| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * Minimal file system backend for holding eBPF maps and programs, |
| 4 | * used by bpf(2) object pinning. |
| 5 | * |
| 6 | * Authors: |
| 7 | * |
| 8 | * Daniel Borkmann <daniel@iogearbox.net> |
| 9 | */ |
| 10 | |
| 11 | #include <linux/init.h> |
| 12 | #include <linux/magic.h> |
| 13 | #include <linux/major.h> |
| 14 | #include <linux/mount.h> |
| 15 | #include <linux/namei.h> |
| 16 | #include <linux/fs.h> |
| 17 | #include <linux/fs_context.h> |
| 18 | #include <linux/fs_parser.h> |
| 19 | #include <linux/kdev_t.h> |
| 20 | #include <linux/filter.h> |
| 21 | #include <linux/bpf.h> |
| 22 | #include <linux/bpf_trace.h> |
| 23 | #include <linux/kstrtox.h> |
| 24 | #include "preload/bpf_preload.h" |
| 25 | |
| 26 | enum bpf_type { |
| 27 | BPF_TYPE_UNSPEC = 0, |
| 28 | BPF_TYPE_PROG, |
| 29 | BPF_TYPE_MAP, |
| 30 | BPF_TYPE_LINK, |
| 31 | }; |
| 32 | |
| 33 | static void *bpf_any_get(void *raw, enum bpf_type type) |
| 34 | { |
| 35 | switch (type) { |
| 36 | case BPF_TYPE_PROG: |
| 37 | bpf_prog_inc(prog: raw); |
| 38 | break; |
| 39 | case BPF_TYPE_MAP: |
| 40 | bpf_map_inc_with_uref(map: raw); |
| 41 | break; |
| 42 | case BPF_TYPE_LINK: |
| 43 | bpf_link_inc(link: raw); |
| 44 | break; |
| 45 | default: |
| 46 | WARN_ON_ONCE(1); |
| 47 | break; |
| 48 | } |
| 49 | |
| 50 | return raw; |
| 51 | } |
| 52 | |
| 53 | static void bpf_any_put(void *raw, enum bpf_type type) |
| 54 | { |
| 55 | switch (type) { |
| 56 | case BPF_TYPE_PROG: |
| 57 | bpf_prog_put(prog: raw); |
| 58 | break; |
| 59 | case BPF_TYPE_MAP: |
| 60 | bpf_map_put_with_uref(map: raw); |
| 61 | break; |
| 62 | case BPF_TYPE_LINK: |
| 63 | bpf_link_put(link: raw); |
| 64 | break; |
| 65 | default: |
| 66 | WARN_ON_ONCE(1); |
| 67 | break; |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type) |
| 72 | { |
| 73 | void *raw; |
| 74 | |
| 75 | raw = bpf_map_get_with_uref(ufd); |
| 76 | if (!IS_ERR(ptr: raw)) { |
| 77 | *type = BPF_TYPE_MAP; |
| 78 | return raw; |
| 79 | } |
| 80 | |
| 81 | raw = bpf_prog_get(ufd); |
| 82 | if (!IS_ERR(ptr: raw)) { |
| 83 | *type = BPF_TYPE_PROG; |
| 84 | return raw; |
| 85 | } |
| 86 | |
| 87 | raw = bpf_link_get_from_fd(ufd); |
| 88 | if (!IS_ERR(ptr: raw)) { |
| 89 | *type = BPF_TYPE_LINK; |
| 90 | return raw; |
| 91 | } |
| 92 | |
| 93 | return ERR_PTR(error: -EINVAL); |
| 94 | } |
| 95 | |
| 96 | static const struct inode_operations bpf_dir_iops; |
| 97 | |
| 98 | static const struct inode_operations bpf_prog_iops = { }; |
| 99 | static const struct inode_operations bpf_map_iops = { }; |
| 100 | static const struct inode_operations bpf_link_iops = { }; |
| 101 | |
| 102 | struct inode *bpf_get_inode(struct super_block *sb, |
| 103 | const struct inode *dir, |
| 104 | umode_t mode) |
| 105 | { |
| 106 | struct inode *inode; |
| 107 | |
| 108 | switch (mode & S_IFMT) { |
| 109 | case S_IFDIR: |
| 110 | case S_IFREG: |
| 111 | case S_IFLNK: |
| 112 | break; |
| 113 | default: |
| 114 | return ERR_PTR(error: -EINVAL); |
| 115 | } |
| 116 | |
| 117 | inode = new_inode(sb); |
| 118 | if (!inode) |
| 119 | return ERR_PTR(error: -ENOSPC); |
| 120 | |
| 121 | inode->i_ino = get_next_ino(); |
| 122 | simple_inode_init_ts(inode); |
| 123 | |
| 124 | inode_init_owner(idmap: &nop_mnt_idmap, inode, dir, mode); |
| 125 | |
| 126 | return inode; |
| 127 | } |
| 128 | |
| 129 | static int bpf_inode_type(const struct inode *inode, enum bpf_type *type) |
| 130 | { |
| 131 | *type = BPF_TYPE_UNSPEC; |
| 132 | if (inode->i_op == &bpf_prog_iops) |
| 133 | *type = BPF_TYPE_PROG; |
| 134 | else if (inode->i_op == &bpf_map_iops) |
| 135 | *type = BPF_TYPE_MAP; |
| 136 | else if (inode->i_op == &bpf_link_iops) |
| 137 | *type = BPF_TYPE_LINK; |
| 138 | else |
| 139 | return -EACCES; |
| 140 | |
| 141 | return 0; |
| 142 | } |
| 143 | |
| 144 | static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode, |
| 145 | struct inode *dir) |
| 146 | { |
| 147 | d_make_persistent(dentry, inode); |
| 148 | |
| 149 | inode_set_mtime_to_ts(inode: dir, ts: inode_set_ctime_current(inode: dir)); |
| 150 | } |
| 151 | |
| 152 | static struct dentry *bpf_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
| 153 | struct dentry *dentry, umode_t mode) |
| 154 | { |
| 155 | struct inode *inode; |
| 156 | |
