return link;
11895 }
11896
11897 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11898 {
11899 *link = bpf_program__attach_iter(prog, NULL);
11900 return libbpf_get_error(*link);
11901 }
11902
11903 struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
11904 {
11905 struct bpf_link *link = NULL;
11906 int err;
11907
11908 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
11909 return libbpf_err_ptr(-EOPNOTSUPP);
11910
11911 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link);
11912 if (err)
11913 return libbpf_err_ptr(err);
11914
11915 /* When calling bpf_program__attach() explicitly, auto-attach support
11916 * is expected to work, so NULL returned link is considered an error.
11917 * This is different for skeleton's attach, see comment in
11918 * bpf_object__attach_skeleton().
11919 */
11920 if (!link)
11921 return libbpf_err_ptr(-EOPNOTSUPP);
11922
11923 return link;
11924 }
11925
11926 static int bpf_link__detach_struct_ops(struct bpf_link *link)
11927 {
11928 __u32 zero = 0;
11929
11930 if (bpf_map_delete_elem(link->fd, &zero))
11931 return -errno;
11932
11933 return 0;
11934 }
11935
11936 struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
11937 {
11938 struct bpf_struct_ops *st_ops;
11939 struct bpf_link *link;
11940 __u32 i, zero = 0;
11941 int err;
11942
11943 if (!bpf_map__is_struct_ops(map) || map->fd == -1)
11944 return libbpf_err_ptr(-EINVAL);
11945
11946 link = calloc(1, sizeof(*link));
11947 if (!link)
11948 return libbpf_err_ptr(-EINVAL);
11949
11950 st_ops = map->st_ops;
11951 for (i = 0; i < btf_vlen(st_ops->type); i++) {
11952 struct bpf_program *prog = st_ops->progs[i];
11953 void *kern_data;
11954 int prog_fd;
11955
11956 if (!prog)
11957 continue;
11958
11959 prog_fd = bpf_program__fd(prog);
11960 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
11961 *(unsigned long *)kern_data = prog_fd;
11962 }
11963
11964 err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
11965 if (err) {
11966 err = -errno;
11967 free(link);
11968 return libbpf_err_ptr(err);
11969 }
11970
11971 link->detach = bpf_link__detach_struct_ops;
11972 link->fd = map->fd;
11973
11974 return link;
11975 }
11976
11977 typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
11978 void *private_data);
11979
11980 static enum bpf_perf_event_ret
11981 perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
11982 void **copy_mem, size_t *copy_size,
11983 bpf_perf_event_print_t fn, void *private_data)
11984 {
11985 struct perf_event_mmap_page *header = mmap_mem;
11986 __u64 data_head = ring_buffer_read_head(header);
11987 __u64 data_tail = header->data_tail;
11988 void *base = ((__u8 *)header) + page_size;
11989 int ret = LIBBPF_PERF_EVENT_CONT;
11990 struct perf_event_header *ehdr;
11991 size_t ehdr_size;
11992
11993 while (data_head != data_tail) {
11994 ehdr = base + (data_tail & (mmap_size - 1));
11995 ehdr_size = ehdr->size;
11996
11997 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
11998 void *copy_start = ehdr;
11999 size_t len_first = base + mmap_size - copy_start;
12000 size_t len_secnd = ehdr_size - len_first;
12001
12002 if (*copy_size < ehdr_size) {
12003 free(*copy_mem);
12004 *copy_mem = malloc(ehdr_size);
12005 if (!*copy_mem) {
12006 *copy_size = 0;
12007 ret = LIBBPF_PERF_EVENT_ERROR;
12008 break;
12009 }
12010 *copy_size = ehdr_size;
12011 }
12012
12013 memcpy(*copy_mem, copy_start, len_first);
12014 memcpy(*copy_mem + len_first, base, len_secnd);
12015 ehdr = *copy_mem;
12016 }
12017
12018 ret = fn(ehdr, private_data);
12019 data_tail += ehdr_size;
12020 if (ret != LIBBPF_PERF_EVENT_CONT)
12021 break;
12022 }
12023
12024 ring_buffer_write_tail(header, data_tail);
12025 return libbpf_err(ret);
12026 }
12027
12028 struct perf_buffer;
12029
12030 struct perf_buffer_params {
12031 struct perf_event_attr *attr;
12032 /* if event_cb is specified, it takes precendence */
12033 perf_buffer_event_fn event_cb;
