1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */
16 #include <linux/limits.h>
17 #include <linux/magic.h>
19 #include <sys/mount.h>
20 #include <sys/resource.h>
25 #include <bpf/libbpf.h> /* libbpf_num_possible_cpus */
30 #define BPF_FS_MAGIC 0xcafe4a11
33 const 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",
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",
72 void p_err(const char *fmt, ...)
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);
83 fprintf(stderr, "Error: ");
84 vfprintf(stderr, fmt, ap);
85 fprintf(stderr, "\n");
90 void p_info(const char *fmt, ...)
98 vfprintf(stderr, fmt, ap);
99 fprintf(stderr, "\n");
103 static bool is_bpffs(char *path)
107 if (statfs(path, &st_fs) < 0)
110 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC;
113 void set_max_rlimit(void)
115 struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY };
117 setrlimit(RLIMIT_MEMLOCK, &rinf);
121 mnt_fs(const char *target, const char *type, char *buff, size_t bufflen)
123 bool bind_done = false;
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));
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));
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));
152 int mount_tracefs(const char *target)
154 char err_str[ERR_MAX_LEN];
157 err = mnt_fs(target, "tracefs", err_str, ERR_MAX_LEN);
159 err_str[ERR_MAX_LEN - 1] = '\0';
160 p_err("can't mount tracefs: %s", err_str);
166 int open_obj_pinned(const char *path, bool quiet)
171 pname = strdup(path);
174 p_err("mem alloc failed");
178 fd = bpf_obj_get(pname);
181 p_err("bpf obj get (%s): %s", pname,
182 errno == EACCES && !is_bpffs(dirname(pname)) ?
183 "directory not in bpf file system (bpffs)" :
194 int open_obj_pinned_any(const char *path, enum bpf_obj_type exp_type)
196 enum bpf_obj_type type;
199 fd = open_obj_pinned(path, false);
203 type = get_fd_type(fd);
208 if (type != exp_type) {
209 p_err("incorrect object type: %s", get_fd_type_name(type));
217 int mount_bpffs_for_pin(const char *name)
219 char err_str[ERR_MAX_LEN];
224 file = malloc(strlen(name) + 1);
226 p_err("mem alloc failed");
234 /* nothing to do if already mounted */
238 p_err("no BPF file system found, not mounting it due to --nomount option");
243 err = mnt_fs(dir, "bpf", err_str, ERR_MAX_LEN);
245 err_str[ERR_MAX_LEN - 1] = '\0';
246 p_err("can't mount BPF file system to pin the object (%s): %s",
255 int do_pin_fd(int fd, const char *name)
259 err = mount_bpffs_for_pin(name);
263 err = bpf_obj_pin(fd, name);
265 p_err("can't pin the object (%s): %s", name, strerror(errno));
270 int do_pin_any(int argc, char **argv, int (*get_fd)(int *, char ***))
275 fd = get_fd(&argc, &argv);
279 err = do_pin_fd(fd, *argv);
285 const char *get_fd_type_name(enum bpf_obj_type type)
287 static const char * const names[] = {
288 [BPF_OBJ_UNKNOWN] = "unknown",
289 [BPF_OBJ_PROG] = "prog",
290 [BPF_OBJ_MAP] = "map",
293 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type])
294 return names[BPF_OBJ_UNKNOWN];
299 int get_fd_type(int fd)
305 snprintf(path, sizeof(path), "/proc/self/fd/%d", fd);
307 n = readlink(path, buf, sizeof(buf));
309 p_err("can't read link type: %s", strerror(errno));
312 if (n == sizeof(path)) {
313 p_err("can't read link type: path too long!");
317 if (strstr(buf, "bpf-map"))
319 else if (strstr(buf, "bpf-prog"))
321 else if (strstr(buf, "bpf-link"))
324 return BPF_OBJ_UNKNOWN;
327 char *get_fdinfo(int fd, const char *key)
335 snprintf(path, sizeof(path), "/proc/self/fdinfo/%d", fd);
337 fdi = fopen(path, "r");
341 while ((n = getline(&line, &line_n, fdi)) > 0) {
345 if (!strstr(line, key))
350 value = strchr(line, '\t');
351 if (!value || !value[1]) {
358 memmove(line, value, len);
359 line[len - 1] = '\0';
369 void print_data_json(uint8_t *data, size_t len)
