1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */
3
4 #ifndef _GNU_SOURCE
5 #define _GNU_SOURCE
6 #endif
7 #include <assert.h>
8 #include <ctype.h>
9 #include <errno.h>
10 #include <fcntl.h>
11 #include <ftw.h>
12 #include <libgen.h>
13 #include <mntent.h>
14 #include <stdbool.h>
15 #include <stdio.h>
16 #include <stdlib.h>
17 #include <string.h>
18 #include <unistd.h>
19 #include <net/if.h>
20 #include <sys/mount.h>
21 #include <sys/resource.h>
22 #include <sys/stat.h>
23 #include <sys/vfs.h>
24
25 #include <linux/filter.h>
26 #include <linux/limits.h>
27 #include <linux/magic.h>
28 #include <linux/unistd.h>
29
30 #include <bpf/bpf.h>
31 #include <bpf/hashmap.h>
32 #include <bpf/libbpf.h> /* libbpf_num_possible_cpus */
33 #include <bpf/btf.h>
34
35 #include "main.h"
36
37 #ifndef BPF_FS_MAGIC
38 #define BPF_FS_MAGIC 0xcafe4a11
39 #endif
40
p_err(const char * fmt,...)41 void p_err(const char *fmt, ...)
42 {
43 va_list ap;
44
45 va_start(ap, fmt);
46 if (json_output) {
47 jsonw_start_object(json_wtr);
48 jsonw_name(json_wtr, "error");
49 jsonw_vprintf_enquote(json_wtr, fmt, ap);
50 jsonw_end_object(json_wtr);
51 } else {
52 fprintf(stderr, "Error: ");
53 vfprintf(stderr, fmt, ap);
54 fprintf(stderr, "\n");
55 }
56 va_end(ap);
57 }
58
p_info(const char * fmt,...)59 void p_info(const char *fmt, ...)
60 {
61 va_list ap;
62
63 if (json_output)
64 return;
65
66 va_start(ap, fmt);
67 vfprintf(stderr, fmt, ap);
68 fprintf(stderr, "\n");
69 va_end(ap);
70 }
71
is_bpffs(const char * path)72 static bool is_bpffs(const char *path)
73 {
74 struct statfs st_fs;
75
76 if (statfs(path, &st_fs) < 0)
77 return false;
78
79 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC;
80 }
81
82 /* Probe whether kernel switched from memlock-based (RLIMIT_MEMLOCK) to
83 * memcg-based memory accounting for BPF maps and programs. This was done in
84 * commit 97306be45fbe ("Merge branch 'switch to memcg-based memory
85 * accounting'"), in Linux 5.11.
86 *
87 * Libbpf also offers to probe for memcg-based accounting vs rlimit, but does
88 * so by checking for the availability of a given BPF helper and this has
89 * failed on some kernels with backports in the past, see commit 6b4384ff1088
90 * ("Revert "bpftool: Use libbpf 1.0 API mode instead of RLIMIT_MEMLOCK"").
91 * Instead, we can probe by lowering the process-based rlimit to 0, trying to
92 * load a BPF object, and resetting the rlimit. If the load succeeds then
93 * memcg-based accounting is supported.
94 *
95 * This would be too dangerous to do in the library, because multithreaded
96 * applications might attempt to load items while the rlimit is at 0. Given
97 * that bpftool is single-threaded, this is fine to do here.
98 */
known_to_need_rlimit(void)99 static bool known_to_need_rlimit(void)
100 {
101 struct rlimit rlim_init, rlim_cur_zero = {};
102 struct bpf_insn insns[] = {
103 BPF_MOV64_IMM(BPF_REG_0, 0),
104 BPF_EXIT_INSN(),
105 };
106 size_t insn_cnt = ARRAY_SIZE(insns);
107 union bpf_attr attr;
108 int prog_fd, err;
109
110 memset(&attr, 0, sizeof(attr));
111 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
112 attr.insns = ptr_to_u64(insns);
113 attr.insn_cnt = insn_cnt;
114 attr.license = ptr_to_u64("GPL");
115
116 if (getrlimit(RLIMIT_MEMLOCK, &rlim_init))
117 return false;
118
119 /* Drop the soft limit to zero. We maintain the hard limit to its
120 * current value, because lowering it would be a permanent operation
121 * for unprivileged users.
122 */
123 rlim_cur_zero.rlim_max = rlim_init.rlim_max;
124 if (setrlimit(RLIMIT_MEMLOCK, &rlim_cur_zero))
125 return false;
126
127 /* Do not use bpf_prog_load() from libbpf here, because it calls
128 * bump_rlimit_memlock(), interfering with the current probe.