| 157 | inode = bpf_get_inode(sb: dir->i_sb, dir, mode: mode | S_IFDIR); |
| 158 | if (IS_ERR(ptr: inode)) |
| 159 | return ERR_CAST(ptr: inode); |
| 160 | |
| 161 | inode->i_op = &bpf_dir_iops; |
| 162 | inode->i_fop = &simple_dir_operations; |
| 163 | |
| 164 | inc_nlink(inode); |
| 165 | inc_nlink(inode: dir); |
| 166 | |
| 167 | bpf_dentry_finalize(dentry, inode, dir); |
| 168 | return NULL; |
| 169 | } |
| 170 | |
| 171 | struct map_iter { |
| 172 | void *key; |
| 173 | bool done; |
| 174 | }; |
| 175 | |
| 176 | static struct map_iter *map_iter(struct seq_file *m) |
| 177 | { |
| 178 | return m->private; |
| 179 | } |
| 180 | |
| 181 | static struct bpf_map *seq_file_to_map(struct seq_file *m) |
| 182 | { |
| 183 | return file_inode(f: m->file)->i_private; |
| 184 | } |
| 185 | |
| 186 | static void map_iter_free(struct map_iter *iter) |
| 187 | { |
| 188 | if (iter) { |
| 189 | kfree(objp: iter->key); |
| 190 | kfree(objp: iter); |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | static struct map_iter *map_iter_alloc(struct bpf_map *map) |
| 195 | { |
| 196 | struct map_iter *iter; |
| 197 | |
| 198 | iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN); |
| 199 | if (!iter) |
| 200 | goto error; |
| 201 | |
| 202 | iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN); |
| 203 | if (!iter->key) |
| 204 | goto error; |
| 205 | |
| 206 | return iter; |
| 207 | |
| 208 | error: |
| 209 | map_iter_free(iter); |
| 210 | return NULL; |
| 211 | } |
| 212 | |
| 213 | static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos) |
| 214 | { |
| 215 | struct bpf_map *map = seq_file_to_map(m); |
| 216 | void *key = map_iter(m)->key; |
| 217 | void *prev_key; |
| 218 | |
| 219 | (*pos)++; |
| 220 | if (map_iter(m)->done) |
| 221 | return NULL; |
| 222 | |
| 223 | if (unlikely(v == SEQ_START_TOKEN)) |
| 224 | prev_key = NULL; |
| 225 | else |
| 226 | prev_key = key; |
| 227 | |
| 228 | rcu_read_lock(); |
| 229 | if (map->ops->map_get_next_key(map, prev_key, key)) { |
| 230 | map_iter(m)->done = true; |
| 231 | key = NULL; |
| 232 | } |
| 233 | rcu_read_unlock(); |
| 234 | return key; |
| 235 | } |
| 236 | |
| 237 | static void *map_seq_start(struct seq_file *m, loff_t *pos) |
| 238 | { |
| 239 | if (map_iter(m)->done) |
| 240 | return NULL; |
| 241 | |
| 242 | return *pos ? map_iter(m)->key : SEQ_START_TOKEN; |
| 243 | } |
| 244 | |
| 245 | static void map_seq_stop(struct seq_file *m, void *v) |
| 246 | { |
| 247 | } |
| 248 | |
| 249 | static int map_seq_show(struct seq_file *m, void *v) |
| 250 | { |
| 251 | struct bpf_map *map = seq_file_to_map(m); |
| 252 | void *key = map_iter(m)->key; |
| 253 | |
| 254 | if (unlikely(v == SEQ_START_TOKEN)) { |
| 255 | seq_puts(m, s: "# WARNING!! The output is for debug purpose only\n" ); |
| 256 | seq_puts(m, s: "# WARNING!! The output format will change\n" ); |
| 257 | } else { |
| 258 | map->ops->map_seq_show_elem(map, key, m); |
| 259 | } |
| 260 | |
| 261 | return 0; |
| 262 | } |
| 263 | |
| 264 | static const struct seq_operations bpffs_map_seq_ops = { |
| 265 | .start = map_seq_start, |
| 266 | .next = map_seq_next, |
| 267 | .show = map_seq_show, |
| 268 | .stop = map_seq_stop, |
| 269 | }; |
| 270 | |
| 271 | static int bpffs_map_open(struct inode *inode, struct file *file) |
| 272 | { |
| 273 | struct bpf_map *map = inode->i_private; |
| 274 | struct map_iter *iter; |
| 275 | struct seq_file *m; |
| 276 | int err; |
| 277 | |
| 278 | iter = map_iter_alloc(map); |
| 279 | if (!iter) |
| 280 | return -ENOMEM; |
| 281 | |
| 282 | err = seq_open(file, &bpffs_map_seq_ops); |
| 283 | if (err) { |
| 284 | map_iter_free(iter); |
| 285 | return err; |
| 286 | } |
| 287 | |
| 288 | m = file->private_data; |
| 289 | m->private = iter; |
| 290 | |
| 291 | return 0; |
| 292 | } |
| 293 | |
| 294 | static int bpffs_map_release(struct inode *inode, struct file *file) |
| 295 | { |
| 296 | struct seq_file *m = file->private_data; |
| 297 | |
| 298 | map_iter_free(iter: map_iter(m)); |
| 299 | |
| 300 | return seq_release(inode, file); |
| 301 | } |
| 302 | |
| 303 | /* bpffs_map_fops should only implement the basic |
| 304 | * read operation for a BPF map. The purpose is to |
| 305 | * provide a simple user intuitive way to do |
| 306 | * "cat bpffs/pathto/a-pinned-map". |
| 307 | * |
