| 1 | |
| 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 <[email protected]>
|
| 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(raw);
|
| 38 | break;
|
| 39 | case BPF_TYPE_MAP:
|
| 40 | bpf_map_inc_with_uref(raw);
|
| 41 | break;
|
| 42 | case BPF_TYPE_LINK:
|
| 43 | bpf_link_inc(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(raw);
|
| 58 | break;
|
| 59 | case BPF_TYPE_MAP:
|
| 60 | bpf_map_put_with_uref(raw);
|
| 61 | break;
|
| 62 | case BPF_TYPE_LINK:
|
| 63 | bpf_link_put(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(raw)) {
|
| 77 | *type = BPF_TYPE_MAP;
|
| 78 | return raw;
|
| 79 | }
|
| 80 |
|
| 81 | raw = bpf_prog_get(ufd);
|
| 82 | if (!IS_ERR(raw)) {
|
| 83 | *type = BPF_TYPE_PROG;
|
| 84 | return raw;
|
| 85 | }
|
| 86 |
|
| 87 | raw = bpf_link_get_from_fd(ufd);
|
| 88 | if (!IS_ERR(raw)) {
|
| 89 | *type = BPF_TYPE_LINK;
|
| 90 | return raw;
|
| 91 | }
|
| 92 |
|
| 93 | return ERR_PTR(-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(-EINVAL);
|
| 115 | }
|
| 116 |
|
| 117 | inode = new_inode(sb);
|
| 118 | if (!inode)
|
| 119 | return ERR_PTR(-ENOSPC);
|
| 120 |
|
| 121 | inode->i_ino = get_next_ino();
|
| 122 | simple_inode_init_ts(inode);
|
| 123 |
|
| 124 | inode_init_owner(&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_instantiate(dentry, inode);
|
| 148 | dget(dentry);
|
| 149 |
|
| 150 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
|
| 151 | }
|
| 152 |
|
| 153 | static int bpf_mkdir(struct mnt_idmap *idmap, struct inode *dir,
|
| 154 | struct dentry *dentry, umode_t mode)
|
| 155 | {
|
| 156 | struct inode *inode;
|
| 157 |
|
| 158 | inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
|
| 159 | if (IS_ERR(inode))
|
| 160 | return PTR_ERR(inode);
|
| 161 |
|
| 162 | inode->i_op = &bpf_dir_iops;
|
| 163 | inode->i_fop = &simple_dir_operations;
|
| 164 |
|
| 165 | inc_nlink(inode);
|
| 166 | inc_nlink(dir);
|
| 167 |
|
| 168 | bpf_dentry_finalize(dentry, inode, dir);
|
| 169 | return 0;
|
| 170 | }
|
| 171 |
|
| 172 | struct map_iter {
|
| 173 | void *key;
|
| 174 | bool done;
|
| 175 | };
|
| 176 |
|
| 177 | static struct map_iter *map_iter(struct seq_file *m)
|
| 178 | {
|
| 179 | return m->private;
|
| 180 | }
|
| 181 |
|
| 182 | static struct bpf_map *seq_file_to_map(struct seq_file *m)
|
| 183 | {
|
| 184 | return file_inode(m->file)->i_private;
|
| 185 | }
|
| 186 |
|
| 187 | static void map_iter_free(struct map_iter *iter)
|
| 188 | {
|
| 189 | if (iter) {
|
| 190 | kfree(iter->key);
|
| 191 | kfree(iter);
|
| 192 | }
|
| 193 | }
|
| 194 |
|
| 195 | static struct map_iter *map_iter_alloc(struct bpf_map *map)
|
| 196 | {
|
| 197 | struct map_iter *iter;
|
| 198 |
|
| 199 | iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
|
| 200 | if (!iter)
|
| 201 | goto error;
|
| 202 |
|
| 203 | iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
|
| 204 | if (!iter->key)
|
| 205 | goto error;
|
| 206 |
|
| 207 | return iter;
|
| 208 |
|
| 209 | error:
|
| 210 | map_iter_free(iter);
|
| 211 | return NULL;
|
| 212 | }
|
| 213 |
|
| 214 | static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
|
| 215 | {
|
| 216 | struct bpf_map *map = seq_file_to_map(m);
|
| 217 | void *key = map_iter(m)->key;
|
| 218 | void *prev_key;
|
| 219 |
|
| 220 | (*pos)++;
|
| 221 | if (map_iter(m)->done)
|
| 222 | return NULL;
|
| 223 |
|
| 224 | if (unlikely(v == SEQ_START_TOKEN))
|
| 225 | prev_key = NULL;
|
| 226 | else
|
| 227 | prev_key = key;
|
| 228 |
|
| 229 | rcu_read_lock();
|
| 230 | if (map->ops->map_get_next_key(map, prev_key, key)) {
|
| 231 | map_iter(m)->done = true;
|
| 232 | key = NULL;
|
| 233 | }
|
| 234 | rcu_read_unlock();
|
| 235 | return key;
|
| 236 | }
|
| 237 |
|
| 238 | static void *map_seq_start(struct seq_file *m, loff_t *pos)
|
| 239 | {
|
| 240 | if (map_iter(m)->done)
|
| 241 | return NULL;
|
| 242 |
|
| 243 | return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
|
| 244 | }
|
| 245 |
|
| 246 | static void map_seq_stop(struct seq_file *m, void *v)
|
| 247 | {
|
| 248 | }
|
| 249 |
|
