mm, memory_hotplug: try to migrate full pfn range
[linux/fpc-iii.git] / samples / bpf / bpf_load.c
blobeae7b635343d13182a4d458d3280464f6ffee4ca
1 // SPDX-License-Identifier: GPL-2.0
2 #include <stdio.h>
3 #include <sys/types.h>
4 #include <sys/stat.h>
5 #include <fcntl.h>
6 #include <libelf.h>
7 #include <gelf.h>
8 #include <errno.h>
9 #include <unistd.h>
10 #include <string.h>
11 #include <stdbool.h>
12 #include <stdlib.h>
13 #include <linux/bpf.h>
14 #include <linux/filter.h>
15 #include <linux/perf_event.h>
16 #include <linux/netlink.h>
17 #include <linux/rtnetlink.h>
18 #include <linux/types.h>
19 #include <sys/socket.h>
20 #include <sys/syscall.h>
21 #include <sys/ioctl.h>
22 #include <sys/mman.h>
23 #include <poll.h>
24 #include <ctype.h>
25 #include <assert.h>
26 #include <bpf/bpf.h>
27 #include "bpf_load.h"
28 #include "perf-sys.h"
30 #define DEBUGFS "/sys/kernel/debug/tracing/"
32 static char license[128];
33 static int kern_version;
34 static bool processed_sec[128];
35 char bpf_log_buf[BPF_LOG_BUF_SIZE];
36 int map_fd[MAX_MAPS];
37 int prog_fd[MAX_PROGS];
38 int event_fd[MAX_PROGS];
39 int prog_cnt;
40 int prog_array_fd = -1;
42 struct bpf_map_data map_data[MAX_MAPS];
43 int map_data_count = 0;
45 static int populate_prog_array(const char *event, int prog_fd)
47 int ind = atoi(event), err;
49 err = bpf_map_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY);
50 if (err < 0) {
51 printf("failed to store prog_fd in prog_array\n");
52 return -1;
54 return 0;
57 static int write_kprobe_events(const char *val)
59 int fd, ret, flags;
61 if (val == NULL)
62 return -1;
63 else if (val[0] == '\0')
64 flags = O_WRONLY | O_TRUNC;
65 else
66 flags = O_WRONLY | O_APPEND;
68 fd = open("/sys/kernel/debug/tracing/kprobe_events", flags);
70 ret = write(fd, val, strlen(val));
71 close(fd);
73 return ret;
76 static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
78 bool is_socket = strncmp(event, "socket", 6) == 0;
79 bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
80 bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
81 bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
82 bool is_raw_tracepoint = strncmp(event, "raw_tracepoint/", 15) == 0;
83 bool is_xdp = strncmp(event, "xdp", 3) == 0;
84 bool is_perf_event = strncmp(event, "perf_event", 10) == 0;
85 bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0;
86 bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0;
87 bool is_sockops = strncmp(event, "sockops", 7) == 0;
88 bool is_sk_skb = strncmp(event, "sk_skb", 6) == 0;
89 bool is_sk_msg = strncmp(event, "sk_msg", 6) == 0;
90 size_t insns_cnt = size / sizeof(struct bpf_insn);
91 enum bpf_prog_type prog_type;
92 char buf[256];
93 int fd, efd, err, id;
94 struct perf_event_attr attr = {};
96 attr.type = PERF_TYPE_TRACEPOINT;
97 attr.sample_type = PERF_SAMPLE_RAW;
98 attr.sample_period = 1;
99 attr.wakeup_events = 1;
101 if (is_socket) {
102 prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
103 } else if (is_kprobe || is_kretprobe) {
104 prog_type = BPF_PROG_TYPE_KPROBE;
105 } else if (is_tracepoint) {
106 prog_type = BPF_PROG_TYPE_TRACEPOINT;
107 } else if (is_raw_tracepoint) {
108 prog_type = BPF_PROG_TYPE_RAW_TRACEPOINT;
109 } else if (is_xdp) {
110 prog_type = BPF_PROG_TYPE_XDP;
111 } else if (is_perf_event) {
112 prog_type = BPF_PROG_TYPE_PERF_EVENT;
113 } else if (is_cgroup_skb) {
114 prog_type = BPF_PROG_TYPE_CGROUP_SKB;
115 } else if (is_cgroup_sk) {
116 prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
117 } else if (is_sockops) {
118 prog_type = BPF_PROG_TYPE_SOCK_OPS;
119 } else if (is_sk_skb) {