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);
379 void print_hex_data_json(uint8_t *data, size_t len)
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);
389 /* extra params for nftw cb */
390 static struct pinned_obj_table *build_fn_table;
391 static enum bpf_obj_type build_fn_type;
393 static int do_build_table_cb(const char *fpath, const struct stat *sb,
394 int typeflag, struct FTW *ftwbuf)
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;
402 if (typeflag != FTW_F)
405 fd = open_obj_pinned(fpath, true);
409 objtype = get_fd_type(fd);
410 if (objtype != build_fn_type)
413 memset(&pinned_info, 0, sizeof(pinned_info));
414 if (bpf_obj_get_info_by_fd(fd, &pinned_info, &len))
417 obj_node = calloc(1, sizeof(*obj_node));
423 obj_node->id = pinned_info.id;
424 obj_node->path = strdup(fpath);
425 if (!obj_node->path) {
431 hash_add(build_fn_table->table, &obj_node->hash, obj_node->id);
438 int build_pinned_obj_table(struct pinned_obj_table *tab,
439 enum bpf_obj_type type)
441 struct mntent *mntent = NULL;
442 FILE *mntfile = NULL;
443 int flags = FTW_PHYS;
447 mntfile = setmntent("/proc/mounts", "r");
451 build_fn_table = tab;
452 build_fn_type = type;
454 while ((mntent = getmntent(mntfile))) {
455 char *path = mntent->mnt_dir;
457 if (strncmp(mntent->mnt_type, "bpf", 3) != 0)
459 err = nftw(path, do_build_table_cb, nopenfd, flags);
467 void delete_pinned_obj_table(struct pinned_obj_table *tab)
469 struct pinned_obj *obj;
470 struct hlist_node *tmp;
473 hash_for_each_safe(tab->table, bkt, tmp, obj, hash) {
474 hash_del(&obj->hash);
480 unsigned int get_page_size(void)
485 result = getpagesize();
489 unsigned int get_possible_cpus(void)
491 int cpus = libbpf_num_possible_cpus();
494 p_err("Can't get # of possible cpus: %s", strerror(-cpus));
501 ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf)
506 err = stat("/proc/self/ns/net", &st);
508 p_err("Can't stat /proc/self: %s", strerror(errno));
512 if (st.st_dev != ns_dev || st.st_ino != ns_ino)
515 return if_indextoname(ifindex, buf);
518 static int read_sysfs_hex_int(char *path)
520 char vendor_id_buf[8];
524 fd = open(path, O_RDONLY);
526 p_err("Can't open %s: %s", path, strerror(errno));
530 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf));
533 p_err("Can't read %s: %s", path, strerror(errno));
536 if (len >= (int)sizeof(vendor_id_buf)) {
537 p_err("Value in %s too long", path);
541 vendor_id_buf[len] = 0;
543 return strtol(vendor_id_buf, NULL, 0);
546 static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name)
550 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s",
551 devname, entry_name);
553 return read_sysfs_hex_int(full_path);
557 ifindex_to_bfd_params(__u32 ifindex, __u64 ns_dev, __u64 ns_ino,
560 char devname[IF_NAMESIZE];
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,
570 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor");
572 p_err("Can't get device vendor id for %s", devname);
578 device_id = read_sysfs_netdev_hex_int(devname, "device");
579 if (device_id != 0x4000 &&
580 device_id != 0x6000 &&
582 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch");
586 p_err("Can't get bfd arch name for device vendor id 0x%04x",
592 void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
594 char name[IF_NAMESIZE];
599 printf(" offloaded_to ");
600 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
603 printf("ifindex %u ns_dev %llu ns_ino %llu",
604 ifindex, ns_dev, ns_inode);
607 void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
609 char name[IF_NAMESIZE];
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);
624 int parse_u32_arg(int *argc, char ***argv, __u32 *val, const char *what)
631 p_err("%s already specified", what);
635 *val = strtoul(**argv, &endptr, 0);
