Loading...
1// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2/* Copyright (C) 2017-2018 Netronome Systems, Inc. */
3
4#define _GNU_SOURCE
5#include <ctype.h>
6#include <errno.h>
7#include <fcntl.h>
8#include <ftw.h>
9#include <libgen.h>
10#include <mntent.h>
11#include <stdbool.h>
12#include <stdio.h>
13#include <stdlib.h>
14#include <string.h>
15#include <unistd.h>
16#include <linux/limits.h>
17#include <linux/magic.h>
18#include <net/if.h>
19#include <sys/mount.h>
20#include <sys/resource.h>
21#include <sys/stat.h>
22#include <sys/vfs.h>
23
24#include <bpf/bpf.h>
25#include <bpf/libbpf.h> /* libbpf_num_possible_cpus */
26
27#include "main.h"
28
29#ifndef BPF_FS_MAGIC
30#define BPF_FS_MAGIC 0xcafe4a11
31#endif
32
33const char * const attach_type_name[__MAX_BPF_ATTACH_TYPE] = {
34 [BPF_CGROUP_INET_INGRESS] = "ingress",
35 [BPF_CGROUP_INET_EGRESS] = "egress",
36 [BPF_CGROUP_INET_SOCK_CREATE] = "sock_create",
37 [BPF_CGROUP_INET_SOCK_RELEASE] = "sock_release",
38 [BPF_CGROUP_SOCK_OPS] = "sock_ops",
39 [BPF_CGROUP_DEVICE] = "device",
40 [BPF_CGROUP_INET4_BIND] = "bind4",
41 [BPF_CGROUP_INET6_BIND] = "bind6",
42 [BPF_CGROUP_INET4_CONNECT] = "connect4",
43 [BPF_CGROUP_INET6_CONNECT] = "connect6",
44 [BPF_CGROUP_INET4_POST_BIND] = "post_bind4",
45 [BPF_CGROUP_INET6_POST_BIND] = "post_bind6",
46 [BPF_CGROUP_INET4_GETPEERNAME] = "getpeername4",
47 [BPF_CGROUP_INET6_GETPEERNAME] = "getpeername6",
48 [BPF_CGROUP_INET4_GETSOCKNAME] = "getsockname4",
49 [BPF_CGROUP_INET6_GETSOCKNAME] = "getsockname6",
50 [BPF_CGROUP_UDP4_SENDMSG] = "sendmsg4",
51 [BPF_CGROUP_UDP6_SENDMSG] = "sendmsg6",
52 [BPF_CGROUP_SYSCTL] = "sysctl",
53 [BPF_CGROUP_UDP4_RECVMSG] = "recvmsg4",
54 [BPF_CGROUP_UDP6_RECVMSG] = "recvmsg6",
55 [BPF_CGROUP_GETSOCKOPT] = "getsockopt",
56 [BPF_CGROUP_SETSOCKOPT] = "setsockopt",
57
58 [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser",
59 [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict",
60 [BPF_SK_SKB_VERDICT] = "sk_skb_verdict",
61 [BPF_SK_MSG_VERDICT] = "sk_msg_verdict",
62 [BPF_LIRC_MODE2] = "lirc_mode2",
63 [BPF_FLOW_DISSECTOR] = "flow_dissector",
64 [BPF_TRACE_RAW_TP] = "raw_tp",
65 [BPF_TRACE_FENTRY] = "fentry",
66 [BPF_TRACE_FEXIT] = "fexit",
67 [BPF_MODIFY_RETURN] = "mod_ret",
68 [BPF_LSM_MAC] = "lsm_mac",
69 [BPF_SK_LOOKUP] = "sk_lookup",
70};
71
72void p_err(const char *fmt, ...)
73{
74 va_list ap;
75
76 va_start(ap, fmt);
77 if (json_output) {
78 jsonw_start_object(json_wtr);
79 jsonw_name(json_wtr, "error");
80 jsonw_vprintf_enquote(json_wtr, fmt, ap);
81 jsonw_end_object(json_wtr);
82 } else {
83 fprintf(stderr, "Error: ");
84 vfprintf(stderr, fmt, ap);
85 fprintf(stderr, "\n");
86 }
87 va_end(ap);
88}
89
90void p_info(const char *fmt, ...)