120 prog_type = BPF_PROG_TYPE_SK_SKB;
121 } else if (is_sk_msg) {
122 prog_type = BPF_PROG_TYPE_SK_MSG;
123 } else {
124 printf("Unknown event '%s'\n", event);
125 return -1;
128 if (prog_cnt == MAX_PROGS)
129 return -1;
131 fd = bpf_load_program(prog_type, prog, insns_cnt, license, kern_version,
132 bpf_log_buf, BPF_LOG_BUF_SIZE);
133 if (fd < 0) {
134 printf("bpf_load_program() err=%d\n%s", errno, bpf_log_buf);
135 return -1;
138 prog_fd[prog_cnt++] = fd;
140 if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk)
141 return 0;
143 if (is_socket || is_sockops || is_sk_skb || is_sk_msg) {
144 if (is_socket)
145 event += 6;
146 else
147 event += 7;
148 if (*event != '/')
149 return 0;
150 event++;
151 if (!isdigit(*event)) {
152 printf("invalid prog number\n");
153 return -1;
155 return populate_prog_array(event, fd);
158 if (is_raw_tracepoint) {
159 efd = bpf_raw_tracepoint_open(event + 15, fd);
160 if (efd < 0) {
161 printf("tracepoint %s %s\n", event + 15, strerror(errno));
162 return -1;
164 event_fd[prog_cnt - 1] = efd;
165 return 0;
168 if (is_kprobe || is_kretprobe) {
169 bool need_normal_check = true;
170 const char *event_prefix = "";
172 if (is_kprobe)
173 event += 7;
174 else
175 event += 10;
177 if (*event == 0) {
178 printf("event name cannot be empty\n");
179 return -1;
182 if (isdigit(*event))
183 return populate_prog_array(event, fd);
185 #ifdef __x86_64__
186 if (strncmp(event, "sys_", 4) == 0) {
187 snprintf(buf, sizeof(buf), "%c:__x64_%s __x64_%s",
188 is_kprobe ? 'p' : 'r', event, event);
189 err = write_kprobe_events(buf);
190 if (err >= 0) {
191 need_normal_check = false;
192 event_prefix = "__x64_";
195 #endif
196 if (need_normal_check) {
197 snprintf(buf, sizeof(buf), "%c:%s %s",
198 is_kprobe ? 'p' : 'r', event, event);
199 err = write_kprobe_events(buf);
200 if (err < 0) {
201 printf("failed to create kprobe '%s' error '%s'\n",
202 event, strerror(errno));
203 return -1;
207 strcpy(buf, DEBUGFS);
208 strcat(buf, "events/kprobes/");
209 strcat(buf, event_prefix);
210 strcat(buf, event);
211 strcat(buf, "/id");
212 } else if (is_tracepoint) {
213 event += 11;
215 if (*event == 0) {
216 printf("event name cannot be empty\n");
217 return -1;
219 strcpy(buf, DEBUGFS);
220 strcat(buf, "events/");
221 strcat(buf, event);
222 strcat(buf, "/id");
225 efd = open(buf, O_RDONLY, 0);
226 if (efd < 0) {
227 printf("failed to open event %s\n", event);
228 return -1;
231 err = read(efd, buf, sizeof(buf));
232 if (err < 0 || err >= sizeof(buf)) {
233 printf("read from '%s' failed '%s'\n", event, strerror(errno));
234 return -1;
237 close(efd);
239 buf[err] = 0;
240 id = atoi(buf);
241 attr.config = id;
243 efd = sys_perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
244 if (efd < 0) {
245 printf("event %d fd %d err %s\n", id, efd, strerror(errno));
246 return -1;
248 event_fd[prog_cnt - 1] = efd;
249 err = ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
250 if (err < 0) {
251 printf("ioctl PERF_EVENT_IOC_ENABLE failed err %s\n",
252 strerror(errno));
253 return -1;
255 err = ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
256 if (err < 0) {
257 printf("ioctl PERF_EVENT_IOC_SET_BPF failed err %s\n",
258 strerror(errno));
259 return -1;
262 return 0;
265 static int load_maps(struct bpf_map_data *maps, int nr_maps,
266 fixup_map_cb fixup_map)
268 int i, numa_node;
270 for (i = 0; i < nr_maps; i++) {
271 if (fixup_map) {
272 fixup_map(&maps[i], i);
273 /* Allow userspace to assign map FD prior to creation */
274 if (maps[i].fd != -1) {
275 map_fd[i] = maps[i].fd;
276 continue;
280 numa_node = maps[i].def.map_flags & BPF_F_NUMA_NODE ?
