xref: /openbmc/linux/tools/lib/bpf/libbpf.c (revision b68fc09be48edbc47de1a0f3d42ef8adf6c0ac55)
1 // SPDX-License-Identifier: LGPL-2.1
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  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU Lesser General Public
13  * License as published by the Free Software Foundation;
14  * version 2.1 of the License (not later!)
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  * GNU Lesser General Public License for more details.
20  *
21  * You should have received a copy of the GNU Lesser General Public
22  * License along with this program; if not,  see <http://www.gnu.org/licenses>
23  */
24 
25 #define _GNU_SOURCE
26 #include <stdlib.h>
27 #include <stdio.h>
28 #include <stdarg.h>
29 #include <libgen.h>
30 #include <inttypes.h>
31 #include <string.h>
32 #include <unistd.h>
33 #include <fcntl.h>
34 #include <errno.h>
35 #include <perf-sys.h>
36 #include <asm/unistd.h>
37 #include <linux/err.h>
38 #include <linux/kernel.h>
39 #include <linux/bpf.h>
40 #include <linux/btf.h>
41 #include <linux/list.h>
42 #include <linux/limits.h>
43 #include <sys/stat.h>
44 #include <sys/types.h>
45 #include <sys/vfs.h>
46 #include <tools/libc_compat.h>
47 #include <libelf.h>
48 #include <gelf.h>
49 
50 #include "libbpf.h"
51 #include "bpf.h"
52 #include "btf.h"
53 #include "str_error.h"
54 
55 #ifndef EM_BPF
56 #define EM_BPF 247
57 #endif
58 
59 #ifndef BPF_FS_MAGIC
60 #define BPF_FS_MAGIC		0xcafe4a11
61 #endif
62 
63 #define __printf(a, b)	__attribute__((format(printf, a, b)))
64 
65 __printf(1, 2)
66 static int __base_pr(const char *format, ...)
67 {
68 	va_list args;
69 	int err;
70 
71 	va_start(args, format);
72 	err = vfprintf(stderr, format, args);
73 	va_end(args);
74 	return err;
75 }
76 
77 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
78 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
79 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
80 
81 #define __pr(func, fmt, ...)	\
82 do {				\
83 	if ((func))		\
84 		(func)("libbpf: " fmt, ##__VA_ARGS__); \
85 } while (0)
86 
87 #define pr_warning(fmt, ...)	__pr(__pr_warning, fmt, ##__VA_ARGS__)
88 #define pr_info(fmt, ...)	__pr(__pr_info, fmt, ##__VA_ARGS__)
89 #define pr_debug(fmt, ...)	__pr(__pr_debug, fmt, ##__VA_ARGS__)
90 
91 void libbpf_set_print(libbpf_print_fn_t warn,
92 		      libbpf_print_fn_t info,
93 		      libbpf_print_fn_t debug)
94 {
95 	__pr_warning = warn;
96 	__pr_info = info;
97 	__pr_debug = debug;
98 }
99 
100 #define STRERR_BUFSIZE  128
101 
102 #define CHECK_ERR(action, err, out) do {	\
103 	err = action;			\
104 	if (err)			\
105 		goto out;		\
106 } while(0)
107 
108 
109 /* Copied from tools/perf/util/util.h */
110 #ifndef zfree
111 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
112 #endif
113 
114 #ifndef zclose
115 # define zclose(fd) ({			\
116 	int ___err = 0;			\
117 	if ((fd) >= 0)			\
118 		___err = close((fd));	\
119 	fd = -1;			\
120 	___err; })
121 #endif
122 
123 #ifdef HAVE_LIBELF_MMAP_SUPPORT
124 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
125 #else
126 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
127 #endif
128 
129 /*
130  * bpf_prog should be a better name but it has been used in
131  * linux/filter.h.
132  */
133 struct bpf_program {
134 	/* Index in elf obj file, for relocation use. */
135 	int idx;
136 	char *name;
137 	int prog_ifindex;
138 	char *section_name;
139 	struct bpf_insn *insns;
140 	size_t insns_cnt, main_prog_cnt;
141 	enum bpf_prog_type type;
142 
143 	struct reloc_desc {
144 		enum {
145 			RELO_LD64,
146 			RELO_CALL,
147 		} type;
148 		int insn_idx;
149 		union {
150 			int map_idx;
151 			int text_off;
152 		};
153 	} *reloc_desc;
154 	int nr_reloc;
155 
156 	struct {
157 		int nr;
158 		int *fds;
159 	} instances;
160 	bpf_program_prep_t preprocessor;
161 
162 	struct bpf_object *obj;
163 	void *priv;
164 	bpf_program_clear_priv_t clear_priv;
165 
166 	enum bpf_attach_type expected_attach_type;
167 };
168 
169 struct bpf_map {
170 	int fd;
171 	char *name;
172 	size_t offset;
173 	int map_ifindex;
174 	struct bpf_map_def def;
175 	__u32 btf_key_type_id;
176 	__u32 btf_value_type_id;
177 	void *priv;
178 	bpf_map_clear_priv_t clear_priv;
179 };
180 
181 static LIST_HEAD(bpf_objects_list);
182 
183 struct bpf_object {
184 	char license[64];
185 	u32 kern_version;
186 
187 	struct bpf_program *programs;
188 	size_t nr_programs;
189 	struct bpf_map *maps;
190 	size_t nr_maps;
191 
192 	bool loaded;
193 	bool has_pseudo_calls;
194 
195 	/*
196 	 * Information when doing elf related work. Only valid if fd
197 	 * is valid.
198 	 */
199 	struct {
200 		int fd;
201 		void *obj_buf;
202 		size_t obj_buf_sz;
203 		Elf *elf;
204 		GElf_Ehdr ehdr;
205 		Elf_Data *symbols;
206 		size_t strtabidx;
207 		struct {
208 			GElf_Shdr shdr;
209 			Elf_Data *data;
210 		} *reloc;
211 		int nr_reloc;
212 		int maps_shndx;
213 		int text_shndx;
214 	} efile;
215 	/*
216 	 * All loaded bpf_object is linked in a list, which is
217 	 * hidden to caller. bpf_objects__<func> handlers deal with
218 	 * all objects.
219 	 */
220 	struct list_head list;
221 
222 	struct btf *btf;
223 
224 	void *priv;
225 	bpf_object_clear_priv_t clear_priv;
226 
227 	char path[];
228 };
229 #define obj_elf_valid(o)	((o)->efile.elf)
230 
231 static void bpf_program__unload(struct bpf_program *prog)
232 {
233 	int i;
234 
235 	if (!prog)
236 		return;
237 
238 	/*
239 	 * If the object is opened but the program was never loaded,
240 	 * it is possible that prog->instances.nr == -1.
241 	 */
242 	if (prog->instances.nr > 0) {
243 		for (i = 0; i < prog->instances.nr; i++)
244 			zclose(prog->instances.fds[i]);
245 	} else if (prog->instances.nr != -1) {
246 		pr_warning("Internal error: instances.nr is %d\n",
247 			   prog->instances.nr);
248 	}
249 
250 	prog->instances.nr = -1;
251 	zfree(&prog->instances.fds);
252 }
253 
254 static void bpf_program__exit(struct bpf_program *prog)
255 {
256 	if (!prog)
257 		return;
258 
259 	if (prog->clear_priv)
260 		prog->clear_priv(prog, prog->priv);
261 
262 	prog->priv = NULL;
263 	prog->clear_priv = NULL;
264 
265 	bpf_program__unload(prog);
266 	zfree(&prog->name);
267 	zfree(&prog->section_name);
268 	zfree(&prog->insns);
269 	zfree(&prog->reloc_desc);
270 
271 	prog->nr_reloc = 0;
272 	prog->insns_cnt = 0;
273 	prog->idx = -1;
274 }
275 
276 static int
277 bpf_program__init(void *data, size_t size, char *section_name, int idx,
278 		  struct bpf_program *prog)
279 {
280 	if (size < sizeof(struct bpf_insn)) {
281 		pr_warning("corrupted section '%s'\n", section_name);
282 		return -EINVAL;
283 	}
284 
285 	bzero(prog, sizeof(*prog));
286 
287 	prog->section_name = strdup(section_name);
288 	if (!prog->section_name) {
289 		pr_warning("failed to alloc name for prog under section(%d) %s\n",
290 			   idx, section_name);
291 		goto errout;
292 	}
293 
294 	prog->insns = malloc(size);
295 	if (!prog->insns) {
296 		pr_warning("failed to alloc insns for prog under section %s\n",
297 			   section_name);
298 		goto errout;
299 	}
300 	prog->insns_cnt = size / sizeof(struct bpf_insn);
301 	memcpy(prog->insns, data,
302 	       prog->insns_cnt * sizeof(struct bpf_insn));
303 	prog->idx = idx;
304 	prog->instances.fds = NULL;
305 	prog->instances.nr = -1;
306 	prog->type = BPF_PROG_TYPE_KPROBE;
307 
308 	return 0;
309 errout:
310 	bpf_program__exit(prog);
311 	return -ENOMEM;
312 }
313 
314 static int
315 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
316 			char *section_name, int idx)
317 {
318 	struct bpf_program prog, *progs;
319 	int nr_progs, err;
320 
321 	err = bpf_program__init(data, size, section_name, idx, &prog);
322 	if (err)
323 		return err;
324 
325 	progs = obj->programs;
326 	nr_progs = obj->nr_programs;
327 
328 	progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
329 	if (!progs) {
330 		/*
331 		 * In this case the original obj->programs
332 		 * is still valid, so don't need special treat for
333 		 * bpf_close_object().