| 250 | static int map_seq_show(struct seq_file *m, void *v)
|
| 251 | {
|
| 252 | struct bpf_map *map = seq_file_to_map(m);
|
| 253 | void *key = map_iter(m)->key;
|
| 254 |
|
| 255 | if (unlikely(v == SEQ_START_TOKEN)) {
|
| 256 | seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
|
| 257 | seq_puts(m, "# WARNING!! The output format will change\n");
|
| 258 | } else {
|
| 259 | map->ops->map_seq_show_elem(map, key, m);
|
| 260 | }
|
| 261 |
|
| 262 | return 0;
|
| 263 | }
|
| 264 |
|
| 265 | static const struct seq_operations bpffs_map_seq_ops = {
|
| 266 | .start = map_seq_start,
|
| 267 | .next = map_seq_next,
|
| 268 | .show = map_seq_show,
|
| 269 | .stop = map_seq_stop,
|
| 270 | };
|
| 271 |
|
| 272 | static int bpffs_map_open(struct inode *inode, struct file *file)
|
| 273 | {
|
| 274 | struct bpf_map *map = inode->i_private;
|
| 275 | struct map_iter *iter;
|
| 276 | struct seq_file *m;
|
| 277 | int err;
|
| 278 |
|
| 279 | iter = map_iter_alloc(map);
|
| 280 | if (!iter)
|
| 281 | return -ENOMEM;
|
| 282 |
|
| 283 | err = seq_open(file, &bpffs_map_seq_ops);
|
| 284 | if (err) {
|
| 285 | map_iter_free(iter);
|
| 286 | return err;
|
| 287 | }
|
| 288 |
|
| 289 | m = file->private_data;
|
| 290 | m->private = iter;
|
| 291 |
|
| 292 | return 0;
|
| 293 | }
|
| 294 |
|
| 295 | static int bpffs_map_release(struct inode *inode, struct file *file)
|
| 296 | {
|
| 297 | struct seq_file *m = file->private_data;
|
| 298 |
|
| 299 | map_iter_free(map_iter(m));
|
| 300 |
|
| 301 | return seq_release(inode, file);
|
| 302 | }
|
| 303 |
|
| 304 | |
| 305 | * read operation for a BPF map. The purpose is to
|
| 306 | * provide a simple user intuitive way to do
|
| 307 | * "cat bpffs/pathto/a-pinned-map".
|
| 308 | *
|
| 309 | * Other operations (e.g. write, lookup...) should be realized by
|
| 310 | * the userspace tools (e.g. bpftool) through the
|
| 311 | * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
|
| 312 | * interface.
|
| 313 |
|
| 314 | static const struct file_operations bpffs_map_fops = {
|
| 315 | .open = bpffs_map_open,
|
| 316 | .read = seq_read,
|
| 317 | .release = bpffs_map_release,
|
| 318 | };
|
| 319 |
|
| 320 | static int bpffs_obj_open(struct inode *inode, struct file *file)
|
| 321 | {
|
| 322 | return -EIO;
|
| 323 | }
|
| 324 |
|
| 325 | static const struct file_operations bpffs_obj_fops = {
|
| 326 | .open = bpffs_obj_open,
|
| 327 | };
|
| 328 |
|
| 329 | static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
|
| 330 | const struct inode_operations *iops,
|
| 331 | const struct file_operations *fops)
|
| 332 | {
|
| 333 | struct inode *dir = dentry->d_parent->d_inode;
|
| 334 | struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
|
| 335 | if (IS_ERR(inode))
|
| 336 | return PTR_ERR(inode);
|
| 337 |
|
| 338 | inode->i_op = iops;
|
| 339 | inode->i_fop = fops;
|
| 340 | inode->i_private = raw;
|
| 341 |
|
| 342 | bpf_dentry_finalize(dentry, inode, dir);
|
| 343 | return 0;
|
| 344 | }
|
| 345 |
|
| 346 | static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
|
| 347 | {
|
| 348 | return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
|
| 349 | &bpffs_obj_fops);
|
| 350 | }
|
| 351 |
|
| 352 | static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
|
| 353 | {
|
| 354 | struct bpf_map *map = arg;
|
| 355 |
|
| 356 | return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
|
| 357 | bpf_map_support_seq_show(map) ?
|
| 358 | &bpffs_map_fops : &bpffs_obj_fops);
|
| 359 | }
|
| 360 |
|
| 361 | static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
|
| 362 | {
|
| 363 | struct bpf_link *link = arg;
|
| 364 |
|
| 365 | return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
|
| 366 | bpf_link_is_iter(link) ?
|
| 367 | &bpf_iter_fops : &bpffs_obj_fops);
|
| 368 | }
|
| 369 |
|
| 370 | static struct dentry *
|
| 371 | bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
|
| 372 | {
|
| 373 | |
| 374 | * extensions. That allows popoulate_bpffs() create special files.
|
| 375 |
|
| 376 | if ((dir->i_mode & S_IALLUGO) &&
|
| 377 | strchr(dentry->d_name.name, '.'))
