kernel/bpf/inode.c

   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
  24 enum bpf_type {
  25         BPF_TYPE_UNSPEC = 0,
  26         BPF_TYPE_PROG,
  27         BPF_TYPE_MAP,
  28 };
  29
  30 static void *bpf_any_get(void *raw, enum bpf_type type)
  31 {
  32         switch (type) {
  33         case BPF_TYPE_PROG:
  34                 bpf_prog_inc(raw);
  35                 break;
  36         case BPF_TYPE_MAP:
  37                 bpf_map_inc_with_uref(raw);
  38                 break;
  39         default:
  40                 WARN_ON_ONCE(1);
  41                 break;
  42         }
  43
  44         return raw;
  45 }
  46
  47 static void bpf_any_put(void *raw, enum bpf_type type)
  48 {
  49         switch (type) {
  50         case BPF_TYPE_PROG:
  51                 bpf_prog_put(raw);
  52                 break;
  53         case BPF_TYPE_MAP:
  54                 bpf_map_put_with_uref(raw);
  55                 break;
  56         default:
  57                 WARN_ON_ONCE(1);
  58                 break;
  59         }
  60 }
  61
  62 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
  63 {
  64         void *raw;
  65
  66         *type = BPF_TYPE_MAP;
  67         raw = bpf_map_get_with_uref(ufd);
  68         if (IS_ERR(raw)) {
  69                 *type = BPF_TYPE_PROG;
  70                 raw = bpf_prog_get(ufd);
  71         }
  72
  73         return raw;
  74 }
  75
  76 static const struct inode_operations bpf_dir_iops;
  77
  78 static const struct inode_operations bpf_prog_iops = { };
  79 static const struct inode_operations bpf_map_iops  = { };
  80
  81 static struct inode *bpf_get_inode(struct super_block *sb,
  82                                    const struct inode *dir,
  83                                    umode_t mode)
  84 {
  85         struct inode *inode;
  86
  87         switch (mode & S_IFMT) {
  88         case S_IFDIR:
  89         case S_IFREG:
  90         case S_IFLNK:
  91                 break;
  92         default:
  93                 return ERR_PTR(-EINVAL);
  94         }
  95
  96         inode = new_inode(sb);
  97         if (!inode)
  98                 return ERR_PTR(-ENOSPC);
  99
 100         inode->i_ino = get_next_ino();
 101         inode->i_atime = current_time(inode);
 102         inode->i_mtime = inode->i_atime;
 103         inode->i_ctime = inode->i_atime;
 104
 105         inode_init_owner(inode, dir, mode);
 106
 107         return inode;
 108 }
 109
 110 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
 111 {
 112         *type = BPF_TYPE_UNSPEC;
 113         if (inode->i_op == &bpf_prog_iops)
 114                 *type = BPF_TYPE_PROG;
 115         else if (inode->i_op == &bpf_map_iops)
 116                 *type = BPF_TYPE_MAP;
 117         else
 118                 return -EACCES;
 119
 120         return 0;
 121 }
 122
 123 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
 124                                 struct inode *dir)
 125 {
 126         d_instantiate(dentry, inode);
 127         dget(dentry);
 128
 129         dir->i_mtime = current_time(dir);
 130         dir->i_ctime = dir->i_mtime;
 131 }
 132
 133 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 134 {
 135         struct inode *inode;
 136
 137         inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
 138         if (IS_ERR(inode))
 139                 return PTR_ERR(inode);
 140
 141         inode->i_op = &bpf_dir_iops;
 142         inode->i_fop = &simple_dir_operations;
 143
 144         inc_nlink(inode);
 145         inc_nlink(dir);
 146
 147         bpf_dentry_finalize(dentry, inode, dir);
 148         return 0;
 149 }
 150
 151 struct map_iter {
 152         void *key;
 153         bool done;
 154 };
 155
 156 static struct map_iter *map_iter(struct seq_file *m)
 157 {