12034 /* sample_cb and lost_cb are higher-level common-case callbacks */
12035 perf_buffer_sample_fn sample_cb;
12036 perf_buffer_lost_fn lost_cb;
12037 void *ctx;
12038 int cpu_cnt;
12039 int *cpus;
12040 int *map_keys;
12041 };
12042
12043 struct perf_cpu_buf {
12044 struct perf_buffer *pb;
12045 void *base; /* mmap()'ed memory */
12046 void *buf; /* for reconstructing segmented data */
12047 size_t buf_size;
12048 int fd;
12049 int cpu;
12050 int map_key;
12051 };
12052
12053 struct perf_buffer {
12054 perf_buffer_event_fn event_cb;
12055 perf_buffer_sample_fn sample_cb;
12056 perf_buffer_lost_fn lost_cb;
12057 void *ctx; /* passed into callbacks */
12058
12059 size_t page_size;
12060 size_t mmap_size;
12061 struct perf_cpu_buf **cpu_bufs;
12062 struct epoll_event *events;
12063 int cpu_cnt; /* number of allocated CPU buffers */
12064 int epoll_fd; /* perf event FD */
12065 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
12066 };
12067
12068 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
12069 struct perf_cpu_buf *cpu_buf)
12070 {
12071 if (!cpu_buf)
12072 return;
12073 if (cpu_buf->base &&
12074 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
12075 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
12076 if (cpu_buf->fd >= 0) {
12077 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
12078 close(cpu_buf->fd);
12079 }
12080 free(cpu_buf->buf);
12081 free(cpu_buf);
12082 }
12083
12084 void perf_buffer__free(struct perf_buffer *pb)
12085 {
12086 int i;
12087
12088 if (IS_ERR_OR_NULL(pb))
12089 return;
12090 if (pb->cpu_bufs) {
12091 for (i = 0; i < pb->cpu_cnt; i++) {
12092 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
12093
12094 if (!cpu_buf)
12095 continue;
12096
12097 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
12098 perf_buffer__free_cpu_buf(pb, cpu_buf);
12099 }
12100 free(pb->cpu_bufs);
12101 }
12102 if (pb->epoll_fd >= 0)
12103 close(pb->epoll_fd);
12104 free(pb->events);
12105 free(pb);
12106 }
12107
12108 static struct perf_cpu_buf *
12109 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
12110 int cpu, int map_key)
12111 {
12112 struct perf_cpu_buf *cpu_buf;
12113 char msg[STRERR_BUFSIZE];
12114 int err;
12115
12116 cpu_buf = calloc(1, sizeof(*cpu_buf));
12117 if (!cpu_buf)
12118 return ERR_PTR(-ENOMEM);
12119
12120 cpu_buf->pb = pb;
12121 cpu_buf->cpu = cpu;
12122 cpu_buf->map_key = map_key;
12123
12124 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
12125 -1, PERF_FLAG_FD_CLOEXEC);
12126 if (cpu_buf->fd < 0) {
12127 err = -errno;
12128 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
12129 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
12130 goto error;
12131 }
12132
12133 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
12134 PROT_READ | PROT_WRITE, MAP_SHARED,
12135 cpu_buf->fd, 0);
12136 if (cpu_buf->base == MAP_FAILED) {
12137 cpu_buf->base = NULL;
12138 err = -errno;
12139 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
12140 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
12141 goto error;
12142 }
12143
12144 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
12145 err = -errno;
12146 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
12147 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
12148 goto error;
12149 }
12150
12151 return cpu_buf;
12152
12153 error:
12154 perf_buffer__free_cpu_buf(pb, cpu_buf);
12155 return (struct perf_cpu_buf *)ERR_PTR(err);
12156 }
12157
12158 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
12159 struct perf_buffer_params *p);
12160
12161 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
12162 perf_buffer_sample_fn sample_cb,
12163 perf_buffer_lost_fn lost_cb,
12164 void *ctx,
12165 const struct perf_buffer_opts *opts)
12166 {
12167 const size_t attr_sz = sizeof(struct perf_event_attr);
12168 struct perf_buffer_params p = {};
12169 struct perf_event_attr attr;
12170
