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",
70 [BPF_TRACE_ITER] = "trace_iter",
71 [BPF_XDP_DEVMAP] = "xdp_devmap",
72 [BPF_XDP_CPUMAP] = "xdp_cpumap",
74 [BPF_SK_REUSEPORT_SELECT] = "sk_skb_reuseport_select",
75 [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_skb_reuseport_select_or_migrate",
78 void p_err(const char *fmt, ...)
84 jsonw_start_object(json_wtr);
85 jsonw_name(json_wtr, "error");
86 jsonw_vprintf_enquote(json_wtr, fmt, ap);
87 jsonw_end_object(json_wtr);
89 fprintf(stderr, "Error: ");
90 vfprintf(stderr, fmt, ap);
91 fprintf(stderr, "\n");
96 void p_info(const char *fmt, ...)
104 vfprintf(stderr, fmt, ap);
105 fprintf(stderr, "\n");
109 static bool is_bpffs(char *path)
113 if (statfs(path, &st_fs) < 0)
116 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC;
119 void set_max_rlimit(void)
121 struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY };
123 setrlimit(RLIMIT_MEMLOCK, &rinf);
127 mnt_fs(const char *target, const char *type, char *buff, size_t bufflen)
129 bool bind_done = false;
131 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) {
132 if (errno != EINVAL || bind_done) {
133 snprintf(buff, bufflen,
134 "mount --make-private %s failed: %s",
135 target, strerror(errno));
139 if (mount(target, target, "none", MS_BIND, NULL)) {
140 snprintf(buff, bufflen,
141 "mount --bind %s %s failed: %s",
142 target, target, strerror(errno));
149 if (mount(type, target, type, 0, "mode=0700")) {
150 snprintf(buff, bufflen, "mount -t %s %s %s failed: %s",
151 type, type, target, strerror(errno));
158 int mount_tracefs(const char *target)
160 char err_str[ERR_MAX_LEN];
163 err = mnt_fs(target, "tracefs", err_str, ERR_MAX_LEN);
165 err_str[ERR_MAX_LEN - 1] = '\0';
166 p_err("can't mount tracefs: %s", err_str);
172 int open_obj_pinned(const char *path, bool quiet)
177 pname = strdup(path);
180 p_err("mem alloc failed");
184 fd = bpf_obj_get(pname);
187 p_err("bpf obj get (%s): %s", pname,
188 errno == EACCES && !is_bpffs(dirname(pname)) ?
189 "directory not in bpf file system (bpffs)" :
200 int open_obj_pinned_any(const char *path, enum bpf_obj_type exp_type)
202 enum bpf_obj_type type;
205 fd = open_obj_pinned(path, false);
209 type = get_fd_type(fd);
214 if (type != exp_type) {
215 p_err("incorrect object type: %s", get_fd_type_name(type));
223 int mount_bpffs_for_pin(const char *name)
225 char err_str[ERR_MAX_LEN];
230 file = malloc(strlen(name) + 1);
232 p_err("mem alloc failed");
240 /* nothing to do if already mounted */
244 p_err("no BPF file system found, not mounting it due to --nomount option");
249 err = mnt_fs(dir, "bpf", err_str, ERR_MAX_LEN);
251 err_str[ERR_MAX_LEN - 1] = '\0';
252 p_err("can't mount BPF file system to pin the object (%s): %s",
261 int do_pin_fd(int fd, const char *name)
265 err = mount_bpffs_for_pin(name);
269 err = bpf_obj_pin(fd, name);
271 p_err("can't pin the object (%s): %s", name, strerror(errno));
276 int do_pin_any(int argc, char **argv, int (*get_fd)(int *, char ***))
281 fd = get_fd(&argc, &argv);
285 err = do_pin_fd(fd, *argv);
291 const char *get_fd_type_name(enum bpf_obj_type type)
293 static const char * const names[] = {
294 [BPF_OBJ_UNKNOWN] = "unknown",
295 [BPF_OBJ_PROG] = "prog",
296 [BPF_OBJ_MAP] = "map",
299 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type])
300 return names[BPF_OBJ_UNKNOWN];
305 int get_fd_type(int fd)
311 snprintf(path, sizeof(path), "/proc/self/fd/%d", fd);
313 n = readlink(path, buf, sizeof(buf));
315 p_err("can't read link type: %s", strerror(errno));
318 if (n == sizeof(path)) {
319 p_err("can't read link type: path too long!");
323 if (strstr(buf, "bpf-map"))
325 else if (strstr(buf, "bpf-prog"))
327 else if (strstr(buf, "bpf-link"))
330 return BPF_OBJ_UNKNOWN;
333 char *get_fdinfo(int fd, const char *key)
341 snprintf(path, sizeof(path), "/proc/self/fdinfo/%d", fd);
343 fdi = fopen(path, "r");
347 while ((n = getline(&line, &line_n, fdi)) > 0) {
351 if (!strstr(line, key))
356 value = strchr(line, '\t');
357 if (!value || !value[1]) {
364 memmove(line, value, len);
365 line[len - 1] = '\0';