 158         return m->private;
 159 }
 160
 161 static struct bpf_map *seq_file_to_map(struct seq_file *m)
 162 {
 163         return file_inode(m->file)->i_private;
 164 }
 165
 166 static void map_iter_free(struct map_iter *iter)
 167 {
 168         if (iter) {
 169                 kfree(iter->key);
 170                 kfree(iter);
 171         }
 172 }
 173
 174 static struct map_iter *map_iter_alloc(struct bpf_map *map)
 175 {
 176         struct map_iter *iter;
 177
 178         iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
 179         if (!iter)
 180                 goto error;
 181
 182         iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
 183         if (!iter->key)
 184                 goto error;
 185
 186         return iter;
 187
 188 error:
 189         map_iter_free(iter);
 190         return NULL;
 191 }
 192
 193 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
 194 {
 195         struct bpf_map *map = seq_file_to_map(m);
 196         void *key = map_iter(m)->key;
 197         void *prev_key;
 198
 199         if (map_iter(m)->done)
 200                 return NULL;
 201
 202         if (unlikely(v == SEQ_START_TOKEN))
 203                 prev_key = NULL;
 204         else
 205                 prev_key = key;
 206
 207         if (map->ops->map_get_next_key(map, prev_key, key)) {
 208                 map_iter(m)->done = true;
 209                 return NULL;
 210         }
 211
 212         ++(*pos);
 213         return key;
 214 }
 215
 216 static void *map_seq_start(struct seq_file *m, loff_t *pos)
 217 {
 218         if (map_iter(m)->done)
 219                 return NULL;
 220
 221         return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
 222 }
 223
 224 static void map_seq_stop(struct seq_file *m, void *v)
 225 {
 226 }
 227
 228 static int map_seq_show(struct seq_file *m, void *v)
 229 {
 230         struct bpf_map *map = seq_file_to_map(m);
 231         void *key = map_iter(m)->key;
 232
 233         if (unlikely(v == SEQ_START_TOKEN)) {
 234                 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
 235                 seq_puts(m, "# WARNING!! The output format will change\n");
 236         } else {
 237                 map->ops->map_seq_show_elem(map, key, m);
 238         }
 239
 240         return 0;
 241 }
 242
 243 static const struct seq_operations bpffs_map_seq_ops = {
 244         .start  = map_seq_start,
 245         .next   = map_seq_next,
 246         .show   = map_seq_show,
 247         .stop   = map_seq_stop,
 248 };
 249
 250 static int bpffs_map_open(struct inode *inode, struct file *file)
 251 {
 252         struct bpf_map *map = inode->i_private;
 253         struct map_iter *iter;
 254         struct seq_file *m;
 255         int err;
 256
 257         iter = map_iter_alloc(map);
 258         if (!iter)
 259                 return -ENOMEM;
 260
 261         err = seq_open(file, &bpffs_map_seq_ops);
 262         if (err) {
 263                 map_iter_free(iter);
 264                 return err;
 265         }
 266
 267         m = file->private_data;
 268         m->private = iter;
 269
 270         return 0;
 271 }
 272
 273 static int bpffs_map_release(struct inode *inode, struct file *file)
 274 {
 275         struct seq_file *m = file->private_data;
 276
 277         map_iter_free(map_iter(m));
 278
 279         return seq_release(inode, file);
 280 }
 281
 282 /* bpffs_map_fops should only implement the basic
 283  * read operation for a BPF map.  The purpose is to
 284  * provide a simple user intuitive way to do
 285  * "cat bpffs/pathto/a-pinned-map".
 286  *
 287  * Other operations (e.g. write, lookup...) should be realized by
 288  * the userspace tools (e.g. bpftool) through the
 289  * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
 290  * interface.