129 */
130 prog_fd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
131 err = errno;
132
133 /* reset soft rlimit to its initial value */
134 setrlimit(RLIMIT_MEMLOCK, &rlim_init);
135
136 if (prog_fd < 0)
137 return err == EPERM;
138
139 close(prog_fd);
140 return false;
141 }
142
set_max_rlimit(void)143 void set_max_rlimit(void)
144 {
145 struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY };
146
147 if (known_to_need_rlimit())
148 setrlimit(RLIMIT_MEMLOCK, &rinf);
149 }
150
151 static int
mnt_fs(const char * target,const char * type,char * buff,size_t bufflen)152 mnt_fs(const char *target, const char *type, char *buff, size_t bufflen)
153 {
154 bool bind_done = false;
155
156 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) {
157 if (errno != EINVAL || bind_done) {
158 snprintf(buff, bufflen,
159 "mount --make-private %s failed: %s",
160 target, strerror(errno));
161 return -1;
162 }
163
164 if (mount(target, target, "none", MS_BIND, NULL)) {
165 snprintf(buff, bufflen,
166 "mount --bind %s %s failed: %s",
167 target, target, strerror(errno));
168 return -1;
169 }
170
171 bind_done = true;
172 }
173
174 if (mount(type, target, type, 0, "mode=0700")) {
175 snprintf(buff, bufflen, "mount -t %s %s %s failed: %s",
176 type, type, target, strerror(errno));
177 return -1;
178 }
179
180 return 0;
181 }
182
mount_tracefs(const char * target)183 int mount_tracefs(const char *target)
184 {
185 char err_str[ERR_MAX_LEN];
186 int err;
187
188 err = mnt_fs(target, "tracefs", err_str, ERR_MAX_LEN);
189 if (err) {
190 err_str[ERR_MAX_LEN - 1] = '\0';
191 p_err("can't mount tracefs: %s", err_str);
192 }
193
194 return err;
195 }
196
open_obj_pinned(const char * path,bool quiet,const struct bpf_obj_get_opts * opts)197 int open_obj_pinned(const char *path, bool quiet,
198 const struct bpf_obj_get_opts *opts)
199 {
200 char *pname;
201 int fd = -1;
202
203 pname = strdup(path);
204 if (!pname) {
205 if (!quiet)
206 p_err("mem alloc failed");
207 goto out_ret;
208 }
209
210 fd = bpf_obj_get_opts(pname, opts);
211 if (fd < 0) {
212 if (!quiet)
213 p_err("bpf obj get (%s): %s", pname,
214 errno == EACCES && !is_bpffs(dirname(pname)) ?
215 "directory not in bpf file system (bpffs)" :
216 strerror(errno));
217 goto out_free;
218 }
219
220 out_free:
221 free(pname);
222 out_ret:
223 return fd;
224 }
225
open_obj_pinned_any(const char * path,enum bpf_obj_type exp_type,const struct bpf_obj_get_opts * opts)226 int open_obj_pinned_any(const char *path, enum bpf_obj_type exp_type,
227 const struct bpf_obj_get_opts *opts)
228 {
229 enum bpf_obj_type type;
230 int fd;
231
232 fd = open_obj_pinned(path, false, opts);
233 if (fd < 0)
234 return -1;
235
236 type = get_fd_type(fd);
237 if (type < 0) {
238 close(fd);
239 return type;
240 }
241 if (type != exp_type) {
242 p_err("incorrect object type: %s", get_fd_type_name(type));
243 close(fd);
244 return -1;
245 }
246
247 return fd;
248 }
249
create_and_mount_bpffs_dir(const char * dir_name)250 int create_and_mount_bpffs_dir(const char *dir_name)
251 {
252 char err_str[ERR_MAX_LEN];
253 bool dir_exists;
254 int err = 0;
255
256 if (is_bpffs(dir_name))
257 return err;
258
259 dir_exists = access(dir_name, F_OK) == 0;
260
261 if (!dir_exists) {
262 char *temp_name;
263 char *parent_name;
264
265 temp_name = strdup(dir_name);
266 if (!temp_name) {
267 p_err("mem alloc failed");
268 return -1;
269 }
270
271 parent_name = dirname(temp_name);
272
273 if (is_bpffs(parent_name)) {
274 /* nothing to do if already mounted */
275 free(temp_name);
276 return err;
277 }
278
279 if (access(parent_name, F_OK) == -1) {
280 p_err("can't create dir '%s' to pin BPF object: parent dir '%s' doesn't exist",
281 dir_name, parent_name);
282 free(temp_name);
283 return -1;
284 }
285
286 free(temp_name);
287 }
288
289 if (block_mount) {
290 p_err("no BPF file system found, not mounting it due to --nomount option");
291 return -1;
292 }
