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         (*pos)++;
 200         if (map_iter(m)->done)
 201                 return NULL;
 202
 203         if (unlikely(v == SEQ_START_TOKEN))
 204                 prev_key = NULL;
 205         else
 206                 prev_key = key;
 207
 208         if (map->ops->map_get_next_key(map, prev_key, key)) {
 209                 map_iter(m)->done = true;
 210                 return NULL;
 211         }
 212         return key;
 213 }
 214
 215 static void *map_seq_start(struct seq_file *m, loff_t *pos)
 216 {
 217         if (map_iter(m)->done)
 218                 return NULL;
 219
 220         return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
 221 }
 222
 223 static void map_seq_stop(struct seq_file *m, void *v)
 224 {
 225 }
 226
 227 static int map_seq_show(struct seq_file *m, void *v)
 228 {
 229         struct bpf_map *map = seq_file_to_map(m);
 230         void *key = map_iter(m)->key;
 231
 232         if (unlikely(v == SEQ_START_TOKEN)) {
 233                 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
 234                 seq_puts(m, "# WARNING!! The output format will change\n");
 235         } else {
 236                 map->ops->map_seq_show_elem(map, key, m);
 237         }
 238
 239         return 0;
 240 }
 241
 242 static const struct seq_operations bpffs_map_seq_ops = {
 243         .start  = map_seq_start,
 244         .next   = map_seq_next,
 245         .show   = map_seq_show,
 246         .stop   = map_seq_stop,
 247 };
 248
 249 static int bpffs_map_open(struct inode *inode, struct file *file)
 250 {
 251         struct bpf_map *map = inode->i_private;
 252         struct map_iter *iter;
 253         struct seq_file *m;
 254         int err;
 255
 256         iter = map_iter_alloc(map);
 257         if (!iter)
 258                 return -ENOMEM;
 259
 260         err = seq_open(file, &bpffs_map_seq_ops);
 261         if (err) {
 262                 map_iter_free(iter);
 263                 return err;
 264         }
 265
 266         m = file->private_data;
 267         m->private = iter;
 268
 269         return 0;
 270 }
 271
 272 static int bpffs_map_release(struct inode *inode, struct file *file)
 273 {
 274         struct seq_file *m = file->private_data;
 275
 276         map_iter_free(map_iter(m));
 277
 278         return seq_release(inode, file);
 279 }
 280
 281 /* bpffs_map_fops should only implement the basic
 282  * read operation for a BPF map.  The purpose is to
 283  * provide a simple user intuitive way to do
 284  * "cat bpffs/pathto/a-pinned-map".
 285  *
 286  * Other operations (e.g. write, lookup...) should be realized by
 287  * the userspace tools (e.g. bpftool) through the
 288  * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
 289  * interface.
 290  */
 291 static const struct file_operations bpffs_map_fops = {
 292         .open           = bpffs_map_open,
 293         .read           = seq_read,
 294         .release        = bpffs_map_release,
 295 };
 296
 297 static int bpffs_obj_open(struct inode *inode, struct file *file)
 298 {
 299         return -EIO;
 300 }
 301
 302 static const struct file_operations bpffs_obj_fops = {
 303         .open           = bpffs_obj_open,
 304 };
 305
 306 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
 307                          const struct inode_operations *iops,
 308                          const struct file_operations *fops)
 309 {
 310         struct inode *dir = dentry->d_parent->d_inode;
 311         struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
 312         if (IS_ERR(inode))
 313                 return PTR_ERR(inode);
 314
 315         inode->i_op = iops;
 316         inode->i_fop = fops;
 317         inode->i_private = raw;
 318
 319         bpf_dentry_finalize(dentry, inode, dir);
 320         return 0;
 321 }
 322
 323 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
 324 {
 325         return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
 326                              &bpffs_obj_fops);
 327 }
 328
 329 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
 330 {
 331         struct bpf_map *map = arg;
 332
 333         return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
 334                              bpf_map_support_seq_show(map) ?
 335                              &bpffs_map_fops : &bpffs_obj_fops);
 336 }
 337
 338 static struct dentry *
 339 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
 340 {
 341         /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
 342          * extensions.
 343          */
 344         if (strchr(dentry->d_name.name, '.'))