91{
92 va_list ap;
93
94 if (json_output)
95 return;
96
97 va_start(ap, fmt);
98 vfprintf(stderr, fmt, ap);
99 fprintf(stderr, "\n");
100 va_end(ap);
101}
102
103static bool is_bpffs(char *path)
104{
105 struct statfs st_fs;
106
107 if (statfs(path, &st_fs) < 0)
108 return false;
109
110 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC;
111}
112
113void set_max_rlimit(void)
114{
115 struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY };
116
117 setrlimit(RLIMIT_MEMLOCK, &rinf);
118}
119
120static int
121mnt_fs(const char *target, const char *type, char *buff, size_t bufflen)
122{
123 bool bind_done = false;
124
125 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) {
126 if (errno != EINVAL || bind_done) {
127 snprintf(buff, bufflen,
128 "mount --make-private %s failed: %s",
129 target, strerror(errno));
130 return -1;
131 }
132
133 if (mount(target, target, "none", MS_BIND, NULL)) {
134 snprintf(buff, bufflen,
135 "mount --bind %s %s failed: %s",
136 target, target, strerror(errno));
137 return -1;
138 }
139
140 bind_done = true;
141 }
142
143 if (mount(type, target, type, 0, "mode=0700")) {
144 snprintf(buff, bufflen, "mount -t %s %s %s failed: %s",
145 type, type, target, strerror(errno));
146 return -1;
147 }
148
149 return 0;
150}
151
152int mount_tracefs(const char *target)
153{
154 char err_str[ERR_MAX_LEN];
155 int err;
156
157 err = mnt_fs(target, "tracefs", err_str, ERR_MAX_LEN);
158 if (err) {
159 err_str[ERR_MAX_LEN - 1] = '\0';
160 p_err("can't mount tracefs: %s", err_str);
161 }
162
163 return err;
164}
165
166int open_obj_pinned(const char *path, bool quiet)
167{
168 char *pname;
169 int fd = -1;
170
171 pname = strdup(path);
172 if (!pname) {
173 if (!quiet)
174 p_err("mem alloc failed");
175 goto out_ret;
176 }
177
178 fd = bpf_obj_get(pname);
179 if (fd < 0) {
180 if (!quiet)
181 p_err("bpf obj get (%s): %s", pname,
182 errno == EACCES && !is_bpffs(dirname(pname)) ?
183 "directory not in bpf file system (bpffs)" :
184 strerror(errno));
185 goto out_free;
186 }
187
188out_free:
189 free(pname);
190out_ret:
191 return fd;
192}
193
194int open_obj_pinned_any(const char *path, enum bpf_obj_type exp_type)
195{
196 enum bpf_obj_type type;
197 int fd;
198
199 fd = open_obj_pinned(path, false);
200 if (fd < 0)
201 return -1;
202
203 type = get_fd_type(fd);
204 if (type < 0) {
205 close(fd);
206 return type;
207 }
208 if (type != exp_type) {
209 p_err("incorrect object type: %s", get_fd_type_name(type));
210 close(fd);
211 return -1;
212 }
213
214 return fd;
215}
216
217int mount_bpffs_for_pin(const char *name)
218{
219 char err_str[ERR_MAX_LEN];
220 char *file;
221 char *dir;
222 int err = 0;
223
224 file = malloc(strlen(name) + 1);
225 if (!file) {
226 p_err("mem alloc failed");
227 return -1;
228 }
229
230 strcpy(file, name);
231 dir = dirname(file);
232
233 if (is_bpffs(dir))
234 /* nothing to do if already mounted */
235 goto out_free;
236
237 if (block_mount) {
238 p_err("no BPF file system found, not mounting it due to --nomount option");
239 err = -1;
240 goto out_free;
241 }
242
243 err = mnt_fs(dir, "bpf", err_str, ERR_MAX_LEN);
244 if (err) {
245 err_str[ERR_MAX_LEN - 1] = '\0';
246 p_err("can't mount BPF file system to pin the object (%s): %s",
247 name, err_str);
248 }
249
250out_free:
251 free(file);
252 return err;
253}
254
255int do_pin_fd(int fd, const char *name)
256{
257 int err;
258
259 err = mount_bpffs_for_pin(name);
260 if (err)
261 return err;
262
263 err = bpf_obj_pin(fd, name);
264 if (err)
265 p_err("can't pin the object (%s): %s", name, strerror(errno));
266
267 return err;
268}
269
270int do_pin_any(int argc, char **argv, int (*get_fd)(int *, char ***))
271{
272 int err;
273 int fd;
274
275 fd = get_fd(&argc, &argv);
276 if (fd < 0)
277 return fd;
278
279 err = do_pin_fd(fd, *argv);
280
281 close(fd);
282 return err;
283}
284
285const char *get_fd_type_name(enum bpf_obj_type type)
286{
287 static const char * const names[] = {
288 [BPF_OBJ_UNKNOWN] = "unknown",
289 [BPF_OBJ_PROG] = "prog",
290 [BPF_OBJ_MAP] = "map",
291 };
292
293 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type])
294 return names[BPF_OBJ_UNKNOWN];
295
296 return names[type];
297}
298
299int get_fd_type(int fd)
300{
301 char path[PATH_MAX];
302 char buf[512];
303 ssize_t n;
304
305 snprintf(path, sizeof(path), "/proc/self/fd/%d", fd);
306
307 n = readlink(path, buf, sizeof(buf));
308 if (n < 0) {
309 p_err("can't read link type: %s", strerror(errno));
310 return -1;
311 }
312 if (n == sizeof(path)) {
313 p_err("can't read link type: path too long!");
314 return -1;
315 }
316