334 		 */
335 		pr_warning("failed to alloc a new program under section '%s'\n",
336 			   section_name);
337 		bpf_program__exit(&prog);
338 		return -ENOMEM;
339 	}
340 
341 	pr_debug("found program %s\n", prog.section_name);
342 	obj->programs = progs;
343 	obj->nr_programs = nr_progs + 1;
344 	prog.obj = obj;
345 	progs[nr_progs] = prog;
346 	return 0;
347 }
348 
349 static int
350 bpf_object__init_prog_names(struct bpf_object *obj)
351 {
352 	Elf_Data *symbols = obj->efile.symbols;
353 	struct bpf_program *prog;
354 	size_t pi, si;
355 
356 	for (pi = 0; pi < obj->nr_programs; pi++) {
357 		const char *name = NULL;
358 
359 		prog = &obj->programs[pi];
360 
361 		for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
362 		     si++) {
363 			GElf_Sym sym;
364 
365 			if (!gelf_getsym(symbols, si, &sym))
366 				continue;
367 			if (sym.st_shndx != prog->idx)
368 				continue;
369 			if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
370 				continue;
371 
372 			name = elf_strptr(obj->efile.elf,
373 					  obj->efile.strtabidx,
374 					  sym.st_name);
375 			if (!name) {
376 				pr_warning("failed to get sym name string for prog %s\n",
377 					   prog->section_name);
378 				return -LIBBPF_ERRNO__LIBELF;
379 			}
380 		}
381 
382 		if (!name && prog->idx == obj->efile.text_shndx)
383 			name = ".text";
384 
385 		if (!name) {
386 			pr_warning("failed to find sym for prog %s\n",
387 				   prog->section_name);
388 			return -EINVAL;
389 		}
390 
391 		prog->name = strdup(name);
392 		if (!prog->name) {
393 			pr_warning("failed to allocate memory for prog sym %s\n",
394 				   name);
395 			return -ENOMEM;
396 		}
397 	}
398 
399 	return 0;
400 }
401 
402 static struct bpf_object *bpf_object__new(const char *path,
403 					  void *obj_buf,
404 					  size_t obj_buf_sz)
405 {
406 	struct bpf_object *obj;
407 
408 	obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
409 	if (!obj) {
410 		pr_warning("alloc memory failed for %s\n", path);
411 		return ERR_PTR(-ENOMEM);
412 	}
413 
414 	strcpy(obj->path, path);
415 	obj->efile.fd = -1;
416 
417 	/*
418 	 * Caller of this function should also calls
419 	 * bpf_object__elf_finish() after data collection to return
420 	 * obj_buf to user. If not, we should duplicate the buffer to
421 	 * avoid user freeing them before elf finish.
422 	 */
423 	obj->efile.obj_buf = obj_buf;
424 	obj->efile.obj_buf_sz = obj_buf_sz;
425 	obj->efile.maps_shndx = -1;
426 
427 	obj->loaded = false;
428 
429 	INIT_LIST_HEAD(&obj->list);
430 	list_add(&obj->list, &bpf_objects_list);
431 	return obj;
432 }
433 
434 static void bpf_object__elf_finish(struct bpf_object *obj)
435 {
436 	if (!obj_elf_valid(obj))
437 		return;
438 
439 	if (obj->efile.elf) {
440 		elf_end(obj->efile.elf);
441 		obj->efile.elf = NULL;
442 	}
443 	obj->efile.symbols = NULL;
444 
445 	zfree(&obj->efile.reloc);
446 	obj->efile.nr_reloc = 0;
447 	zclose(obj->efile.fd);
448 	obj->efile.obj_buf = NULL;
449 	obj->efile.obj_buf_sz = 0;
450 }
451 
452 static int bpf_object__elf_init(struct bpf_object *obj)
453 {
454 	int err = 0;
455 	GElf_Ehdr *ep;
456 
457 	if (obj_elf_valid(obj)) {
458 		pr_warning("elf init: internal error\n");
459 		return -LIBBPF_ERRNO__LIBELF;
460 	}
461 
462 	if (obj->efile.obj_buf_sz > 0) {
463 		/*
464 		 * obj_buf should have been validated by
465 		 * bpf_object__open_buffer().
466 		 */
467 		obj->efile.elf = elf_memory(obj->efile.obj_buf,
468 					    obj->efile.obj_buf_sz);
469 	} else {
470 		obj->efile.fd = open(obj->path, O_RDONLY);
471 		if (obj->efile.fd < 0) {
472 			char errmsg[STRERR_BUFSIZE];
473 			char *cp = str_error(errno, errmsg, sizeof(errmsg));
474 
475 			pr_warning("failed to open %s: %s\n", obj->path, cp);
476 			return -errno;
477 		}
478 
479 		obj->efile.elf = elf_begin(obj->efile.fd,
480 				LIBBPF_ELF_C_READ_MMAP,
481 				NULL);
482 	}
483 
484 	if (!obj->efile.elf) {
485 		pr_warning("failed to open %s as ELF file\n",
486 				obj->path);
487 		err = -LIBBPF_ERRNO__LIBELF;
488 		goto errout;
489 	}
490 
491 	if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
492 		pr_warning("failed to get EHDR from %s\n",
493 				obj->path);
494 		err = -LIBBPF_ERRNO__FORMAT;
495 		goto errout;
496 	}
497 	ep = &obj->efile.ehdr;
498 
499 	/* Old LLVM set e_machine to EM_NONE */
500 	if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
501 		pr_warning("%s is not an eBPF object file\n",
502 			obj->path);
503 		err = -LIBBPF_ERRNO__FORMAT;
504 		goto errout;
505 	}
506 
507 	return 0;
508 errout:
509 	bpf_object__elf_finish(obj);
510 	return err;
511 }
512 
513 static int
514 bpf_object__check_endianness(struct bpf_object *obj)
515 {
516 	static unsigned int const endian = 1;
517 
518 	switch (obj->efile.ehdr.e_ident[EI_DATA]) {
519 	case ELFDATA2LSB:
520 		/* We are big endian, BPF obj is little endian. */
521 		if (*(unsigned char const *)&endian != 1)
522 			goto mismatch;
523 		break;
524 
525 	case ELFDATA2MSB:
526 		/* We are little endian, BPF obj is big endian. */
527 		if (*(unsigned char const *)&endian != 0)
528 			goto mismatch;
529 		break;
530 	default:
531 		return -LIBBPF_ERRNO__ENDIAN;
532 	}
533 
534 	return 0;
535 
536 mismatch:
537 	pr_warning("Error: endianness mismatch.\n");
538 	return -LIBBPF_ERRNO__ENDIAN;
539 }
540 
541 static int
542 bpf_object__init_license(struct bpf_object *obj,
543 			 void *data, size_t size)
544 {
545 	memcpy(obj->license, data,
546 	       min(size, sizeof(obj->license) - 1));
547 	pr_debug("license of %s is %s\n", obj->path, obj->license);
548 	return 0;
549 }
550 
551 static int
552 bpf_object__init_kversion(struct bpf_object *obj,
553 			  void *data, size_t size)
554 {
555 	u32 kver;
556 
557 	if (size != sizeof(kver)) {
558 		pr_warning("invalid kver section in %s\n", obj->path);
559 		return -LIBBPF_ERRNO__FORMAT;
560 	}
561 	memcpy(&kver, data, sizeof(kver));
562 	obj->kern_version = kver;
563 	pr_debug("kernel version of %s is %x\n", obj->path,
564 		 obj->kern_version);
565 	return 0;
566 }
567 
568 static int compare_bpf_map(const void *_a, const void *_b)
569 {
570 	const struct bpf_map *a = _a;
571 	const struct bpf_map *b = _b;
572 
573 	return a->offset - b->offset;
574 }
575 
576 static int
577 bpf_object__init_maps(struct bpf_object *obj)
578 {
579 	int i, map_idx, map_def_sz, nr_maps = 0;
580 	Elf_Scn *scn;
581 	Elf_Data *data;
582 	Elf_Data *symbols = obj->efile.symbols;
583 
584 	if (obj->efile.maps_shndx < 0)
585 		return -EINVAL;
586 	if (!symbols)
587 		return -EINVAL;
588 
589 	scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
590 	if (scn)
591 		data = elf_getdata(scn, NULL);
592 	if (!scn || !data) {
593 		pr_warning("failed to get Elf_Data from map section %d\n",
594 			   obj->efile.maps_shndx);
595 		return -EINVAL;
596 	}
597 
598 	/*
599 	 * Count number of maps. Each map has a name.
600 	 * Array of maps is not supported: only the first element is
601 	 * considered.
602 	 *
603 	 * TODO: Detect array of map and report error.
604 	 */
605 	for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
606 		GElf_Sym sym;
607 
608 		if (!gelf_getsym(symbols, i, &sym))
609 			continue;
610 		if (sym.st_shndx != obj->efile.maps_shndx)
611 			continue;
612 		nr_maps++;
613 	}
614 
615 	/* Alloc obj->maps and fill nr_maps. */
616 	pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
617 		 nr_maps, data->d_size);
618 
619 	if (!nr_maps)
620 		return 0;
621 
622 	/* Assume equally sized map definitions */
623 	map_def_sz = data->d_size / nr_maps;
624 	if (!data->d_size || (data->d_size % nr_maps) != 0) {
625 		pr_warning("unable to determine map definition size "
626 			   "section %s, %d maps in %zd bytes\n",
627 			   obj->path, nr_maps, data->d_size);
628 		return -EINVAL;
629 	}
630 
631 	obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
632 	if (!obj->maps) {
633 		pr_warning("alloc maps for object failed\n");
634 		return -ENOMEM;
635 	}
636 	obj->nr_maps = nr_maps;
637 
638 	/*
639 	 * fill all fd with -1 so won't close incorrect
640 	 * fd (fd=0 is stdin) when failure (zclose won't close
641 	 * negative fd)).
642 	 */
643 	for (i = 0; i < nr_maps; i++)
644 		obj->maps[i].fd = -1;
645 
646 	/*
647 	 * Fill obj->maps using data in "maps" section.
648 	 */
649 	for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
650 		GElf_Sym sym;
651 		const char *map_name;
652 		struct bpf_map_def *def;
653 
654 		if (!gelf_getsym(symbols, i, &sym))
655 			continue;
656 		if (sym.st_shndx != obj->efile.maps_shndx)
657 			continue;
658 
659 		map_name = elf_strptr(obj->efile.elf,
660 				      obj->efile.strtabidx,
661 				      sym.st_name);
662 		obj->maps[map_idx].offset = sym.st_value;
663 		if (sym.st_value + map_def_sz > data->d_size) {
664 			pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
665 				   obj->path, map_name);
666 			return -EINVAL;
667 		}
668 
669 		obj->maps[map_idx].name = strdup(map_name);
670 		if (!obj->maps[map_idx].name) {
671 			pr_warning("failed to alloc map name\n");
672 			return -ENOMEM;
673 		}
674 		pr_debug("map %d is \"%s\"\n", map_idx,
675 			 obj->maps[map_idx].name);
676 		def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
677 		/*
678 		 * If the definition of the map in the object file fits in
679 		 * bpf_map_def, copy it.  Any extra fields in our version
680 		 * of bpf_map_def will default to zero as a result of the
681 		 * calloc above.