637 p_err("can't parse %s as %s", **argv, what);
646 print_all_levels(__maybe_unused enum libbpf_print_level level,
647 const char *format, va_list args)
649 return vfprintf(stderr, format, args);
652 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
660 struct bpf_prog_info info = {};
661 __u32 len = sizeof(info);
663 err = bpf_prog_get_next_id(id, &id);
665 if (errno != ENOENT) {
666 p_err("%s", strerror(errno));
672 fd = bpf_prog_get_fd_by_id(id);
674 p_err("can't get prog by id (%u): %s",
675 id, strerror(errno));
679 err = bpf_obj_get_info_by_fd(fd, &info, &len);
681 p_err("can't get prog info (%u): %s",
682 id, strerror(errno));
686 if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
687 (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
693 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
695 p_err("failed to realloc");
700 (*fds)[nb_fds++] = fd;
706 while (--nb_fds >= 0)
707 close((*fds)[nb_fds]);
711 int prog_parse_fds(int *argc, char ***argv, int **fds)
713 if (is_prefix(**argv, "id")) {
719 id = strtoul(**argv, &endptr, 0);
721 p_err("can't parse %s as ID", **argv);
726 (*fds)[0] = bpf_prog_get_fd_by_id(id);
728 p_err("get by id (%u): %s", id, strerror(errno));
732 } else if (is_prefix(**argv, "tag")) {
733 unsigned char tag[BPF_TAG_SIZE];
737 if (sscanf(**argv, BPF_TAG_FMT, tag, tag + 1, tag + 2,
738 tag + 3, tag + 4, tag + 5, tag + 6, tag + 7)
740 p_err("can't parse tag");
745 return prog_fd_by_nametag(tag, fds, true);
746 } else if (is_prefix(**argv, "name")) {
752 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
753 p_err("can't parse name");
758 return prog_fd_by_nametag(name, fds, false);
759 } else if (is_prefix(**argv, "pinned")) {
767 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_PROG);
773 p_err("expected 'id', 'tag', 'name' or 'pinned', got: '%s'?", **argv);
777 int prog_parse_fd(int *argc, char ***argv)
782 fds = malloc(sizeof(int));
784 p_err("mem alloc failed");
787 nb_fds = prog_parse_fds(argc, argv, &fds);
790 p_err("several programs match this handle");
804 static int map_fd_by_name(char *name, int **fds)
812 struct bpf_map_info info = {};
813 __u32 len = sizeof(info);
815 err = bpf_map_get_next_id(id, &id);
817 if (errno != ENOENT) {
818 p_err("%s", strerror(errno));
824 fd = bpf_map_get_fd_by_id(id);
826 p_err("can't get map by id (%u): %s",
827 id, strerror(errno));
831 err = bpf_obj_get_info_by_fd(fd, &info, &len);
833 p_err("can't get map info (%u): %s",
834 id, strerror(errno));
838 if (strncmp(name, info.name, BPF_OBJ_NAME_LEN)) {
844 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
846 p_err("failed to realloc");
851 (*fds)[nb_fds++] = fd;
857 while (--nb_fds >= 0)
858 close((*fds)[nb_fds]);
862 int map_parse_fds(int *argc, char ***argv, int **fds)
864 if (is_prefix(**argv, "id")) {
870 id = strtoul(**argv, &endptr, 0);
872 p_err("can't parse %s as ID", **argv);
877 (*fds)[0] = bpf_map_get_fd_by_id(id);
879 p_err("get map by id (%u): %s", id, strerror(errno));
883 } else if (is_prefix(**argv, "name")) {
889 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
890 p_err("can't parse name");
895 return map_fd_by_name(name, fds);
896 } else if (is_prefix(**argv, "pinned")) {
904 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_MAP);
910 p_err("expected 'id', 'name' or 'pinned', got: '%s'?", **argv);
914 int map_parse_fd(int *argc, char ***argv)
919 fds = malloc(sizeof(int));
921 p_err("mem alloc failed");
924 nb_fds = map_parse_fds(argc, argv, &fds);
927 p_err("several maps match this handle");
941 int map_parse_fd_and_info(int *argc, char ***argv, void *info, __u32 *info_len)
946 fd = map_parse_fd(argc, argv);
950 err = bpf_obj_get_info_by_fd(fd, info, info_len);
952 p_err("can't get map info: %s", strerror(errno));