317 if (strstr(buf, "bpf-map"))
318 return BPF_OBJ_MAP;
319 else if (strstr(buf, "bpf-prog"))
320 return BPF_OBJ_PROG;
321 else if (strstr(buf, "bpf-link"))
322 return BPF_OBJ_LINK;
323
324 return BPF_OBJ_UNKNOWN;
325}
326
327char *get_fdinfo(int fd, const char *key)
328{
329 char path[PATH_MAX];
330 char *line = NULL;
331 size_t line_n = 0;
332 ssize_t n;
333 FILE *fdi;
334
335 snprintf(path, sizeof(path), "/proc/self/fdinfo/%d", fd);
336
337 fdi = fopen(path, "r");
338 if (!fdi)
339 return NULL;
340
341 while ((n = getline(&line, &line_n, fdi)) > 0) {
342 char *value;
343 int len;
344
345 if (!strstr(line, key))
346 continue;
347
348 fclose(fdi);
349
350 value = strchr(line, '\t');
351 if (!value || !value[1]) {
352 free(line);
353 return NULL;
354 }
355 value++;
356
357 len = strlen(value);
358 memmove(line, value, len);
359 line[len - 1] = '\0';
360
361 return line;
362 }
363
364 free(line);
365 fclose(fdi);
366 return NULL;
367}
368
369void print_data_json(uint8_t *data, size_t len)
370{
371 unsigned int i;
372
373 jsonw_start_array(json_wtr);
374 for (i = 0; i < len; i++)
375 jsonw_printf(json_wtr, "%d", data[i]);
376 jsonw_end_array(json_wtr);
377}
378
379void print_hex_data_json(uint8_t *data, size_t len)
380{
381 unsigned int i;
382
383 jsonw_start_array(json_wtr);
384 for (i = 0; i < len; i++)
385 jsonw_printf(json_wtr, "\"0x%02hhx\"", data[i]);
386 jsonw_end_array(json_wtr);
387}
388
389/* extra params for nftw cb */
390static struct pinned_obj_table *build_fn_table;
391static enum bpf_obj_type build_fn_type;
392
393static int do_build_table_cb(const char *fpath, const struct stat *sb,
394 int typeflag, struct FTW *ftwbuf)
395{
396 struct bpf_prog_info pinned_info;
397 __u32 len = sizeof(pinned_info);
398 struct pinned_obj *obj_node;
399 enum bpf_obj_type objtype;
400 int fd, err = 0;
401
402 if (typeflag != FTW_F)
403 goto out_ret;
404
405 fd = open_obj_pinned(fpath, true);
406 if (fd < 0)
407 goto out_ret;
408
409 objtype = get_fd_type(fd);
410 if (objtype != build_fn_type)
411 goto out_close;
412
413 memset(&pinned_info, 0, sizeof(pinned_info));
414 if (bpf_obj_get_info_by_fd(fd, &pinned_info, &len))
415 goto out_close;
416
417 obj_node = calloc(1, sizeof(*obj_node));
418 if (!obj_node) {
419 err = -1;
420 goto out_close;
421 }
422
423 obj_node->id = pinned_info.id;
424 obj_node->path = strdup(fpath);
425 if (!obj_node->path) {
426 err = -1;
427 free(obj_node);
428 goto out_close;
429 }
430
431 hash_add(build_fn_table->table, &obj_node->hash, obj_node->id);
432out_close:
433 close(fd);
434out_ret:
435 return err;
436}
437
438int build_pinned_obj_table(struct pinned_obj_table *tab,
439 enum bpf_obj_type type)
440{
441 struct mntent *mntent = NULL;
442 FILE *mntfile = NULL;
443 int flags = FTW_PHYS;
444 int nopenfd = 16;
445 int err = 0;
446
447 mntfile = setmntent("/proc/mounts", "r");
448 if (!mntfile)
449 return -1;
450
451 build_fn_table = tab;
452 build_fn_type = type;
453
454 while ((mntent = getmntent(mntfile))) {
455 char *path = mntent->mnt_dir;
456
457 if (strncmp(mntent->mnt_type, "bpf", 3) != 0)
458 continue;
459 err = nftw(path, do_build_table_cb, nopenfd, flags);
460 if (err)
461 break;
462 }
463 fclose(mntfile);
464 return err;
465}
466
467void delete_pinned_obj_table(struct pinned_obj_table *tab)
468{
469 struct pinned_obj *obj;
470 struct hlist_node *tmp;
471 unsigned int bkt;
472
473 hash_for_each_safe(tab->table, bkt, tmp, obj, hash) {
474 hash_del(&obj->hash);
475 free(obj->path);
476 free(obj);
477 }
478}
479
480unsigned int get_page_size(void)
481{
482 static int result;
483
484 if (!result)
485 result = getpagesize();
486 return result;
487}
488
489unsigned int get_possible_cpus(void)
490{
491 int cpus = libbpf_num_possible_cpus();
492
493 if (cpus < 0) {
494 p_err("Can't get # of possible cpus: %s", strerror(-cpus));
495 exit(-1);
496 }
497 return cpus;
498}
499
500static char *
501ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf)
502{
503 struct stat st;
504 int err;
505
506 err = stat("/proc/self/ns/net", &st);
507 if (err) {
508 p_err("Can't stat /proc/self: %s", strerror(errno));
509 return NULL;
510 }
511
512 if (st.st_dev != ns_dev || st.st_ino != ns_ino)
513 return NULL;
514
515 return if_indextoname(ifindex, buf);
516}
517
518static int read_sysfs_hex_int(char *path)
519{
520 char vendor_id_buf[8];
521 int len;
522 int fd;
523
524 fd = open(path, O_RDONLY);
525 if (fd < 0) {
526 p_err("Can't open %s: %s", path, strerror(errno));
527 return -1;
528 }
529
530 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf));
531 close(fd);
532 if (len < 0) {