281 maps[i].def.numa_node : -1;
283 if (maps[i].def.type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
284 maps[i].def.type == BPF_MAP_TYPE_HASH_OF_MAPS) {
285 int inner_map_fd = map_fd[maps[i].def.inner_map_idx];
287 map_fd[i] = bpf_create_map_in_map_node(maps[i].def.type,
288 maps[i].name,
289 maps[i].def.key_size,
290 inner_map_fd,
291 maps[i].def.max_entries,
292 maps[i].def.map_flags,
293 numa_node);
294 } else {
295 map_fd[i] = bpf_create_map_node(maps[i].def.type,
296 maps[i].name,
297 maps[i].def.key_size,
298 maps[i].def.value_size,
299 maps[i].def.max_entries,
300 maps[i].def.map_flags,
301 numa_node);
303 if (map_fd[i] < 0) {
304 printf("failed to create map %d (%s): %d %s\n",
305 i, maps[i].name, errno, strerror(errno));
306 return 1;
308 maps[i].fd = map_fd[i];
310 if (maps[i].def.type == BPF_MAP_TYPE_PROG_ARRAY)
311 prog_array_fd = map_fd[i];
313 return 0;
316 static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
317 GElf_Shdr *shdr, Elf_Data **data)
319 Elf_Scn *scn;
321 scn = elf_getscn(elf, i);
322 if (!scn)
323 return 1;
325 if (gelf_getshdr(scn, shdr) != shdr)
326 return 2;
328 *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
329 if (!*shname || !shdr->sh_size)
330 return 3;
332 *data = elf_getdata(scn, 0);
333 if (!*data || elf_getdata(scn, *data) != NULL)
334 return 4;
336 return 0;
339 static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
340 GElf_Shdr *shdr, struct bpf_insn *insn,
341 struct bpf_map_data *maps, int nr_maps)
343 int i, nrels;
345 nrels = shdr->sh_size / shdr->sh_entsize;
347 for (i = 0; i < nrels; i++) {
348 GElf_Sym sym;
349 GElf_Rel rel;
350 unsigned int insn_idx;
351 bool match = false;
352 int j, map_idx;
354 gelf_getrel(data, i, &rel);
356 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
358 gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
360 if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
361 printf("invalid relo for insn[%d].code 0x%x\n",
362 insn_idx, insn[insn_idx].code);
363 return 1;
365 insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
367 /* Match FD relocation against recorded map_data[] offset */
368 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
369 if (maps[map_idx].elf_offset == sym.st_value) {
370 match = true;
371 break;
374 if (match) {
375 insn[insn_idx].imm = maps[map_idx].fd;
376 } else {
377 printf("invalid relo for insn[%d] no map_data match\n",
378 insn_idx);
379 return 1;
383 return 0;
386 static int cmp_symbols(const void *l, const void *r)
388 const GElf_Sym *lsym = (const GElf_Sym *)l;
389 const GElf_Sym *rsym = (const GElf_Sym *)r;
391 if (lsym->st_value < rsym->st_value)
392 return -1;
393 else if (lsym->st_value > rsym->st_value)
394 return 1;
395 else
396 return 0;
399 static int load_elf_maps_section(struct bpf_map_data *maps, int maps_shndx,
400 Elf *elf, Elf_Data *symbols, int strtabidx)
402 int map_sz_elf, map_sz_copy;
403 bool validate_zero = false;
404 Elf_Data *data_maps;
405 int i, nr_maps;
406 GElf_Sym *sym;
407 Elf_Scn *scn;
408 int copy_sz;
410 if (maps_shndx < 0)
411 return -EINVAL;
412 if (!symbols)
413 return -EINVAL;
415 /* Get data for maps section via elf index */
416 scn = elf_getscn(elf, maps_shndx);
417 if (scn)
418 data_maps = elf_getdata(scn, NULL);
419 if (!scn || !data_maps) {
420 printf("Failed to get Elf_Data from maps section %d\n",
421 maps_shndx);
422 return -EINVAL;
425 /* For each map get corrosponding symbol table entry */
426 sym = calloc(MAX_MAPS+1, sizeof(GElf_Sym));
427 for (i = 0, nr_maps = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
428 assert(nr_maps < MAX_MAPS+1);
429 if (!gelf_getsym(symbols, i, &sym[nr_maps]))
430 continue;
431 if (sym[nr_maps].st_shndx != maps_shndx)
432 continue;
433 /* Only increment iif maps section */
434 nr_maps++;
437 /* Align to map_fd[] order, via sort on offset in sym.st_value */
438 qsort(sym, nr_maps, sizeof(GElf_Sym), cmp_symbols);
440 /* Keeping compatible with ELF maps section changes
441 * ------------------------------------------------
442 * The program size of struct bpf_load_map_def is known by loader
443 * code, but struct stored in ELF file can be different.