 291  */
 292 static const struct file_operations bpffs_map_fops = {
 293         .open           = bpffs_map_open,
 294         .read           = seq_read,
 295         .release        = bpffs_map_release,
 296 };
 297
 298 static int bpffs_obj_open(struct inode *inode, struct file *file)
 299 {
 300         return -EIO;
 301 }
 302
 303 static const struct file_operations bpffs_obj_fops = {
 304         .open           = bpffs_obj_open,
 305 };
 306
 307 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
 308                          const struct inode_operations *iops,
 309                          const struct file_operations *fops)
 310 {
 311         struct inode *dir = dentry->d_parent->d_inode;
 312         struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
 313         if (IS_ERR(inode))
 314                 return PTR_ERR(inode);
 315
 316         inode->i_op = iops;
 317         inode->i_fop = fops;
 318         inode->i_private = raw;
 319
 320         bpf_dentry_finalize(dentry, inode, dir);
 321         return 0;
 322 }
 323
 324 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
 325 {
 326         return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
 327                              &bpffs_obj_fops);
 328 }
 329
 330 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
 331 {
 332         struct bpf_map *map = arg;
 333
 334         return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
 335                              bpf_map_support_seq_show(map) ?
 336                              &bpffs_map_fops : &bpffs_obj_fops);
 337 }
 338
 339 static struct dentry *
 340 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
 341 {
 342         /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
 343          * extensions.
 344          */
 345         if (strchr(dentry->d_name.name, '.'))
 346                 return ERR_PTR(-EPERM);
 347
 348         return simple_lookup(dir, dentry, flags);
 349 }
 350
 351 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
 352                        const char *target)
 353 {
 354         char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
 355         struct inode *inode;
 356
 357         if (!link)
 358                 return -ENOMEM;
 359
 360         inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
 361         if (IS_ERR(inode)) {
 362                 kfree(link);
 363                 return PTR_ERR(inode);
 364         }
 365
 366         inode->i_op = &simple_symlink_inode_operations;
 367         inode->i_link = link;
 368
 369         bpf_dentry_finalize(dentry, inode, dir);
 370         return 0;
 371 }
 372
 373 static const struct inode_operations bpf_dir_iops = {
 374         .lookup         = bpf_lookup,
 375         .mkdir          = bpf_mkdir,
 376         .symlink        = bpf_symlink,
 377         .rmdir          = simple_rmdir,
 378         .rename         = simple_rename,
 379         .link           = simple_link,
 380         .unlink         = simple_unlink,
 381 };
 382
 383 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
 384                           enum bpf_type type)
 385 {
 386         struct dentry *dentry;
 387         struct inode *dir;
 388         struct path path;
 389         umode_t mode;
 390         int ret;
 391
 392         dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
 393         if (IS_ERR(dentry))
 394                 return PTR_ERR(dentry);
 395
 396         mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
 397
 398         ret = security_path_mknod(&path, dentry, mode, 0);
 399         if (ret)
 400                 goto out;
 401
 402         dir = d_inode(path.dentry);
 403         if (dir->i_op != &bpf_dir_iops) {
 404                 ret = -EPERM;
 405                 goto out;
 406         }
 407
 408         switch (type) {
 409         case BPF_TYPE_PROG:
 410                 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
 411                 break;
 412         case BPF_TYPE_MAP:
 413                 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
 414                 break;
 415         default:
 416                 ret = -EPERM;
 417         }
 418 out:
 419         done_path_create(&path, dentry);
 420         return ret;
 421 }
 422
 423 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
 424 {
 425         struct filename *pname;
 426         enum bpf_type type;
 427         void *raw;
 428         int ret;
 429
 430         pname = getname(pathname);
 431         if (IS_ERR(pname))
 432                 return PTR_ERR(pname);
 433
 434         raw = bpf_fd_probe_obj(ufd, &type);
 435         if (IS_ERR(raw)) {
 436                 ret = PTR_ERR(raw);
 437                 goto out;
 438         }
 439
 440         ret = bpf_obj_do_pin(pname, raw, type);
 441         if (ret != 0)
 442                 bpf_any_put(raw, type);
 443 out:
 444         putname(pname);
 445         return ret;
 446 }
 447
 448 static void *bpf_obj_do_get(const struct filename *pathname,
 449                             enum bpf_type *type, int flags)
 450 {
 451         struct inode *inode;
 452         struct path path;
 453         void *raw;
 454         int ret;
 455
 456         ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
 457         if (ret)
 458                 return ERR_PTR(ret);
 459
 460         inode = d_backing_inode(path.dentry);
 461         ret = inode_permission(inode, ACC_MODE(flags));
 462         if (ret)
 463                 goto out;
 464
 465         ret = bpf_inode_type(inode, type);
 466         if (ret)
 467                 goto out;
 468
 469         raw = bpf_any_get(inode->i_private, *type);
 470         if (!IS_ERR(raw))
 471                 touch_atime(&path);
 472
 473         path_put(&path);
 474         return raw;
 475 out:
 476         path_put(&path);
 477         return ERR_PTR(ret);
 478 }
 479
 480 int bpf_obj_get_user(const char __user *pathname, int flags)
 481 {
 482         enum bpf_type type = BPF_TYPE_UNSPEC;
 483         struct filename *pname;
 484         int ret = -ENOENT;
 485         int f_flags;
 486         void *raw;
 487
 488         f_flags = bpf_get_file_flag(flags);
 489         if (f_flags < 0)
 490                 return f_flags;
 491
 492         pname = getname(pathname);
 493         if (IS_ERR(pname))
 494                 return PTR_ERR(pname);
 495
 496         raw = bpf_obj_do_get(pname, &type, f_flags);
 497         if (IS_ERR(raw)) {
 498                 ret = PTR_ERR(raw);
 499                 goto out;
 500         }
 501
 502         if (type == BPF_TYPE_PROG)
 503                 ret = bpf_prog_new_fd(raw);
 504         else if (type == BPF_TYPE_MAP)
 505                 ret = bpf_map_new_fd(raw, f_flags);
 506         else
 507                 goto out;
 508
 509         if (ret < 0)
 510                 bpf_any_put(raw, type);
 511 out:
 512         putname(pname);
 513         return ret;
 514 }
 515
 516 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
 517 {
 518         struct bpf_prog *prog;
 519         int ret = inode_permission(inode, MAY_READ);
 520         if (ret)
 521                 return ERR_PTR(ret);
 522
 523         if (inode->i_op == &bpf_map_iops)
 524                 return ERR_PTR(-EINVAL);
 525         if (inode->i_op != &bpf_prog_iops)
 526                 return ERR_PTR(-EACCES);
 527
 528         prog = inode->i_private;
 529
 530         ret = security_bpf_prog(prog);
 531         if (ret < 0)
 532                 return ERR_PTR(ret);
 533
 534         if (!bpf_prog_get_ok(prog, &type, false))
 535                 return ERR_PTR(-EINVAL);
 536
 537         bpf_prog_inc(prog);
 538         return prog;
 539 }
 540
 541 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
 542 {
 543         struct bpf_prog *prog;
 544         struct path path;
 545         int ret = kern_path(name, LOOKUP_FOLLOW, &path);
 546         if (ret)
 547                 return ERR_PTR(ret);
 548         prog = __get_prog_inode(d_backing_inode(path.dentry), type);
 549         if (!IS_ERR(prog))
 550                 touch_atime(&path);
 551         path_put(&path);
 552         return prog;
 553 }
 554 EXPORT_SYMBOL(bpf_prog_get_type_path);
 555
 556 /*
 557  * Display the mount options in /proc/mounts.