533 p_err("Can't read %s: %s", path, strerror(errno));
534 return -1;
535 }
536 if (len >= (int)sizeof(vendor_id_buf)) {
537 p_err("Value in %s too long", path);
538 return -1;
539 }
540
541 vendor_id_buf[len] = 0;
542
543 return strtol(vendor_id_buf, NULL, 0);
544}
545
546static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name)
547{
548 char full_path[64];
549
550 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s",
551 devname, entry_name);
552
553 return read_sysfs_hex_int(full_path);
554}
555
556const char *
557ifindex_to_bfd_params(__u32 ifindex, __u64 ns_dev, __u64 ns_ino,
558 const char **opt)
559{
560 char devname[IF_NAMESIZE];
561 int vendor_id;
562 int device_id;
563
564 if (!ifindex_to_name_ns(ifindex, ns_dev, ns_ino, devname)) {
565 p_err("Can't get net device name for ifindex %d: %s", ifindex,
566 strerror(errno));
567 return NULL;
568 }
569
570 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor");
571 if (vendor_id < 0) {
572 p_err("Can't get device vendor id for %s", devname);
573 return NULL;
574 }
575
576 switch (vendor_id) {
577 case 0x19ee:
578 device_id = read_sysfs_netdev_hex_int(devname, "device");
579 if (device_id != 0x4000 &&
580 device_id != 0x6000 &&
581 device_id != 0x6003)
582 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch");
583 *opt = "ctx4";
584 return "NFP-6xxx";
585 default:
586 p_err("Can't get bfd arch name for device vendor id 0x%04x",
587 vendor_id);
588 return NULL;
589 }
590}
591
592void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
593{
594 char name[IF_NAMESIZE];
595
596 if (!ifindex)
597 return;
598
599 printf(" offloaded_to ");
600 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
601 printf("%s", name);
602 else
603 printf("ifindex %u ns_dev %llu ns_ino %llu",
604 ifindex, ns_dev, ns_inode);
605}
606
607void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
608{
609 char name[IF_NAMESIZE];
610
611 if (!ifindex)
612 return;
613
614 jsonw_name(json_wtr, "dev");
615 jsonw_start_object(json_wtr);
616 jsonw_uint_field(json_wtr, "ifindex", ifindex);
617 jsonw_uint_field(json_wtr, "ns_dev", ns_dev);
618 jsonw_uint_field(json_wtr, "ns_inode", ns_inode);
619 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
620 jsonw_string_field(json_wtr, "ifname", name);
621 jsonw_end_object(json_wtr);
622}
623
624int parse_u32_arg(int *argc, char ***argv, __u32 *val, const char *what)
625{
626 char *endptr;
627
628 NEXT_ARGP();
629
630 if (*val) {
631 p_err("%s already specified", what);
632 return -1;
633 }
634
635 *val = strtoul(**argv, &endptr, 0);
636 if (*endptr) {
637 p_err("can't parse %s as %s", **argv, what);
638 return -1;
639 }
640 NEXT_ARGP();
641
642 return 0;
643}
644
645int __printf(2, 0)
646print_all_levels(__maybe_unused enum libbpf_print_level level,
647 const char *format, va_list args)
648{
649 return vfprintf(stderr, format, args);
650}
651
652static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
653{
654 unsigned int id = 0;
655 int fd, nb_fds = 0;
656 void *tmp;
657 int err;
658
659 while (true) {
660 struct bpf_prog_info info = {};
661 __u32 len = sizeof(info);
662
663 err = bpf_prog_get_next_id(id, &id);
664 if (err) {
665 if (errno != ENOENT) {
666 p_err("%s", strerror(errno));
667 goto err_close_fds;
668 }
669 return nb_fds;
670 }
671
672 fd = bpf_prog_get_fd_by_id(id);
673 if (fd < 0) {
674 p_err("can't get prog by id (%u): %s",
675 id, strerror(errno));
676 goto err_close_fds;
677 }
678
679 err = bpf_obj_get_info_by_fd(fd, &info, &len);
680 if (err) {
681 p_err("can't get prog info (%u): %s",
682 id, strerror(errno));
683 goto err_close_fd;
684 }
685
686 if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
687 (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
688 close(fd);
689 continue;
690 }
691
692 if (nb_fds > 0) {
693 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
694 if (!tmp) {
695 p_err("failed to realloc");
696 goto err_close_fd;
697 }
698 *fds = tmp;
699 }
700 (*fds)[nb_fds++] = fd;
701 }
702
703err_close_fd:
704 close(fd);
705err_close_fds:
706 while (--nb_fds >= 0)
707 close((*fds)[nb_fds]);
708 return -1;
709}
710
711int prog_parse_fds(int *argc, char ***argv, int **fds)
712{
713 if (is_prefix(**argv, "id")) {
714 unsigned int id;
715 char *endptr;
716
717 NEXT_ARGP();
718
719 id = strtoul(**argv, &endptr, 0);
720 if (*endptr) {
721 p_err("can't parse %s as ID", **argv);
722 return -1;
723 }
724 NEXT_ARGP();
725
726 (*fds)[0] = bpf_prog_get_fd_by_id(id);
727 if ((*fds)[0] < 0) {
728 p_err("get by id (%u): %s", id, strerror(errno));
729 return -1;
730 }
731 return 1;
732 } else if (is_prefix(**argv, "tag")) {