|
| 378 | return ERR_PTR(-EPERM);
|
| 379 |
|
| 380 | return simple_lookup(dir, dentry, flags);
|
| 381 | }
|
| 382 |
|
| 383 | static int bpf_symlink(struct mnt_idmap *idmap, struct inode *dir,
|
| 384 | struct dentry *dentry, const char *target)
|
| 385 | {
|
| 386 | char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
|
| 387 | struct inode *inode;
|
| 388 |
|
| 389 | if (!link)
|
| 390 | return -ENOMEM;
|
| 391 |
|
| 392 | inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
|
| 393 | if (IS_ERR(inode)) {
|
| 394 | kfree(link);
|
| 395 | return PTR_ERR(inode);
|
| 396 | }
|
| 397 |
|
| 398 | inode->i_op = &simple_symlink_inode_operations;
|
| 399 | inode->i_link = link;
|
| 400 |
|
| 401 | bpf_dentry_finalize(dentry, inode, dir);
|
| 402 | return 0;
|
| 403 | }
|
| 404 |
|
| 405 | static const struct inode_operations bpf_dir_iops = {
|
| 406 | .lookup = bpf_lookup,
|
| 407 | .mkdir = bpf_mkdir,
|
| 408 | .symlink = bpf_symlink,
|
| 409 | .rmdir = simple_rmdir,
|
| 410 | .rename = simple_rename,
|
| 411 | .link = simple_link,
|
| 412 | .unlink = simple_unlink,
|
| 413 | };
|
| 414 |
|
| 415 |
|
| 416 | static int bpf_iter_link_pin_kernel(struct dentry *parent,
|
| 417 | const char *name, struct bpf_link *link)
|
| 418 | {
|
| 419 | umode_t mode = S_IFREG | S_IRUSR;
|
| 420 | struct dentry *dentry;
|
| 421 | int ret;
|
| 422 |
|
| 423 | inode_lock(parent->d_inode);
|
| 424 | dentry = lookup_one_len(name, parent, strlen(name));
|
| 425 | if (IS_ERR(dentry)) {
|
| 426 | inode_unlock(parent->d_inode);
|
| 427 | return PTR_ERR(dentry);
|
| 428 | }
|
| 429 | ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
|
| 430 | &bpf_iter_fops);
|
| 431 | dput(dentry);
|
| 432 | inode_unlock(parent->d_inode);
|
| 433 | return ret;
|
| 434 | }
|
| 435 |
|
| 436 | static int bpf_obj_do_pin(int path_fd, const char __user *pathname, void *raw,
|
| 437 | enum bpf_type type)
|
| 438 | {
|
| 439 | struct dentry *dentry;
|
| 440 | struct inode *dir;
|
| 441 | struct path path;
|
| 442 | umode_t mode;
|
| 443 | int ret;
|
| 444 |
|
| 445 | dentry = user_path_create(path_fd, pathname, &path, 0);
|
| 446 | if (IS_ERR(dentry))
|
| 447 | return PTR_ERR(dentry);
|
| 448 |
|
| 449 | dir = d_inode(path.dentry);
|
| 450 | if (dir->i_op != &bpf_dir_iops) {
|
| 451 | ret = -EPERM;
|
| 452 | goto out;
|
| 453 | }
|
| 454 |
|
| 455 | mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
|
| 456 | ret = security_path_mknod(&path, dentry, mode, 0);
|
| 457 | if (ret)
|
| 458 | goto out;
|
| 459 |
|
| 460 | switch (type) {
|
| 461 | case BPF_TYPE_PROG:
|
| 462 | ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
|
| 463 | break;
|
| 464 | case BPF_TYPE_MAP:
|
| 465 | ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
|
| 466 | break;
|
| 467 | case BPF_TYPE_LINK:
|
| 468 | ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
|
| 469 | break;
|
| 470 | default:
|
| 471 | ret = -EPERM;
|
| 472 | }
|
| 473 | out:
|
| 474 | done_path_create(&path, dentry);
|
| 475 | return ret;
|
| 476 | }
|
| 477 |
|
| 478 | int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname)
|
| 479 | {
|
| 480 | enum bpf_type type;
|
| 481 | void *raw;
|
| 482 | int ret;
|
| 483 |
|
| 484 | raw = bpf_fd_probe_obj(ufd, &type);
|
| 485 | if (IS_ERR(raw))
|
| 486 | return PTR_ERR(raw);
|
| 487 |
|
| 488 | ret = bpf_obj_do_pin(path_fd, pathname, raw, type);
|
| 489 | if (ret != 0)
|
| 490 | bpf_any_put(raw, type);
|
| 491 |
|
| 492 | return ret;