293
294 if (!dir_exists) {
295 err = mkdir(dir_name, S_IRWXU);
296 if (err) {
297 p_err("failed to create dir '%s': %s", dir_name, strerror(errno));
298 return err;
299 }
300 }
301
302 err = mnt_fs(dir_name, "bpf", err_str, ERR_MAX_LEN);
303 if (err) {
304 err_str[ERR_MAX_LEN - 1] = '\0';
305 p_err("can't mount BPF file system on given dir '%s': %s",
306 dir_name, err_str);
307
308 if (!dir_exists)
309 rmdir(dir_name);
310 }
311
312 return err;
313 }
314
mount_bpffs_for_file(const char * file_name)315 int mount_bpffs_for_file(const char *file_name)
316 {
317 char err_str[ERR_MAX_LEN];
318 char *temp_name;
319 char *dir;
320 int err = 0;
321
322 if (access(file_name, F_OK) != -1) {
323 p_err("can't pin BPF object: path '%s' already exists", file_name);
324 return -1;
325 }
326
327 temp_name = strdup(file_name);
328 if (!temp_name) {
329 p_err("mem alloc failed");
330 return -1;
331 }
332
333 dir = dirname(temp_name);
334
335 if (is_bpffs(dir))
336 /* nothing to do if already mounted */
337 goto out_free;
338
339 if (access(dir, F_OK) == -1) {
340 p_err("can't pin BPF object: dir '%s' doesn't exist", dir);
341 err = -1;
342 goto out_free;
343 }
344
345 if (block_mount) {
346 p_err("no BPF file system found, not mounting it due to --nomount option");
347 err = -1;
348 goto out_free;
349 }
350
351 err = mnt_fs(dir, "bpf", err_str, ERR_MAX_LEN);
352 if (err) {
353 err_str[ERR_MAX_LEN - 1] = '\0';
354 p_err("can't mount BPF file system to pin the object '%s': %s",
355 file_name, err_str);
356 }
357
358 out_free:
359 free(temp_name);
360 return err;
361 }
362
do_pin_fd(int fd,const char * name)363 int do_pin_fd(int fd, const char *name)
364 {
365 int err;
366
367 err = mount_bpffs_for_file(name);
368 if (err)
369 return err;
370
371 err = bpf_obj_pin(fd, name);
372 if (err)
373 p_err("can't pin the object (%s): %s", name, strerror(errno));
374
375 return err;
376 }
377
do_pin_any(int argc,char ** argv,int (* get_fd)(int *,char ***))378 int do_pin_any(int argc, char **argv, int (*get_fd)(int *, char ***))
379 {
380 int err;
381 int fd;
382
383 if (!REQ_ARGS(3))
384 return -EINVAL;
385
386 fd = get_fd(&argc, &argv);
387 if (fd < 0)
388 return fd;
389
390 err = do_pin_fd(fd, *argv);
391
392 close(fd);
393 return err;
394 }
395
get_fd_type_name(enum bpf_obj_type type)396 const char *get_fd_type_name(enum bpf_obj_type type)
397 {
398 static const char * const names[] = {
399 [BPF_OBJ_UNKNOWN] = "unknown",
400 [BPF_OBJ_PROG] = "prog",
401 [BPF_OBJ_MAP] = "map",
402 [BPF_OBJ_LINK] = "link",
403 };
404
405 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type])
406 return names[BPF_OBJ_UNKNOWN];
407
408 return names[type];
409 }
410
get_prog_full_name(const struct bpf_prog_info * prog_info,int prog_fd,char * name_buff,size_t buff_len)411 void get_prog_full_name(const struct bpf_prog_info *prog_info, int prog_fd,
412 char *name_buff, size_t buff_len)
413 {
414 const char *prog_name = prog_info->name;
415 const struct btf_type *func_type;
416 struct bpf_func_info finfo = {};
417 struct bpf_prog_info info = {};
418 __u32 info_len = sizeof(info);
419 struct btf *prog_btf = NULL;
420
421 if (buff_len <= BPF_OBJ_NAME_LEN ||
422 strlen(prog_info->name) < BPF_OBJ_NAME_LEN - 1)
423 goto copy_name;
424
425 if (!prog_info->btf_id || prog_info->nr_func_info == 0)
426 goto copy_name;
427
428 info.nr_func_info = 1;
429 info.func_info_rec_size = prog_info->func_info_rec_size;
430 if (info.func_info_rec_size > sizeof(finfo))
431 info.func_info_rec_size = sizeof(finfo);
432 info.func_info = ptr_to_u64(&finfo);
433
434 if (bpf_prog_get_info_by_fd(prog_fd, &info, &info_len))
435 goto copy_name;
436
437 prog_btf = btf__load_from_kernel_by_id(info.btf_id);
438 if (!prog_btf)
439 goto copy_name;
440
441 func_type = btf__type_by_id(prog_btf, finfo.type_id);
442 if (!func_type || !btf_is_func(func_type))
443 goto copy_name;
444
445 prog_name = btf__name_by_offset(prog_btf, func_type->name_off);