733 unsigned char tag[BPF_TAG_SIZE];
734
735 NEXT_ARGP();
736
737 if (sscanf(**argv, BPF_TAG_FMT, tag, tag + 1, tag + 2,
738 tag + 3, tag + 4, tag + 5, tag + 6, tag + 7)
739 != BPF_TAG_SIZE) {
740 p_err("can't parse tag");
741 return -1;
742 }
743 NEXT_ARGP();
744
745 return prog_fd_by_nametag(tag, fds, true);
746 } else if (is_prefix(**argv, "name")) {
747 char *name;
748
749 NEXT_ARGP();
750
751 name = **argv;
752 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
753 p_err("can't parse name");
754 return -1;
755 }
756 NEXT_ARGP();
757
758 return prog_fd_by_nametag(name, fds, false);
759 } else if (is_prefix(**argv, "pinned")) {
760 char *path;
761
762 NEXT_ARGP();
763
764 path = **argv;
765 NEXT_ARGP();
766
767 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_PROG);
768 if ((*fds)[0] < 0)
769 return -1;
770 return 1;
771 }
772
773 p_err("expected 'id', 'tag', 'name' or 'pinned', got: '%s'?", **argv);
774 return -1;
775}
776
777int prog_parse_fd(int *argc, char ***argv)
778{
779 int *fds = NULL;
780 int nb_fds, fd;
781
782 fds = malloc(sizeof(int));
783 if (!fds) {
784 p_err("mem alloc failed");
785 return -1;
786 }
787 nb_fds = prog_parse_fds(argc, argv, &fds);
788 if (nb_fds != 1) {
789 if (nb_fds > 1) {
790 p_err("several programs match this handle");
791 while (nb_fds--)
792 close(fds[nb_fds]);
793 }
794 fd = -1;
795 goto exit_free;
796 }
797
798 fd = fds[0];
799exit_free:
800 free(fds);
801 return fd;
802}
803
804static int map_fd_by_name(char *name, int **fds)
805{
806 unsigned int id = 0;
807 int fd, nb_fds = 0;
808 void *tmp;
809 int err;
810
811 while (true) {
812 struct bpf_map_info info = {};
813 __u32 len = sizeof(info);
814
815 err = bpf_map_get_next_id(id, &id);
816 if (err) {
817 if (errno != ENOENT) {
818 p_err("%s", strerror(errno));
819 goto err_close_fds;
820 }
821 return nb_fds;
822 }
823
824 fd = bpf_map_get_fd_by_id(id);
825 if (fd < 0) {
826 p_err("can't get map by id (%u): %s",
827 id, strerror(errno));
828 goto err_close_fds;
829 }
830
831 err = bpf_obj_get_info_by_fd(fd, &info, &len);
832 if (err) {
833 p_err("can't get map info (%u): %s",
834 id, strerror(errno));
835 goto err_close_fd;
836 }
837
838 if (strncmp(name, info.name, BPF_OBJ_NAME_LEN)) {
839 close(fd);
840 continue;
841 }
842
843 if (nb_fds > 0) {
844 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
845 if (!tmp) {
846 p_err("failed to realloc");
847 goto err_close_fd;
848 }
849 *fds = tmp;
850 }
851 (*fds)[nb_fds++] = fd;
852 }
853
854err_close_fd:
855 close(fd);
856err_close_fds:
857 while (--nb_fds >= 0)
858 close((*fds)[nb_fds]);
859 return -1;
860}
861
862int map_parse_fds(int *argc, char ***argv, int **fds)
863{
864 if (is_prefix(**argv, "id")) {
865 unsigned int id;
866 char *endptr;
867
868 NEXT_ARGP();
869
870 id = strtoul(**argv, &endptr, 0);
871 if (*endptr) {
872 p_err("can't parse %s as ID", **argv);
873 return -1;
874 }
875 NEXT_ARGP();
876
877 (*fds)[0] = bpf_map_get_fd_by_id(id);
878 if ((*fds)[0] < 0) {
879 p_err("get map by id (%u): %s", id, strerror(errno));
880 return -1;
881 }
882 return 1;
883 } else if (is_prefix(**argv, "name")) {
884 char *name;
885
886 NEXT_ARGP();
887
888 name = **argv;
889 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
890 p_err("can't parse name");
891 return -1;
892 }
893 NEXT_ARGP();
894
895 return map_fd_by_name(name, fds);
896 } else if (is_prefix(**argv, "pinned")) {
897 char *path;
898
899 NEXT_ARGP();
900
901 path = **argv;
902 NEXT_ARGP();
903
904 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_MAP);
905 if ((*fds)[0] < 0)
906 return -1;
907 return 1;
908 }
909
910 p_err("expected 'id', 'name' or 'pinned', got: '%s'?", **argv);
911 return -1;
912}
913
914int map_parse_fd(int *argc, char ***argv)
915{
916 int *fds = NULL;
917 int nb_fds, fd;
918
919 fds = malloc(sizeof(int));
920 if (!fds) {
921 p_err("mem alloc failed");
922 return -1;
923 }
924 nb_fds = map_parse_fds(argc, argv, &fds);
925 if (nb_fds != 1) {
926 if (nb_fds > 1) {
927 p_err("several maps match this handle");
928 while (nb_fds--)
929 close(fds[nb_fds]);
930 }
931 fd = -1;
932 goto exit_free;
933 }
934
935 fd = fds[0];
936exit_free:
937 free(fds);
938 return fd;
939}
940
941int map_parse_fd_and_info(int *argc, char ***argv, void *info, __u32 *info_len)
942{
943 int err;
944 int fd;
945
946 fd = map_parse_fd(argc, argv);
947 if (fd < 0)
948 return -1;
949
950 err = bpf_obj_get_info_by_fd(fd, info, info_len);
951 if (err) {
952 p_err("can't get map info: %s", strerror(errno));
953 close(fd);
954 return err;
955 }
956
957 return fd;
958}
1/*
2 * Copyright (C) 2017 Netronome Systems, Inc.