12171 if (!OPTS_VALID(opts, perf_buffer_opts))
12172 return libbpf_err_ptr(-EINVAL);
12173
12174 memset(&attr, 0, attr_sz);
12175 attr.size = attr_sz;
12176 attr.config = PERF_COUNT_SW_BPF_OUTPUT;
12177 attr.type = PERF_TYPE_SOFTWARE;
12178 attr.sample_type = PERF_SAMPLE_RAW;
12179 attr.sample_period = 1;
12180 attr.wakeup_events = 1;
12181
12182 p.attr = &attr;
12183 p.sample_cb = sample_cb;
12184 p.lost_cb = lost_cb;
12185 p.ctx = ctx;
12186
12187 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
12188 }
12189
12190 struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt,
12191 struct perf_event_attr *attr,
12192 perf_buffer_event_fn event_cb, void *ctx,
12193 const struct perf_buffer_raw_opts *opts)
12194 {
12195 struct perf_buffer_params p = {};
12196
12197 if (!attr)
12198 return libbpf_err_ptr(-EINVAL);
12199
12200 if (!OPTS_VALID(opts, perf_buffer_raw_opts))
12201 return libbpf_err_ptr(-EINVAL);
12202
12203 p.attr = attr;
12204 p.event_cb = event_cb;
12205 p.ctx = ctx;
12206 p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
12207 p.cpus = OPTS_GET(opts, cpus, NULL);
12208 p.map_keys = OPTS_GET(opts, map_keys, NULL);
12209
12210 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
12211 }
12212
12213 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
12214 struct perf_buffer_params *p)
12215 {
12216 const char *online_cpus_file = "/sys/devices/system/cpu/online";
12217 struct bpf_map_info map;
12218 char msg[STRERR_BUFSIZE];
12219 struct perf_buffer *pb;
12220 bool *online = NULL;
12221 __u32 map_info_len;
12222 int err, i, j, n;
12223
12224 if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) {
12225 pr_warn("page count should be power of two, but is %zu\n",
12226 page_cnt);
12227 return ERR_PTR(-EINVAL);
12228 }
12229
12230 /* best-effort sanity checks */
12231 memset(&map, 0, sizeof(map));
12232 map_info_len = sizeof(map);
12233 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
12234 if (err) {
12235 err = -errno;
12236 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
12237 * -EBADFD, -EFAULT, or -E2BIG on real error
12238 */
12239 if (err != -EINVAL) {
12240 pr_warn("failed to get map info for map FD %d: %s\n",
12241 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
12242 return ERR_PTR(err);
12243 }
12244 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
12245 map_fd);
12246 } else {
12247 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
12248 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
12249 map.name);
12250 return ERR_PTR(-EINVAL);
12251 }
12252 }
12253
12254 pb = calloc(1, sizeof(*pb));
12255 if (!pb)
12256 return ERR_PTR(-ENOMEM);
12257
12258 pb->event_cb = p->event_cb;
12259 pb->sample_cb = p->sample_cb;
12260 pb->lost_cb = p->lost_cb;
12261 pb->ctx = p->ctx;
12262
12263 pb->page_size = getpagesize();
12264 pb->mmap_size = pb->page_size * page_cnt;
12265 pb->map_fd = map_fd;
12266
12267 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
12268 if (pb->epoll_fd < 0) {
12269 err = -errno;
12270 pr_warn("failed to create epoll instance: %s\n",
12271 libbpf_strerror_r(err, msg, sizeof(msg)));
12272 goto error;
12273 }
12274
12275 if (p->cpu_cnt > 0) {
12276 pb->cpu_cnt = p->cpu_cnt;
12277 } else {
12278 pb->cpu_cnt = libbpf_num_possible_cpus();
12279 if (pb->cpu_cnt < 0) {
12280 err = pb->cpu_cnt;
12281 goto error;
12282 }
12283 if (map.max_entries && map.max_entries < pb->cpu_cnt)
12284 pb->cpu_cnt = map.max_entries;
12285 }
12286
12287 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
12288 if (!pb->events) {
12289 err = -ENOMEM;
12290 pr_warn("failed to allocate events: out of memory\n");
12291 goto error;
12292 }
12293 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
12294 if (!pb->cpu_bufs) {
12295 err = -ENOMEM;
12296 pr_warn("failed to allocate buffers: out of memory\n");
12297 goto error;
12298 }
12299
12300 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
12301 if (err) {