 345                 return ERR_PTR(-EPERM);
 346
 347         return simple_lookup(dir, dentry, flags);
 348 }
 349
 350 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
 351                        const char *target)
 352 {
 353         char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
 354         struct inode *inode;
 355
 356         if (!link)
 357                 return -ENOMEM;
 358
 359         inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
 360         if (IS_ERR(inode)) {
 361                 kfree(link);
 362                 return PTR_ERR(inode);
 363         }
 364
 365         inode->i_op = &simple_symlink_inode_operations;
 366         inode->i_link = link;
 367
 368         bpf_dentry_finalize(dentry, inode, dir);
 369         return 0;
 370 }
 371
 372 static const struct inode_operations bpf_dir_iops = {
 373         .lookup         = bpf_lookup,
 374         .mkdir          = bpf_mkdir,
 375         .symlink        = bpf_symlink,
 376         .rmdir          = simple_rmdir,
 377         .rename         = simple_rename,
 378         .link           = simple_link,
 379         .unlink         = simple_unlink,
 380 };
 381
 382 static int bpf_obj_do_pin(const char __user *pathname, void *raw,
 383                           enum bpf_type type)
 384 {
 385         struct dentry *dentry;
 386         struct inode *dir;
 387         struct path path;
 388         umode_t mode;
 389         int ret;
 390
 391         dentry = user_path_create(AT_FDCWD, pathname, &path, 0);
 392         if (IS_ERR(dentry))
 393                 return PTR_ERR(dentry);
 394
 395         mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
 396
 397         ret = security_path_mknod(&path, dentry, mode, 0);
 398         if (ret)
 399                 goto out;
 400
 401         dir = d_inode(path.dentry);
 402         if (dir->i_op != &bpf_dir_iops) {
 403                 ret = -EPERM;
 404                 goto out;
 405         }
 406
 407         switch (type) {
 408         case BPF_TYPE_PROG:
 409                 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
 410                 break;
 411         case BPF_TYPE_MAP:
 412                 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
 413                 break;
 414         default:
 415                 ret = -EPERM;
 416         }
 417 out:
 418         done_path_create(&path, dentry);
 419         return ret;
 420 }
 421
 422 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
 423 {
 424         enum bpf_type type;
 425         void *raw;
 426         int ret;
 427
 428         raw = bpf_fd_probe_obj(ufd, &type);
 429         if (IS_ERR(raw))
 430                 return PTR_ERR(raw);
 431
 432         ret = bpf_obj_do_pin(pathname, raw, type);
 433         if (ret != 0)
 434                 bpf_any_put(raw, type);
 435
 436         return ret;
 437 }
 438
 439 static void *bpf_obj_do_get(const char __user *pathname,
 440                             enum bpf_type *type, int flags)
 441 {
 442         struct inode *inode;
 443         struct path path;
 444         void *raw;
 445         int ret;
 446
 447         ret = user_path_at(AT_FDCWD, pathname, LOOKUP_FOLLOW, &path);
 448         if (ret)
 449                 return ERR_PTR(ret);
 450
 451         inode = d_backing_inode(path.dentry);
 452         ret = inode_permission(inode, ACC_MODE(flags));
 453         if (ret)
 454                 goto out;
 455
 456         ret = bpf_inode_type(inode, type);
 457         if (ret)
 458                 goto out;
 459
 460         raw = bpf_any_get(inode->i_private, *type);
 461         if (!IS_ERR(raw))
 462                 touch_atime(&path);
 463
 464         path_put(&path);
 465         return raw;
 466 out:
 467         path_put(&path);
 468         return ERR_PTR(ret);
 469 }
 470
 471 int bpf_obj_get_user(const char __user *pathname, int flags)
 472 {
 473         enum bpf_type type = BPF_TYPE_UNSPEC;
 474         int f_flags;
 475         void *raw;
 476         int ret;
 477
 478         f_flags = bpf_get_file_flag(flags);
 479         if (f_flags < 0)
 480                 return f_flags;
 481
 482         raw = bpf_obj_do_get(pathname, &type, f_flags);
 483         if (IS_ERR(raw))
 484                 return PTR_ERR(raw);
 485
 486         if (type == BPF_TYPE_PROG)
 487                 ret = bpf_prog_new_fd(raw);
 488         else if (type == BPF_TYPE_MAP)
 489                 ret = bpf_map_new_fd(raw, f_flags);
 490         else
 491                 return -ENOENT;
 492
 493         if (ret < 0)
 494                 bpf_any_put(raw, type);
 495         return ret;
 496 }
 497
 498 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
 499 {
 500         struct bpf_prog *prog;
 501         int ret = inode_permission(inode, MAY_READ);
 502         if (ret)
 503                 return ERR_PTR(ret);
 504
 505         if (inode->i_op == &bpf_map_iops)
 506                 return ERR_PTR(-EINVAL);
 507         if (inode->i_op != &bpf_prog_iops)
 508                 return ERR_PTR(-EACCES);
 509
 510         prog = inode->i_private;
 511
 512         ret = security_bpf_prog(prog);
 513         if (ret < 0)
 514                 return ERR_PTR(ret);
 515
 516         if (!bpf_prog_get_ok(prog, &type, false))
 517                 return ERR_PTR(-EINVAL);
 518
 519         bpf_prog_inc(prog);
 520         return prog;
 521 }
 522
 523 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
 524 {
 525         struct bpf_prog *prog;
 526         struct path path;
 527         int ret = kern_path(name, LOOKUP_FOLLOW, &path);
 528         if (ret)
 529                 return ERR_PTR(ret);
 530         prog = __get_prog_inode(d_backing_inode(path.dentry), type);
 531         if (!IS_ERR(prog))
 532                 touch_atime(&path);
 533         path_put(&path);
 534         return prog;
 535 }
 536 EXPORT_SYMBOL(bpf_prog_get_type_path);
 537
 538 /*
 539  * Display the mount options in /proc/mounts.