3 *
4 * This software is dual licensed under the GNU General License Version 2,
5 * June 1991 as shown in the file COPYING in the top-level directory of this
6 * source tree or the BSD 2-Clause License provided below. You have the
7 * option to license this software under the complete terms of either license.
8 *
9 * The BSD 2-Clause License:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * 1. Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * 2. Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34/* Author: Jakub Kicinski <kubakici@wp.pl> */
35
36#include <errno.h>
37#include <fcntl.h>
38#include <fts.h>
39#include <libgen.h>
40#include <mntent.h>
41#include <stdbool.h>
42#include <stdio.h>
43#include <stdlib.h>
44#include <string.h>
45#include <unistd.h>
46#include <linux/limits.h>
47#include <linux/magic.h>
48#include <net/if.h>
49#include <sys/mount.h>
50#include <sys/stat.h>
51#include <sys/types.h>
52#include <sys/vfs.h>
53
54#include <bpf.h>
55
56#include "main.h"
57
58#ifndef BPF_FS_MAGIC
59#define BPF_FS_MAGIC 0xcafe4a11
60#endif
61
62void p_err(const char *fmt, ...)
63{
64 va_list ap;
65
66 va_start(ap, fmt);
67 if (json_output) {
68 jsonw_start_object(json_wtr);
69 jsonw_name(json_wtr, "error");
70 jsonw_vprintf_enquote(json_wtr, fmt, ap);
71 jsonw_end_object(json_wtr);
72 } else {
73 fprintf(stderr, "Error: ");
74 vfprintf(stderr, fmt, ap);
75 fprintf(stderr, "\n");
76 }
77 va_end(ap);
78}
79
80void p_info(const char *fmt, ...)
81{
82 va_list ap;
83
84 if (json_output)
85 return;
86
87 va_start(ap, fmt);
88 vfprintf(stderr, fmt, ap);
89 fprintf(stderr, "\n");
90 va_end(ap);
91}
92
93static bool is_bpffs(char *path)
94{
95 struct statfs st_fs;
96
97 if (statfs(path, &st_fs) < 0)
98 return false;
99
100 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC;
101}
102
103static int mnt_bpffs(const char *target, char *buff, size_t bufflen)
104{
105 bool bind_done = false;
106
107 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) {
108 if (errno != EINVAL || bind_done) {
109 snprintf(buff, bufflen,
110 "mount --make-private %s failed: %s",
111 target, strerror(errno));
112 return -1;
113 }
114
115 if (mount(target, target, "none", MS_BIND, NULL)) {
116 snprintf(buff, bufflen,
117 "mount --bind %s %s failed: %s",
118 target, target, strerror(errno));
119 return -1;
120 }
121
122 bind_done = true;
123 }
124
125 if (mount("bpf", target, "bpf", 0, "mode=0700")) {
126 snprintf(buff, bufflen, "mount -t bpf bpf %s failed: %s",
127 target, strerror(errno));
128 return -1;
129 }
130
131 return 0;
132}
133
134int open_obj_pinned(char *path)
135{
136 int fd;
137
138 fd = bpf_obj_get(path);
139 if (fd < 0) {
140 p_err("bpf obj get (%s): %s", path,
141 errno == EACCES && !is_bpffs(dirname(path)) ?
