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/kdev_t.h>
  18 #include <linux/parser.h>
  19 #include <linux/filter.h>
  20 #include <linux/bpf.h>
  21 #include <linux/bpf_trace.h>
  22
  23 enum bpf_type {
  24         BPF_TYPE_UNSPEC = 0,
  25         BPF_TYPE_PROG,
  26         BPF_TYPE_MAP,
  27 };
  28
  29 static void *bpf_any_get(void *raw, enum bpf_type type)
  30 {
  31         switch (type) {
  32         case BPF_TYPE_PROG:
  33                 raw = bpf_prog_inc(raw);
  34                 break;
  35         case BPF_TYPE_MAP:
  36                 raw = bpf_map_inc(raw, true);
  37                 break;
  38         default:
  39                 WARN_ON_ONCE(1);
  40                 break;
  41         }
  42
  43         return raw;
  44 }
  45
  46 static void bpf_any_put(void *raw, enum bpf_type type)
  47 {
  48         switch (type) {
  49         case BPF_TYPE_PROG:
  50                 bpf_prog_put(raw);
  51                 break;
  52         case BPF_TYPE_MAP:
  53                 bpf_map_put_with_uref(raw);
  54                 break;
  55         default:
  56                 WARN_ON_ONCE(1);
  57                 break;
  58         }
  59 }
  60
  61 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
  62 {
  63         void *raw;
  64
  65         *type = BPF_TYPE_MAP;
  66         raw = bpf_map_get_with_uref(ufd);
  67         if (IS_ERR(raw)) {
  68                 *type = BPF_TYPE_PROG;
  69                 raw = bpf_prog_get(ufd);
  70         }
  71
  72         return raw;
  73 }
  74
  75 static const struct inode_operations bpf_dir_iops;
  76
  77 static const struct inode_operations bpf_prog_iops = { };
  78 static const struct inode_operations bpf_map_iops  = { };
  79
  80 static struct inode *bpf_get_inode(struct super_block *sb,
  81                                    const struct inode *dir,
  82                                    umode_t mode)
  83 {
  84         struct inode *inode;
  85
  86         switch (mode & S_IFMT) {
  87         case S_IFDIR:
  88         case S_IFREG:
  89         case S_IFLNK:
  90                 break;
  91         default:
  92                 return ERR_PTR(-EINVAL);
  93         }
  94
  95         inode = new_inode(sb);
  96         if (!inode)
  97                 return ERR_PTR(-ENOSPC);
  98
  99         inode->i_ino = get_next_ino();
 100         inode->i_atime = current_time(inode);
 101         inode->i_mtime = inode->i_atime;
 102         inode->i_ctime = inode->i_atime;
 103
 104         inode_init_owner(inode, dir, mode);
 105
 106         return inode;
 107 }
 108
 109 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
 110 {
 111         *type = BPF_TYPE_UNSPEC;
 112         if (inode->i_op == &bpf_prog_iops)
 113                 *type = BPF_TYPE_PROG;
 114         else if (inode->i_op == &bpf_map_iops)
 115                 *type = BPF_TYPE_MAP;
 116         else
 117                 return -EACCES;
 118
 119         return 0;
 120 }
 121
 122 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
 123                                 struct inode *dir)
 124 {
 125         d_instantiate(dentry, inode);
 126         dget(dentry);
 127
 128         dir->i_mtime = current_time(dir);
 129         dir->i_ctime = dir->i_mtime;
 130 }
 131
 132 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 133 {
 134         struct inode *inode;
 135
 136         inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
 137         if (IS_ERR(inode))
 138                 return PTR_ERR(inode);
 139
 140         inode->i_op = &bpf_dir_iops;
 141         inode->i_fop = &simple_dir_operations;
 142
 143         inc_nlink(inode);
 144         inc_nlink(dir);
 145
 146         bpf_dentry_finalize(dentry, inode, dir);
 147         return 0;
 148 }
 149
 150 struct map_iter {
 151         void *key;
 152         bool done;
 153 };
 154
 155 static struct map_iter *map_iter(struct seq_file *m)
 156 {
 157         return m->private;
 158 }
 159
 160 static struct bpf_map *seq_file_to_map(struct seq_file *m)
 161 {
 162         return file_inode(m->file)->i_private;
 163 }
 164
 165 static void map_iter_free(struct map_iter *iter)
 166 {
 167         if (iter) {
 168                 kfree(iter->key);
 169                 kfree(iter);
 170         }
 171 }
 172
 173 static struct map_iter *map_iter_alloc(struct bpf_map *map)
 174 {
 175         struct map_iter *iter;
 176
 177         iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
 178         if (!iter)
 179                 goto error;
 180
 181         iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
 182         if (!iter->key)
 183                 goto error;
 184
 185         return iter;
 186
 187 error:
 188         map_iter_free(iter);
 189         return NULL;