446
447 copy_name:
448 snprintf(name_buff, buff_len, "%s", prog_name);
449
450 if (prog_btf)
451 btf__free(prog_btf);
452 }
453
get_fd_type(int fd)454 int get_fd_type(int fd)
455 {
456 char path[PATH_MAX];
457 char buf[512];
458 ssize_t n;
459
460 snprintf(path, sizeof(path), "/proc/self/fd/%d", fd);
461
462 n = readlink(path, buf, sizeof(buf));
463 if (n < 0) {
464 p_err("can't read link type: %s", strerror(errno));
465 return -1;
466 }
467 if (n == sizeof(buf)) {
468 p_err("can't read link type: path too long!");
469 return -1;
470 }
471 buf[n] = '\0';
472
473 if (strstr(buf, "bpf-map"))
474 return BPF_OBJ_MAP;
475 else if (strstr(buf, "bpf-prog"))
476 return BPF_OBJ_PROG;
477 else if (strstr(buf, "bpf-link"))
478 return BPF_OBJ_LINK;
479
480 return BPF_OBJ_UNKNOWN;
481 }
482
get_fdinfo(int fd,const char * key)483 char *get_fdinfo(int fd, const char *key)
484 {
485 char path[PATH_MAX];
486 char *line = NULL;
487 size_t line_n = 0;
488 ssize_t n;
489 FILE *fdi;
490
491 snprintf(path, sizeof(path), "/proc/self/fdinfo/%d", fd);
492
493 fdi = fopen(path, "r");
494 if (!fdi)
495 return NULL;
496
497 while ((n = getline(&line, &line_n, fdi)) > 0) {
498 char *value;
499 int len;
500
501 if (!strstr(line, key))
502 continue;
503
504 fclose(fdi);
505
506 value = strchr(line, '\t');
507 if (!value || !value[1]) {
508 free(line);
509 return NULL;
510 }
511 value++;
512
513 len = strlen(value);
514 memmove(line, value, len);
515 line[len - 1] = '\0';
516
517 return line;
518 }
519
520 free(line);
521 fclose(fdi);
522 return NULL;
523 }
524
print_data_json(uint8_t * data,size_t len)525 void print_data_json(uint8_t *data, size_t len)
526 {
527 unsigned int i;
528
529 jsonw_start_array(json_wtr);
530 for (i = 0; i < len; i++)
531 jsonw_printf(json_wtr, "%d", data[i]);
532 jsonw_end_array(json_wtr);
533 }
534
print_hex_data_json(uint8_t * data,size_t len)535 void print_hex_data_json(uint8_t *data, size_t len)
536 {
537 unsigned int i;
538
539 jsonw_start_array(json_wtr);
540 for (i = 0; i < len; i++)
541 jsonw_printf(json_wtr, "\"0x%02hhx\"", data[i]);
542 jsonw_end_array(json_wtr);
543 }
544
545 /* extra params for nftw cb */
546 static struct hashmap *build_fn_table;
547 static enum bpf_obj_type build_fn_type;
548
do_build_table_cb(const char * fpath,const struct stat * sb,int typeflag,struct FTW * ftwbuf)549 static int do_build_table_cb(const char *fpath, const struct stat *sb,
550 int typeflag, struct FTW *ftwbuf)
551 {
552 struct bpf_prog_info pinned_info;
553 __u32 len = sizeof(pinned_info);
554 enum bpf_obj_type objtype;
555 int fd, err = 0;
556 char *path;
557
558 if (typeflag != FTW_F)
559 goto out_ret;
560
561 fd = open_obj_pinned(fpath, true, NULL);
562 if (fd < 0)
563 goto out_ret;
564
565 objtype = get_fd_type(fd);
566 if (objtype != build_fn_type)
567 goto out_close;
568
569 memset(&pinned_info, 0, sizeof(pinned_info));
570 if (bpf_prog_get_info_by_fd(fd, &pinned_info, &len))
571 goto out_close;
572
573 path = strdup(fpath);
574 if (!path) {
575 err = -1;
576 goto out_close;
577 }
578
579 err = hashmap__append(build_fn_table, pinned_info.id, path);
580 if (err) {
581 p_err("failed to append entry to hashmap for ID %u, path '%s': %s",
582 pinned_info.id, path, strerror(errno));
583 free(path);
584 goto out_close;
585 }
586
587 out_close:
588 close(fd);
589 out_ret:
590 return err;
591 }
592
build_pinned_obj_table(struct hashmap * tab,enum bpf_obj_type type)593 int build_pinned_obj_table(struct hashmap *tab,
594 enum bpf_obj_type type)
595 {
596 struct mntent *mntent = NULL;
597 FILE *mntfile = NULL;
598 int flags = FTW_PHYS;
599 int nopenfd = 16;
600 int err = 0;
601
602 mntfile = setmntent("/proc/mounts", "r");
603 if (!mntfile)
604 return -1;
605
606 build_fn_table = tab;
607 build_fn_type = type;
608
609 while ((mntent = getmntent(mntfile))) {
610 char *path = mntent->mnt_dir;
611
612 if (strncmp(mntent->mnt_type, "bpf", 3) != 0)