|
| 493 | }
|
| 494 |
|
| 495 | static void *bpf_obj_do_get(int path_fd, const char __user *pathname,
|
| 496 | enum bpf_type *type, int flags)
|
| 497 | {
|
| 498 | struct inode *inode;
|
| 499 | struct path path;
|
| 500 | void *raw;
|
| 501 | int ret;
|
| 502 |
|
| 503 | ret = user_path_at(path_fd, pathname, LOOKUP_FOLLOW, &path);
|
| 504 | if (ret)
|
| 505 | return ERR_PTR(ret);
|
| 506 |
|
| 507 | inode = d_backing_inode(path.dentry);
|
| 508 | ret = path_permission(&path, ACC_MODE(flags));
|
| 509 | if (ret)
|
| 510 | goto out;
|
| 511 |
|
| 512 | ret = bpf_inode_type(inode, type);
|
| 513 | if (ret)
|
| 514 | goto out;
|
| 515 |
|
| 516 | raw = bpf_any_get(inode->i_private, *type);
|
| 517 | if (!IS_ERR(raw))
|
| 518 | touch_atime(&path);
|
| 519 |
|
| 520 | path_put(&path);
|
| 521 | return raw;
|
| 522 | out:
|
| 523 | path_put(&path);
|
| 524 | return ERR_PTR(ret);
|
| 525 | }
|
| 526 |
|
| 527 | int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags)
|
| 528 | {
|
| 529 | enum bpf_type type = BPF_TYPE_UNSPEC;
|
| 530 | int f_flags;
|
| 531 | void *raw;
|
| 532 | int ret;
|
| 533 |
|
| 534 | f_flags = bpf_get_file_flag(flags);
|
| 535 | if (f_flags < 0)
|
| 536 | return f_flags;
|
| 537 |
|
| 538 | raw = bpf_obj_do_get(path_fd, pathname, &type, f_flags);
|
| 539 | if (IS_ERR(raw))
|
| 540 | return PTR_ERR(raw);
|
| 541 |
|
| 542 | if (type == BPF_TYPE_PROG)
|
| 543 | ret = bpf_prog_new_fd(raw);
|
| 544 | else if (type == BPF_TYPE_MAP)
|
| 545 | ret = bpf_map_new_fd(raw, f_flags);
|
| 546 | else if (type == BPF_TYPE_LINK)
|
| 547 | ret = (f_flags != O_RDWR) ? -EINVAL : bpf_link_new_fd(raw);
|
| 548 | else
|
| 549 | return -ENOENT;
|
| 550 |
|
| 551 | if (ret < 0)
|
| 552 | bpf_any_put(raw, type);
|
| 553 | return ret;
|
| 554 | }
|
| 555 |
|
| 556 | static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
|
| 557 | {
|
| 558 | struct bpf_prog *prog;
|
| 559 | int ret = inode_permission(&nop_mnt_idmap, inode, MAY_READ);
|
| 560 | if (ret)
|
| 561 | return ERR_PTR(ret);
|
| 562 |
|
| 563 | if (inode->i_op == &bpf_map_iops)
|
| 564 | return ERR_PTR(-EINVAL);
|
| 565 | if (inode->i_op == &bpf_link_iops)
|
| 566 | return ERR_PTR(-EINVAL);
|
| 567 | if (inode->i_op != &bpf_prog_iops)
|
| 568 | return ERR_PTR(-EACCES);
|
| 569 |
|
| 570 | prog = inode->i_private;
|
| 571 |
|
| 572 | ret = security_bpf_prog(prog);
|
| 573 | if (ret < 0)
|
| 574 | return ERR_PTR(ret);
|
| 575 |
|
| 576 | if (!bpf_prog_get_ok(prog, &type, false))
|
| 577 | return ERR_PTR(-EINVAL);
|
| 578 |
|
| 579 | bpf_prog_inc(prog);
|
| 580 | return prog;
|
| 581 | }
|
| 582 |
|
| 583 | struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
|
| 584 | {
|
| 585 | struct bpf_prog *prog;
|
| 586 | struct path path;
|
| 587 | int ret = kern_path(name, LOOKUP_FOLLOW, &path);
|
| 588 | if (ret)
|
| 589 | return ERR_PTR(ret);
|
| 590 | prog = __get_prog_inode(d_backing_inode(path.dentry), type);
|
| 591 | if (!IS_ERR(prog))
|
| 592 | touch_atime(&path);
|
| 593 | path_put(&path);
|
| 594 | return prog;
|
| 595 | }
|
| 596 | EXPORT_SYMBOL(bpf_prog_get_type_path);
|
| 597 |
|
| 598 | struct bpffs_btf_enums {
|
| 599 | const struct btf *btf;
|
| 600 | const struct btf_type *cmd_t;
|
| 601 | const struct btf_type *map_t;
|
| 602 | const struct btf_type *prog_t;
|
| 603 | const struct btf_type *attach_t;
|
| 604 | };
|
| 605 |
|
| 606 | static int find_bpffs_btf_enums(struct bpffs_btf_enums *info)
|
| 607 | {
|
| 608 | const struct btf *btf;