12302 pr_warn("failed to get online CPU mask: %d\n", err);
12303 goto error;
12304 }
12305
12306 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
12307 struct perf_cpu_buf *cpu_buf;
12308 int cpu, map_key;
12309
12310 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
12311 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
12312
12313 /* in case user didn't explicitly requested particular CPUs to
12314 * be attached to, skip offline/not present CPUs
12315 */
12316 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
12317 continue;
12318
12319 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
12320 if (IS_ERR(cpu_buf)) {
12321 err = PTR_ERR(cpu_buf);
12322 goto error;
12323 }
12324
12325 pb->cpu_bufs[j] = cpu_buf;
12326
12327 err = bpf_map_update_elem(pb->map_fd, &map_key,
12328 &cpu_buf->fd, 0);
12329 if (err) {
12330 err = -errno;
12331 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
12332 cpu, map_key, cpu_buf->fd,
12333 libbpf_strerror_r(err, msg, sizeof(msg)));
12334 goto error;
12335 }
12336
12337 pb->events[j].events = EPOLLIN;
12338 pb->events[j].data.ptr = cpu_buf;
12339 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
12340 &pb->events[j]) < 0) {
12341 err = -errno;
12342 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
12343 cpu, cpu_buf->fd,
12344 libbpf_strerror_r(err, msg, sizeof(msg)));
12345 goto error;
12346 }
12347 j++;
12348 }
12349 pb->cpu_cnt = j;
12350 free(online);
12351
12352 return pb;
12353
12354 error:
12355 free(online);
12356 if (pb)
12357 perf_buffer__free(pb);
12358 return ERR_PTR(err);
12359 }
12360
12361 struct perf_sample_raw {
12362 struct perf_event_header header;
12363 uint32_t size;
12364 char data[];
12365 };
12366
12367 struct perf_sample_lost {
12368 struct perf_event_header header;
12369 uint64_t id;
12370 uint64_t lost;
12371 uint64_t sample_id;
12372 };
12373
12374 static enum bpf_perf_event_ret
12375 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
12376 {
12377 struct perf_cpu_buf *cpu_buf = ctx;
12378 struct perf_buffer *pb = cpu_buf->pb;
12379 void *data = e;
12380
12381 /* user wants full control over parsing perf event */
12382 if (pb->event_cb)
12383 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
12384
12385 switch (e->type) {
12386 case PERF_RECORD_SAMPLE: {
12387 struct perf_sample_raw *s = data;
12388
12389 if (pb->sample_cb)
12390 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
12391 break;
12392 }
12393 case PERF_RECORD_LOST: {
12394 struct perf_sample_lost *s = data;
12395
12396 if (pb->lost_cb)
12397 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
12398 break;
12399 }
12400 default:
12401 pr_warn("unknown perf sample type %d\n", e->type);
12402 return LIBBPF_PERF_EVENT_ERROR;
12403 }
12404 return LIBBPF_PERF_EVENT_CONT;
12405 }
12406
12407 static int perf_buffer__process_records(struct perf_buffer *pb,
12408 struct perf_cpu_buf *cpu_buf)
12409 {
12410 enum bpf_perf_event_ret ret;
12411
12412 ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
12413 pb->page_size, &cpu_buf->buf,
12414 &cpu_buf->buf_size,
12415 perf_buffer__process_record, cpu_buf);
12416 if (ret != LIBBPF_PERF_EVENT_CONT)
12417 return ret;
12418 return 0;
12419 }
12420
12421 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
12422 {
12423 return pb->epoll_fd;
12424 }
12425
12426 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
12427 {
12428 int i, cnt, err;
12429
12430 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
12431 if (cnt < 0)
12432 return -errno;
12433
12434 for (i = 0; i < cnt; i++) {
12435 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
12436
12437 err = perf_buffer__process_records(pb, cpu_buf);
12438 if (err) {
12439 pr_warn("error while processing records: %d\n", err);
12440 return libbpf_err(err);
12441 }
12442 }
12443 return cnt;
12444 }
12445
12446 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
12447 * manager.