613 continue;
614 err = nftw(path, do_build_table_cb, nopenfd, flags);
615 if (err)
616 break;
617 }
618 fclose(mntfile);
619 return err;
620 }
621
delete_pinned_obj_table(struct hashmap * map)622 void delete_pinned_obj_table(struct hashmap *map)
623 {
624 struct hashmap_entry *entry;
625 size_t bkt;
626
627 if (!map)
628 return;
629
630 hashmap__for_each_entry(map, entry, bkt)
631 free(entry->pvalue);
632
633 hashmap__free(map);
634 }
635
get_page_size(void)636 unsigned int get_page_size(void)
637 {
638 static int result;
639
640 if (!result)
641 result = getpagesize();
642 return result;
643 }
644
get_possible_cpus(void)645 unsigned int get_possible_cpus(void)
646 {
647 int cpus = libbpf_num_possible_cpus();
648
649 if (cpus < 0) {
650 p_err("Can't get # of possible cpus: %s", strerror(-cpus));
651 exit(-1);
652 }
653 return cpus;
654 }
655
656 static char *
ifindex_to_name_ns(__u32 ifindex,__u32 ns_dev,__u32 ns_ino,char * buf)657 ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf)
658 {
659 struct stat st;
660 int err;
661
662 err = stat("/proc/self/ns/net", &st);
663 if (err) {
664 p_err("Can't stat /proc/self: %s", strerror(errno));
665 return NULL;
666 }
667
668 if (st.st_dev != ns_dev || st.st_ino != ns_ino)
669 return NULL;
670
671 return if_indextoname(ifindex, buf);
672 }
673
read_sysfs_hex_int(char * path)674 static int read_sysfs_hex_int(char *path)
675 {
676 char vendor_id_buf[8];
677 int len;
678 int fd;
679
680 fd = open(path, O_RDONLY);
681 if (fd < 0) {
682 p_err("Can't open %s: %s", path, strerror(errno));
683 return -1;
684 }
685
686 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf));
687 close(fd);
688 if (len < 0) {
689 p_err("Can't read %s: %s", path, strerror(errno));
690 return -1;
691 }
692 if (len >= (int)sizeof(vendor_id_buf)) {
693 p_err("Value in %s too long", path);
694 return -1;
695 }
696
697 vendor_id_buf[len] = 0;
698
699 return strtol(vendor_id_buf, NULL, 0);
700 }
701
read_sysfs_netdev_hex_int(char * devname,const char * entry_name)702 static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name)
703 {
704 char full_path[64];
705
706 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s",
707 devname, entry_name);
708
709 return read_sysfs_hex_int(full_path);
710 }
711
712 const char *
ifindex_to_arch(__u32 ifindex,__u64 ns_dev,__u64 ns_ino,const char ** opt)713 ifindex_to_arch(__u32 ifindex, __u64 ns_dev, __u64 ns_ino, const char **opt)
714 {
715 __maybe_unused int device_id;
716 char devname[IF_NAMESIZE];
717 int vendor_id;
718
719 if (!ifindex_to_name_ns(ifindex, ns_dev, ns_ino, devname)) {
720 p_err("Can't get net device name for ifindex %u: %s", ifindex,
721 strerror(errno));
722 return NULL;
723 }
724
725 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor");
726 if (vendor_id < 0) {
727 p_err("Can't get device vendor id for %s", devname);
728 return NULL;
729 }
730
731 switch (vendor_id) {
732 #ifdef HAVE_LIBBFD_SUPPORT
733 case 0x19ee:
734 device_id = read_sysfs_netdev_hex_int(devname, "device");
735 if (device_id != 0x4000 &&
736 device_id != 0x6000 &&
737 device_id != 0x6003)
738 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch");
739 *opt = "ctx4";
740 return "NFP-6xxx";
741 #endif /* HAVE_LIBBFD_SUPPORT */
742 /* No NFP support in LLVM, we have no valid triple to return. */
743 default:
744 p_err("Can't get arch name for device vendor id 0x%04x",
745 (unsigned int)vendor_id);
746 return NULL;
747 }
748 }
749
print_dev_plain(__u32 ifindex,__u64 ns_dev,__u64 ns_inode)750 void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
751 {
752 char name[IF_NAMESIZE];
753
754 if (!ifindex)
755 return;
756
757 printf(" offloaded_to ");
758 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
759 printf("%s", name);
760 else
761 printf("ifindex %u ns_dev %llu ns_ino %llu",
762 ifindex, ns_dev, ns_inode);
763 }
764