 558  */
 559 static int bpf_show_options(struct seq_file *m, struct dentry *root)
 560 {
 561         umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
 562
 563         if (mode != S_IRWXUGO)
 564                 seq_printf(m, ",mode=%o", mode);
 565         return 0;
 566 }
 567
 568 static void bpf_free_inode(struct inode *inode)
 569 {
 570         enum bpf_type type;
 571
 572         if (S_ISLNK(inode->i_mode))
 573                 kfree(inode->i_link);
 574         if (!bpf_inode_type(inode, &type))
 575                 bpf_any_put(inode->i_private, type);
 576         free_inode_nonrcu(inode);
 577 }
 578
 579 static const struct super_operations bpf_super_ops = {
 580         .statfs         = simple_statfs,
 581         .drop_inode     = generic_delete_inode,
 582         .show_options   = bpf_show_options,
 583         .free_inode     = bpf_free_inode,
 584 };
 585
 586 enum {
 587         OPT_MODE,
 588 };
 589
 590 static const struct fs_parameter_spec bpf_param_specs[] = {
 591         fsparam_u32oct  ("mode",                        OPT_MODE),
 592         {}
 593 };
 594
 595 static const struct fs_parameter_description bpf_fs_parameters = {
 596         .name           = "bpf",
 597         .specs          = bpf_param_specs,
 598 };
 599
 600 struct bpf_mount_opts {
 601         umode_t mode;
 602 };
 603
 604 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
 605 {
 606         struct bpf_mount_opts *opts = fc->fs_private;
 607         struct fs_parse_result result;
 608         int opt;
 609
 610         opt = fs_parse(fc, &bpf_fs_parameters, param, &result);
 611         if (opt < 0)
 612                 /* We might like to report bad mount options here, but
 613                  * traditionally we've ignored all mount options, so we'd
 614                  * better continue to ignore non-existing options for bpf.
 615                  */
 616                 return opt == -ENOPARAM ? 0 : opt;
 617
 618         switch (opt) {
 619         case OPT_MODE:
 620                 opts->mode = result.uint_32 & S_IALLUGO;
 621                 break;
 622         }
 623
 624         return 0;
 625 }
 626
 627 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
 628 {
 629         static const struct tree_descr bpf_rfiles[] = { { "" } };
 630         struct bpf_mount_opts *opts = fc->fs_private;
 631         struct inode *inode;
 632         int ret;
 633
 634         ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
 635         if (ret)
 636                 return ret;
 637
 638         sb->s_op = &bpf_super_ops;
 639
 640         inode = sb->s_root->d_inode;
 641         inode->i_op = &bpf_dir_iops;
 642         inode->i_mode &= ~S_IALLUGO;
 643         inode->i_mode |= S_ISVTX | opts->mode;
 644
 645         return 0;
 646 }
 647
 648 static int bpf_get_tree(struct fs_context *fc)
 649 {
 650         return get_tree_nodev(fc, bpf_fill_super);
 651 }
 652
 653 static void bpf_free_fc(struct fs_context *fc)
 654 {
 655         kfree(fc->fs_private);
 656 }
 657
 658 static const struct fs_context_operations bpf_context_ops = {
 659         .free           = bpf_free_fc,
 660         .parse_param    = bpf_parse_param,
 661         .get_tree       = bpf_get_tree,
 662 };
 663
 664 /*
 665  * Set up the filesystem mount context.
 666  */
 667 static int bpf_init_fs_context(struct fs_context *fc)
 668 {
 669         struct bpf_mount_opts *opts;
 670
 671         opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
 672         if (!opts)
 673                 return -ENOMEM;
 674
 675         opts->mode = S_IRWXUGO;
 676
 677         fc->fs_private = opts;
 678         fc->ops = &bpf_context_ops;
 679         return 0;
 680 }
 681
 682 static struct file_system_type bpf_fs_type = {
 683         .owner          = THIS_MODULE,
 684         .name           = "bpf",
 685         .init_fs_context = bpf_init_fs_context,
 686         .parameters     = &bpf_fs_parameters,
 687         .kill_sb        = kill_litter_super,
 688 };
 689
 690 static int __init bpf_init(void)
 691 {
 692         int ret;
 693
 694         ret = sysfs_create_mount_point(fs_kobj, "bpf");
 695         if (ret)
 696                 return ret;
 697
 698         ret = register_filesystem(&bpf_fs_type);
 699         if (ret)
 700                 sysfs_remove_mount_point(fs_kobj, "bpf");
 701
 702         return ret;
 703 }
 704 fs_initcall(bpf_init);