kernel/bpf/inode.c

   1 /*
   2  * Minimal file system backend for holding eBPF maps and programs,
   3  * used by bpf(2) object pinning.
   4  *
   5  * Authors:
   6  *
   7  *      Daniel Borkmann <daniel@iogearbox.net>
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License
  11  * version 2 as published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/init.h>
  15 #include <linux/magic.h>
  16 #include <linux/major.h>
  17 #include <linux/mount.h>
  18 #include <linux/namei.h>
  19 #include <linux/fs.h>
  20 #include <linux/kdev_t.h>
  21 #include <linux/parser.h>
  22 #include <linux/filter.h>
  23 #include <linux/bpf.h>
  24 #include <linux/bpf_trace.h>
  25
  26 enum bpf_type {
  27         BPF_TYPE_UNSPEC = 0,
  28         BPF_TYPE_PROG,
  29         BPF_TYPE_MAP,
  30 };
  31
  32 static void *bpf_any_get(void *raw, enum bpf_type type)
  33 {
  34         switch (type) {
  35         case BPF_TYPE_PROG:
  36                 raw = bpf_prog_inc(raw);
  37                 break;
  38         case BPF_TYPE_MAP:
  39                 raw = bpf_map_inc(raw, true);
  40                 break;
  41         default:
  42                 WARN_ON_ONCE(1);
  43                 break;
  44         }
  45
  46         return raw;
  47 }
  48
  49 static void bpf_any_put(void *raw, enum bpf_type type)
  50 {
  51         switch (type) {
  52         case BPF_TYPE_PROG:
  53                 bpf_prog_put(raw);
  54                 break;
  55         case BPF_TYPE_MAP:
  56                 bpf_map_put_with_uref(raw);
  57                 break;
  58         default:
  59                 WARN_ON_ONCE(1);
  60                 break;
  61         }
  62 }
  63
  64 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
  65 {
  66         void *raw;
  67
  68         *type = BPF_TYPE_MAP;
  69         raw = bpf_map_get_with_uref(ufd);
  70         if (IS_ERR(raw)) {
  71                 *type = BPF_TYPE_PROG;
  72                 raw = bpf_prog_get(ufd);
  73         }
  74
  75         return raw;
  76 }
  77
  78 static const struct inode_operations bpf_dir_iops;
  79
  80 static const struct inode_operations bpf_prog_iops = { };
  81 static const struct inode_operations bpf_map_iops  = { };
  82
  83 static struct inode *bpf_get_inode(struct super_block *sb,
  84                                    const struct inode *dir,
  85                                    umode_t mode)
  86 {
  87         struct inode *inode;
  88
  89         switch (mode & S_IFMT) {
  90         case S_IFDIR:
  91         case S_IFREG:
  92         case S_IFLNK:
  93                 break;
  94         default:
  95                 return ERR_PTR(-EINVAL);
  96         }
  97
  98         inode = new_inode(sb);
  99         if (!inode)
 100                 return ERR_PTR(-ENOSPC);
 101
 102         inode->i_ino = get_next_ino();
 103         inode->i_atime = current_time(inode);
 104         inode->i_mtime = inode->i_atime;
 105         inode->i_ctime = inode->i_atime;
 106
 107         inode_init_owner(inode, dir, mode);
 108
 109         return inode;
 110 }
 111
 112 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
 113 {
 114         *type = BPF_TYPE_UNSPEC;
 115         if (inode->i_op == &bpf_prog_iops)
 116                 *type = BPF_TYPE_PROG;
 117         else if (inode->i_op == &bpf_map_iops)
 118                 *type = BPF_TYPE_MAP;
 119         else
 120                 return -EACCES;
 121
 122         return 0;
 123 }
 124
 125 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
 126                                 struct inode *dir)
 127 {
 128         d_instantiate(dentry, inode);
 129         dget(dentry);
 130
 131         dir->i_mtime = current_time(dir);
 132         dir->i_ctime = dir->i_mtime;
 133 }
 134
 135 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 136 {
 137         struct inode *inode;
 138
 139         inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
 140         if (IS_ERR(inode))
 141                 return PTR_ERR(inode);
 142
 143         inode->i_op = &bpf_dir_iops;
 144         inode->i_fop = &simple_dir_operations;
 145
 146         inc_nlink(inode);
 147         inc_nlink(dir);
 148
 149         bpf_dentry_finalize(dentry, inode, dir);
 150         return 0;
 151 }
 152
 153 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
 154                          const struct inode_operations *iops)
 155 {
 156         struct inode *dir = dentry->d_parent->d_inode;
 157         struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
 158         if (IS_ERR(inode))
 159                 return PTR_ERR(inode);
 160
 161         inode->i_op = iops;
 162         inode->i_private = raw;
 163
 164         bpf_dentry_finalize(dentry, inode, dir);
 165         return 0;
 166 }
 167
 168 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
 169 {
 170         return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops);
 171 }
 172
 173 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
 174 {
 175         return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops);
 176 }
 177
 178 static struct dentry *
 179 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
 180 {
 181         /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
 182          * extensions.