print_dev_json(__u32 ifindex,__u64 ns_dev,__u64 ns_inode)765 void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
766 {
767 char name[IF_NAMESIZE];
768
769 if (!ifindex)
770 return;
771
772 jsonw_name(json_wtr, "dev");
773 jsonw_start_object(json_wtr);
774 jsonw_uint_field(json_wtr, "ifindex", ifindex);
775 jsonw_uint_field(json_wtr, "ns_dev", ns_dev);
776 jsonw_uint_field(json_wtr, "ns_inode", ns_inode);
777 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
778 jsonw_string_field(json_wtr, "ifname", name);
779 jsonw_end_object(json_wtr);
780 }
781
parse_u32_arg(int * argc,char *** argv,__u32 * val,const char * what)782 int parse_u32_arg(int *argc, char ***argv, __u32 *val, const char *what)
783 {
784 char *endptr;
785
786 NEXT_ARGP();
787
788 if (*val) {
789 p_err("%s already specified", what);
790 return -1;
791 }
792
793 *val = strtoul(**argv, &endptr, 0);
794 if (*endptr) {
795 p_err("can't parse %s as %s", **argv, what);
796 return -1;
797 }
798 NEXT_ARGP();
799
800 return 0;
801 }
802
803 int __printf(2, 0)
print_all_levels(__maybe_unused enum libbpf_print_level level,const char * format,va_list args)804 print_all_levels(__maybe_unused enum libbpf_print_level level,
805 const char *format, va_list args)
806 {
807 return vfprintf(stderr, format, args);
808 }
809
prog_fd_by_nametag(void * nametag,int ** fds,bool tag)810 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
811 {
812 char prog_name[MAX_PROG_FULL_NAME];
813 unsigned int id = 0;
814 int fd, nb_fds = 0;
815 void *tmp;
816 int err;
817
818 while (true) {
819 struct bpf_prog_info info = {};
820 __u32 len = sizeof(info);
821
822 err = bpf_prog_get_next_id(id, &id);
823 if (err) {
824 if (errno != ENOENT) {
825 p_err("%s", strerror(errno));
826 goto err_close_fds;
827 }
828 return nb_fds;
829 }
830
831 fd = bpf_prog_get_fd_by_id(id);
832 if (fd < 0) {
833 p_err("can't get prog by id (%u): %s",
834 id, strerror(errno));
835 goto err_close_fds;
836 }
837
838 err = bpf_prog_get_info_by_fd(fd, &info, &len);
839 if (err) {
840 p_err("can't get prog info (%u): %s",
841 id, strerror(errno));
842 goto err_close_fd;
843 }
844
845 if (tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) {
846 close(fd);
847 continue;
848 }
849
850 if (!tag) {
851 get_prog_full_name(&info, fd, prog_name,
852 sizeof(prog_name));
853 if (strncmp(nametag, prog_name, sizeof(prog_name))) {
854 close(fd);
855 continue;
856 }
857 }
858
859 if (nb_fds > 0) {
860 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
861 if (!tmp) {
862 p_err("failed to realloc");
863 goto err_close_fd;
864 }
865 *fds = tmp;
866 }
867 (*fds)[nb_fds++] = fd;
868 }
869
870 err_close_fd:
871 close(fd);
872 err_close_fds:
873 while (--nb_fds >= 0)
874 close((*fds)[nb_fds]);
875 return -1;
876 }
877
prog_parse_fds(int * argc,char *** argv,int ** fds)878 int prog_parse_fds(int *argc, char ***argv, int **fds)
879 {
880 if (is_prefix(**argv, "id")) {
881 unsigned int id;
882 char *endptr;
883
884 NEXT_ARGP();
885
886 id = strtoul(**argv, &endptr, 0);
887 if (*endptr) {
888 p_err("can't parse %s as ID", **argv);
889 return -1;
890 }
891 NEXT_ARGP();
892
893 (*fds)[0] = bpf_prog_get_fd_by_id(id);
894 if ((*fds)[0] < 0) {
895 p_err("get by id (%u): %s", id, strerror(errno));
896 return -1;
897 }
898 return 1;
899 } else if (is_prefix(**argv, "tag")) {
900 unsigned char tag[BPF_TAG_SIZE];
901
902 NEXT_ARGP();
903
904 if (sscanf(**argv, BPF_TAG_FMT, tag, tag + 1, tag + 2,
905 tag + 3, tag + 4, tag + 5, tag + 6, tag + 7)
906 != BPF_TAG_SIZE) {
907 p_err("can't parse tag");
908 return -1;
909 }
910 NEXT_ARGP();
911
912 return prog_fd_by_nametag(tag, fds, true);
913 } else if (is_prefix(**argv, "name")) {
914 char *name;
915
916 NEXT_ARGP();
917
918 name = **argv;
919 if (strlen(name) > MAX_PROG_FULL_NAME - 1) {
920 p_err("can't parse name");
921 return -1;
922 }
923 NEXT_ARGP();
924
925 return prog_fd_by_nametag(name, fds, false);
926 } else if (is_prefix(**argv, "pinned")) {
927 char *path;