| 308 | * Other operations (e.g. write, lookup...) should be realized by |
| 309 | * the userspace tools (e.g. bpftool) through the |
| 310 | * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update |
| 311 | * interface. |
| 312 | */ |
| 313 | static const struct file_operations bpffs_map_fops = { |
| 314 | .open = bpffs_map_open, |
| 315 | .read = seq_read, |
| 316 | .release = bpffs_map_release, |
| 317 | }; |
| 318 | |
| 319 | static int bpffs_obj_open(struct inode *inode, struct file *file) |
| 320 | { |
| 321 | return -EIO; |
| 322 | } |
| 323 | |
| 324 | static const struct file_operations bpffs_obj_fops = { |
| 325 | .open = bpffs_obj_open, |
| 326 | }; |
| 327 | |
| 328 | static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw, |
| 329 | const struct inode_operations *iops, |
| 330 | const struct file_operations *fops) |
| 331 | { |
| 332 | struct inode *dir = dentry->d_parent->d_inode; |
| 333 | struct inode *inode = bpf_get_inode(sb: dir->i_sb, dir, mode); |
| 334 | if (IS_ERR(ptr: inode)) |
| 335 | return PTR_ERR(ptr: inode); |
| 336 | |
| 337 | inode->i_op = iops; |
| 338 | inode->i_fop = fops; |
| 339 | inode->i_private = raw; |
| 340 | |
| 341 | bpf_dentry_finalize(dentry, inode, dir); |
| 342 | return 0; |
| 343 | } |
| 344 | |
| 345 | static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg) |
| 346 | { |
| 347 | return bpf_mkobj_ops(dentry, mode, raw: arg, iops: &bpf_prog_iops, |
| 348 | fops: &bpffs_obj_fops); |
| 349 | } |
| 350 | |
| 351 | static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg) |
| 352 | { |
| 353 | struct bpf_map *map = arg; |
| 354 | |
| 355 | return bpf_mkobj_ops(dentry, mode, raw: arg, iops: &bpf_map_iops, |
| 356 | fops: bpf_map_support_seq_show(map) ? |
| 357 | &bpffs_map_fops : &bpffs_obj_fops); |
| 358 | } |
| 359 | |
| 360 | static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg) |
| 361 | { |
| 362 | struct bpf_link *link = arg; |
| 363 | |
| 364 | return bpf_mkobj_ops(dentry, mode, raw: arg, iops: &bpf_link_iops, |
| 365 | fops: bpf_link_is_iter(link) ? |
| 366 | &bpf_iter_fops : &bpffs_obj_fops); |
| 367 | } |
| 368 | |
| 369 | static struct dentry * |
| 370 | bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags) |
| 371 | { |
| 372 | /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future |
| 373 | * extensions. That allows popoulate_bpffs() create special files. |
| 374 | */ |
| 375 | if ((dir->i_mode & S_IALLUGO) && |
| 376 | strchr(dentry->d_name.name, '.')) |
| 377 | return ERR_PTR(error: -EPERM); |
| 378 | |
| 379 | return simple_lookup(dir, dentry, flags); |
| 380 | } |
| 381 | |
| 382 | static int bpf_symlink(struct mnt_idmap *idmap, struct inode *dir, |
| 383 | struct dentry *dentry, const char *target) |
| 384 | { |
| 385 | char *link = kstrdup(s: target, GFP_USER | __GFP_NOWARN); |
| 386 | struct inode *inode; |
| 387 | |
| 388 | if (!link) |
| 389 | return -ENOMEM; |
| 390 | |
| 391 | inode = bpf_get_inode(sb: dir->i_sb, dir, S_IRWXUGO | S_IFLNK); |
| 392 | if (IS_ERR(ptr: inode)) { |
| 393 | kfree(objp: link); |
| 394 | return PTR_ERR(ptr: inode); |
| 395 | } |
| 396 | |
| 397 | inode->i_op = &simple_symlink_inode_operations; |
| 398 | inode->i_link = link; |
| 399 | |
| 400 | bpf_dentry_finalize(dentry, inode, dir); |
| 401 | return 0; |
| 402 | } |
| 403 | |
| 404 | static const struct inode_operations bpf_dir_iops = { |
| 405 | .lookup = bpf_lookup, |
| 406 | .mkdir = bpf_mkdir, |
| 407 | .symlink = bpf_symlink, |
| 408 | .rmdir = simple_rmdir, |
| 409 | .rename = simple_rename, |
| 410 | .link = simple_link, |
| 411 | .unlink = simple_unlink, |
| 412 | }; |
| 413 | |
| 414 | /* pin iterator link into bpffs */ |
| 415 | static int bpf_iter_link_pin_kernel(struct dentry *parent, |
| 416 | const char *name, struct bpf_link *link) |
| 417 | { |
| 418 | umode_t mode = S_IFREG | S_IRUSR; |
| 419 | struct dentry *dentry; |
| 420 | int ret; |
| 421 | |
| 422 | dentry = simple_start_creating(parent, name); |
| 423 | if (IS_ERR(ptr: dentry)) |
| 424 | return PTR_ERR(ptr: dentry); |
| 425 | ret = bpf_mkobj_ops(dentry, mode, raw: link, iops: &bpf_link_iops, |
| 426 | fops: &bpf_iter_fops); |
| 427 | simple_done_creating(dentry); |
| 428 | return ret; |
| 429 | } |
| 430 | |
| 431 | static int bpf_obj_do_pin(int path_fd, const char __user *pathname, void *raw, |