375 void print_data_json(uint8_t *data, size_t len)
379 jsonw_start_array(json_wtr);
380 for (i = 0; i < len; i++)
381 jsonw_printf(json_wtr, "%d", data[i]);
382 jsonw_end_array(json_wtr);
385 void print_hex_data_json(uint8_t *data, size_t len)
389 jsonw_start_array(json_wtr);
390 for (i = 0; i < len; i++)
391 jsonw_printf(json_wtr, "\"0x%02hhx\"", data[i]);
392 jsonw_end_array(json_wtr);
395 /* extra params for nftw cb */
396 static struct pinned_obj_table *build_fn_table;
397 static enum bpf_obj_type build_fn_type;
399 static int do_build_table_cb(const char *fpath, const struct stat *sb,
400 int typeflag, struct FTW *ftwbuf)
402 struct bpf_prog_info pinned_info;
403 __u32 len = sizeof(pinned_info);
404 struct pinned_obj *obj_node;
405 enum bpf_obj_type objtype;
408 if (typeflag != FTW_F)
411 fd = open_obj_pinned(fpath, true);
415 objtype = get_fd_type(fd);
416 if (objtype != build_fn_type)
419 memset(&pinned_info, 0, sizeof(pinned_info));
420 if (bpf_obj_get_info_by_fd(fd, &pinned_info, &len))
423 obj_node = calloc(1, sizeof(*obj_node));
429 obj_node->id = pinned_info.id;
430 obj_node->path = strdup(fpath);
431 if (!obj_node->path) {
437 hash_add(build_fn_table->table, &obj_node->hash, obj_node->id);
444 int build_pinned_obj_table(struct pinned_obj_table *tab,
445 enum bpf_obj_type type)
447 struct mntent *mntent = NULL;
448 FILE *mntfile = NULL;
449 int flags = FTW_PHYS;
453 mntfile = setmntent("/proc/mounts", "r");
457 build_fn_table = tab;
458 build_fn_type = type;
460 while ((mntent = getmntent(mntfile))) {
461 char *path = mntent->mnt_dir;
463 if (strncmp(mntent->mnt_type, "bpf", 3) != 0)
465 err = nftw(path, do_build_table_cb, nopenfd, flags);
473 void delete_pinned_obj_table(struct pinned_obj_table *tab)
475 struct pinned_obj *obj;
476 struct hlist_node *tmp;
479 hash_for_each_safe(tab->table, bkt, tmp, obj, hash) {
480 hash_del(&obj->hash);
486 unsigned int get_page_size(void)
491 result = getpagesize();
495 unsigned int get_possible_cpus(void)
497 int cpus = libbpf_num_possible_cpus();
500 p_err("Can't get # of possible cpus: %s", strerror(-cpus));
507 ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf)
512 err = stat("/proc/self/ns/net", &st);
514 p_err("Can't stat /proc/self: %s", strerror(errno));
518 if (st.st_dev != ns_dev || st.st_ino != ns_ino)
521 return if_indextoname(ifindex, buf);
524 static int read_sysfs_hex_int(char *path)
526 char vendor_id_buf[8];
530 fd = open(path, O_RDONLY);
532 p_err("Can't open %s: %s", path, strerror(errno));
536 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf));
539 p_err("Can't read %s: %s", path, strerror(errno));
542 if (len >= (int)sizeof(vendor_id_buf)) {
543 p_err("Value in %s too long", path);
547 vendor_id_buf[len] = 0;
549 return strtol(vendor_id_buf, NULL, 0);
552 static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name)
556 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s",
557 devname, entry_name);
559 return read_sysfs_hex_int(full_path);
563 ifindex_to_bfd_params(__u32 ifindex, __u64 ns_dev, __u64 ns_ino,
566 char devname[IF_NAMESIZE];
570 if (!ifindex_to_name_ns(ifindex, ns_dev, ns_ino, devname)) {
571 p_err("Can't get net device name for ifindex %d: %s", ifindex,
576 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor");
578 p_err("Can't get device vendor id for %s", devname);
584 device_id = read_sysfs_netdev_hex_int(devname, "device");
585 if (device_id != 0x4000 &&
586 device_id != 0x6000 &&
588 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch");
592 p_err("Can't get bfd arch name for device vendor id 0x%04x",
598 void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
600 char name[IF_NAMESIZE];
605 printf(" offloaded_to ");
606 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
609 printf("ifindex %u ns_dev %llu ns_ino %llu",
610 ifindex, ns_dev, ns_inode);
613 void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
615 char name[IF_NAMESIZE];
620 jsonw_name(json_wtr, "dev");
621 jsonw_start_object(json_wtr);
622 jsonw_uint_field(json_wtr, "ifindex", ifindex);
623 jsonw_uint_field(json_wtr, "ns_dev", ns_dev);