|
| 609 | const struct btf_type *t;
|
| 610 | const char *name;
|
| 611 | int i, n;
|
| 612 |
|
| 613 | memset(info, 0, sizeof(*info));
|
| 614 |
|
| 615 | btf = bpf_get_btf_vmlinux();
|
| 616 | if (IS_ERR(btf))
|
| 617 | return PTR_ERR(btf);
|
| 618 | if (!btf)
|
| 619 | return -ENOENT;
|
| 620 |
|
| 621 | info->btf = btf;
|
| 622 |
|
| 623 | for (i = 1, n = btf_nr_types(btf); i < n; i++) {
|
| 624 | t = btf_type_by_id(btf, i);
|
| 625 | if (!btf_type_is_enum(t))
|
| 626 | continue;
|
| 627 |
|
| 628 | name = btf_name_by_offset(btf, t->name_off);
|
| 629 | if (!name)
|
| 630 | continue;
|
| 631 |
|
| 632 | if (strcmp(name, "bpf_cmd") == 0)
|
| 633 | info->cmd_t = t;
|
| 634 | else if (strcmp(name, "bpf_map_type") == 0)
|
| 635 | info->map_t = t;
|
| 636 | else if (strcmp(name, "bpf_prog_type") == 0)
|
| 637 | info->prog_t = t;
|
| 638 | else if (strcmp(name, "bpf_attach_type") == 0)
|
| 639 | info->attach_t = t;
|
| 640 | else
|
| 641 | continue;
|
| 642 |
|
| 643 | if (info->cmd_t && info->map_t && info->prog_t && info->attach_t)
|
| 644 | return 0;
|
| 645 | }
|
| 646 |
|
| 647 | return -ESRCH;
|
| 648 | }
|
| 649 |
|
| 650 | static bool find_btf_enum_const(const struct btf *btf, const struct btf_type *enum_t,
|
| 651 | const char *prefix, const char *str, int *value)
|
| 652 | {
|
| 653 | const struct btf_enum *e;
|
| 654 | const char *name;
|
| 655 | int i, n, pfx_len = strlen(prefix);
|
| 656 |
|
| 657 | *value = 0;
|
| 658 |
|
| 659 | if (!btf || !enum_t)
|
| 660 | return false;
|
| 661 |
|
| 662 | for (i = 0, n = btf_vlen(enum_t); i < n; i++) {
|
| 663 | e = &btf_enum(enum_t)[i];
|
| 664 |
|
| 665 | name = btf_name_by_offset(btf, e->name_off);
|
| 666 | if (!name || strncasecmp(name, prefix, pfx_len) != 0)
|
| 667 | continue;
|
| 668 |
|
| 669 |
|
| 670 | if (strcasecmp(name + pfx_len, str) == 0) {
|
| 671 | *value = e->val;
|
| 672 | return true;
|
| 673 | }
|
| 674 | }
|
| 675 |
|
| 676 | return false;
|
| 677 | }
|
| 678 |
|
| 679 | static void seq_print_delegate_opts(struct seq_file *m,
|
| 680 | const char *opt_name,
|
| 681 | const struct btf *btf,
|
| 682 | const struct btf_type *enum_t,
|
| 683 | const char *prefix,
|
| 684 | u64 delegate_msk, u64 any_msk)
|
| 685 | {
|
| 686 | const struct btf_enum *e;
|
| 687 | bool first = true;
|
| 688 | const char *name;
|
| 689 | u64 msk;
|
| 690 | int i, n, pfx_len = strlen(prefix);
|
| 691 |
|
| 692 | delegate_msk &= any_msk;
|
| 693 |
|
| 694 | if (delegate_msk == 0)
|
| 695 | return;
|
| 696 |
|
| 697 | seq_printf(m, ",%s", opt_name);
|
| 698 | if (delegate_msk == any_msk) {
|
| 699 | seq_printf(m, "=any");
|
| 700 | return;
|
| 701 | }
|
| 702 |
|
| 703 | if (btf && enum_t) {
|
| 704 | for (i = 0, n = btf_vlen(enum_t); i < n; i++) {
|
| 705 | e = &btf_enum(enum_t)[i];
|
| 706 | name = btf_name_by_offset(btf, e->name_off);
|
| 707 | if (!name || strncasecmp(name, prefix, pfx_len) != 0)
|
| 708 | continue;
|
| 709 | msk = 1ULL << e->val;
|
| 710 | if (delegate_msk & msk) {
|
| 711 |
|
| 712 | seq_putc(m, first ? '=' : ':');
|
| 713 | name += pfx_len;
|
| 714 | while (*name) {
|
| 715 | seq_putc(m, tolower(*name));
|
| 716 | name++;
|
| 717 | }
|
| 718 |
|
| 719 | delegate_msk &= ~msk;
|
| 720 | first = false;
|
| 721 | }
|
| 722 | }
|
| 723 | }
|
| 724 | if (delegate_msk)
|
| 725 | seq_printf(m, "%c0x%llx", first ? '=' : ':', delegate_msk);
|
| 726 | }
|
| 727 |
|
| 728 | |
| 729 | * Display the mount options in /proc/mounts.