682 		 */
683 		if (map_def_sz <= sizeof(struct bpf_map_def)) {
684 			memcpy(&obj->maps[map_idx].def, def, map_def_sz);
685 		} else {
686 			/*
687 			 * Here the map structure being read is bigger than what
688 			 * we expect, truncate if the excess bits are all zero.
689 			 * If they are not zero, reject this map as
690 			 * incompatible.
691 			 */
692 			char *b;
693 			for (b = ((char *)def) + sizeof(struct bpf_map_def);
694 			     b < ((char *)def) + map_def_sz; b++) {
695 				if (*b != 0) {
696 					pr_warning("maps section in %s: \"%s\" "
697 						   "has unrecognized, non-zero "
698 						   "options\n",
699 						   obj->path, map_name);
700 					return -EINVAL;
701 				}
702 			}
703 			memcpy(&obj->maps[map_idx].def, def,
704 			       sizeof(struct bpf_map_def));
705 		}
706 		map_idx++;
707 	}
708 
709 	qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
710 	return 0;
711 }
712 
713 static bool section_have_execinstr(struct bpf_object *obj, int idx)
714 {
715 	Elf_Scn *scn;
716 	GElf_Shdr sh;
717 
718 	scn = elf_getscn(obj->efile.elf, idx);
719 	if (!scn)
720 		return false;
721 
722 	if (gelf_getshdr(scn, &sh) != &sh)
723 		return false;
724 
725 	if (sh.sh_flags & SHF_EXECINSTR)
726 		return true;
727 
728 	return false;
729 }
730 
731 static int bpf_object__elf_collect(struct bpf_object *obj)
732 {
733 	Elf *elf = obj->efile.elf;
734 	GElf_Ehdr *ep = &obj->efile.ehdr;
735 	Elf_Scn *scn = NULL;
736 	int idx = 0, err = 0;
737 
738 	/* Elf is corrupted/truncated, avoid calling elf_strptr. */
739 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
740 		pr_warning("failed to get e_shstrndx from %s\n",
741 			   obj->path);
742 		return -LIBBPF_ERRNO__FORMAT;
743 	}
744 
745 	while ((scn = elf_nextscn(elf, scn)) != NULL) {
746 		char *name;
747 		GElf_Shdr sh;
748 		Elf_Data *data;
749 
750 		idx++;
751 		if (gelf_getshdr(scn, &sh) != &sh) {
752 			pr_warning("failed to get section(%d) header from %s\n",
753 				   idx, obj->path);
754 			err = -LIBBPF_ERRNO__FORMAT;
755 			goto out;
756 		}
757 
758 		name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
759 		if (!name) {
760 			pr_warning("failed to get section(%d) name from %s\n",
761 				   idx, obj->path);
762 			err = -LIBBPF_ERRNO__FORMAT;
763 			goto out;
764 		}
765 
766 		data = elf_getdata(scn, 0);
767 		if (!data) {
768 			pr_warning("failed to get section(%d) data from %s(%s)\n",
769 				   idx, name, obj->path);
770 			err = -LIBBPF_ERRNO__FORMAT;
771 			goto out;
772 		}
773 		pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
774 			 idx, name, (unsigned long)data->d_size,
775 			 (int)sh.sh_link, (unsigned long)sh.sh_flags,
776 			 (int)sh.sh_type);
777 
778 		if (strcmp(name, "license") == 0)
779 			err = bpf_object__init_license(obj,
780 						       data->d_buf,
781 						       data->d_size);
782 		else if (strcmp(name, "version") == 0)
783 			err = bpf_object__init_kversion(obj,
784 							data->d_buf,
785 							data->d_size);
786 		else if (strcmp(name, "maps") == 0)
787 			obj->efile.maps_shndx = idx;
788 		else if (strcmp(name, BTF_ELF_SEC) == 0) {
789 			obj->btf = btf__new(data->d_buf, data->d_size,
790 					    __pr_debug);
791 			if (IS_ERR(obj->btf)) {
792 				pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
793 					   BTF_ELF_SEC, PTR_ERR(obj->btf));
794 				obj->btf = NULL;
795 			}
796 		} else if (sh.sh_type == SHT_SYMTAB) {
797 			if (obj->efile.symbols) {
798 				pr_warning("bpf: multiple SYMTAB in %s\n",
799 					   obj->path);
800 				err = -LIBBPF_ERRNO__FORMAT;
801 			} else {
802 				obj->efile.symbols = data;
803 				obj->efile.strtabidx = sh.sh_link;
804 			}
805 		} else if ((sh.sh_type == SHT_PROGBITS) &&
806 			   (sh.sh_flags & SHF_EXECINSTR) &&
807 			   (data->d_size > 0)) {
808 			if (strcmp(name, ".text") == 0)
809 				obj->efile.text_shndx = idx;
810 			err = bpf_object__add_program(obj, data->d_buf,
811 						      data->d_size, name, idx);
812 			if (err) {
813 				char errmsg[STRERR_BUFSIZE];
814 				char *cp = str_error(-err, errmsg, sizeof(errmsg));
815 
816 				pr_warning("failed to alloc program %s (%s): %s",
817 					   name, obj->path, cp);
818 			}
819 		} else if (sh.sh_type == SHT_REL) {
820 			void *reloc = obj->efile.reloc;
821 			int nr_reloc = obj->efile.nr_reloc + 1;
822 			int sec = sh.sh_info; /* points to other section */
823 
824 			/* Only do relo for section with exec instructions */
825 			if (!section_have_execinstr(obj, sec)) {
826 				pr_debug("skip relo %s(%d) for section(%d)\n",
827 					 name, idx, sec);
828 				continue;
829 			}
830 
831 			reloc = reallocarray(reloc, nr_reloc,
832 					     sizeof(*obj->efile.reloc));
833 			if (!reloc) {
834 				pr_warning("realloc failed\n");
835 				err = -ENOMEM;
836 			} else {
837 				int n = nr_reloc - 1;
838 
839 				obj->efile.reloc = reloc;
840 				obj->efile.nr_reloc = nr_reloc;
841 
842 				obj->efile.reloc[n].shdr = sh;
843 				obj->efile.reloc[n].data = data;
844 			}
845 		} else {
846 			pr_debug("skip section(%d) %s\n", idx, name);
847 		}
848 		if (err)
849 			goto out;
850 	}
851 
852 	if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
853 		pr_warning("Corrupted ELF file: index of strtab invalid\n");
854 		return LIBBPF_ERRNO__FORMAT;
855 	}
856 	if (obj->efile.maps_shndx >= 0) {
857 		err = bpf_object__init_maps(obj);
858 		if (err)
859 			goto out;
860 	}
861 	err = bpf_object__init_prog_names(obj);
862 out:
863 	return err;
864 }
865 
866 static struct bpf_program *
867 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
868 {
869 	struct bpf_program *prog;
870 	size_t i;
871 
872 	for (i = 0; i < obj->nr_programs; i++) {
873 		prog = &obj->programs[i];
874 		if (prog->idx == idx)
875 			return prog;
876 	}
877 	return NULL;
878 }
879 
880 struct bpf_program *
881 bpf_object__find_program_by_title(struct bpf_object *obj, const char *title)
882 {
883 	struct bpf_program *pos;
884 
885 	bpf_object__for_each_program(pos, obj) {
886 		if (pos->section_name && !strcmp(pos->section_name, title))
887 			return pos;
888 	}
889 	return NULL;
890 }
891 
892 static int
893 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
894 			   Elf_Data *data, struct bpf_object *obj)
895 {
896 	Elf_Data *symbols = obj->efile.symbols;
897 	int text_shndx = obj->efile.text_shndx;
898 	int maps_shndx = obj->efile.maps_shndx;
899 	struct bpf_map *maps = obj->maps;
900 	size_t nr_maps = obj->nr_maps;
901 	int i, nrels;
902 
903 	pr_debug("collecting relocating info for: '%s'\n",
904 		 prog->section_name);
905 	nrels = shdr->sh_size / shdr->sh_entsize;
906 
907 	prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
908 	if (!prog->reloc_desc) {
909 		pr_warning("failed to alloc memory in relocation\n");
910 		return -ENOMEM;
911 	}
912 	prog->nr_reloc = nrels;
913 
914 	for (i = 0; i < nrels; i++) {
915 		GElf_Sym sym;
916 		GElf_Rel rel;
917 		unsigned int insn_idx;
918 		struct bpf_insn *insns = prog->insns;
919 		size_t map_idx;
920 
921 		if (!gelf_getrel(data, i, &rel)) {
922 			pr_warning("relocation: failed to get %d reloc\n", i);
923 			return -LIBBPF_ERRNO__FORMAT;
924 		}
925 
926 		if (!gelf_getsym(symbols,
927 				 GELF_R_SYM(rel.r_info),
928 				 &sym)) {
929 			pr_warning("relocation: symbol %"PRIx64" not found\n",
930 				   GELF_R_SYM(rel.r_info));
931 			return -LIBBPF_ERRNO__FORMAT;
932 		}
933 		pr_debug("relo for %lld value %lld name %d\n",
934 			 (long long) (rel.r_info >> 32),
935 			 (long long) sym.st_value, sym.st_name);
936 
937 		if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
938 			pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
939 				   prog->section_name, sym.st_shndx);
940 			return -LIBBPF_ERRNO__RELOC;
941 		}
942 
943 		insn_idx = rel.r_offset / sizeof(struct bpf_insn);
944 		pr_debug("relocation: insn_idx=%u\n", insn_idx);
945 
946 		if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
947 			if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
948 				pr_warning("incorrect bpf_call opcode\n");
949 				return -LIBBPF_ERRNO__RELOC;
950 			}
951 			prog->reloc_desc[i].type = RELO_CALL;
952 			prog->reloc_desc[i].insn_idx = insn_idx;
953 			prog->reloc_desc[i].text_off = sym.st_value;
954 			obj->has_pseudo_calls = true;
955 			continue;
956 		}
957 