 540  */
 541 static int bpf_show_options(struct seq_file *m, struct dentry *root)
 542 {
 543         umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
 544
 545         if (mode != S_IRWXUGO)
 546                 seq_printf(m, ",mode=%o", mode);
 547         return 0;
 548 }
 549
 550 static void bpf_free_inode(struct inode *inode)
 551 {
 552         enum bpf_type type;
 553
 554         if (S_ISLNK(inode->i_mode))
 555                 kfree(inode->i_link);
 556         if (!bpf_inode_type(inode, &type))
 557                 bpf_any_put(inode->i_private, type);
 558         free_inode_nonrcu(inode);
 559 }
 560
 561 static const struct super_operations bpf_super_ops = {
 562         .statfs         = simple_statfs,
 563         .drop_inode     = generic_delete_inode,
 564         .show_options   = bpf_show_options,
 565         .free_inode     = bpf_free_inode,
 566 };
 567
 568 enum {
 569         OPT_MODE,
 570 };
 571
 572 static const struct fs_parameter_spec bpf_fs_parameters[] = {
 573         fsparam_u32oct  ("mode",                        OPT_MODE),
 574         {}
 575 };
 576
 577 struct bpf_mount_opts {
 578         umode_t mode;
 579 };
 580
 581 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
 582 {
 583         struct bpf_mount_opts *opts = fc->fs_private;
 584         struct fs_parse_result result;
 585         int opt;
 586
 587         opt = fs_parse(fc, bpf_fs_parameters, param, &result);
 588         if (opt < 0)
 589                 /* We might like to report bad mount options here, but
 590                  * traditionally we've ignored all mount options, so we'd
 591                  * better continue to ignore non-existing options for bpf.
 592                  */
 593                 return opt == -ENOPARAM ? 0 : opt;
 594
 595         switch (opt) {
 596         case OPT_MODE:
 597                 opts->mode = result.uint_32 & S_IALLUGO;
 598                 break;
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
 605 {
 606         static const struct tree_descr bpf_rfiles[] = { { "" } };
 607         struct bpf_mount_opts *opts = fc->fs_private;
 608         struct inode *inode;
 609         int ret;
 610
 611         ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
 612         if (ret)
 613                 return ret;
 614
 615         sb->s_op = &bpf_super_ops;
 616
 617         inode = sb->s_root->d_inode;
 618         inode->i_op = &bpf_dir_iops;
 619         inode->i_mode &= ~S_IALLUGO;
 620         inode->i_mode |= S_ISVTX | opts->mode;
 621
 622         return 0;
 623 }
 624
 625 static int bpf_get_tree(struct fs_context *fc)
 626 {
 627         return get_tree_nodev(fc, bpf_fill_super);
 628 }
 629
 630 static void bpf_free_fc(struct fs_context *fc)
 631 {
 632         kfree(fc->fs_private);
 633 }
 634
 635 static const struct fs_context_operations bpf_context_ops = {
 636         .free           = bpf_free_fc,
 637         .parse_param    = bpf_parse_param,
 638         .get_tree       = bpf_get_tree,
 639 };
 640
 641 /*
 642  * Set up the filesystem mount context.
 643  */
 644 static int bpf_init_fs_context(struct fs_context *fc)
 645 {
 646         struct bpf_mount_opts *opts;
 647
 648         opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
 649         if (!opts)
 650                 return -ENOMEM;
 651
 652         opts->mode = S_IRWXUGO;
 653
 654         fc->fs_private = opts;
 655         fc->ops = &bpf_context_ops;
 656         return 0;
 657 }
 658
 659 static struct file_system_type bpf_fs_type = {
 660         .owner          = THIS_MODULE,
 661         .name           = "bpf",
 662         .init_fs_context = bpf_init_fs_context,
 663         .parameters     = bpf_fs_parameters,
 664         .kill_sb        = kill_litter_super,
 665 };
 666
 667 static int __init bpf_init(void)
 668 {
 669         int ret;
 670
 671         ret = sysfs_create_mount_point(fs_kobj, "bpf");
 672         if (ret)
 673                 return ret;
 674
 675         ret = register_filesystem(&bpf_fs_type);
 676         if (ret)
 677                 sysfs_remove_mount_point(fs_kobj, "bpf");
 678
 679         return ret;
 680 }
 681 fs_initcall(bpf_init);