12448 */
12449 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
12450 {
12451 return pb->cpu_cnt;
12452 }
12453
12454 /*
12455 * Return perf_event FD of a ring buffer in *buf_idx* slot of
12456 * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
12457 * select()/poll()/epoll() Linux syscalls.
12458 */
12459 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
12460 {
12461 struct perf_cpu_buf *cpu_buf;
12462
12463 if (buf_idx >= pb->cpu_cnt)
12464 return libbpf_err(-EINVAL);
12465
12466 cpu_buf = pb->cpu_bufs[buf_idx];
12467 if (!cpu_buf)
12468 return libbpf_err(-ENOENT);
12469
12470 return cpu_buf->fd;
12471 }
12472
12473 int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size)
12474 {
12475 struct perf_cpu_buf *cpu_buf;
12476
12477 if (buf_idx >= pb->cpu_cnt)
12478 return libbpf_err(-EINVAL);
12479
12480 cpu_buf = pb->cpu_bufs[buf_idx];
12481 if (!cpu_buf)
12482 return libbpf_err(-ENOENT);
12483
12484 *buf = cpu_buf->base;
12485 *buf_size = pb->mmap_size;
12486 return 0;
12487 }
12488
12489 /*
12490 * Consume data from perf ring buffer corresponding to slot *buf_idx* in
12491 * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
12492 * consume, do nothing and return success.
12493 * Returns:
12494 * - 0 on success;
12495 * - <0 on failure.
12496 */
12497 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
12498 {
12499 struct perf_cpu_buf *cpu_buf;
12500
12501 if (buf_idx >= pb->cpu_cnt)
12502 return libbpf_err(-EINVAL);
12503
12504 cpu_buf = pb->cpu_bufs[buf_idx];
12505 if (!cpu_buf)
12506 return libbpf_err(-ENOENT);
12507
12508 return perf_buffer__process_records(pb, cpu_buf);
12509 }
12510
12511 int perf_buffer__consume(struct perf_buffer *pb)
12512 {
12513 int i, err;
12514
12515 for (i = 0; i < pb->cpu_cnt; i++) {
12516 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
12517
12518 if (!cpu_buf)
12519 continue;
12520
12521 err = perf_buffer__process_records(pb, cpu_buf);
12522 if (err) {
12523 pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
12524 return libbpf_err(err);
12525 }
12526 }
12527 return 0;
12528 }
12529
12530 int bpf_program__set_attach_target(struct bpf_program *prog,
12531 int attach_prog_fd,
12532 const char *attach_func_name)
12533 {
12534 int btf_obj_fd = 0, btf_id = 0, err;
12535
12536 if (!prog || attach_prog_fd < 0)
12537 return libbpf_err(-EINVAL);
12538
12539 if (prog->obj->loaded)
12540 return libbpf_err(-EINVAL);
12541
12542 if (attach_prog_fd && !attach_func_name) {
12543 /* remember attach_prog_fd and let bpf_program__load() find
12544 * BTF ID during the program load
12545 */
12546 prog->attach_prog_fd = attach_prog_fd;
12547 return 0;
12548 }
12549
12550 if (attach_prog_fd) {
12551 btf_id = libbpf_find_prog_btf_id(attach_func_name,
12552 attach_prog_fd);
12553 if (btf_id < 0)
12554 return libbpf_err(btf_id);
12555 } else {
12556 if (!attach_func_name)
12557 return libbpf_err(-EINVAL);
12558
12559 /* load btf_vmlinux, if not yet */
12560 err = bpf_object__load_vmlinux_btf(prog->obj, true);
12561 if (err)
12562 return libbpf_err(err);
12563 err = find_kernel_btf_id(prog->obj, attach_func_name,
12564 prog->expected_attach_type,
12565 &btf_obj_fd, &btf_id);
12566 if (err)
12567 return libbpf_err(err);
12568 }
12569
12570 prog->attach_btf_id = btf_id;
12571 prog->attach_btf_obj_fd = btf_obj_fd;
12572 prog->attach_prog_fd = attach_prog_fd;
12573 return 0;
12574 }
12575
12576 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
12577 {
12578 int err = 0, n, len, start, end = -1;
12579 bool *tmp;
12580
12581 *mask = NULL;