445 * Unfortunately sym[i].st_size is zero. To calculate the
446 * struct size stored in the ELF file, assume all struct have
447 * the same size, and simply divide with number of map
448 * symbols.
450 map_sz_elf = data_maps->d_size / nr_maps;
451 map_sz_copy = sizeof(struct bpf_load_map_def);
452 if (map_sz_elf < map_sz_copy) {
454 * Backward compat, loading older ELF file with
455 * smaller struct, keeping remaining bytes zero.
457 map_sz_copy = map_sz_elf;
458 } else if (map_sz_elf > map_sz_copy) {
460 * Forward compat, loading newer ELF file with larger
461 * struct with unknown features. Assume zero means
462 * feature not used. Thus, validate rest of struct
463 * data is zero.
465 validate_zero = true;
468 /* Memcpy relevant part of ELF maps data to loader maps */
469 for (i = 0; i < nr_maps; i++) {
470 struct bpf_load_map_def *def;
471 unsigned char *addr, *end;
472 const char *map_name;
473 size_t offset;
475 map_name = elf_strptr(elf, strtabidx, sym[i].st_name);
476 maps[i].name = strdup(map_name);
477 if (!maps[i].name) {
478 printf("strdup(%s): %s(%d)\n", map_name,
479 strerror(errno), errno);
480 free(sym);
481 return -errno;
484 /* Symbol value is offset into ELF maps section data area */
485 offset = sym[i].st_value;
486 def = (struct bpf_load_map_def *)(data_maps->d_buf + offset);
487 maps[i].elf_offset = offset;
488 memset(&maps[i].def, 0, sizeof(struct bpf_load_map_def));
489 memcpy(&maps[i].def, def, map_sz_copy);
491 /* Verify no newer features were requested */
492 if (validate_zero) {
493 addr = (unsigned char*) def + map_sz_copy;
494 end = (unsigned char*) def + map_sz_elf;
495 for (; addr < end; addr++) {
496 if (*addr != 0) {
497 free(sym);
498 return -EFBIG;
504 free(sym);
505 return nr_maps;
508 static int do_load_bpf_file(const char *path, fixup_map_cb fixup_map)
510 int fd, i, ret, maps_shndx = -1, strtabidx = -1;
511 Elf *elf;
512 GElf_Ehdr ehdr;
513 GElf_Shdr shdr, shdr_prog;
514 Elf_Data *data, *data_prog, *data_maps = NULL, *symbols = NULL;
515 char *shname, *shname_prog;
516 int nr_maps = 0;
518 /* reset global variables */
519 kern_version = 0;
520 memset(license, 0, sizeof(license));
521 memset(processed_sec, 0, sizeof(processed_sec));
523 if (elf_version(EV_CURRENT) == EV_NONE)
524 return 1;
526 fd = open(path, O_RDONLY, 0);
527 if (fd < 0)
528 return 1;
530 elf = elf_begin(fd, ELF_C_READ, NULL);
532 if (!elf)
533 return 1;
535 if (gelf_getehdr(elf, &ehdr) != &ehdr)
536 return 1;
538 /* clear all kprobes */
539 i = write_kprobe_events("");
541 /* scan over all elf sections to get license and map info */
542 for (i = 1; i < ehdr.e_shnum; i++) {
544 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
545 continue;
547 if (0) /* helpful for llvm debugging */
548 printf("section %d:%s data %p size %zd link %d flags %d\n",
549 i, shname, data->d_buf, data->d_size,
550 shdr.sh_link, (int) shdr.sh_flags);
552 if (strcmp(shname, "license") == 0) {
553 processed_sec[i] = true;
554 memcpy(license, data->d_buf, data->d_size);