| 432 | enum bpf_type type) |
| 433 | { |
| 434 | struct dentry *dentry; |
| 435 | struct inode *dir; |
| 436 | struct path path; |
| 437 | umode_t mode; |
| 438 | int ret; |
| 439 | |
| 440 | dentry = start_creating_user_path(path_fd, pathname, &path, 0); |
| 441 | if (IS_ERR(ptr: dentry)) |
| 442 | return PTR_ERR(ptr: dentry); |
| 443 | |
| 444 | dir = d_inode(dentry: path.dentry); |
| 445 | if (dir->i_op != &bpf_dir_iops) { |
| 446 | ret = -EPERM; |
| 447 | goto out; |
| 448 | } |
| 449 | |
| 450 | mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask()); |
| 451 | ret = security_path_mknod(dir: &path, dentry, mode, dev: 0); |
| 452 | if (ret) |
| 453 | goto out; |
| 454 | |
| 455 | switch (type) { |
| 456 | case BPF_TYPE_PROG: |
| 457 | ret = vfs_mkobj(dentry, mode, f: bpf_mkprog, raw); |
| 458 | break; |
| 459 | case BPF_TYPE_MAP: |
| 460 | ret = vfs_mkobj(dentry, mode, f: bpf_mkmap, raw); |
| 461 | break; |
| 462 | case BPF_TYPE_LINK: |
| 463 | ret = vfs_mkobj(dentry, mode, f: bpf_mklink, raw); |
| 464 | break; |
| 465 | default: |
| 466 | ret = -EPERM; |
| 467 | } |
| 468 | out: |
| 469 | end_creating_path(&path, dentry); |
| 470 | return ret; |
| 471 | } |
| 472 | |
| 473 | int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname) |
| 474 | { |
| 475 | enum bpf_type type; |
| 476 | void *raw; |
| 477 | int ret; |
| 478 | |
| 479 | raw = bpf_fd_probe_obj(ufd, type: &type); |
| 480 | if (IS_ERR(ptr: raw)) |
| 481 | return PTR_ERR(ptr: raw); |
| 482 | |
| 483 | ret = bpf_obj_do_pin(path_fd, pathname, raw, type); |
| 484 | if (ret != 0) |
| 485 | bpf_any_put(raw, type); |
| 486 | |
| 487 | return ret; |
| 488 | } |
| 489 | |
| 490 | static void *bpf_obj_do_get(int path_fd, const char __user *pathname, |
| 491 | enum bpf_type *type, int flags) |
| 492 | { |
| 493 | struct inode *inode; |
| 494 | struct path path; |
| 495 | void *raw; |
| 496 | int ret; |
| 497 | |
| 498 | ret = user_path_at(path_fd, pathname, LOOKUP_FOLLOW, &path); |
| 499 | if (ret) |
| 500 | return ERR_PTR(error: ret); |
| 501 | |
| 502 | inode = d_backing_inode(upper: path.dentry); |
| 503 | ret = path_permission(path: &path, ACC_MODE(flags)); |
| 504 | if (ret) |
| 505 | goto out; |
| 506 | |
| 507 | ret = bpf_inode_type(inode, type); |
| 508 | if (ret) |
| 509 | goto out; |
| 510 | |
| 511 | raw = bpf_any_get(raw: inode->i_private, type: *type); |
| 512 | if (!IS_ERR(ptr: raw)) |
| 513 | touch_atime(&path); |
| 514 | |
| 515 | path_put(&path); |
| 516 | return raw; |
| 517 | out: |
| 518 | path_put(&path); |
| 519 | return ERR_PTR(error: ret); |
| 520 | } |
| 521 | |
| 522 | int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags) |
| 523 | { |
| 524 | enum bpf_type type = BPF_TYPE_UNSPEC; |
| 525 | int f_flags; |
| 526 | void *raw; |
| 527 | int ret; |
| 528 | |
| 529 | f_flags = bpf_get_file_flag(flags); |
| 530 | if (f_flags < 0) |
| 531 | return f_flags; |
| 532 | |
| 533 | raw = bpf_obj_do_get(path_fd, pathname, type: &type, flags: f_flags); |
| 534 | if (IS_ERR(ptr: raw)) |
| 535 | return PTR_ERR(ptr: raw); |
| 536 | |
| 537 | if (type == BPF_TYPE_PROG) |
| 538 | ret = bpf_prog_new_fd(prog: raw); |
| 539 | else if (type == BPF_TYPE_MAP) |
| 540 | ret = bpf_map_new_fd(map: raw, flags: f_flags); |
| 541 | else if (type == BPF_TYPE_LINK) |
| 542 | ret = (f_flags != O_RDWR) ? -EINVAL : bpf_link_new_fd(link: raw); |
| 543 | else |
| 544 | return -ENOENT; |
| 545 | |
| 546 | if (ret < 0) |
| 547 | bpf_any_put(raw, type); |
| 548 | return ret; |
| 549 | } |
| 550 | |
| 551 | static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type) |
| 552 | { |
| 553 | struct bpf_prog *prog; |
| 554 | int ret = inode_permission(&nop_mnt_idmap, inode, MAY_READ); |
| 555 | if (ret) |
| 556 | return ERR_PTR(error: ret); |
| 557 | |
| 558 | if (inode->i_op == &bpf_map_iops) |
| 559 | return ERR_PTR(error: -EINVAL); |
| 560 | if (inode->i_op == &bpf_link_iops) |
| 561 | return ERR_PTR(error: -EINVAL); |
| 562 | if (inode->i_op != &bpf_prog_iops) |
| 563 | return ERR_PTR(error: -EACCES); |
| 564 | |
| 565 | prog = inode->i_private; |
| 566 | |
| 567 | ret = security_bpf_prog(prog); |
| 568 | if (ret < 0) |
| 569 | return ERR_PTR(error: ret); |
| 570 | |
| 571 | if (!bpf_prog_get_ok(prog, &type, false)) |