142 "directory not in bpf file system (bpffs)" :
143 strerror(errno));
144 return -1;
145 }
146
147 return fd;
148}
149
150int open_obj_pinned_any(char *path, enum bpf_obj_type exp_type)
151{
152 enum bpf_obj_type type;
153 int fd;
154
155 fd = open_obj_pinned(path);
156 if (fd < 0)
157 return -1;
158
159 type = get_fd_type(fd);
160 if (type < 0) {
161 close(fd);
162 return type;
163 }
164 if (type != exp_type) {
165 p_err("incorrect object type: %s", get_fd_type_name(type));
166 close(fd);
167 return -1;
168 }
169
170 return fd;
171}
172
173int do_pin_fd(int fd, const char *name)
174{
175 char err_str[ERR_MAX_LEN];
176 char *file;
177 char *dir;
178 int err = 0;
179
180 err = bpf_obj_pin(fd, name);
181 if (!err)
182 goto out;
183
184 file = malloc(strlen(name) + 1);
185 strcpy(file, name);
186 dir = dirname(file);
187
188 if (errno != EPERM || is_bpffs(dir)) {
189 p_err("can't pin the object (%s): %s", name, strerror(errno));
190 goto out_free;
191 }
192
193 /* Attempt to mount bpffs, then retry pinning. */
194 err = mnt_bpffs(dir, err_str, ERR_MAX_LEN);
195 if (!err) {
196 err = bpf_obj_pin(fd, name);
197 if (err)
198 p_err("can't pin the object (%s): %s", name,
199 strerror(errno));
200 } else {
201 err_str[ERR_MAX_LEN - 1] = '\0';
202 p_err("can't mount BPF file system to pin the object (%s): %s",
203 name, err_str);
204 }
205
206out_free:
207 free(file);
208out:
209 return err;
210}
211
212int do_pin_any(int argc, char **argv, int (*get_fd_by_id)(__u32))
213{
214 unsigned int id;
215 char *endptr;
216 int err;
217 int fd;
218
219 if (!is_prefix(*argv, "id")) {
220 p_err("expected 'id' got %s", *argv);
221 return -1;
222 }
223 NEXT_ARG();
224
225 id = strtoul(*argv, &endptr, 0);
226 if (*endptr) {
227 p_err("can't parse %s as ID", *argv);
228 return -1;
229 }
230 NEXT_ARG();
231
232 if (argc != 1)
233 usage();
234
235 fd = get_fd_by_id(id);
236 if (fd < 0) {
237 p_err("can't get prog by id (%u): %s", id, strerror(errno));
238 return -1;
239 }
240
241 err = do_pin_fd(fd, *argv);
242
243 close(fd);
244 return err;
245}
246
247const char *get_fd_type_name(enum bpf_obj_type type)
248{
249 static const char * const names[] = {
250 [BPF_OBJ_UNKNOWN] = "unknown",
251 [BPF_OBJ_PROG] = "prog",
252 [BPF_OBJ_MAP] = "map",
253 };
254
255 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type])
256 return names[BPF_OBJ_UNKNOWN];
257
258 return names[type];
259}
260
261int get_fd_type(int fd)
262{
263 char path[PATH_MAX];
264 char buf[512];
265 ssize_t n;
266
267 snprintf(path, sizeof(path), "/proc/%d/fd/%d", getpid(), fd);
268
269 n = readlink(path, buf, sizeof(buf));
270 if (n < 0) {
271 p_err("can't read link type: %s", strerror(errno));
272 return -1;
273 }
274 if (n == sizeof(path)) {
275 p_err("can't read link type: path too long!");
276 return -1;
277 }
278
279 if (strstr(buf, "bpf-map"))
280 return BPF_OBJ_MAP;
281 else if (strstr(buf, "bpf-prog"))
282 return BPF_OBJ_PROG;
283
284 return BPF_OBJ_UNKNOWN;
285}
286
287char *get_fdinfo(int fd, const char *key)
288{
289 char path[PATH_MAX];
290 char *line = NULL;
291 size_t line_n = 0;
292 ssize_t n;
293 FILE *fdi;
294
295 snprintf(path, sizeof(path), "/proc/%d/fdinfo/%d", getpid(), fd);
296
297 fdi = fopen(path, "r");
298 if (!fdi) {
299 p_err("can't open fdinfo: %s", strerror(errno));
300 return NULL;
301 }
302
303 while ((n = getline(&line, &line_n, fdi))) {
304 char *value;
305 int len;
306
307 if (!strstr(line, key))
308 continue;
309
310 fclose(fdi);
311
312 value = strchr(line, '\t');
313 if (!value || !value[1]) {
314 p_err("malformed fdinfo!?");
315 free(line);
316 return NULL;
317 }
318 value++;
319
320 len = strlen(value);
321 memmove(line, value, len);
322 line[len - 1] = '\0';
323
324 return line;
325 }
326
327 p_err("key '%s' not found in fdinfo", key);
328 free(line);
329 fclose(fdi);
330 return NULL;
331}
332
333void print_hex_data_json(uint8_t *data, size_t len)
334{
335 unsigned int i;
336
337 jsonw_start_array(json_wtr);
338 for (i = 0; i < len; i++)
339 jsonw_printf(json_wtr, "\"0x%02hhx\"", data[i]);
340 jsonw_end_array(json_wtr);
341}
342
343int build_pinned_obj_table(struct pinned_obj_table *tab,
344 enum bpf_obj_type type)
345{
346 struct bpf_prog_info pinned_info = {};
347 struct pinned_obj *obj_node = NULL;
348 __u32 len = sizeof(pinned_info);
349 struct mntent *mntent = NULL;
350 enum bpf_obj_type objtype;
351 FILE *mntfile = NULL;
352 FTSENT *ftse = NULL;
353 FTS *fts = NULL;