555 } else if (strcmp(shname, "version") == 0) {
556 processed_sec[i] = true;
557 if (data->d_size != sizeof(int)) {
558 printf("invalid size of version section %zd\n",
559 data->d_size);
560 return 1;
562 memcpy(&kern_version, data->d_buf, sizeof(int));
563 } else if (strcmp(shname, "maps") == 0) {
564 int j;
566 maps_shndx = i;
567 data_maps = data;
568 for (j = 0; j < MAX_MAPS; j++)
569 map_data[j].fd = -1;
570 } else if (shdr.sh_type == SHT_SYMTAB) {
571 strtabidx = shdr.sh_link;
572 symbols = data;
576 ret = 1;
578 if (!symbols) {
579 printf("missing SHT_SYMTAB section\n");
580 goto done;
583 if (data_maps) {
584 nr_maps = load_elf_maps_section(map_data, maps_shndx,
585 elf, symbols, strtabidx);
586 if (nr_maps < 0) {
587 printf("Error: Failed loading ELF maps (errno:%d):%s\n",
588 nr_maps, strerror(-nr_maps));
589 goto done;
591 if (load_maps(map_data, nr_maps, fixup_map))
592 goto done;
593 map_data_count = nr_maps;
595 processed_sec[maps_shndx] = true;
598 /* process all relo sections, and rewrite bpf insns for maps */
599 for (i = 1; i < ehdr.e_shnum; i++) {
600 if (processed_sec[i])
601 continue;
603 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
604 continue;
606 if (shdr.sh_type == SHT_REL) {
607 struct bpf_insn *insns;
609 /* locate prog sec that need map fixup (relocations) */
610 if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
611 &shdr_prog, &data_prog))
612 continue;
614 if (shdr_prog.sh_type != SHT_PROGBITS ||
615 !(shdr_prog.sh_flags & SHF_EXECINSTR))
616 continue;
618 insns = (struct bpf_insn *) data_prog->d_buf;
619 processed_sec[i] = true; /* relo section */
621 if (parse_relo_and_apply(data, symbols, &shdr, insns,
622 map_data, nr_maps))
623 continue;
627 /* load programs */
628 for (i = 1; i < ehdr.e_shnum; i++) {
630 if (processed_sec[i])
631 continue;
633 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
634 continue;
636 if (memcmp(shname, "kprobe/", 7) == 0 ||
637 memcmp(shname, "kretprobe/", 10) == 0 ||
638 memcmp(shname, "tracepoint/", 11) == 0 ||
639 memcmp(shname, "raw_tracepoint/", 15) == 0 ||
640 memcmp(shname, "xdp", 3) == 0 ||
641 memcmp(shname, "perf_event", 10) == 0 ||
642 memcmp(shname, "socket", 6) == 0 ||
643 memcmp(shname, "cgroup/", 7) == 0 ||
644 memcmp(shname, "sockops", 7) == 0 ||
645 memcmp(shname, "sk_skb", 6) == 0 ||
646 memcmp(shname, "sk_msg", 6) == 0) {
647 ret = load_and_attach(shname, data->d_buf,
648 data->d_size);
649 if (ret != 0)
650 goto done;
654 done:
655 close(fd);
656 return ret;
659 int load_bpf_file(char *path)
661 return do_load_bpf_file(path, NULL);
664 int load_bpf_file_fixup_map(const char *path, fixup_map_cb fixup_map)
666 return do_load_bpf_file(path, fixup_map);
669 void read_trace_pipe(void)
671 int trace_fd;
673 trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
674 if (trace_fd < 0)
675 return;
677 while (1) {
678 static char buf[4096];
679 ssize_t sz;
681 sz = read(trace_fd, buf, sizeof(buf));
682 if (sz > 0) {
683 buf[sz] = 0;
684 puts(buf);