| 572 | return ERR_PTR(error: -EINVAL); |
| 573 | |
| 574 | bpf_prog_inc(prog); |
| 575 | return prog; |
| 576 | } |
| 577 | |
| 578 | struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type) |
| 579 | { |
| 580 | struct bpf_prog *prog; |
| 581 | struct path path; |
| 582 | int ret = kern_path(name, LOOKUP_FOLLOW, &path); |
| 583 | if (ret) |
| 584 | return ERR_PTR(error: ret); |
| 585 | prog = __get_prog_inode(inode: d_backing_inode(upper: path.dentry), type); |
| 586 | if (!IS_ERR(ptr: prog)) |
| 587 | touch_atime(&path); |
| 588 | path_put(&path); |
| 589 | return prog; |
| 590 | } |
| 591 | EXPORT_SYMBOL(bpf_prog_get_type_path); |
| 592 | |
| 593 | struct bpffs_btf_enums { |
| 594 | const struct btf *btf; |
| 595 | const struct btf_type *cmd_t; |
| 596 | const struct btf_type *map_t; |
| 597 | const struct btf_type *prog_t; |
| 598 | const struct btf_type *attach_t; |
| 599 | }; |
| 600 | |
| 601 | static int find_bpffs_btf_enums(struct bpffs_btf_enums *info) |
| 602 | { |
| 603 | const struct btf *btf; |
| 604 | const struct btf_type *t; |
| 605 | const char *name; |
| 606 | int i, n; |
| 607 | |
| 608 | memset(info, 0, sizeof(*info)); |
| 609 | |
| 610 | btf = bpf_get_btf_vmlinux(); |
| 611 | if (IS_ERR(ptr: btf)) |
| 612 | return PTR_ERR(ptr: btf); |
| 613 | if (!btf) |
| 614 | return -ENOENT; |
| 615 | |
| 616 | info->btf = btf; |
| 617 | |
| 618 | for (i = 1, n = btf_nr_types(btf); i < n; i++) { |
| 619 | t = btf_type_by_id(btf, type_id: i); |
| 620 | if (!btf_type_is_enum(t)) |
| 621 | continue; |
| 622 | |
| 623 | name = btf_name_by_offset(btf, offset: t->name_off); |
| 624 | if (!name) |
| 625 | continue; |
| 626 | |
| 627 | if (strcmp(name, "bpf_cmd" ) == 0) |
| 628 | info->cmd_t = t; |
| 629 | else if (strcmp(name, "bpf_map_type" ) == 0) |
| 630 | info->map_t = t; |
| 631 | else if (strcmp(name, "bpf_prog_type" ) == 0) |
| 632 | info->prog_t = t; |
| 633 | else if (strcmp(name, "bpf_attach_type" ) == 0) |
| 634 | info->attach_t = t; |
| 635 | else |
| 636 | continue; |
| 637 | |
| 638 | if (info->cmd_t && info->map_t && info->prog_t && info->attach_t) |
| 639 | return 0; |
| 640 | } |
| 641 | |
| 642 | return -ESRCH; |
| 643 | } |
| 644 | |
| 645 | static bool find_btf_enum_const(const struct btf *btf, const struct btf_type *enum_t, |
| 646 | const char *prefix, const char *str, int *value) |
| 647 | { |
| 648 | const struct btf_enum *e; |
| 649 | const char *name; |
| 650 | int i, n, pfx_len = strlen(prefix); |
| 651 | |
| 652 | *value = 0; |
| 653 | |
| 654 | if (!btf || !enum_t) |
| 655 | return false; |
| 656 | |
| 657 | for (i = 0, n = btf_vlen(t: enum_t); i < n; i++) { |
| 658 | e = &btf_enum(t: enum_t)[i]; |
| 659 | |
| 660 | name = btf_name_by_offset(btf, offset: e->name_off); |
| 661 | if (!name || strncasecmp(s1: name, s2: prefix, n: pfx_len) != 0) |
| 662 | continue; |
| 663 | |
| 664 | /* match symbolic name case insensitive and ignoring prefix */ |
| 665 | if (strcasecmp(s1: name + pfx_len, s2: str) == 0) { |
| 666 | *value = e->val; |
| 667 | return true; |
| 668 | } |
| 669 | } |
| 670 | |
| 671 | return false; |
| 672 | } |
| 673 | |
| 674 | static void seq_print_delegate_opts(struct seq_file *m, |
| 675 | const char *opt_name, |
| 676 | const struct btf *btf, |
| 677 | const struct btf_type *enum_t, |
| 678 | const char *prefix, |
| 679 | u64 delegate_msk, u64 any_msk) |
| 680 | { |
| 681 | const struct btf_enum *e; |
| 682 | bool first = true; |
| 683 | const char *name; |
| 684 | u64 msk; |
| 685 | int i, n, pfx_len = strlen(prefix); |
| 686 | |
| 687 | delegate_msk &= any_msk; /* clear unknown bits */ |
| 688 | |
| 689 | if (delegate_msk == 0) |
| 690 | return; |
| 691 | |
| 692 | seq_printf(m, fmt: ",%s" , opt_name); |
| 693 | if (delegate_msk == any_msk) { |
| 694 | seq_printf(m, fmt: "=any" ); |
| 695 | return; |
| 696 | } |
| 697 | |
| 698 | if (btf && enum_t) { |
| 699 | for (i = 0, n = btf_vlen(t: enum_t); i < n; i++) { |
| 700 | e = &btf_enum(t: enum_t)[i]; |
| 701 | name = btf_name_by_offset(btf, offset: e->name_off); |
| 702 | if (!name || strncasecmp(s1: name, s2: prefix, n: pfx_len) != 0) |
| 703 | continue; |
| 704 | msk = 1ULL << e->val; |
| 705 | if (delegate_msk & msk) { |
| 706 | /* emit lower-case name without prefix */ |
| 707 | seq_putc(m, c: first ? '=' : ':'); |
| 708 | name += pfx_len; |