354 int fd, err;
355
356 mntfile = setmntent("/proc/mounts", "r");
357 if (!mntfile)
358 return -1;
359
360 while ((mntent = getmntent(mntfile))) {
361 char *path[] = { mntent->mnt_dir, NULL };
362
363 if (strncmp(mntent->mnt_type, "bpf", 3) != 0)
364 continue;
365
366 fts = fts_open(path, 0, NULL);
367 if (!fts)
368 continue;
369
370 while ((ftse = fts_read(fts))) {
371 if (!(ftse->fts_info & FTS_F))
372 continue;
373 fd = open_obj_pinned(ftse->fts_path);
374 if (fd < 0)
375 continue;
376
377 objtype = get_fd_type(fd);
378 if (objtype != type) {
379 close(fd);
380 continue;
381 }
382 memset(&pinned_info, 0, sizeof(pinned_info));
383 err = bpf_obj_get_info_by_fd(fd, &pinned_info, &len);
384 if (err) {
385 close(fd);
386 continue;
387 }
388
389 obj_node = malloc(sizeof(*obj_node));
390 if (!obj_node) {
391 close(fd);
392 fts_close(fts);
393 fclose(mntfile);
394 return -1;
395 }
396
397 memset(obj_node, 0, sizeof(*obj_node));
398 obj_node->id = pinned_info.id;
399 obj_node->path = strdup(ftse->fts_path);
400 hash_add(tab->table, &obj_node->hash, obj_node->id);
401
402 close(fd);
403 }
404 fts_close(fts);
405 }
406 fclose(mntfile);
407 return 0;
408}
409
410void delete_pinned_obj_table(struct pinned_obj_table *tab)
411{
412 struct pinned_obj *obj;
413 struct hlist_node *tmp;
414 unsigned int bkt;
415
416 hash_for_each_safe(tab->table, bkt, tmp, obj, hash) {
417 hash_del(&obj->hash);
418 free(obj->path);
419 free(obj);
420 }
421}
422
423static char *
424ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf)
425{
426 struct stat st;
427 int err;
428
429 err = stat("/proc/self/ns/net", &st);
430 if (err) {
431 p_err("Can't stat /proc/self: %s", strerror(errno));
432 return NULL;
433 }
434
435 if (st.st_dev != ns_dev || st.st_ino != ns_ino)
436 return NULL;
437
438 return if_indextoname(ifindex, buf);
439}
440
441static int read_sysfs_hex_int(char *path)
442{
443 char vendor_id_buf[8];
444 int len;
445 int fd;
446
447 fd = open(path, O_RDONLY);
448 if (fd < 0) {
449 p_err("Can't open %s: %s", path, strerror(errno));
450 return -1;
451 }
452
453 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf));
454 close(fd);
455 if (len < 0) {
456 p_err("Can't read %s: %s", path, strerror(errno));
457 return -1;
458 }
459 if (len >= (int)sizeof(vendor_id_buf)) {
460 p_err("Value in %s too long", path);
461 return -1;
462 }
463
464 vendor_id_buf[len] = 0;
465
466 return strtol(vendor_id_buf, NULL, 0);
467}
468
469static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name)
470{
471 char full_path[64];
472
473 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s",
474 devname, entry_name);
475
476 return read_sysfs_hex_int(full_path);
477}
478
479const char *ifindex_to_bfd_name_ns(__u32 ifindex, __u64 ns_dev, __u64 ns_ino)
480{
481 char devname[IF_NAMESIZE];
482 int vendor_id;
483 int device_id;
484
485 if (!ifindex_to_name_ns(ifindex, ns_dev, ns_ino, devname)) {
486 p_err("Can't get net device name for ifindex %d: %s", ifindex,
487 strerror(errno));
488 return NULL;
489 }
490
491 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor");
492 if (vendor_id < 0) {
493 p_err("Can't get device vendor id for %s", devname);
494 return NULL;
495 }
496
497 switch (vendor_id) {
498 case 0x19ee:
499 device_id = read_sysfs_netdev_hex_int(devname, "device");
500 if (device_id != 0x4000 &&
501 device_id != 0x6000 &&
502 device_id != 0x6003)
503 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch");
504 return "NFP-6xxx";
505 default:
506 p_err("Can't get bfd arch name for device vendor id 0x%04x",
507 vendor_id);
508 return NULL;
509 }
510}
511
512void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
513{
514 char name[IF_NAMESIZE];
515
516 if (!ifindex)
517 return;
518
519 printf(" dev ");
520 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
521 printf("%s", name);
522 else
523 printf("ifindex %u ns_dev %llu ns_ino %llu",
524 ifindex, ns_dev, ns_inode);
525}
526
527void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
528{
529 char name[IF_NAMESIZE];
530
531 if (!ifindex)
532 return;
533
534 jsonw_name(json_wtr, "dev");
535 jsonw_start_object(json_wtr);
536 jsonw_uint_field(json_wtr, "ifindex", ifindex);
537 jsonw_uint_field(json_wtr, "ns_dev", ns_dev);
538 jsonw_uint_field(json_wtr, "ns_inode", ns_inode);
539 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
540 jsonw_string_field(json_wtr, "ifname", name);
541 jsonw_end_object(json_wtr);
542}