|
| 730 |
|
| 731 | static int bpf_show_options(struct seq_file *m, struct dentry *root)
|
| 732 | {
|
| 733 | struct inode *inode = d_inode(root);
|
| 734 | umode_t mode = inode->i_mode & S_IALLUGO & ~S_ISVTX;
|
| 735 | struct bpf_mount_opts *opts = root->d_sb->s_fs_info;
|
| 736 | u64 mask;
|
| 737 |
|
| 738 | if (!uid_eq(inode->i_uid, GLOBAL_ROOT_UID))
|
| 739 | seq_printf(m, ",uid=%u",
|
| 740 | from_kuid_munged(&init_user_ns, inode->i_uid));
|
| 741 | if (!gid_eq(inode->i_gid, GLOBAL_ROOT_GID))
|
| 742 | seq_printf(m, ",gid=%u",
|
| 743 | from_kgid_munged(&init_user_ns, inode->i_gid));
|
| 744 | if (mode != S_IRWXUGO)
|
| 745 | seq_printf(m, ",mode=%o", mode);
|
| 746 |
|
| 747 | if (opts->delegate_cmds || opts->delegate_maps ||
|
| 748 | opts->delegate_progs || opts->delegate_attachs) {
|
| 749 | struct bpffs_btf_enums info;
|
| 750 |
|
| 751 |
|
| 752 | (void)find_bpffs_btf_enums(&info);
|
| 753 |
|
| 754 | mask = (1ULL << __MAX_BPF_CMD) - 1;
|
| 755 | seq_print_delegate_opts(m, "delegate_cmds",
|
| 756 | info.btf, info.cmd_t, "BPF_",
|
| 757 | opts->delegate_cmds, mask);
|
| 758 |
|
| 759 | mask = (1ULL << __MAX_BPF_MAP_TYPE) - 1;
|
| 760 | seq_print_delegate_opts(m, "delegate_maps",
|
| 761 | info.btf, info.map_t, "BPF_MAP_TYPE_",
|
| 762 | opts->delegate_maps, mask);
|
| 763 |
|
| 764 | mask = (1ULL << __MAX_BPF_PROG_TYPE) - 1;
|
| 765 | seq_print_delegate_opts(m, "delegate_progs",
|
| 766 | info.btf, info.prog_t, "BPF_PROG_TYPE_",
|
| 767 | opts->delegate_progs, mask);
|
| 768 |
|
| 769 | mask = (1ULL << __MAX_BPF_ATTACH_TYPE) - 1;
|
| 770 | seq_print_delegate_opts(m, "delegate_attachs",
|
| 771 | info.btf, info.attach_t, "BPF_",
|
| 772 | opts->delegate_attachs, mask);
|
| 773 | }
|
| 774 |
|
| 775 | return 0;
|
| 776 | }
|
| 777 |
|
| 778 | static void bpf_free_inode(struct inode *inode)
|
| 779 | {
|
| 780 | enum bpf_type type;
|
| 781 |
|
| 782 | if (S_ISLNK(inode->i_mode))
|
| 783 | kfree(inode->i_link);
|
| 784 | if (!bpf_inode_type(inode, &type))
|
| 785 | bpf_any_put(inode->i_private, type);
|
| 786 | free_inode_nonrcu(inode);
|
| 787 | }
|
| 788 |
|
| 789 | const struct super_operations bpf_super_ops = {
|
| 790 | .statfs = simple_statfs,
|
| 791 | .drop_inode = generic_delete_inode,
|
| 792 | .show_options = bpf_show_options,
|
| 793 | .free_inode = bpf_free_inode,
|
| 794 | };
|
| 795 |
|
| 796 | enum {
|
| 797 | OPT_UID,
|
| 798 | OPT_GID,
|
| 799 | OPT_MODE,
|
| 800 | OPT_DELEGATE_CMDS,
|
| 801 | OPT_DELEGATE_MAPS,
|
| 802 | OPT_DELEGATE_PROGS,
|
| 803 | OPT_DELEGATE_ATTACHS,
|
| 804 | };
|
| 805 |
|
| 806 | static const struct fs_parameter_spec bpf_fs_parameters[] = {
|
| 807 | fsparam_u32 ("uid", OPT_UID),
|
| 808 | fsparam_u32 ("gid", OPT_GID),
|
| 809 | fsparam_u32oct ("mode", OPT_MODE),
|
| 810 | fsparam_string ("delegate_cmds", OPT_DELEGATE_CMDS),
|
| 811 | fsparam_string ("delegate_maps", OPT_DELEGATE_MAPS),
|
| 812 | fsparam_string ("delegate_progs", OPT_DELEGATE_PROGS),
|
| 813 | fsparam_string ("delegate_attachs", OPT_DELEGATE_ATTACHS),
|
| 814 | {}
|
| 815 | };
|
| 816 |
|
| 817 | static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
|
| 818 | {
|
| 819 | struct bpf_mount_opts *opts = fc->s_fs_info;
|
| 820 | struct fs_parse_result result;
|
| 821 | kuid_t uid;
|
| 822 | kgid_t gid;
|
| 823 | int opt, err;
|
| 824 |
|
| 825 | opt = fs_parse(fc, bpf_fs_parameters, param, &result);
|
| 826 | if (opt < 0) {
|
| 827 | |
| 828 | * traditionally we've ignored all mount options, so we'd
|
| 829 | * better continue to ignore non-existing options for bpf.
|
| 830 |
|
| 831 | if (opt == -ENOPARAM) {
|
| 832 | opt = vfs_parse_fs_param_source(fc, param);
|
| 833 | if (opt != -ENOPARAM)
|
| 834 | return opt;
|
| 835 |
|
| 836 | return 0;
|
| 837 | }
|
| 838 |
|
| 839 | if (opt < 0)
|
| 840 | return opt;
|
| 841 | }
|
| 842 |
|
| 843 | switch (opt) {
|
| 844 | case OPT_UID:
|
| 845 | uid = make_kuid(current_user_ns(), result.uint_32);
|
| 846 | if (!uid_valid(uid))
|
| 847 | goto bad_value;
|
| 848 |
|
| 849 | |
| 850 | * The requested uid must be representable in the
|
| 851 | * filesystem's idmapping.
|
| 852 |
|
| 853 | if (!kuid_has_mapping(fc->user_ns, uid))
|
| 854 | goto bad_value;
|
| 855 |
|
| 856 | opts->uid = uid;
|
| 857 | break;
|
| 858 | case OPT_GID:
|
| 859 | gid = make_kgid(current_user_ns(), result.uint_32);
|
| 860 | if (!gid_valid(gid))
|
| 861 | goto bad_value;
|
| 862 |
|
| 863 | |
| 864 | * The requested gid must be representable in the
|
| 865 | * filesystem's idmapping.