 190 }
 191
 192 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
 193 {
 194         struct bpf_map *map = seq_file_to_map(m);
 195         void *key = map_iter(m)->key;
 196         void *prev_key;
 197
 198         if (map_iter(m)->done)
 199                 return NULL;
 200
 201         if (unlikely(v == SEQ_START_TOKEN))
 202                 prev_key = NULL;
 203         else
 204                 prev_key = key;
 205
 206         if (map->ops->map_get_next_key(map, prev_key, key)) {
 207                 map_iter(m)->done = true;
 208                 return NULL;
 209         }
 210
 211         ++(*pos);
 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 struct filename *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 = kern_path_create(AT_FDCWD, pathname->name, &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         struct filename *pname;
 425         enum bpf_type type;
 426         void *raw;
 427         int ret;
 428
 429         pname = getname(pathname);
 430         if (IS_ERR(pname))
 431                 return PTR_ERR(pname);
 432
 433         raw = bpf_fd_probe_obj(ufd, &type);
 434         if (IS_ERR(raw)) {
 435                 ret = PTR_ERR(raw);
 436                 goto out;
 437         }
 438
 439         ret = bpf_obj_do_pin(pname, raw, type);
 440         if (ret != 0)
 441                 bpf_any_put(raw, type);
 442 out:
 443         putname(pname);
 444         return ret;
 445 }
 446
 447 static void *bpf_obj_do_get(const struct filename *pathname,
 448                             enum bpf_type *type, int flags)
 449 {
 450         struct inode *inode;
 451         struct path path;
 452         void *raw;
 453         int ret;
 454
 455         ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
 456         if (ret)
 457                 return ERR_PTR(ret);
 458
 459         inode = d_backing_inode(path.dentry);
 460         ret = inode_permission(inode, ACC_MODE(flags));
 461         if (ret)
 462                 goto out;
 463
 464         ret = bpf_inode_type(inode, type);
 465         if (ret)
 466                 goto out;
 467
 468         raw = bpf_any_get(inode->i_private, *type);
 469         if (!IS_ERR(raw))
 470                 touch_atime(&path);
 471
 472         path_put(&path);
 473         return raw;
 474 out:
 475         path_put(&path);
 476         return ERR_PTR(ret);
 477 }
 478
 479 int bpf_obj_get_user(const char __user *pathname, int flags)
 480 {
 481         enum bpf_type type = BPF_TYPE_UNSPEC;
 482         struct filename *pname;
 483         int ret = -ENOENT;
 484         int f_flags;
 485         void *raw;
 486
 487         f_flags = bpf_get_file_flag(flags);
 488         if (f_flags < 0)
 489                 return f_flags;
 490
 491         pname = getname(pathname);
 492         if (IS_ERR(pname))
 493                 return PTR_ERR(pname);
 494
 495         raw = bpf_obj_do_get(pname, &type, f_flags);
 496         if (IS_ERR(raw)) {
 497                 ret = PTR_ERR(raw);
 498                 goto out;
 499         }
 500
 501         if (type == BPF_TYPE_PROG)
 502                 ret = bpf_prog_new_fd(raw);
 503         else if (type == BPF_TYPE_MAP)
 504                 ret = bpf_map_new_fd(raw, f_flags);
 505         else
 506                 goto out;
 507
 508         if (ret < 0)
 509                 bpf_any_put(raw, type);
 510 out:
 511         putname(pname);
 512         return ret;
 513 }
 514
 515 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
 516 {
 517         struct bpf_prog *prog;
 518         int ret = inode_permission(inode, MAY_READ);
 519         if (ret)
 520                 return ERR_PTR(ret);
 521
 522         if (inode->i_op == &bpf_map_iops)
 523                 return ERR_PTR(-EINVAL);
 524         if (inode->i_op != &bpf_prog_iops)
 525                 return ERR_PTR(-EACCES);
 526
 527         prog = inode->i_private;
 528
 529         ret = security_bpf_prog(prog);
 530         if (ret < 0)
 531                 return ERR_PTR(ret);
 532
 533         if (!bpf_prog_get_ok(prog, &type, false))
 534                 return ERR_PTR(-EINVAL);
 535
 536         return bpf_prog_inc(prog);
 537 }
 538
 539 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
 540 {
 541         struct bpf_prog *prog;
 542         struct path path;
 543         int ret = kern_path(name, LOOKUP_FOLLOW, &path);
 544         if (ret)
 545                 return ERR_PTR(ret);
 546         prog = __get_prog_inode(d_backing_inode(path.dentry), type);
 547         if (!IS_ERR(prog))
 548                 touch_atime(&path);
 549         path_put(&path);
 550         return prog;
 551 }
 552 EXPORT_SYMBOL(bpf_prog_get_type_path);
 553
 554 /*
 555  * Display the mount options in /proc/mounts.