12582 *mask_sz = 0;
12583
12584 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
12585 while (*s) {
12586 if (*s == ',' || *s == '\n') {
12587 s++;
12588 continue;
12589 }
12590 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
12591 if (n <= 0 || n > 2) {
12592 pr_warn("Failed to get CPU range %s: %d\n", s, n);
12593 err = -EINVAL;
12594 goto cleanup;
12595 } else if (n == 1) {
12596 end = start;
12597 }
12598 if (start < 0 || start > end) {
12599 pr_warn("Invalid CPU range [%d,%d] in %s\n",
12600 start, end, s);
12601 err = -EINVAL;
12602 goto cleanup;
12603 }
12604 tmp = realloc(*mask, end + 1);
12605 if (!tmp) {
12606 err = -ENOMEM;
12607 goto cleanup;
12608 }
12609 *mask = tmp;
12610 memset(tmp + *mask_sz, 0, start - *mask_sz);
12611 memset(tmp + start, 1, end - start + 1);
12612 *mask_sz = end + 1;
12613 s += len;
12614 }
12615 if (!*mask_sz) {
12616 pr_warn("Empty CPU range\n");
12617 return -EINVAL;
12618 }
12619 return 0;
12620 cleanup:
12621 free(*mask);
12622 *mask = NULL;
12623 return err;
12624 }
12625
12626 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
12627 {
12628 int fd, err = 0, len;
12629 char buf[128];
12630
12631 fd = open(fcpu, O_RDONLY | O_CLOEXEC);
12632 if (fd < 0) {
12633 err = -errno;
12634 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
12635 return err;
12636 }
12637 len = read(fd, buf, sizeof(buf));
12638 close(fd);
12639 if (len <= 0) {
12640 err = len ? -errno : -EINVAL;
12641 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
12642 return err;
12643 }
12644 if (len >= sizeof(buf)) {
12645 pr_warn("CPU mask is too big in file %s\n", fcpu);
12646 return -E2BIG;
12647 }
12648 buf[len] = '\0';
12649
12650 return parse_cpu_mask_str(buf, mask, mask_sz);
12651 }
12652
12653 int libbpf_num_possible_cpus(void)
12654 {
12655 static const char *fcpu = "/sys/devices/system/cpu/possible";
12656 static int cpus;
12657 int err, n, i, tmp_cpus;
12658 bool *mask;
12659
12660 tmp_cpus = READ_ONCE(cpus);
12661 if (tmp_cpus > 0)
12662 return tmp_cpus;
12663
12664 err = parse_cpu_mask_file(fcpu, &mask, &n);
12665 if (err)
12666 return libbpf_err(err);
12667
12668 tmp_cpus = 0;
12669 for (i = 0; i < n; i++) {
12670 if (mask[i])
12671 tmp_cpus++;
12672 }
12673 free(mask);
12674
12675 WRITE_ONCE(cpus, tmp_cpus);
12676 return tmp_cpus;
12677 }
12678
12679 static int populate_skeleton_maps(const struct bpf_object *obj,
12680 struct bpf_map_skeleton *maps,
12681 size_t map_cnt)
12682 {
12683 int i;
12684
12685 for (i = 0; i < map_cnt; i++) {
12686 struct bpf_map **map = maps[i].map;
12687 const char *name = maps[i].name;
12688 void **mmaped = maps[i].mmaped;
12689
12690 *map = bpf_object__find_map_by_name(obj, name);
12691 if (!*map) {
12692 pr_warn("failed to find skeleton map '%s'\n", name);
12693 return -ESRCH;
12694 }
12695
12696 /* externs shouldn't be pre-setup from user code */
12697 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
12698 *mmaped = (*map)->mmaped;
12699 }
12700 return 0;
12701 }
12702
12703 static int populate_skeleton_progs(const struct bpf_object *obj,
12704 struct bpf_prog_skeleton *progs,
12705 size_t prog_cnt)
12706 {
12707 int i;
12708
12709 for (i = 0; i < prog_cnt; i++) {
12710 struct bpf_program **prog = progs[i].prog;
12711 const char *name = progs[i].name;
12712
12713 *prog = bpf_object__find_program_by_name(obj, name);
12714 if (!*prog) {
12715 pr_warn("failed to find skeleton program '%s'\n", name);
12716 return -ESRCH;
12717 }
12718 }
12719 return 0;
12720 }
12721
12722 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