958 		if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
959 			pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
960 				   insn_idx, insns[insn_idx].code);
961 			return -LIBBPF_ERRNO__RELOC;
962 		}
963 
964 		/* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
965 		for (map_idx = 0; map_idx < nr_maps; map_idx++) {
966 			if (maps[map_idx].offset == sym.st_value) {
967 				pr_debug("relocation: find map %zd (%s) for insn %u\n",
968 					 map_idx, maps[map_idx].name, insn_idx);
969 				break;
970 			}
971 		}
972 
973 		if (map_idx >= nr_maps) {
974 			pr_warning("bpf relocation: map_idx %d large than %d\n",
975 				   (int)map_idx, (int)nr_maps - 1);
976 			return -LIBBPF_ERRNO__RELOC;
977 		}
978 
979 		prog->reloc_desc[i].type = RELO_LD64;
980 		prog->reloc_desc[i].insn_idx = insn_idx;
981 		prog->reloc_desc[i].map_idx = map_idx;
982 	}
983 	return 0;
984 }
985 
986 static int bpf_map_find_btf_info(struct bpf_map *map, const struct btf *btf)
987 {
988 	const struct btf_type *container_type;
989 	const struct btf_member *key, *value;
990 	struct bpf_map_def *def = &map->def;
991 	const size_t max_name = 256;
992 	char container_name[max_name];
993 	__s64 key_size, value_size;
994 	__s32 container_id;
995 
996 	if (snprintf(container_name, max_name, "____btf_map_%s", map->name) ==
997 	    max_name) {
998 		pr_warning("map:%s length of '____btf_map_%s' is too long\n",
999 			   map->name, map->name);
1000 		return -EINVAL;
1001 	}
1002 
1003 	container_id = btf__find_by_name(btf, container_name);
1004 	if (container_id < 0) {
1005 		pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1006 			 map->name, container_name);
1007 		return container_id;
1008 	}
1009 
1010 	container_type = btf__type_by_id(btf, container_id);
1011 	if (!container_type) {
1012 		pr_warning("map:%s cannot find BTF type for container_id:%u\n",
1013 			   map->name, container_id);
1014 		return -EINVAL;
1015 	}
1016 
1017 	if (BTF_INFO_KIND(container_type->info) != BTF_KIND_STRUCT ||
1018 	    BTF_INFO_VLEN(container_type->info) < 2) {
1019 		pr_warning("map:%s container_name:%s is an invalid container struct\n",
1020 			   map->name, container_name);
1021 		return -EINVAL;
1022 	}
1023 
1024 	key = (struct btf_member *)(container_type + 1);
1025 	value = key + 1;
1026 
1027 	key_size = btf__resolve_size(btf, key->type);
1028 	if (key_size < 0) {
1029 		pr_warning("map:%s invalid BTF key_type_size\n",
1030 			   map->name);
1031 		return key_size;
1032 	}
1033 
1034 	if (def->key_size != key_size) {
1035 		pr_warning("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1036 			   map->name, (__u32)key_size, def->key_size);
1037 		return -EINVAL;
1038 	}
1039 
1040 	value_size = btf__resolve_size(btf, value->type);
1041 	if (value_size < 0) {
1042 		pr_warning("map:%s invalid BTF value_type_size\n", map->name);
1043 		return value_size;
1044 	}
1045 
1046 	if (def->value_size != value_size) {
1047 		pr_warning("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1048 			   map->name, (__u32)value_size, def->value_size);
1049 		return -EINVAL;
1050 	}
1051 
1052 	map->btf_key_type_id = key->type;
1053 	map->btf_value_type_id = value->type;
1054 
1055 	return 0;
1056 }
1057 
1058 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1059 {
1060 	struct bpf_map_info info = {};
1061 	__u32 len = sizeof(info);
1062 	int new_fd, err;
1063 	char *new_name;
1064 
1065 	err = bpf_obj_get_info_by_fd(fd, &info, &len);
1066 	if (err)
1067 		return err;
1068 
1069 	new_name = strdup(info.name);
1070 	if (!new_name)
1071 		return -errno;
1072 
1073 	new_fd = open("/", O_RDONLY | O_CLOEXEC);
1074 	if (new_fd < 0)
1075 		goto err_free_new_name;
1076 
1077 	new_fd = dup3(fd, new_fd, O_CLOEXEC);
1078 	if (new_fd < 0)
1079 		goto err_close_new_fd;
1080 
1081 	err = zclose(map->fd);
1082 	if (err)
1083 		goto err_close_new_fd;
1084 	free(map->name);
1085 
1086 	map->fd = new_fd;
1087 	map->name = new_name;
1088 	map->def.type = info.type;
1089 	map->def.key_size = info.key_size;
1090 	map->def.value_size = info.value_size;
1091 	map->def.max_entries = info.max_entries;
1092 	map->def.map_flags = info.map_flags;
1093 	map->btf_key_type_id = info.btf_key_type_id;
1094 	map->btf_value_type_id = info.btf_value_type_id;
1095 
1096 	return 0;
1097 
1098 err_close_new_fd:
1099 	close(new_fd);
1100 err_free_new_name:
1101 	free(new_name);
1102 	return -errno;
1103 }
1104 
1105 static int
1106 bpf_object__create_maps(struct bpf_object *obj)
1107 {
1108 	struct bpf_create_map_attr create_attr = {};
1109 	unsigned int i;
1110 	int err;
1111 
1112 	for (i = 0; i < obj->nr_maps; i++) {
1113 		struct bpf_map *map = &obj->maps[i];
1114 		struct bpf_map_def *def = &map->def;
1115 		char *cp, errmsg[STRERR_BUFSIZE];
1116 		int *pfd = &map->fd;
1117 
1118 		if (map->fd >= 0) {
1119 			pr_debug("skip map create (preset) %s: fd=%d\n",
1120 				 map->name, map->fd);
1121 			continue;
1122 		}
1123 
1124 		create_attr.name = map->name;
1125 		create_attr.map_ifindex = map->map_ifindex;
1126 		create_attr.map_type = def->type;
1127 		create_attr.map_flags = def->map_flags;
1128 		create_attr.key_size = def->key_size;
1129 		create_attr.value_size = def->value_size;
1130 		create_attr.max_entries = def->max_entries;
1131 		create_attr.btf_fd = 0;
1132 		create_attr.btf_key_type_id = 0;
1133 		create_attr.btf_value_type_id = 0;
1134 
1135 		if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
1136 			create_attr.btf_fd = btf__fd(obj->btf);
1137 			create_attr.btf_key_type_id = map->btf_key_type_id;
1138 			create_attr.btf_value_type_id = map->btf_value_type_id;
1139 		}
1140 
1141 		*pfd = bpf_create_map_xattr(&create_attr);
1142 		if (*pfd < 0 && create_attr.btf_key_type_id) {
1143 			cp = str_error(errno, errmsg, sizeof(errmsg));
1144 			pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
1145 				   map->name, cp, errno);
1146 			create_attr.btf_fd = 0;
1147 			create_attr.btf_key_type_id = 0;
1148 			create_attr.btf_value_type_id = 0;
1149 			map->btf_key_type_id = 0;
1150 			map->btf_value_type_id = 0;
1151 			*pfd = bpf_create_map_xattr(&create_attr);
1152 		}
1153 
1154 		if (*pfd < 0) {
1155 			size_t j;
1156 
1157 			err = *pfd;
1158 			cp = str_error(errno, errmsg, sizeof(errmsg));
1159 			pr_warning("failed to create map (name: '%s'): %s\n",
1160 				   map->name, cp);
1161 			for (j = 0; j < i; j++)
1162 				zclose(obj->maps[j].fd);
1163 			return err;
1164 		}
1165 		pr_debug("create map %s: fd=%d\n", map->name, *pfd);
1166 	}
1167 
1168 	return 0;
1169 }
1170 
1171 static int
1172 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1173 			struct reloc_desc *relo)
1174 {
1175 	struct bpf_insn *insn, *new_insn;
1176 	struct bpf_program *text;
1177 	size_t new_cnt;
1178 
1179 	if (relo->type != RELO_CALL)
1180 		return -LIBBPF_ERRNO__RELOC;
1181 
1182 	if (prog->idx == obj->efile.text_shndx) {
1183 		pr_warning("relo in .text insn %d into off %d\n",
1184 			   relo->insn_idx, relo->text_off);
1185 		return -LIBBPF_ERRNO__RELOC;
1186 	}
1187 
1188 	if (prog->main_prog_cnt == 0) {
1189 		text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1190 		if (!text) {
1191 			pr_warning("no .text section found yet relo into text exist\n");
1192 			return -LIBBPF_ERRNO__RELOC;
1193 		}
1194 		new_cnt = prog->insns_cnt + text->insns_cnt;
1195 		new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
1196 		if (!new_insn) {
1197 			pr_warning("oom in prog realloc\n");
1198 			return -ENOMEM;
1199 		}
1200 		memcpy(new_insn + prog->insns_cnt, text->insns,
1201 		       text->insns_cnt * sizeof(*insn));
1202 		prog->insns = new_insn;
1203 		prog->main_prog_cnt = prog->insns_cnt;
1204 		prog->insns_cnt = new_cnt;
1205 		pr_debug("added %zd insn from %s to prog %s\n",
1206 			 text->insns_cnt, text->section_name,
1207 			 prog->section_name);
1208 	}
1209 	insn = &prog->insns[relo->insn_idx];
1210 	insn->imm += prog->main_prog_cnt - relo->insn_idx;
1211 	return 0;
1212 }
1213 
1214 static int
1215 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1216 {
1217 	int i, err;
1218 
1219 	if (!prog || !prog->reloc_desc)
1220 		return 0;
1221 
1222 	for (i = 0; i < prog->nr_reloc; i++) {
1223 		if (prog->reloc_desc[i].type == RELO_LD64) {
1224 			struct bpf_insn *insns = prog->insns;