| 709 | while (*name) { |
| 710 | seq_putc(m, tolower(*name)); |
| 711 | name++; |
| 712 | } |
| 713 | |
| 714 | delegate_msk &= ~msk; |
| 715 | first = false; |
| 716 | } |
| 717 | } |
| 718 | } |
| 719 | if (delegate_msk) |
| 720 | seq_printf(m, fmt: "%c0x%llx" , first ? '=' : ':', delegate_msk); |
| 721 | } |
| 722 | |
| 723 | /* |
| 724 | * Display the mount options in /proc/mounts. |
| 725 | */ |
| 726 | static int bpf_show_options(struct seq_file *m, struct dentry *root) |
| 727 | { |
| 728 | struct inode *inode = d_inode(dentry: root); |
| 729 | umode_t mode = inode->i_mode & S_IALLUGO & ~S_ISVTX; |
| 730 | struct bpf_mount_opts *opts = root->d_sb->s_fs_info; |
| 731 | u64 mask; |
| 732 | |
| 733 | if (!uid_eq(left: inode->i_uid, GLOBAL_ROOT_UID)) |
| 734 | seq_printf(m, ",uid=%u" , |
| 735 | from_kuid_munged(&init_user_ns, inode->i_uid)); |
| 736 | if (!gid_eq(inode->i_gid, GLOBAL_ROOT_GID)) |
| 737 | seq_printf(m, ",gid=%u" , |
| 738 | from_kgid_munged(&init_user_ns, inode->i_gid)); |
| 739 | if (mode != S_IRWXUGO) |
| 740 | seq_printf(m, ",mode=%o" , mode); |
| 741 | |
| 742 | if (opts->delegate_cmds || opts->delegate_maps || |
| 743 | opts->delegate_progs || opts->delegate_attachs) { |
| 744 | struct bpffs_btf_enums info; |
| 745 | |
| 746 | /* ignore errors, fallback to hex */ |
| 747 | (void)find_bpffs_btf_enums(&info); |
| 748 | |
| 749 | mask = (1ULL << __MAX_BPF_CMD) - 1; |
| 750 | seq_print_delegate_opts(m, "delegate_cmds" , |
| 751 | info.btf, info.cmd_t, "BPF_" , |
| 752 | opts->delegate_cmds, mask); |
| 753 | |
| 754 | mask = (1ULL << __MAX_BPF_MAP_TYPE) - 1; |
| 755 | seq_print_delegate_opts(m, "delegate_maps" , |
| 756 | info.btf, info.map_t, "BPF_MAP_TYPE_" , |
| 757 | opts->delegate_maps, mask); |
| 758 | |
| 759 | mask = (1ULL << __MAX_BPF_PROG_TYPE) - 1; |
| 760 | seq_print_delegate_opts(m, "delegate_progs" , |
| 761 | info.btf, info.prog_t, "BPF_PROG_TYPE_" , |
| 762 | opts->delegate_progs, mask); |
| 763 | |
| 764 | mask = (1ULL << __MAX_BPF_ATTACH_TYPE) - 1; |
| 765 | seq_print_delegate_opts(m, "delegate_attachs" , |
| 766 | info.btf, info.attach_t, "BPF_" , |
| 767 | opts->delegate_attachs, mask); |
| 768 | } |
| 769 | |
| 770 | return 0; |
| 771 | } |
| 772 | |
| 773 | static void bpf_destroy_inode(struct inode *inode) |
| 774 | { |
| 775 | enum bpf_type type; |
| 776 | |
| 777 | if (S_ISLNK(inode->i_mode)) |
| 778 | kfree(objp: inode->i_link); |
| 779 | if (!bpf_inode_type(inode, type: &type)) |
| 780 | bpf_any_put(raw: inode->i_private, type); |
| 781 | free_inode_nonrcu(inode); |
| 782 | } |
| 783 | |
| 784 | const struct super_operations bpf_super_ops = { |
| 785 | .statfs = simple_statfs, |
| 786 | .drop_inode = inode_just_drop, |
| 787 | .show_options = bpf_show_options, |
| 788 | .destroy_inode = bpf_destroy_inode, |
| 789 | }; |
| 790 | |
| 791 | enum { |
| 792 | OPT_UID, |
| 793 | OPT_GID, |
| 794 | OPT_MODE, |
| 795 | OPT_DELEGATE_CMDS, |
| 796 | OPT_DELEGATE_MAPS, |
| 797 | OPT_DELEGATE_PROGS, |
| 798 | OPT_DELEGATE_ATTACHS, |
| 799 | }; |
| 800 | |
| 801 | static const struct fs_parameter_spec bpf_fs_parameters[] = { |
| 802 | fsparam_u32 ("uid" , OPT_UID), |
| 803 | fsparam_u32 ("gid" , OPT_GID), |
| 804 | fsparam_u32oct ("mode" , OPT_MODE), |
| 805 | fsparam_string ("delegate_cmds" , OPT_DELEGATE_CMDS), |
| 806 | fsparam_string ("delegate_maps" , OPT_DELEGATE_MAPS), |
| 807 | fsparam_string ("delegate_progs" , OPT_DELEGATE_PROGS), |
| 808 | fsparam_string ("delegate_attachs" , OPT_DELEGATE_ATTACHS), |
| 809 | {} |
| 810 | }; |
| 811 | |
| 812 | static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param) |
| 813 | { |
| 814 | struct bpf_mount_opts *opts = fc->s_fs_info; |
| 815 | struct fs_parse_result result; |
| 816 | kuid_t uid; |
| 817 | kgid_t gid; |
| 818 | int opt, err; |
| 819 | |
| 820 | opt = fs_parse(fc, desc: bpf_fs_parameters, param, result: &result); |
| 821 | if (opt < 0) { |
| 822 | /* We might like to report bad mount options here, but |
| 823 | * traditionally we've ignored all mount options, so we'd |
| 824 | * better continue to ignore non-existing options for bpf. |
| 825 | */ |
| 826 | if (opt == -ENOPARAM) { |
| 827 | opt = vfs_parse_fs_param_source(fc, param); |
| 828 | if (opt != -ENOPARAM) |
| 829 | return opt; |
| 830 | |
| 831 | return 0; |