 556  */
 557 static int bpf_show_options(struct seq_file *m, struct dentry *root)
 558 {
 559         umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
 560
 561         if (mode != S_IRWXUGO)
 562                 seq_printf(m, ",mode=%o", mode);
 563         return 0;
 564 }
 565
 566 static void bpf_free_inode(struct inode *inode)
 567 {
 568         enum bpf_type type;
 569
 570         if (S_ISLNK(inode->i_mode))
 571                 kfree(inode->i_link);
 572         if (!bpf_inode_type(inode, &type))
 573                 bpf_any_put(inode->i_private, type);
 574         free_inode_nonrcu(inode);
 575 }
 576
 577 static const struct super_operations bpf_super_ops = {
 578         .statfs         = simple_statfs,
 579         .drop_inode     = generic_delete_inode,
 580         .show_options   = bpf_show_options,
 581         .free_inode     = bpf_free_inode,
 582 };
 583
 584 enum {
 585         OPT_MODE,
 586         OPT_ERR,
 587 };
 588
 589 static const match_table_t bpf_mount_tokens = {
 590         { OPT_MODE, "mode=%o" },
 591         { OPT_ERR, NULL },
 592 };
 593
 594 struct bpf_mount_opts {
 595         umode_t mode;
 596 };
 597
 598 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
 599 {
 600         substring_t args[MAX_OPT_ARGS];
 601         int option, token;
 602         char *ptr;
 603
 604         opts->mode = S_IRWXUGO;
 605
 606         while ((ptr = strsep(&data, ",")) != NULL) {
 607                 if (!*ptr)
 608                         continue;
 609
 610                 token = match_token(ptr, bpf_mount_tokens, args);
 611                 switch (token) {
 612                 case OPT_MODE:
 613                         if (match_octal(&args[0], &option))
 614                                 return -EINVAL;
 615                         opts->mode = option & S_IALLUGO;
 616                         break;
 617                 /* We might like to report bad mount options here, but
 618                  * traditionally we've ignored all mount options, so we'd
 619                  * better continue to ignore non-existing options for bpf.
 620                  */
 621                 }
 622         }
 623
 624         return 0;
 625 }
 626
 627 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
 628 {
 629         static const struct tree_descr bpf_rfiles[] = { { "" } };
 630         struct bpf_mount_opts opts;
 631         struct inode *inode;
 632         int ret;
 633
 634         ret = bpf_parse_options(data, &opts);
 635         if (ret)
 636                 return ret;
 637
 638         ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
 639         if (ret)
 640                 return ret;
 641
 642         sb->s_op = &bpf_super_ops;
 643
 644         inode = sb->s_root->d_inode;
 645         inode->i_op = &bpf_dir_iops;
 646         inode->i_mode &= ~S_IALLUGO;
 647         inode->i_mode |= S_ISVTX | opts.mode;
 648
 649         return 0;
 650 }
 651
 652 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
 653                                 const char *dev_name, void *data)
 654 {
 655         return mount_nodev(type, flags, data, bpf_fill_super);
 656 }
 657
 658 static struct file_system_type bpf_fs_type = {
 659         .owner          = THIS_MODULE,
 660         .name           = "bpf",
 661         .mount          = bpf_mount,
 662         .kill_sb        = kill_litter_super,
 663 };
 664
 665 static int __init bpf_init(void)
 666 {
 667         int ret;
 668
 669         ret = sysfs_create_mount_point(fs_kobj, "bpf");
 670         if (ret)
 671                 return ret;
 672
 673         ret = register_filesystem(&bpf_fs_type);
 674         if (ret)
 675                 sysfs_remove_mount_point(fs_kobj, "bpf");
 676
 677         return ret;
 678 }
 679 fs_initcall(bpf_init);