12723 const struct bpf_object_open_opts *opts)
12724 {
12725 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
12726 .object_name = s->name,
12727 );
12728 struct bpf_object *obj;
12729 int err;
12730
12731 /* Attempt to preserve opts->object_name, unless overriden by user
12732 * explicitly. Overwriting object name for skeletons is discouraged,
12733 * as it breaks global data maps, because they contain object name
12734 * prefix as their own map name prefix. When skeleton is generated,
12735 * bpftool is making an assumption that this name will stay the same.
12736 */
12737 if (opts) {
12738 memcpy(&skel_opts, opts, sizeof(*opts));
12739 if (!opts->object_name)
12740 skel_opts.object_name = s->name;
12741 }
12742
12743 obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
12744 err = libbpf_get_error(obj);
12745 if (err) {
12746 pr_warn("failed to initialize skeleton BPF object '%s': %d\n",
12747 s->name, err);
12748 return libbpf_err(err);
12749 }
12750
12751 *s->obj = obj;
12752 err = populate_skeleton_maps(obj, s->maps, s->map_cnt);
12753 if (err) {
12754 pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err);
12755 return libbpf_err(err);
12756 }
12757
12758 err = populate_skeleton_progs(obj, s->progs, s->prog_cnt);
12759 if (err) {
12760 pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err);
12761 return libbpf_err(err);
12762 }
12763
12764 return 0;
12765 }
12766
12767 int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s)
12768 {
12769 int err, len, var_idx, i;
12770 const char *var_name;
12771 const struct bpf_map *map;
12772 struct btf *btf;
12773 __u32 map_type_id;
12774 const struct btf_type *map_type, *var_type;
12775 const struct bpf_var_skeleton *var_skel;
12776 struct btf_var_secinfo *var;
12777
12778 if (!s->obj)
12779 return libbpf_err(-EINVAL);
12780
12781 btf = bpf_object__btf(s->obj);
12782 if (!btf) {
12783 pr_warn("subskeletons require BTF at runtime (object %s)\n",
12784 bpf_object__name(s->obj));
12785 return libbpf_err(-errno);
12786 }
12787
12788 err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt);
12789 if (err) {
12790 pr_warn("failed to populate subskeleton maps: %d\n", err);
12791 return libbpf_err(err);
12792 }
12793
12794 err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt);
12795 if (err) {
12796 pr_warn("failed to populate subskeleton maps: %d\n", err);
12797 return libbpf_err(err);
12798 }
12799
12800 for (var_idx = 0; var_idx < s->var_cnt; var_idx++) {
12801 var_skel = &s->vars[var_idx];
12802 map = *var_skel->map;
12803 map_type_id = bpf_map__btf_value_type_id(map);
12804 map_type = btf__type_by_id(btf, map_type_id);
12805
12806 if (!btf_is_datasec(map_type)) {
12807 pr_warn("type for map '%1$s' is not a datasec: %2$s",
12808 bpf_map__name(map),
12809 __btf_kind_str(btf_kind(map_type)));
12810 return libbpf_err(-EINVAL);
12811 }
12812
12813 len = btf_vlen(map_type);
12814 var = btf_var_secinfos(map_type);
12815 for (i = 0; i < len; i++, var++) {
12816 var_type = btf__type_by_id(btf, var->type);
12817 var_name = btf__name_by_offset(btf, var_type->name_off);
12818 if (strcmp(var_name, var_skel->name) == 0) {
12819 *var_skel->addr = map->mmaped + var->offset;
12820 break;
12821 }
12822 }
12823 }
12824 return 0;
12825 }
12826
12827 void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s)
12828 {
12829 if (!s)
12830 return;
12831 free(s->maps);
12832 free(s->progs);
12833 free(s->vars);
12834 free(s);
12835 }
12836
12837 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