| 832 | } |
| 833 | |
| 834 | if (opt < 0) |
| 835 | return opt; |
| 836 | } |
| 837 | |
| 838 | switch (opt) { |
| 839 | case OPT_UID: |
| 840 | uid = make_kuid(current_user_ns(), uid: result.uint_32); |
| 841 | if (!uid_valid(uid)) |
| 842 | goto bad_value; |
| 843 | |
| 844 | /* |
| 845 | * The requested uid must be representable in the |
| 846 | * filesystem's idmapping. |
| 847 | */ |
| 848 | if (!kuid_has_mapping(ns: fc->user_ns, uid)) |
| 849 | goto bad_value; |
| 850 | |
| 851 | opts->uid = uid; |
| 852 | break; |
| 853 | case OPT_GID: |
| 854 | gid = make_kgid(current_user_ns(), gid: result.uint_32); |
| 855 | if (!gid_valid(gid)) |
| 856 | goto bad_value; |
| 857 | |
| 858 | /* |
| 859 | * The requested gid must be representable in the |
| 860 | * filesystem's idmapping. |
| 861 | */ |
| 862 | if (!kgid_has_mapping(ns: fc->user_ns, gid)) |
| 863 | goto bad_value; |
| 864 | |
| 865 | opts->gid = gid; |
| 866 | break; |
| 867 | case OPT_MODE: |
| 868 | opts->mode = result.uint_32 & S_IALLUGO; |
| 869 | break; |
| 870 | case OPT_DELEGATE_CMDS: |
| 871 | case OPT_DELEGATE_MAPS: |
| 872 | case OPT_DELEGATE_PROGS: |
| 873 | case OPT_DELEGATE_ATTACHS: { |
| 874 | struct bpffs_btf_enums info; |
| 875 | const struct btf_type *enum_t; |
| 876 | const char *enum_pfx; |
| 877 | u64 *delegate_msk, msk = 0; |
| 878 | char *p, *str; |
| 879 | int val; |
| 880 | |
| 881 | /* ignore errors, fallback to hex */ |
| 882 | (void)find_bpffs_btf_enums(info: &info); |
| 883 | |
| 884 | switch (opt) { |
| 885 | case OPT_DELEGATE_CMDS: |
| 886 | delegate_msk = &opts->delegate_cmds; |
| 887 | enum_t = info.cmd_t; |
| 888 | enum_pfx = "BPF_" ; |
| 889 | break; |
| 890 | case OPT_DELEGATE_MAPS: |
| 891 | delegate_msk = &opts->delegate_maps; |
| 892 | enum_t = info.map_t; |
| 893 | enum_pfx = "BPF_MAP_TYPE_" ; |
| 894 | break; |
| 895 | case OPT_DELEGATE_PROGS: |
| 896 | delegate_msk = &opts->delegate_progs; |
| 897 | enum_t = info.prog_t; |
| 898 | enum_pfx = "BPF_PROG_TYPE_" ; |
| 899 | break; |
| 900 | case OPT_DELEGATE_ATTACHS: |
| 901 | delegate_msk = &opts->delegate_attachs; |
| 902 | enum_t = info.attach_t; |
| 903 | enum_pfx = "BPF_" ; |
| 904 | break; |
| 905 | default: |
| 906 | return -EINVAL; |
| 907 | } |
| 908 | |
| 909 | str = param->string; |
| 910 | while ((p = strsep(&str, ":" ))) { |
| 911 | if (strcmp(p, "any" ) == 0) { |
| 912 | msk |= ~0ULL; |
| 913 | } else if (find_btf_enum_const(btf: info.btf, enum_t, prefix: enum_pfx, str: p, value: &val)) { |
| 914 | msk |= 1ULL << val; |
| 915 | } else { |
| 916 | err = kstrtou64(s: p, base: 0, res: &msk); |
| 917 | if (err) |
| 918 | return err; |
| 919 | } |
| 920 | } |
| 921 | |
| 922 | /* Setting delegation mount options requires privileges */ |
| 923 | if (msk && !capable(CAP_SYS_ADMIN)) |
| 924 | return -EPERM; |
| 925 | |
| 926 | *delegate_msk |= msk; |
| 927 | break; |
| 928 | } |
| 929 | default: |
| 930 | /* ignore unknown mount options */ |
| 931 | break; |
| 932 | } |
| 933 | |
| 934 | return 0; |
| 935 | bad_value: |
| 936 | return invalfc(fc, "Bad value for '%s'" , param->key); |
| 937 | } |
| 938 | |
| 939 | struct bpf_preload_ops *bpf_preload_ops; |
| 940 | EXPORT_SYMBOL_GPL(bpf_preload_ops); |
| 941 | |
| 942 | static bool bpf_preload_mod_get(void) |
| 943 | { |
| 944 | /* If bpf_preload.ko wasn't loaded earlier then load it now. |
| 945 | * When bpf_preload is built into vmlinux the module's __init |
| 946 | * function will populate it. |
| 947 | */ |
| 948 | if (!bpf_preload_ops) { |
| 949 | request_module("bpf_preload" ); |
| 950 | if (!bpf_preload_ops) |
| 951 | return false; |
| 952 | } |
| 953 | /* And grab the reference, so the module doesn't disappear while the |
| 954 | * kernel is interacting with the kernel module and its UMD. |
| 955 | */ |
| 956 | if (!try_module_get(module: bpf_preload_ops->owner)) { |
| 957 | pr_err("bpf_preload module get failed.\n" ); |
| 958 | return false; |
| 959 | } |
| 960 | return true; |
| 961 | } |
| 962 | |
| 963 | static void bpf_preload_mod_put(void) |
| 964 | { |
| 965 | if (bpf_preload_ops) |
| 966 | /* now user can "rmmod bpf_preload" if necessary */ |
| 967 | module_put(module: bpf_preload_ops->owner); |
| 968 | } |
| 969 | |
| 970 | static DEFINE_MUTEX(bpf_preload_lock); |
| 971 | |
| 972 | static int populate_bpffs(struct dentry *parent) |