|
| 866 |
|
| 867 | if (!kgid_has_mapping(fc->user_ns, gid))
|
| 868 | goto bad_value;
|
| 869 |
|
| 870 | opts->gid = gid;
|
| 871 | break;
|
| 872 | case OPT_MODE:
|
| 873 | opts->mode = result.uint_32 & S_IALLUGO;
|
| 874 | break;
|
| 875 | case OPT_DELEGATE_CMDS:
|
| 876 | case OPT_DELEGATE_MAPS:
|
| 877 | case OPT_DELEGATE_PROGS:
|
| 878 | case OPT_DELEGATE_ATTACHS: {
|
| 879 | struct bpffs_btf_enums info;
|
| 880 | const struct btf_type *enum_t;
|
| 881 | const char *enum_pfx;
|
| 882 | u64 *delegate_msk, msk = 0;
|
| 883 | char *p, *str;
|
| 884 | int val;
|
| 885 |
|
| 886 |
|
| 887 | (void)find_bpffs_btf_enums(&info);
|
| 888 |
|
| 889 | switch (opt) {
|
| 890 | case OPT_DELEGATE_CMDS:
|
| 891 | delegate_msk = &opts->delegate_cmds;
|
| 892 | enum_t = info.cmd_t;
|
| 893 | enum_pfx = "BPF_";
|
| 894 | break;
|
| 895 | case OPT_DELEGATE_MAPS:
|
| 896 | delegate_msk = &opts->delegate_maps;
|
| 897 | enum_t = info.map_t;
|
| 898 | enum_pfx = "BPF_MAP_TYPE_";
|
| 899 | break;
|
| 900 | case OPT_DELEGATE_PROGS:
|
| 901 | delegate_msk = &opts->delegate_progs;
|
| 902 | enum_t = info.prog_t;
|
| 903 | enum_pfx = "BPF_PROG_TYPE_";
|
| 904 | break;
|
| 905 | case OPT_DELEGATE_ATTACHS:
|
| 906 | delegate_msk = &opts->delegate_attachs;
|
| 907 | enum_t = info.attach_t;
|
| 908 | enum_pfx = "BPF_";
|
| 909 | break;
|
| 910 | default:
|
| 911 | return -EINVAL;
|
| 912 | }
|
| 913 |
|
| 914 | str = param->string;
|
| 915 | while ((p = strsep(&str, ":"))) {
|
| 916 | if (strcmp(p, "any") == 0) {
|
| 917 | msk |= ~0ULL;
|
| 918 | } else if (find_btf_enum_const(info.btf, enum_t, enum_pfx, p, &val)) {
|
| 919 | msk |= 1ULL << val;
|
| 920 | } else {
|
| 921 | err = kstrtou64(p, 0, &msk);
|
| 922 | if (err)
|
| 923 | return err;
|
| 924 | }
|
| 925 | }
|
| 926 |
|
| 927 |
|
| 928 | if (msk && !capable(CAP_SYS_ADMIN))
|
| 929 | return -EPERM;
|
| 930 |
|
| 931 | *delegate_msk |= msk;
|
| 932 | break;
|
| 933 | }
|
| 934 | default:
|
| 935 |
|
| 936 | break;
|
| 937 | }
|
| 938 |
|
| 939 | return 0;
|
| 940 | bad_value:
|
| 941 | return invalfc(fc, "Bad value for '%s'", param->key);
|
| 942 | }
|
| 943 |
|
| 944 | struct bpf_preload_ops *bpf_preload_ops;
|
| 945 | EXPORT_SYMBOL_GPL(bpf_preload_ops);
|
| 946 |
|
| 947 | static bool bpf_preload_mod_get(void)
|
| 948 | {
|
| 949 | |
| 950 | * When bpf_preload is built into vmlinux the module's __init
|
| 951 | * function will populate it.
|
| 952 |
|
| 953 | if (!bpf_preload_ops) {
|
| 954 | request_module("bpf_preload");
|
| 955 | if (!bpf_preload_ops)
|
| 956 | return false;
|
| 957 | }
|
| 958 | |
| 959 | * kernel is interacting with the kernel module and its UMD.