12838 {
12839 int i, err;
12840
12841 err = bpf_object__load(*s->obj);
12842 if (err) {
12843 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
12844 return libbpf_err(err);
12845 }
12846
12847 for (i = 0; i < s->map_cnt; i++) {
12848 struct bpf_map *map = *s->maps[i].map;
12849 size_t mmap_sz = bpf_map_mmap_sz(map);
12850 int prot, map_fd = bpf_map__fd(map);
12851 void **mmaped = s->maps[i].mmaped;
12852
12853 if (!mmaped)
12854 continue;
12855
12856 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
12857 *mmaped = NULL;
12858 continue;
12859 }
12860
12861 if (map->def.map_flags & BPF_F_RDONLY_PROG)
12862 prot = PROT_READ;
12863 else
12864 prot = PROT_READ | PROT_WRITE;
12865
12866 /* Remap anonymous mmap()-ed "map initialization image" as
12867 * a BPF map-backed mmap()-ed memory, but preserving the same
12868 * memory address. This will cause kernel to change process'
12869 * page table to point to a different piece of kernel memory,
12870 * but from userspace point of view memory address (and its
12871 * contents, being identical at this point) will stay the
12872 * same. This mapping will be released by bpf_object__close()
12873 * as per normal clean up procedure, so we don't need to worry
12874 * about it from skeleton's clean up perspective.
12875 */
12876 *mmaped = mmap(map->mmaped, mmap_sz, prot,
12877 MAP_SHARED | MAP_FIXED, map_fd, 0);
12878 if (*mmaped == MAP_FAILED) {
12879 err = -errno;
12880 *mmaped = NULL;
12881 pr_warn("failed to re-mmap() map '%s': %d\n",
12882 bpf_map__name(map), err);
12883 return libbpf_err(err);
12884 }
12885 }
12886
12887 return 0;
12888 }
12889
12890 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
12891 {
12892 int i, err;
12893
12894 for (i = 0; i < s->prog_cnt; i++) {
12895 struct bpf_program *prog = *s->progs[i].prog;
12896 struct bpf_link **link = s->progs[i].link;
12897
12898 if (!prog->autoload || !prog->autoattach)
12899 continue;
12900
12901 /* auto-attaching not supported for this program */
12902 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
12903 continue;
12904
12905 /* if user already set the link manually, don't attempt auto-attach */
12906 if (*link)
12907 continue;
12908
12909 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link);
12910 if (err) {
12911 pr_warn("prog '%s': failed to auto-attach: %d\n",
12912 bpf_program__name(prog), err);
12913 return libbpf_err(err);
12914 }
12915
12916 /* It's possible that for some SEC() definitions auto-attach
12917 * is supported in some cases (e.g., if definition completely
12918 * specifies target information), but is not in other cases.
12919 * SEC("uprobe") is one such case. If user specified target
12920 * binary and function name, such BPF program can be
12921 * auto-attached. But if not, it shouldn't trigger skeleton's
12922 * attach to fail. It should just be skipped.
12923 * attach_fn signals such case with returning 0 (no error) and
12924 * setting link to NULL.
12925 */
12926 }
12927
12928 return 0;
12929 }
12930
12931 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
12932 {
12933 int i;
12934
12935 for (i = 0; i < s->prog_cnt; i++) {
12936 struct bpf_link **link = s->progs[i].link;
12937
12938 bpf_link__destroy(*link);
12939 *link = NULL;
12940 }
12941 }
12942
12943 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
12944 {
12945 if (!s)
12946 return;
12947
12948 if (s->progs)
12949 bpf_object__detach_skeleton(s);
12950 if (s->obj)
12951 bpf_object__close(*s->obj);
12952 free(s->maps);
12953 free(s->progs);
12954 free(s);
12955 }
12956