928
929 NEXT_ARGP();
930
931 path = **argv;
932 NEXT_ARGP();
933
934 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_PROG, NULL);
935 if ((*fds)[0] < 0)
936 return -1;
937 return 1;
938 }
939
940 p_err("expected 'id', 'tag', 'name' or 'pinned', got: '%s'?", **argv);
941 return -1;
942 }
943
prog_parse_fd(int * argc,char *** argv)944 int prog_parse_fd(int *argc, char ***argv)
945 {
946 int *fds = NULL;
947 int nb_fds, fd;
948
949 fds = malloc(sizeof(int));
950 if (!fds) {
951 p_err("mem alloc failed");
952 return -1;
953 }
954 nb_fds = prog_parse_fds(argc, argv, &fds);
955 if (nb_fds != 1) {
956 if (nb_fds > 1) {
957 p_err("several programs match this handle");
958 while (nb_fds--)
959 close(fds[nb_fds]);
960 }
961 fd = -1;
962 goto exit_free;
963 }
964
965 fd = fds[0];
966 exit_free:
967 free(fds);
968 return fd;
969 }
970
map_fd_by_name(char * name,int ** fds,const struct bpf_get_fd_by_id_opts * opts)971 static int map_fd_by_name(char *name, int **fds,
972 const struct bpf_get_fd_by_id_opts *opts)
973 {
974 unsigned int id = 0;
975 int fd, nb_fds = 0;
976 void *tmp;
977 int err;
978
979 while (true) {
980 LIBBPF_OPTS(bpf_get_fd_by_id_opts, opts_ro);
981 struct bpf_map_info info = {};
982 __u32 len = sizeof(info);
983
984 err = bpf_map_get_next_id(id, &id);
985 if (err) {
986 if (errno != ENOENT) {
987 p_err("%s", strerror(errno));
988 goto err_close_fds;
989 }
990 return nb_fds;
991 }
992
993 /* Request a read-only fd to query the map info */
994 opts_ro.open_flags = BPF_F_RDONLY;
995 fd = bpf_map_get_fd_by_id_opts(id, &opts_ro);
996 if (fd < 0) {
997 p_err("can't get map by id (%u): %s",
998 id, strerror(errno));
999 goto err_close_fds;
1000 }
1001
1002 err = bpf_map_get_info_by_fd(fd, &info, &len);
1003 if (err) {
1004 p_err("can't get map info (%u): %s",
1005 id, strerror(errno));
1006 goto err_close_fd;
1007 }
1008
1009 if (strncmp(name, info.name, BPF_OBJ_NAME_LEN)) {
1010 close(fd);
1011 continue;
1012 }
1013
1014 /* Get an fd with the requested options, if they differ
1015 * from the read-only options used to get the fd above.
1016 */
1017 if (memcmp(opts, &opts_ro, sizeof(opts_ro))) {
1018 close(fd);
1019 fd = bpf_map_get_fd_by_id_opts(id, opts);
1020 if (fd < 0) {
1021 p_err("can't get map by id (%u): %s", id,
1022 strerror(errno));
1023 goto err_close_fds;
1024 }
1025 }
1026
1027 if (nb_fds > 0) {
1028 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
1029 if (!tmp) {
1030 p_err("failed to realloc");
1031 goto err_close_fd;
1032 }
1033 *fds = tmp;
1034 }
1035 (*fds)[nb_fds++] = fd;
1036 }
1037
1038 err_close_fd:
1039 close(fd);
1040 err_close_fds:
1041 while (--nb_fds >= 0)
1042 close((*fds)[nb_fds]);
1043 return -1;
1044 }
1045
map_parse_fds(int * argc,char *** argv,int ** fds,__u32 open_flags)1046 int map_parse_fds(int *argc, char ***argv, int **fds, __u32 open_flags)
1047 {
1048 LIBBPF_OPTS(bpf_get_fd_by_id_opts, opts);
1049
1050 assert((open_flags & ~BPF_F_RDONLY) == 0);
1051 opts.open_flags = open_flags;
1052
1053 if (is_prefix(**argv, "id")) {
1054 unsigned int id;
1055 char *endptr;
1056
1057 NEXT_ARGP();
1058
1059 id = strtoul(**argv, &endptr, 0);
1060 if (*endptr) {
1061 p_err("can't parse %s as ID", **argv);
1062 return -1;
1063 }
1064 NEXT_ARGP();
1065
1066 (*fds)[0] = bpf_map_get_fd_by_id_opts(id, &opts);
1067 if ((*fds)[0] < 0) {
1068 p_err("get map by id (%u): %s", id, strerror(errno));
1069 return -1;
1070 }
1071 return 1;
1072 } else if (is_prefix(**argv, "name")) {
1073 char *name;
1074
1075 NEXT_ARGP();
1076
1077 name = **argv;
1078 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
1079 p_err("can't parse name");
1080 return -1;
1081 }
1082 NEXT_ARGP();
1083
1084 return map_fd_by_name(name, fds, &opts);
1085 } else if (is_prefix(**argv, "pinned")) {
1086 char *path;
1087 LIBBPF_OPTS(bpf_obj_get_opts, get_opts);
1088 get_opts.file_flags = open_flags;
1089
1090 NEXT_ARGP();
1091
1092 path = **argv;