1225 			int insn_idx, map_idx;
1226 
1227 			insn_idx = prog->reloc_desc[i].insn_idx;
1228 			map_idx = prog->reloc_desc[i].map_idx;
1229 
1230 			if (insn_idx >= (int)prog->insns_cnt) {
1231 				pr_warning("relocation out of range: '%s'\n",
1232 					   prog->section_name);
1233 				return -LIBBPF_ERRNO__RELOC;
1234 			}
1235 			insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1236 			insns[insn_idx].imm = obj->maps[map_idx].fd;
1237 		} else {
1238 			err = bpf_program__reloc_text(prog, obj,
1239 						      &prog->reloc_desc[i]);
1240 			if (err)
1241 				return err;
1242 		}
1243 	}
1244 
1245 	zfree(&prog->reloc_desc);
1246 	prog->nr_reloc = 0;
1247 	return 0;
1248 }
1249 
1250 
1251 static int
1252 bpf_object__relocate(struct bpf_object *obj)
1253 {
1254 	struct bpf_program *prog;
1255 	size_t i;
1256 	int err;
1257 
1258 	for (i = 0; i < obj->nr_programs; i++) {
1259 		prog = &obj->programs[i];
1260 
1261 		err = bpf_program__relocate(prog, obj);
1262 		if (err) {
1263 			pr_warning("failed to relocate '%s'\n",
1264 				   prog->section_name);
1265 			return err;
1266 		}
1267 	}
1268 	return 0;
1269 }
1270 
1271 static int bpf_object__collect_reloc(struct bpf_object *obj)
1272 {
1273 	int i, err;
1274 
1275 	if (!obj_elf_valid(obj)) {
1276 		pr_warning("Internal error: elf object is closed\n");
1277 		return -LIBBPF_ERRNO__INTERNAL;
1278 	}
1279 
1280 	for (i = 0; i < obj->efile.nr_reloc; i++) {
1281 		GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1282 		Elf_Data *data = obj->efile.reloc[i].data;
1283 		int idx = shdr->sh_info;
1284 		struct bpf_program *prog;
1285 
1286 		if (shdr->sh_type != SHT_REL) {
1287 			pr_warning("internal error at %d\n", __LINE__);
1288 			return -LIBBPF_ERRNO__INTERNAL;
1289 		}
1290 
1291 		prog = bpf_object__find_prog_by_idx(obj, idx);
1292 		if (!prog) {
1293 			pr_warning("relocation failed: no section(%d)\n", idx);
1294 			return -LIBBPF_ERRNO__RELOC;
1295 		}
1296 
1297 		err = bpf_program__collect_reloc(prog,
1298 						 shdr, data,
1299 						 obj);
1300 		if (err)
1301 			return err;
1302 	}
1303 	return 0;
1304 }
1305 
1306 static int
1307 load_program(enum bpf_prog_type type, enum bpf_attach_type expected_attach_type,
1308 	     const char *name, struct bpf_insn *insns, int insns_cnt,
1309 	     char *license, u32 kern_version, int *pfd, int prog_ifindex)
1310 {
1311 	struct bpf_load_program_attr load_attr;
1312 	char *cp, errmsg[STRERR_BUFSIZE];
1313 	char *log_buf;
1314 	int ret;
1315 
1316 	memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
1317 	load_attr.prog_type = type;
1318 	load_attr.expected_attach_type = expected_attach_type;
1319 	load_attr.name = name;
1320 	load_attr.insns = insns;
1321 	load_attr.insns_cnt = insns_cnt;
1322 	load_attr.license = license;
1323 	load_attr.kern_version = kern_version;
1324 	load_attr.prog_ifindex = prog_ifindex;
1325 
1326 	if (!load_attr.insns || !load_attr.insns_cnt)
1327 		return -EINVAL;
1328 
1329 	log_buf = malloc(BPF_LOG_BUF_SIZE);
1330 	if (!log_buf)
1331 		pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1332 
1333 	ret = bpf_load_program_xattr(&load_attr, log_buf, BPF_LOG_BUF_SIZE);
1334 
1335 	if (ret >= 0) {
1336 		*pfd = ret;
1337 		ret = 0;
1338 		goto out;
1339 	}
1340 
1341 	ret = -LIBBPF_ERRNO__LOAD;
1342 	cp = str_error(errno, errmsg, sizeof(errmsg));
1343 	pr_warning("load bpf program failed: %s\n", cp);
1344 
1345 	if (log_buf && log_buf[0] != '\0') {
1346 		ret = -LIBBPF_ERRNO__VERIFY;
1347 		pr_warning("-- BEGIN DUMP LOG ---\n");
1348 		pr_warning("\n%s\n", log_buf);
1349 		pr_warning("-- END LOG --\n");
1350 	} else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
1351 		pr_warning("Program too large (%zu insns), at most %d insns\n",
1352 			   load_attr.insns_cnt, BPF_MAXINSNS);
1353 		ret = -LIBBPF_ERRNO__PROG2BIG;
1354 	} else {
1355 		/* Wrong program type? */
1356 		if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
1357 			int fd;
1358 
1359 			load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
1360 			load_attr.expected_attach_type = 0;
1361 			fd = bpf_load_program_xattr(&load_attr, NULL, 0);
1362 			if (fd >= 0) {
1363 				close(fd);
1364 				ret = -LIBBPF_ERRNO__PROGTYPE;
1365 				goto out;
1366 			}
1367 		}
1368 
1369 		if (log_buf)
1370 			ret = -LIBBPF_ERRNO__KVER;
1371 	}
1372 
1373 out:
1374 	free(log_buf);
1375 	return ret;
1376 }
1377 
1378 static int
1379 bpf_program__load(struct bpf_program *prog,
1380 		  char *license, u32 kern_version)
1381 {
1382 	int err = 0, fd, i;
1383 
1384 	if (prog->instances.nr < 0 || !prog->instances.fds) {
1385 		if (prog->preprocessor) {
1386 			pr_warning("Internal error: can't load program '%s'\n",
1387 				   prog->section_name);
1388 			return -LIBBPF_ERRNO__INTERNAL;
1389 		}
1390 
1391 		prog->instances.fds = malloc(sizeof(int));
1392 		if (!prog->instances.fds) {
1393 			pr_warning("Not enough memory for BPF fds\n");
1394 			return -ENOMEM;
1395 		}
1396 		prog->instances.nr = 1;
1397 		prog->instances.fds[0] = -1;
1398 	}
1399 
1400 	if (!prog->preprocessor) {
1401 		if (prog->instances.nr != 1) {
1402 			pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1403 				   prog->section_name, prog->instances.nr);
1404 		}
1405 		err = load_program(prog->type, prog->expected_attach_type,
1406 				   prog->name, prog->insns, prog->insns_cnt,
1407 				   license, kern_version, &fd,
1408 				   prog->prog_ifindex);
1409 		if (!err)
1410 			prog->instances.fds[0] = fd;
1411 		goto out;
1412 	}
1413 
1414 	for (i = 0; i < prog->instances.nr; i++) {
1415 		struct bpf_prog_prep_result result;
1416 		bpf_program_prep_t preprocessor = prog->preprocessor;
1417 
1418 		bzero(&result, sizeof(result));
1419 		err = preprocessor(prog, i, prog->insns,
1420 				   prog->insns_cnt, &result);
1421 		if (err) {
1422 			pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1423 				   i, prog->section_name);
1424 			goto out;
1425 		}
1426 
1427 		if (!result.new_insn_ptr || !result.new_insn_cnt) {
1428 			pr_debug("Skip loading the %dth instance of program '%s'\n",
1429 				 i, prog->section_name);
1430 			prog->instances.fds[i] = -1;
1431 			if (result.pfd)
1432 				*result.pfd = -1;
1433 			continue;
1434 		}
1435 
1436 		err = load_program(prog->type, prog->expected_attach_type,
1437 				   prog->name, result.new_insn_ptr,
1438 				   result.new_insn_cnt,
1439 				   license, kern_version, &fd,
1440 				   prog->prog_ifindex);
1441 
1442 		if (err) {
1443 			pr_warning("Loading the %dth instance of program '%s' failed\n",
1444 					i, prog->section_name);
1445 			goto out;
1446 		}
1447 
1448 		if (result.pfd)
1449 			*result.pfd = fd;
1450 		prog->instances.fds[i] = fd;
1451 	}
1452 out:
1453 	if (err)
1454 		pr_warning("failed to load program '%s'\n",
1455 			   prog->section_name);
1456 	zfree(&prog->insns);
1457 	prog->insns_cnt = 0;
1458 	return err;
1459 }
1460 
1461 static bool bpf_program__is_function_storage(struct bpf_program *prog,
1462 					     struct bpf_object *obj)
1463 {
1464 	return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
1465 }
1466 
1467 static int
1468 bpf_object__load_progs(struct bpf_object *obj)
1469 {
1470 	size_t i;
1471 	int err;
1472 
1473 	for (i = 0; i < obj->nr_programs; i++) {
1474 		if (bpf_program__is_function_storage(&obj->programs[i], obj))
1475 			continue;
1476 		err = bpf_program__load(&obj->programs[i],
1477 					obj->license,
1478 					obj->kern_version);
1479 		if (err)
1480 			return err;
1481 	}
1482 	return 0;
1483 }
1484 
1485 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
1486 {
1487 	switch (type) {
1488 	case BPF_PROG_TYPE_SOCKET_FILTER:
1489 	case BPF_PROG_TYPE_SCHED_CLS:
1490 	case BPF_PROG_TYPE_SCHED_ACT:
1491 	case BPF_PROG_TYPE_XDP:
1492 	case BPF_PROG_TYPE_CGROUP_SKB:
1493 	case BPF_PROG_TYPE_CGROUP_SOCK:
1494 	case BPF_PROG_TYPE_LWT_IN:
1495 	case BPF_PROG_TYPE_LWT_OUT:
1496 	case BPF_PROG_TYPE_LWT_XMIT:
1497 	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
1498 	case BPF_PROG_TYPE_SOCK_OPS:
1499 	case BPF_PROG_TYPE_SK_SKB:
1500 	case BPF_PROG_TYPE_CGROUP_DEVICE:
1501 	case BPF_PROG_TYPE_SK_MSG:
1502 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1503 	case BPF_PROG_TYPE_LIRC_MODE2:
1504 	case BPF_PROG_TYPE_SK_REUSEPORT:
1505 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
1506 		return false;
1507 	case BPF_PROG_TYPE_UNSPEC:
1508 	case BPF_PROG_TYPE_KPROBE:
1509 	case BPF_PROG_TYPE_TRACEPOINT:
1510 	case BPF_PROG_TYPE_PERF_EVENT:
1511 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
1512 	default:
1513 		return true;
1514 	}
1515 }
1516 
1517 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
1518 {
1519 	if (needs_kver && obj->kern_version == 0) {
1520 		pr_warning("%s doesn't provide kernel version\n",
1521 			   obj->path);
1522 		return -LIBBPF_ERRNO__KVERSION;
1523 	}
1524 	return 0;
1525 }
1526 
1527 static struct bpf_object *
1528 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
1529 		   bool needs_kver)
1530 {
1531 	struct bpf_object *obj;
1532 	int err;
1533 
1534 	if (elf_version(EV_CURRENT) == EV_NONE) {
1535 		pr_warning("failed to init libelf for %s\n", path);
1536 		return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1537 	}
1538 
1539 	obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1540 	if (IS_ERR(obj))
1541 		return obj;
1542 
1543 	CHECK_ERR(bpf_object__elf_init(obj), err, out);
1544 	CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1545 	CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1546 	CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1547 	CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
1548 
1549 	bpf_object__elf_finish(obj);
1550 	return obj;
1551 out:
1552 	bpf_object__close(obj);
1553 	return ERR_PTR(err);
1554 }
1555 
1556 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
1557 {
1558 	/* param validation */
1559 	if (!attr->file)
1560 		return NULL;
1561 
1562 	pr_debug("loading %s\n", attr->file);
1563 
1564 	return __bpf_object__open(attr->file, NULL, 0,
1565 				  bpf_prog_type__needs_kver(attr->prog_type));
1566 }
1567 
1568 struct bpf_object *bpf_object__open(const char *path)
1569 {
1570 	struct bpf_object_open_attr attr = {
1571 		.file		= path,
1572 		.prog_type	= BPF_PROG_TYPE_UNSPEC,
1573 	};
1574 
1575 	return bpf_object__open_xattr(&attr);
1576 }
1577 
1578 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1579 					   size_t obj_buf_sz,
1580 					   const char *name)
1581 {
1582 	char tmp_name[64];
1583 
1584 	/* param validation */
1585 	if (!obj_buf || obj_buf_sz <= 0)
1586 		return NULL;
1587 
1588 	if (!name) {
1589 		snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1590 			 (unsigned long)obj_buf,
1591 			 (unsigned long)obj_buf_sz);
1592 		tmp_name[sizeof(tmp_name) - 1] = '\0';
1593 		name = tmp_name;
1594 	}
1595 	pr_debug("loading object '%s' from buffer\n",
1596 		 name);
1597 
1598 	return __bpf_object__open(name, obj_buf, obj_buf_sz, true);
1599 }
1600 
1601 int bpf_object__unload(struct bpf_object *obj)
1602 {
1603 	size_t i;
1604 
1605 	if (!obj)
1606 		return -EINVAL;
1607 
1608 	for (i = 0; i < obj->nr_maps; i++)
1609 		zclose(obj->maps[i].fd);
1610 
1611 	for (i = 0; i < obj->nr_programs; i++)
1612 		bpf_program__unload(&obj->programs[i]);
1613 
1614 	return 0;
1615 }
1616 
1617 int bpf_object__load(struct bpf_object *obj)
1618 {
1619 	int err;
1620 
1621 	if (!obj)
1622 		return -EINVAL;
1623 
1624 	if (obj->loaded) {
1625 		pr_warning("object should not be loaded twice\n");
1626 		return -EINVAL;
1627 	}
1628 
1629 	obj->loaded = true;
1630 
1631 	CHECK_ERR(bpf_object__create_maps(obj), err, out);
1632 	CHECK_ERR(bpf_object__relocate(obj), err, out);
1633 	CHECK_ERR(bpf_object__load_progs(obj), err, out);
1634 
1635 	return 0;
1636 out:
1637 	bpf_object__unload(obj);
1638 	pr_warning("failed to load object '%s'\n", obj->path);
1639 	return err;
1640 }
1641 
1642 static int check_path(const char *path)
1643 {
1644 	char *cp, errmsg[STRERR_BUFSIZE];
1645 	struct statfs st_fs;
1646 	char *dname, *dir;
1647 	int err = 0;
1648 
1649 	if (path == NULL)
1650 		return -EINVAL;
1651 
1652 	dname = strdup(path);
1653 	if (dname == NULL)
1654 		return -ENOMEM;
1655 
1656 	dir = dirname(dname);
1657 	if (statfs(dir, &st_fs)) {
1658 		cp = str_error(errno, errmsg, sizeof(errmsg));
1659 		pr_warning("failed to statfs %s: %s\n", dir, cp);
1660 		err = -errno;
1661 	}
1662 	free(dname);
1663 
1664 	if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1665 		pr_warning("specified path %s is not on BPF FS\n", path);
1666 		err = -EINVAL;
1667 	}
1668 
1669 	return err;
1670 }
1671 
1672 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1673 			      int instance)
1674 {
1675 	char *cp, errmsg[STRERR_BUFSIZE];
1676 	int err;
1677 
1678 	err = check_path(path);
1679 	if (err)
1680 		return err;
1681 
1682 	if (prog == NULL) {
1683 		pr_warning("invalid program pointer\n");
1684 		return -EINVAL;
1685 	}
1686 
1687 	if (instance < 0 || instance >= prog->instances.nr) {
1688 		pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1689 			   instance, prog->section_name, prog->instances.nr);
1690 		return -EINVAL;
1691 	}
1692 
1693 	if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1694 		cp = str_error(errno, errmsg, sizeof(errmsg));
1695 		pr_warning("failed to pin program: %s\n", cp);
1696 		return -errno;
1697 	}
1698 	pr_debug("pinned program '%s'\n", path);
1699 
1700 	return 0;
1701 }
1702 
1703 static int make_dir(const char *path)
1704 {
1705 	char *cp, errmsg[STRERR_BUFSIZE];
1706 	int err = 0;
1707 
1708 	if (mkdir(path, 0700) && errno != EEXIST)
1709 		err = -errno;
1710 
1711 	if (err) {
1712 		cp = str_error(-err, errmsg, sizeof(errmsg));
1713 		pr_warning("failed to mkdir %s: %s\n", path, cp);
1714 	}
1715 	return err;
1716 }
1717 
1718 int bpf_program__pin(struct bpf_program *prog, const char *path)
1719 {
1720 	int i, err;
1721 
1722 	err = check_path(path);
1723 	if (err)
1724 		return err;
1725 
1726 	if (prog == NULL) {
1727 		pr_warning("invalid program pointer\n");
1728 		return -EINVAL;
1729 	}
1730 
1731 	if (prog->instances.nr <= 0) {
1732 		pr_warning("no instances of prog %s to pin\n",
1733 			   prog->section_name);
1734 		return -EINVAL;
1735 	}
1736 
1737 	err = make_dir(path);
1738 	if (err)
1739 		return err;
1740 
1741 	for (i = 0; i < prog->instances.nr; i++) {
1742 		char buf[PATH_MAX];
1743 		int len;
1744 
1745 		len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1746 		if (len < 0)
1747 			return -EINVAL;
1748 		else if (len >= PATH_MAX)
1749 			return -ENAMETOOLONG;
1750 
1751 		err = bpf_program__pin_instance(prog, buf, i);
1752 		if (err)
1753 			return err;
1754 	}
1755 
1756 	return 0;
1757 }
1758 
1759 int bpf_map__pin(struct bpf_map *map, const char *path)
1760 {
1761 	char *cp, errmsg[STRERR_BUFSIZE];
1762 	int err;
1763 
1764 	err = check_path(path);
1765 	if (err)
1766 		return err;
1767 
1768 	if (map == NULL) {
1769 		pr_warning("invalid map pointer\n");
1770 		return -EINVAL;
1771 	}
1772 
1773 	if (bpf_obj_pin(map->fd, path)) {
1774 		cp = str_error(errno, errmsg, sizeof(errmsg));
1775 		pr_warning("failed to pin map: %s\n", cp);
1776 		return -errno;
1777 	}
1778 
1779 	pr_debug("pinned map '%s'\n", path);
1780 	return 0;
1781 }
1782 
1783 int bpf_object__pin(struct bpf_object *obj, const char *path)
1784 {
1785 	struct bpf_program *prog;
1786 	struct bpf_map *map;
1787 	int err;
1788 
1789 	if (!obj)
1790 		return -ENOENT;
1791 
1792 	if (!obj->loaded) {
1793 		pr_warning("object not yet loaded; load it first\n");
1794 		return -ENOENT;
1795 	}
1796 
1797 	err = make_dir(path);
1798 	if (err)
1799 		return err;
1800 
1801 	bpf_map__for_each(map, obj) {
1802 		char buf[PATH_MAX];
1803 		int len;
1804 
1805 		len = snprintf(buf, PATH_MAX, "%s/%s", path,
1806 			       bpf_map__name(map));
1807 		if (len < 0)
1808 			return -EINVAL;
1809 		else if (len >= PATH_MAX)
1810 			return -ENAMETOOLONG;
1811 
1812 		err = bpf_map__pin(map, buf);
1813 		if (err)
1814 			return err;
1815 	}
1816 
1817 	bpf_object__for_each_program(prog, obj) {
1818 		char buf[PATH_MAX];
1819 		int len;
1820 
1821 		len = snprintf(buf, PATH_MAX, "%s/%s", path,
1822 			       prog->section_name);
1823 		if (len < 0)
1824 			return -EINVAL;
1825 		else if (len >= PATH_MAX)
1826 			return -ENAMETOOLONG;
1827 
1828 		err = bpf_program__pin(prog, buf);
1829 		if (err)
1830 			return err;
1831 	}
1832 
1833 	return 0;
1834 }
1835 
1836 void bpf_object__close(struct bpf_object *obj)
1837 {
1838 	size_t i;