|
| 960 |
|
| 961 | if (!try_module_get(bpf_preload_ops->owner)) {
|
| 962 | pr_err("bpf_preload module get failed.\n");
|
| 963 | return false;
|
| 964 | }
|
| 965 | return true;
|
| 966 | }
|
| 967 |
|
| 968 | static void bpf_preload_mod_put(void)
|
| 969 | {
|
| 970 | if (bpf_preload_ops)
|
| 971 |
|
| 972 | module_put(bpf_preload_ops->owner);
|
| 973 | }
|
| 974 |
|
| 975 | static DEFINE_MUTEX(bpf_preload_lock);
|
| 976 |
|
| 977 | static int populate_bpffs(struct dentry *parent)
|
| 978 | {
|
| 979 | struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
|
| 980 | int err = 0, i;
|
| 981 |
|
| 982 | |
| 983 | * are serialized
|
| 984 |
|
| 985 | mutex_lock(&bpf_preload_lock);
|
| 986 |
|
| 987 |
|
| 988 | if (!bpf_preload_mod_get())
|
| 989 | goto out;
|
| 990 |
|
| 991 | err = bpf_preload_ops->preload(objs);
|
| 992 | if (err)
|
| 993 | goto out_put;
|
| 994 | for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
|
| 995 | bpf_link_inc(objs[i].link);
|
| 996 | err = bpf_iter_link_pin_kernel(parent,
|
| 997 | objs[i].link_name, objs[i].link);
|
| 998 | if (err) {
|
| 999 | bpf_link_put(objs[i].link);
|
| 1000 | goto out_put;
|
| 1001 | }
|
| 1002 | }
|
| 1003 | out_put:
|
| 1004 | bpf_preload_mod_put();
|
| 1005 | out:
|
| 1006 | mutex_unlock(&bpf_preload_lock);
|
| 1007 | return err;
|
| 1008 | }
|
| 1009 |
|
| 1010 | static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
|
| 1011 | {
|
| 1012 | static const struct tree_descr bpf_rfiles[] = { { "" } };
|
| 1013 | struct bpf_mount_opts *opts = sb->s_fs_info;
|
| 1014 | struct inode *inode;
|
| 1015 | int ret;
|
| 1016 |
|
| 1017 |
|
| 1018 | if (fc->user_ns != &init_user_ns && !capable(CAP_SYS_ADMIN))
|
| 1019 | return -EPERM;
|
| 1020 |
|
| 1021 | ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
|
| 1022 | if (ret)
|
| 1023 | return ret;
|
| 1024 |
|
| 1025 | sb->s_op = &bpf_super_ops;
|
| 1026 |
|
| 1027 | inode = sb->s_root->d_inode;
|
| 1028 | inode->i_uid = opts->uid;
|
| 1029 | inode->i_gid = opts->gid;
|
| 1030 | inode->i_op = &bpf_dir_iops;
|
| 1031 | inode->i_mode &= ~S_IALLUGO;
|
| 1032 | populate_bpffs(sb->s_root);
|
| 1033 | inode->i_mode |= S_ISVTX | opts->mode;
|
| 1034 | return 0;
|
| 1035 | }
|
| 1036 |
|
| 1037 | static int bpf_get_tree(struct fs_context *fc)
|
| 1038 | {
|
| 1039 | return get_tree_nodev(fc, bpf_fill_super);
|
| 1040 | }
|
| 1041 |
|
| 1042 | static void bpf_free_fc(struct fs_context *fc)
|
| 1043 | {
|
| 1044 | kfree(fc->s_fs_info);
|
| 1045 | }
|
| 1046 |
|
| 1047 | static const struct fs_context_operations bpf_context_ops = {
|
| 1048 | .free = bpf_free_fc,
|
| 1049 | .parse_param = bpf_parse_param,
|
| 1050 | .get_tree = bpf_get_tree,
|
| 1051 | };
|
| 1052 |
|
| 1053 | |
| 1054 | * Set up the filesystem mount context.
|
| 1055 |
|
| 1056 | static int bpf_init_fs_context(struct fs_context *fc)
|
| 1057 | {
|
| 1058 | struct bpf_mount_opts *opts;
|
| 1059 |
|
| 1060 | opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
|
| 1061 | if (!opts)
|
| 1062 | return -ENOMEM;
|
| 1063 |
|
| 1064 | opts->mode = S_IRWXUGO;
|
| 1065 | opts->uid = current_fsuid();
|
| 1066 | opts->gid = current_fsgid();
|
| 1067 |
|
| 1068 |
|
| 1069 | opts->delegate_cmds = 0;
|
| 1070 | opts->delegate_maps = 0;
|
| 1071 | opts->delegate_progs = 0;
|
| 1072 | opts->delegate_attachs = 0;
|
| 1073 |
|
| 1074 | fc->s_fs_info = opts;
|
| 1075 | fc->ops = &bpf_context_ops;
|
| 1076 | return 0;
|
| 1077 | }
|
| 1078 |
|
| 1079 | static void bpf_kill_super(struct super_block *sb)
|
| 1080 | {
|
| 1081 | struct bpf_mount_opts *opts = sb->s_fs_info;
|
| 1082 |
|
| 1083 | kill_litter_super(sb);
|
| 1084 | kfree(opts);
|
| 1085 | }
|
| 1086 |
|
| 1087 | static struct file_system_type bpf_fs_type = {
|
| 1088 | .owner = THIS_MODULE,
|
| 1089 | .name = "bpf",
|
| 1090 | .init_fs_context = bpf_init_fs_context,
|
| 1091 | .parameters = bpf_fs_parameters,
|
| 1092 | .kill_sb = bpf_kill_super,
|
| 1093 | .fs_flags = FS_USERNS_MOUNT,
|
| 1094 | };
|
| 1095 |
|
| 1096 | static int __init bpf_init(void)
|
| 1097 | {
|
| 1098 | int ret;
|
| 1099 |
|
| 1100 | ret = sysfs_create_mount_point(fs_kobj, "bpf");
|
| 1101 | if (ret)
|
| 1102 | return ret;
|
| 1103 |
|
| 1104 | ret = register_filesystem(&bpf_fs_type);
|
| 1105 | if (ret)
|
| 1106 | sysfs_remove_mount_point(fs_kobj, "bpf");
|
| 1107 |
|
| 1108 | return ret;
|
| 1109 | }
|
| 1110 | fs_initcall(bpf_init);
|