624 jsonw_uint_field(json_wtr, "ns_inode", ns_inode);
625 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
626 jsonw_string_field(json_wtr, "ifname", name);
627 jsonw_end_object(json_wtr);
630 int parse_u32_arg(int *argc, char ***argv, __u32 *val, const char *what)
637 p_err("%s already specified", what);
641 *val = strtoul(**argv, &endptr, 0);
643 p_err("can't parse %s as %s", **argv, what);
652 print_all_levels(__maybe_unused enum libbpf_print_level level,
653 const char *format, va_list args)
655 return vfprintf(stderr, format, args);
658 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
666 struct bpf_prog_info info = {};
667 __u32 len = sizeof(info);
669 err = bpf_prog_get_next_id(id, &id);
671 if (errno != ENOENT) {
672 p_err("%s", strerror(errno));
678 fd = bpf_prog_get_fd_by_id(id);
680 p_err("can't get prog by id (%u): %s",
681 id, strerror(errno));
685 err = bpf_obj_get_info_by_fd(fd, &info, &len);
687 p_err("can't get prog info (%u): %s",
688 id, strerror(errno));
692 if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
693 (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
699 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
701 p_err("failed to realloc");
706 (*fds)[nb_fds++] = fd;
712 while (--nb_fds >= 0)
713 close((*fds)[nb_fds]);
717 int prog_parse_fds(int *argc, char ***argv, int **fds)
719 if (is_prefix(**argv, "id")) {
725 id = strtoul(**argv, &endptr, 0);
727 p_err("can't parse %s as ID", **argv);
732 (*fds)[0] = bpf_prog_get_fd_by_id(id);
734 p_err("get by id (%u): %s", id, strerror(errno));
738 } else if (is_prefix(**argv, "tag")) {
739 unsigned char tag[BPF_TAG_SIZE];
743 if (sscanf(**argv, BPF_TAG_FMT, tag, tag + 1, tag + 2,
744 tag + 3, tag + 4, tag + 5, tag + 6, tag + 7)
746 p_err("can't parse tag");
751 return prog_fd_by_nametag(tag, fds, true);
752 } else if (is_prefix(**argv, "name")) {
758 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
759 p_err("can't parse name");
764 return prog_fd_by_nametag(name, fds, false);
765 } else if (is_prefix(**argv, "pinned")) {
773 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_PROG);
779 p_err("expected 'id', 'tag', 'name' or 'pinned', got: '%s'?", **argv);
783 int prog_parse_fd(int *argc, char ***argv)
788 fds = malloc(sizeof(int));
790 p_err("mem alloc failed");
793 nb_fds = prog_parse_fds(argc, argv, &fds);
796 p_err("several programs match this handle");
810 static int map_fd_by_name(char *name, int **fds)
818 struct bpf_map_info info = {};
819 __u32 len = sizeof(info);
821 err = bpf_map_get_next_id(id, &id);
823 if (errno != ENOENT) {
824 p_err("%s", strerror(errno));
830 fd = bpf_map_get_fd_by_id(id);
832 p_err("can't get map by id (%u): %s",
833 id, strerror(errno));
837 err = bpf_obj_get_info_by_fd(fd, &info, &len);
839 p_err("can't get map info (%u): %s",
840 id, strerror(errno));
844 if (strncmp(name, info.name, BPF_OBJ_NAME_LEN)) {
850 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
852 p_err("failed to realloc");
857 (*fds)[nb_fds++] = fd;
863 while (--nb_fds >= 0)
864 close((*fds)[nb_fds]);
868 int map_parse_fds(int *argc, char ***argv, int **fds)
870 if (is_prefix(**argv, "id")) {
876 id = strtoul(**argv, &endptr, 0);
878 p_err("can't parse %s as ID", **argv);
883 (*fds)[0] = bpf_map_get_fd_by_id(id);
885 p_err("get map by id (%u): %s", id, strerror(errno));
889 } else if (is_prefix(**argv, "name")) {
895 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
896 p_err("can't parse name");
901 return map_fd_by_name(name, fds);
902 } else if (is_prefix(**argv, "pinned")) {
910 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_MAP);
916 p_err("expected 'id', 'name' or 'pinned', got: '%s'?", **argv);
920 int map_parse_fd(int *argc, char ***argv)
925 fds = malloc(sizeof(int));
927 p_err("mem alloc failed");
930 nb_fds = map_parse_fds(argc, argv, &fds);
933 p_err("several maps match this handle");
947 int map_parse_fd_and_info(int *argc, char ***argv, void *info, __u32 *info_len)
952 fd = map_parse_fd(argc, argv);
956 err = bpf_obj_get_info_by_fd(fd, info, info_len);
958 p_err("can't get map info: %s", strerror(errno));