1839 
1840 	if (!obj)
1841 		return;
1842 
1843 	if (obj->clear_priv)
1844 		obj->clear_priv(obj, obj->priv);
1845 
1846 	bpf_object__elf_finish(obj);
1847 	bpf_object__unload(obj);
1848 	btf__free(obj->btf);
1849 
1850 	for (i = 0; i < obj->nr_maps; i++) {
1851 		zfree(&obj->maps[i].name);
1852 		if (obj->maps[i].clear_priv)
1853 			obj->maps[i].clear_priv(&obj->maps[i],
1854 						obj->maps[i].priv);
1855 		obj->maps[i].priv = NULL;
1856 		obj->maps[i].clear_priv = NULL;
1857 	}
1858 	zfree(&obj->maps);
1859 	obj->nr_maps = 0;
1860 
1861 	if (obj->programs && obj->nr_programs) {
1862 		for (i = 0; i < obj->nr_programs; i++)
1863 			bpf_program__exit(&obj->programs[i]);
1864 	}
1865 	zfree(&obj->programs);
1866 
1867 	list_del(&obj->list);
1868 	free(obj);
1869 }
1870 
1871 struct bpf_object *
1872 bpf_object__next(struct bpf_object *prev)
1873 {
1874 	struct bpf_object *next;
1875 
1876 	if (!prev)
1877 		next = list_first_entry(&bpf_objects_list,
1878 					struct bpf_object,
1879 					list);
1880 	else
1881 		next = list_next_entry(prev, list);
1882 
1883 	/* Empty list is noticed here so don't need checking on entry. */
1884 	if (&next->list == &bpf_objects_list)
1885 		return NULL;
1886 
1887 	return next;
1888 }
1889 
1890 const char *bpf_object__name(struct bpf_object *obj)
1891 {
1892 	return obj ? obj->path : ERR_PTR(-EINVAL);
1893 }
1894 
1895 unsigned int bpf_object__kversion(struct bpf_object *obj)
1896 {
1897 	return obj ? obj->kern_version : 0;
1898 }
1899 
1900 int bpf_object__btf_fd(const struct bpf_object *obj)
1901 {
1902 	return obj->btf ? btf__fd(obj->btf) : -1;
1903 }
1904 
1905 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1906 			 bpf_object_clear_priv_t clear_priv)
1907 {
1908 	if (obj->priv && obj->clear_priv)
1909 		obj->clear_priv(obj, obj->priv);
1910 
1911 	obj->priv = priv;
1912 	obj->clear_priv = clear_priv;
1913 	return 0;
1914 }
1915 
1916 void *bpf_object__priv(struct bpf_object *obj)
1917 {
1918 	return obj ? obj->priv : ERR_PTR(-EINVAL);
1919 }
1920 
1921 static struct bpf_program *
1922 __bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1923 {
1924 	size_t idx;
1925 
1926 	if (!obj->programs)
1927 		return NULL;
1928 	/* First handler */
1929 	if (prev == NULL)
1930 		return &obj->programs[0];
1931 
1932 	if (prev->obj != obj) {
1933 		pr_warning("error: program handler doesn't match object\n");
1934 		return NULL;
1935 	}
1936 
1937 	idx = (prev - obj->programs) + 1;
1938 	if (idx >= obj->nr_programs)
1939 		return NULL;
1940 	return &obj->programs[idx];
1941 }
1942 
1943 struct bpf_program *
1944 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1945 {
1946 	struct bpf_program *prog = prev;
1947 
1948 	do {
1949 		prog = __bpf_program__next(prog, obj);
1950 	} while (prog && bpf_program__is_function_storage(prog, obj));
1951 
1952 	return prog;
1953 }
1954 
1955 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1956 			  bpf_program_clear_priv_t clear_priv)
1957 {
1958 	if (prog->priv && prog->clear_priv)
1959 		prog->clear_priv(prog, prog->priv);
1960 
1961 	prog->priv = priv;
1962 	prog->clear_priv = clear_priv;
1963 	return 0;
1964 }
1965 
1966 void *bpf_program__priv(struct bpf_program *prog)
1967 {
1968 	return prog ? prog->priv : ERR_PTR(-EINVAL);
1969 }
1970 
1971 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
1972 {
1973 	prog->prog_ifindex = ifindex;
1974 }
1975 
1976 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1977 {
1978 	const char *title;
1979 
1980 	title = prog->section_name;
1981 	if (needs_copy) {
1982 		title = strdup(title);
1983 		if (!title) {
1984 			pr_warning("failed to strdup program title\n");
1985 			return ERR_PTR(-ENOMEM);
1986 		}
1987 	}
1988 
1989 	return title;
1990 }
1991 
1992 int bpf_program__fd(struct bpf_program *prog)
1993 {
1994 	return bpf_program__nth_fd(prog, 0);
1995 }
1996 
1997 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1998 			  bpf_program_prep_t prep)
1999 {
2000 	int *instances_fds;
2001 
2002 	if (nr_instances <= 0 || !prep)
2003 		return -EINVAL;
2004 
2005 	if (prog->instances.nr > 0 || prog->instances.fds) {
2006 		pr_warning("Can't set pre-processor after loading\n");
2007 		return -EINVAL;
2008 	}
2009 
2010 	instances_fds = malloc(sizeof(int) * nr_instances);
2011 	if (!instances_fds) {
2012 		pr_warning("alloc memory failed for fds\n");
2013 		return -ENOMEM;
2014 	}
2015 
2016 	/* fill all fd with -1 */
2017 	memset(instances_fds, -1, sizeof(int) * nr_instances);
2018 
2019 	prog->instances.nr = nr_instances;
2020 	prog->instances.fds = instances_fds;
2021 	prog->preprocessor = prep;
2022 	return 0;
2023 }
2024 
2025 int bpf_program__nth_fd(struct bpf_program *prog, int n)
2026 {
2027 	int fd;
2028 
2029 	if (!prog)
2030 		return -EINVAL;
2031 
2032 	if (n >= prog->instances.nr || n < 0) {
2033 		pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
2034 			   n, prog->section_name, prog->instances.nr);
2035 		return -EINVAL;
2036 	}
2037 
2038 	fd = prog->instances.fds[n];
2039 	if (fd < 0) {
2040 		pr_warning("%dth instance of program '%s' is invalid\n",
2041 			   n, prog->section_name);
2042 		return -ENOENT;
2043 	}
2044 
2045 	return fd;
2046 }
2047 
2048 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
2049 {
2050 	prog->type = type;
2051 }
2052 
2053 static bool bpf_program__is_type(struct bpf_program *prog,
2054 				 enum bpf_prog_type type)
2055 {
2056 	return prog ? (prog->type == type) : false;
2057 }
2058 
2059 #define BPF_PROG_TYPE_FNS(NAME, TYPE)			\
2060 int bpf_program__set_##NAME(struct bpf_program *prog)	\
2061 {							\
2062 	if (!prog)					\
2063 		return -EINVAL;				\
2064 	bpf_program__set_type(prog, TYPE);		\
2065 	return 0;					\
2066 }							\
2067 							\
2068 bool bpf_program__is_##NAME(struct bpf_program *prog)	\
2069 {							\
2070 	return bpf_program__is_type(prog, TYPE);	\
2071 }							\
2072 
2073 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
2074 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
2075 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
2076 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
2077 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
2078 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
2079 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
2080 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
2081 
2082 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
2083 					   enum bpf_attach_type type)
2084 {
2085 	prog->expected_attach_type = type;
2086 }
2087 
2088 #define BPF_PROG_SEC_FULL(string, ptype, atype) \
2089 	{ string, sizeof(string) - 1, ptype, atype }
2090 
2091 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_FULL(string, ptype, 0)
2092 
2093 #define BPF_S_PROG_SEC(string, ptype) \
2094 	BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK, ptype)
2095 
2096 #define BPF_SA_PROG_SEC(string, ptype) \
2097 	BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK_ADDR, ptype)
2098 
2099 static const struct {
2100 	const char *sec;
2101 	size_t len;
2102 	enum bpf_prog_type prog_type;
2103 	enum bpf_attach_type expected_attach_type;
2104 } section_names[] = {
2105 	BPF_PROG_SEC("socket",		BPF_PROG_TYPE_SOCKET_FILTER),
2106 	BPF_PROG_SEC("kprobe/",		BPF_PROG_TYPE_KPROBE),
2107 	BPF_PROG_SEC("kretprobe/",	BPF_PROG_TYPE_KPROBE),
2108 	BPF_PROG_SEC("classifier",	BPF_PROG_TYPE_SCHED_CLS),
2109 	BPF_PROG_SEC("action",		BPF_PROG_TYPE_SCHED_ACT),
2110 	BPF_PROG_SEC("tracepoint/",	BPF_PROG_TYPE_TRACEPOINT),
2111 	BPF_PROG_SEC("raw_tracepoint/",	BPF_PROG_TYPE_RAW_TRACEPOINT),
2112 	BPF_PROG_SEC("xdp",		BPF_PROG_TYPE_XDP),
2113 	BPF_PROG_SEC("perf_event",	BPF_PROG_TYPE_PERF_EVENT),
2114 	BPF_PROG_SEC("cgroup/skb",	BPF_PROG_TYPE_CGROUP_SKB),
2115 	BPF_PROG_SEC("cgroup/sock",	BPF_PROG_TYPE_CGROUP_SOCK),
2116 	BPF_PROG_SEC("cgroup/dev",	BPF_PROG_TYPE_CGROUP_DEVICE),
2117 	BPF_PROG_SEC("lwt_in",		BPF_PROG_TYPE_LWT_IN),
2118 	BPF_PROG_SEC("lwt_out",		BPF_PROG_TYPE_LWT_OUT),
2119 	BPF_PROG_SEC("lwt_xmit",	BPF_PROG_TYPE_LWT_XMIT),
2120 	BPF_PROG_SEC("lwt_seg6local",	BPF_PROG_TYPE_LWT_SEG6LOCAL),
2121 	BPF_PROG_SEC("sockops",		BPF_PROG_TYPE_SOCK_OPS),
2122 	BPF_PROG_SEC("sk_skb",		BPF_PROG_TYPE_SK_SKB),
2123 	BPF_PROG_SEC("sk_msg",		BPF_PROG_TYPE_SK_MSG),
2124 	BPF_PROG_SEC("lirc_mode2",	BPF_PROG_TYPE_LIRC_MODE2),