1093 NEXT_ARGP();
1094
1095 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_MAP, &get_opts);
1096 if ((*fds)[0] < 0)
1097 return -1;
1098 return 1;
1099 }
1100
1101 p_err("expected 'id', 'name' or 'pinned', got: '%s'?", **argv);
1102 return -1;
1103 }
1104
map_parse_fd(int * argc,char *** argv,__u32 open_flags)1105 int map_parse_fd(int *argc, char ***argv, __u32 open_flags)
1106 {
1107 int *fds = NULL;
1108 int nb_fds, fd;
1109
1110 fds = malloc(sizeof(int));
1111 if (!fds) {
1112 p_err("mem alloc failed");
1113 return -1;
1114 }
1115 nb_fds = map_parse_fds(argc, argv, &fds, open_flags);
1116 if (nb_fds != 1) {
1117 if (nb_fds > 1) {
1118 p_err("several maps match this handle");
1119 while (nb_fds--)
1120 close(fds[nb_fds]);
1121 }
1122 fd = -1;
1123 goto exit_free;
1124 }
1125
1126 fd = fds[0];
1127 exit_free:
1128 free(fds);
1129 return fd;
1130 }
1131
map_parse_fd_and_info(int * argc,char *** argv,struct bpf_map_info * info,__u32 * info_len,__u32 open_flags)1132 int map_parse_fd_and_info(int *argc, char ***argv, struct bpf_map_info *info,
1133 __u32 *info_len, __u32 open_flags)
1134 {
1135 int err;
1136 int fd;
1137
1138 fd = map_parse_fd(argc, argv, open_flags);
1139 if (fd < 0)
1140 return -1;
1141
1142 err = bpf_map_get_info_by_fd(fd, info, info_len);
1143 if (err) {
1144 p_err("can't get map info: %s", strerror(errno));
1145 close(fd);
1146 return err;
1147 }
1148
1149 return fd;
1150 }
1151
hash_fn_for_key_as_id(long key,void * ctx)1152 size_t hash_fn_for_key_as_id(long key, void *ctx)
1153 {
1154 return key;
1155 }
1156
equal_fn_for_key_as_id(long k1,long k2,void * ctx)1157 bool equal_fn_for_key_as_id(long k1, long k2, void *ctx)
1158 {
1159 return k1 == k2;
1160 }
1161
bpf_attach_type_input_str(enum bpf_attach_type t)1162 const char *bpf_attach_type_input_str(enum bpf_attach_type t)
1163 {
1164 switch (t) {
1165 case BPF_CGROUP_INET_INGRESS: return "ingress";
1166 case BPF_CGROUP_INET_EGRESS: return "egress";
1167 case BPF_CGROUP_INET_SOCK_CREATE: return "sock_create";
1168 case BPF_CGROUP_INET_SOCK_RELEASE: return "sock_release";
1169 case BPF_CGROUP_SOCK_OPS: return "sock_ops";
1170 case BPF_CGROUP_DEVICE: return "device";
1171 case BPF_CGROUP_INET4_BIND: return "bind4";
1172 case BPF_CGROUP_INET6_BIND: return "bind6";
1173 case BPF_CGROUP_INET4_CONNECT: return "connect4";
1174 case BPF_CGROUP_INET6_CONNECT: return "connect6";
1175 case BPF_CGROUP_INET4_POST_BIND: return "post_bind4";
1176 case BPF_CGROUP_INET6_POST_BIND: return "post_bind6";
1177 case BPF_CGROUP_INET4_GETPEERNAME: return "getpeername4";
1178 case BPF_CGROUP_INET6_GETPEERNAME: return "getpeername6";
1179 case BPF_CGROUP_INET4_GETSOCKNAME: return "getsockname4";
1180 case BPF_CGROUP_INET6_GETSOCKNAME: return "getsockname6";
1181 case BPF_CGROUP_UDP4_SENDMSG: return "sendmsg4";
1182 case BPF_CGROUP_UDP6_SENDMSG: return "sendmsg6";
1183 case BPF_CGROUP_SYSCTL: return "sysctl";
1184 case BPF_CGROUP_UDP4_RECVMSG: return "recvmsg4";
1185 case BPF_CGROUP_UDP6_RECVMSG: return "recvmsg6";
1186 case BPF_CGROUP_GETSOCKOPT: return "getsockopt";
1187 case BPF_CGROUP_SETSOCKOPT: return "setsockopt";
1188 case BPF_TRACE_RAW_TP: return "raw_tp";
1189 case BPF_TRACE_FENTRY: return "fentry";
1190 case BPF_TRACE_FEXIT: return "fexit";
1191 case BPF_MODIFY_RETURN: return "mod_ret";
1192 case BPF_SK_REUSEPORT_SELECT: return "sk_skb_reuseport_select";
1193 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE: return "sk_skb_reuseport_select_or_migrate";
1194 default: return libbpf_bpf_attach_type_str(t);
1195 }
1196 }
1197
pathname_concat(char * buf,int buf_sz,const char * path,const char * name)1198 int pathname_concat(char *buf, int buf_sz, const char *path,
1199 const char *name)
1200 {
1201 int len;
1202
1203 len = snprintf(buf, buf_sz, "%s/%s", path, name);
1204 if (len < 0)
1205 return -EINVAL;
1206 if (len >= buf_sz)
1207 return -ENAMETOOLONG;
1208
1209 return 0;
1210 }
1211