| 973 | { |
| 974 | struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {}; |
| 975 | int err = 0, i; |
| 976 | |
| 977 | /* grab the mutex to make sure the kernel interactions with bpf_preload |
| 978 | * are serialized |
| 979 | */ |
| 980 | mutex_lock(&bpf_preload_lock); |
| 981 | |
| 982 | /* if bpf_preload.ko wasn't built into vmlinux then load it */ |
| 983 | if (!bpf_preload_mod_get()) |
| 984 | goto out; |
| 985 | |
| 986 | err = bpf_preload_ops->preload(objs); |
| 987 | if (err) |
| 988 | goto out_put; |
| 989 | for (i = 0; i < BPF_PRELOAD_LINKS; i++) { |
| 990 | bpf_link_inc(link: objs[i].link); |
| 991 | err = bpf_iter_link_pin_kernel(parent, |
| 992 | name: objs[i].link_name, link: objs[i].link); |
| 993 | if (err) { |
| 994 | bpf_link_put(link: objs[i].link); |
| 995 | goto out_put; |
| 996 | } |
| 997 | } |
| 998 | out_put: |
| 999 | bpf_preload_mod_put(); |
| 1000 | out: |
| 1001 | mutex_unlock(lock: &bpf_preload_lock); |
| 1002 | return err; |
| 1003 | } |
| 1004 | |
| 1005 | static int bpf_fill_super(struct super_block *sb, struct fs_context *fc) |
| 1006 | { |
| 1007 | static const struct tree_descr bpf_rfiles[] = { { "" } }; |
| 1008 | struct bpf_mount_opts *opts = sb->s_fs_info; |
| 1009 | struct inode *inode; |
| 1010 | int ret; |
| 1011 | |
| 1012 | /* Mounting an instance of BPF FS requires privileges */ |
| 1013 | if (fc->user_ns != &init_user_ns && !capable(CAP_SYS_ADMIN)) |
| 1014 | return -EPERM; |
| 1015 | |
| 1016 | ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles); |
| 1017 | if (ret) |
| 1018 | return ret; |
| 1019 | |
| 1020 | sb->s_op = &bpf_super_ops; |
| 1021 | |
| 1022 | inode = sb->s_root->d_inode; |
| 1023 | inode->i_uid = opts->uid; |
| 1024 | inode->i_gid = opts->gid; |
| 1025 | inode->i_op = &bpf_dir_iops; |
| 1026 | inode->i_mode &= ~S_IALLUGO; |
| 1027 | populate_bpffs(parent: sb->s_root); |
| 1028 | inode->i_mode |= S_ISVTX | opts->mode; |
| 1029 | return 0; |
| 1030 | } |
| 1031 | |
| 1032 | static int bpf_get_tree(struct fs_context *fc) |
| 1033 | { |
| 1034 | return get_tree_nodev(fc, fill_super: bpf_fill_super); |
| 1035 | } |
| 1036 | |
| 1037 | static void bpf_free_fc(struct fs_context *fc) |
| 1038 | { |
| 1039 | kfree(objp: fc->s_fs_info); |
| 1040 | } |
| 1041 | |
| 1042 | static const struct fs_context_operations bpf_context_ops = { |
| 1043 | .free = bpf_free_fc, |
| 1044 | .parse_param = bpf_parse_param, |
| 1045 | .get_tree = bpf_get_tree, |
| 1046 | }; |
| 1047 | |
| 1048 | /* |
| 1049 | * Set up the filesystem mount context. |
| 1050 | */ |
| 1051 | static int bpf_init_fs_context(struct fs_context *fc) |
| 1052 | { |
| 1053 | struct bpf_mount_opts *opts; |
| 1054 | |
| 1055 | opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL); |
| 1056 | if (!opts) |
| 1057 | return -ENOMEM; |
| 1058 | |
| 1059 | opts->mode = S_IRWXUGO; |
| 1060 | opts->uid = current_fsuid(); |
| 1061 | opts->gid = current_fsgid(); |
| 1062 | |
| 1063 | /* start out with no BPF token delegation enabled */ |
| 1064 | opts->delegate_cmds = 0; |
| 1065 | opts->delegate_maps = 0; |
| 1066 | opts->delegate_progs = 0; |
| 1067 | opts->delegate_attachs = 0; |
| 1068 | |
| 1069 | fc->s_fs_info = opts; |
| 1070 | fc->ops = &bpf_context_ops; |
| 1071 | return 0; |
| 1072 | } |
| 1073 | |
| 1074 | static void bpf_kill_super(struct super_block *sb) |
| 1075 | { |
| 1076 | struct bpf_mount_opts *opts = sb->s_fs_info; |
| 1077 | |
| 1078 | kill_anon_super(sb); |
| 1079 | kfree(objp: opts); |
| 1080 | } |
| 1081 | |
| 1082 | static struct file_system_type bpf_fs_type = { |
| 1083 | .owner = THIS_MODULE, |
| 1084 | .name = "bpf" , |
| 1085 | .init_fs_context = bpf_init_fs_context, |
| 1086 | .parameters = bpf_fs_parameters, |
| 1087 | .kill_sb = bpf_kill_super, |
| 1088 | .fs_flags = FS_USERNS_MOUNT, |
| 1089 | }; |
| 1090 | |
| 1091 | static int __init bpf_init(void) |
| 1092 | { |
| 1093 | int ret; |
| 1094 | |
| 1095 | ret = sysfs_create_mount_point(parent_kobj: fs_kobj, name: "bpf" ); |
| 1096 | if (ret) |
| 1097 | return ret; |
| 1098 | |
| 1099 | ret = register_filesystem(&bpf_fs_type); |
| 1100 | if (ret) |
| 1101 | sysfs_remove_mount_point(parent_kobj: fs_kobj, name: "bpf" ); |
| 1102 | |
| 1103 | return ret; |
| 1104 | } |
| 1105 | fs_initcall(bpf_init); |
| 1106 | |