2125 	BPF_PROG_SEC("flow_dissector",	BPF_PROG_TYPE_FLOW_DISSECTOR),
2126 	BPF_SA_PROG_SEC("cgroup/bind4",	BPF_CGROUP_INET4_BIND),
2127 	BPF_SA_PROG_SEC("cgroup/bind6",	BPF_CGROUP_INET6_BIND),
2128 	BPF_SA_PROG_SEC("cgroup/connect4", BPF_CGROUP_INET4_CONNECT),
2129 	BPF_SA_PROG_SEC("cgroup/connect6", BPF_CGROUP_INET6_CONNECT),
2130 	BPF_SA_PROG_SEC("cgroup/sendmsg4", BPF_CGROUP_UDP4_SENDMSG),
2131 	BPF_SA_PROG_SEC("cgroup/sendmsg6", BPF_CGROUP_UDP6_SENDMSG),
2132 	BPF_S_PROG_SEC("cgroup/post_bind4", BPF_CGROUP_INET4_POST_BIND),
2133 	BPF_S_PROG_SEC("cgroup/post_bind6", BPF_CGROUP_INET6_POST_BIND),
2134 };
2135 
2136 #undef BPF_PROG_SEC
2137 #undef BPF_PROG_SEC_FULL
2138 #undef BPF_S_PROG_SEC
2139 #undef BPF_SA_PROG_SEC
2140 
2141 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
2142 			     enum bpf_attach_type *expected_attach_type)
2143 {
2144 	int i;
2145 
2146 	if (!name)
2147 		return -EINVAL;
2148 
2149 	for (i = 0; i < ARRAY_SIZE(section_names); i++) {
2150 		if (strncmp(name, section_names[i].sec, section_names[i].len))
2151 			continue;
2152 		*prog_type = section_names[i].prog_type;
2153 		*expected_attach_type = section_names[i].expected_attach_type;
2154 		return 0;
2155 	}
2156 	return -EINVAL;
2157 }
2158 
2159 static int
2160 bpf_program__identify_section(struct bpf_program *prog,
2161 			      enum bpf_prog_type *prog_type,
2162 			      enum bpf_attach_type *expected_attach_type)
2163 {
2164 	return libbpf_prog_type_by_name(prog->section_name, prog_type,
2165 					expected_attach_type);
2166 }
2167 
2168 int bpf_map__fd(struct bpf_map *map)
2169 {
2170 	return map ? map->fd : -EINVAL;
2171 }
2172 
2173 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
2174 {
2175 	return map ? &map->def : ERR_PTR(-EINVAL);
2176 }
2177 
2178 const char *bpf_map__name(struct bpf_map *map)
2179 {
2180 	return map ? map->name : NULL;
2181 }
2182 
2183 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
2184 {
2185 	return map ? map->btf_key_type_id : 0;
2186 }
2187 
2188 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
2189 {
2190 	return map ? map->btf_value_type_id : 0;
2191 }
2192 
2193 int bpf_map__set_priv(struct bpf_map *map, void *priv,
2194 		     bpf_map_clear_priv_t clear_priv)
2195 {
2196 	if (!map)
2197 		return -EINVAL;
2198 
2199 	if (map->priv) {
2200 		if (map->clear_priv)
2201 			map->clear_priv(map, map->priv);
2202 	}
2203 
2204 	map->priv = priv;
2205 	map->clear_priv = clear_priv;
2206 	return 0;
2207 }
2208 
2209 void *bpf_map__priv(struct bpf_map *map)
2210 {
2211 	return map ? map->priv : ERR_PTR(-EINVAL);
2212 }
2213 
2214 bool bpf_map__is_offload_neutral(struct bpf_map *map)
2215 {
2216 	return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
2217 }
2218 
2219 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
2220 {
2221 	map->map_ifindex = ifindex;
2222 }
2223 
2224 struct bpf_map *
2225 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
2226 {
2227 	size_t idx;
2228 	struct bpf_map *s, *e;
2229 
2230 	if (!obj || !obj->maps)
2231 		return NULL;
2232 
2233 	s = obj->maps;
2234 	e = obj->maps + obj->nr_maps;
2235 
2236 	if (prev == NULL)
2237 		return s;
2238 
2239 	if ((prev < s) || (prev >= e)) {
2240 		pr_warning("error in %s: map handler doesn't belong to object\n",
2241 			   __func__);
2242 		return NULL;
2243 	}
2244 
2245 	idx = (prev - obj->maps) + 1;
2246 	if (idx >= obj->nr_maps)
2247 		return NULL;
2248 	return &obj->maps[idx];
2249 }
2250 
2251 struct bpf_map *
2252 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
2253 {
2254 	struct bpf_map *pos;
2255 
2256 	bpf_map__for_each(pos, obj) {
2257 		if (pos->name && !strcmp(pos->name, name))
2258 			return pos;
2259 	}
2260 	return NULL;
2261 }
2262 
2263 struct bpf_map *
2264 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
2265 {
2266 	int i;
2267 
2268 	for (i = 0; i < obj->nr_maps; i++) {
2269 		if (obj->maps[i].offset == offset)
2270 			return &obj->maps[i];
2271 	}
2272 	return ERR_PTR(-ENOENT);
2273 }
2274 
2275 long libbpf_get_error(const void *ptr)
2276 {
2277 	if (IS_ERR(ptr))
2278 		return PTR_ERR(ptr);
2279 	return 0;
2280 }
2281 
2282 int bpf_prog_load(const char *file, enum bpf_prog_type type,
2283 		  struct bpf_object **pobj, int *prog_fd)
2284 {
2285 	struct bpf_prog_load_attr attr;
2286 
2287 	memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
2288 	attr.file = file;
2289 	attr.prog_type = type;
2290 	attr.expected_attach_type = 0;
2291 
2292 	return bpf_prog_load_xattr(&attr, pobj, prog_fd);
2293 }
2294 
2295 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
2296 			struct bpf_object **pobj, int *prog_fd)
2297 {
2298 	struct bpf_object_open_attr open_attr = {
2299 		.file		= attr->file,
2300 		.prog_type	= attr->prog_type,
2301 	};
2302 	struct bpf_program *prog, *first_prog = NULL;
2303 	enum bpf_attach_type expected_attach_type;
2304 	enum bpf_prog_type prog_type;
2305 	struct bpf_object *obj;
2306 	struct bpf_map *map;
2307 	int err;
2308 
2309 	if (!attr)
2310 		return -EINVAL;
2311 	if (!attr->file)
2312 		return -EINVAL;
2313 
2314 	obj = bpf_object__open_xattr(&open_attr);
2315 	if (IS_ERR_OR_NULL(obj))
2316 		return -ENOENT;
2317 
2318 	bpf_object__for_each_program(prog, obj) {
2319 		/*
2320 		 * If type is not specified, try to guess it based on
2321 		 * section name.
2322 		 */
2323 		prog_type = attr->prog_type;
2324 		prog->prog_ifindex = attr->ifindex;
2325 		expected_attach_type = attr->expected_attach_type;
2326 		if (prog_type == BPF_PROG_TYPE_UNSPEC) {
2327 			err = bpf_program__identify_section(prog, &prog_type,
2328 							    &expected_attach_type);
2329 			if (err < 0) {
2330 				pr_warning("failed to guess program type based on section name %s\n",
2331 					   prog->section_name);
2332 				bpf_object__close(obj);
2333 				return -EINVAL;
2334 			}
2335 		}
2336 
2337 		bpf_program__set_type(prog, prog_type);
2338 		bpf_program__set_expected_attach_type(prog,
2339 						      expected_attach_type);
2340 
2341 		if (!first_prog)
2342 			first_prog = prog;
2343 	}
2344 
2345 	bpf_map__for_each(map, obj) {
2346 		if (!bpf_map__is_offload_neutral(map))
2347 			map->map_ifindex = attr->ifindex;
2348 	}
2349 
2350 	if (!first_prog) {
2351 		pr_warning("object file doesn't contain bpf program\n");
2352 		bpf_object__close(obj);
2353 		return -ENOENT;
2354 	}
2355 
2356 	err = bpf_object__load(obj);
2357 	if (err) {
2358 		bpf_object__close(obj);
2359 		return -EINVAL;
2360 	}
2361 
2362 	*pobj = obj;
2363 	*prog_fd = bpf_program__fd(first_prog);
2364 	return 0;
2365 }
2366 
2367 enum bpf_perf_event_ret
2368 bpf_perf_event_read_simple(void *mem, unsigned long size,
2369 			   unsigned long page_size, void **buf, size_t *buf_len,
2370 			   bpf_perf_event_print_t fn, void *priv)
2371 {
2372 	volatile struct perf_event_mmap_page *header = mem;
2373 	__u64 data_tail = header->data_tail;
2374 	__u64 data_head = header->data_head;
2375 	int ret = LIBBPF_PERF_EVENT_ERROR;
2376 	void *base, *begin, *end;
2377 
2378 	asm volatile("" ::: "memory"); /* in real code it should be smp_rmb() */
2379 	if (data_head == data_tail)
2380 		return LIBBPF_PERF_EVENT_CONT;
2381 
2382 	base = ((char *)header) + page_size;
2383 
2384 	begin = base + data_tail % size;
2385 	end = base + data_head % size;
2386 
2387 	while (begin != end) {
2388 		struct perf_event_header *ehdr;
2389 
2390 		ehdr = begin;
2391 		if (begin + ehdr->size > base + size) {
2392 			long len = base + size - begin;
2393 
2394 			if (*buf_len < ehdr->size) {
2395 				free(*buf);
2396 				*buf = malloc(ehdr->size);
2397 				if (!*buf) {
2398 					ret = LIBBPF_PERF_EVENT_ERROR;
2399 					break;
2400 				}
2401 				*buf_len = ehdr->size;
2402 			}
2403 
2404 			memcpy(*buf, begin, len);
2405 			memcpy(*buf + len, base, ehdr->size - len);
2406 			ehdr = (void *)*buf;
2407 			begin = base + ehdr->size - len;
2408 		} else if (begin + ehdr->size == base + size) {
2409 			begin = base;
2410 		} else {
2411 			begin += ehdr->size;
2412 		}
2413 
2414 		ret = fn(ehdr, priv);
2415 		if (ret != LIBBPF_PERF_EVENT_CONT)
2416 			break;
2417 
2418 		data_tail += ehdr->size;
2419 	}
2420 
2421 	__sync_synchronize(); /* smp_mb() */
2422 	header->data_tail = data_tail;
2423 
2424 	return ret;
2425 }
2426