 183          */
 184         if (strchr(dentry->d_name.name, '.'))
 185                 return ERR_PTR(-EPERM);
 186
 187         return simple_lookup(dir, dentry, flags);
 188 }
 189
 190 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
 191                        const char *target)
 192 {
 193         char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
 194         struct inode *inode;
 195
 196         if (!link)
 197                 return -ENOMEM;
 198
 199         inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
 200         if (IS_ERR(inode)) {
 201                 kfree(link);
 202                 return PTR_ERR(inode);
 203         }
 204
 205         inode->i_op = &simple_symlink_inode_operations;
 206         inode->i_link = link;
 207
 208         bpf_dentry_finalize(dentry, inode, dir);
 209         return 0;
 210 }
 211
 212 static const struct inode_operations bpf_dir_iops = {
 213         .lookup         = bpf_lookup,
 214         .mkdir          = bpf_mkdir,
 215         .symlink        = bpf_symlink,
 216         .rmdir          = simple_rmdir,
 217         .rename         = simple_rename,
 218         .link           = simple_link,
 219         .unlink         = simple_unlink,
 220 };
 221
 222 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
 223                           enum bpf_type type)
 224 {
 225         struct dentry *dentry;
 226         struct inode *dir;
 227         struct path path;
 228         umode_t mode;
 229         int ret;
 230
 231         dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
 232         if (IS_ERR(dentry))
 233                 return PTR_ERR(dentry);
 234
 235         mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
 236
 237         ret = security_path_mknod(&path, dentry, mode, 0);
 238         if (ret)
 239                 goto out;
 240
 241         dir = d_inode(path.dentry);
 242         if (dir->i_op != &bpf_dir_iops) {
 243                 ret = -EPERM;
 244                 goto out;
 245         }
 246
 247         switch (type) {
 248         case BPF_TYPE_PROG:
 249                 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
 250                 break;
 251         case BPF_TYPE_MAP:
 252                 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
 253                 break;
 254         default:
 255                 ret = -EPERM;
 256         }
 257 out:
 258         done_path_create(&path, dentry);
 259         return ret;
 260 }
 261
 262 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
 263 {
 264         struct filename *pname;
 265         enum bpf_type type;
 266         void *raw;
 267         int ret;
 268
 269         pname = getname(pathname);
 270         if (IS_ERR(pname))
 271                 return PTR_ERR(pname);
 272
 273         raw = bpf_fd_probe_obj(ufd, &type);
 274         if (IS_ERR(raw)) {
 275                 ret = PTR_ERR(raw);
 276                 goto out;
 277         }
 278
 279         ret = bpf_obj_do_pin(pname, raw, type);
 280         if (ret != 0)
 281                 bpf_any_put(raw, type);
 282         if ((trace_bpf_obj_pin_prog_enabled() ||
 283              trace_bpf_obj_pin_map_enabled()) && !ret) {
 284                 if (type == BPF_TYPE_PROG)
 285                         trace_bpf_obj_pin_prog(raw, ufd, pname);
 286                 if (type == BPF_TYPE_MAP)
 287                         trace_bpf_obj_pin_map(raw, ufd, pname);
 288         }
 289 out:
 290         putname(pname);
 291         return ret;
 292 }
 293
 294 static void *bpf_obj_do_get(const struct filename *pathname,
 295                             enum bpf_type *type, int flags)
 296 {
 297         struct inode *inode;
 298         struct path path;
 299         void *raw;
 300         int ret;
 301
 302         ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
 303         if (ret)
 304                 return ERR_PTR(ret);
 305
 306         inode = d_backing_inode(path.dentry);
 307         ret = inode_permission(inode, ACC_MODE(flags));
 308         if (ret)
 309                 goto out;
 310
 311         ret = bpf_inode_type(inode, type);
 312         if (ret)
 313                 goto out;
 314
 315         raw = bpf_any_get(inode->i_private, *type);
 316         if (!IS_ERR(raw))
 317                 touch_atime(&path);
 318
 319         path_put(&path);
 320         return raw;
 321 out:
 322         path_put(&path);
 323         return ERR_PTR(ret);
 324 }
 325
 326 int bpf_obj_get_user(const char __user *pathname, int flags)
 327 {
 328         enum bpf_type type = BPF_TYPE_UNSPEC;
 329         struct filename *pname;
 330         int ret = -ENOENT;
 331         int f_flags;
 332         void *raw;
 333
 334         f_flags = bpf_get_file_flag(flags);
 335         if (f_flags < 0)
 336                 return f_flags;
 337
 338         pname = getname(pathname);
 339         if (IS_ERR(pname))
 340                 return PTR_ERR(pname);
 341
 342         raw = bpf_obj_do_get(pname, &type, f_flags);
 343         if (IS_ERR(raw)) {
 344                 ret = PTR_ERR(raw);
 345                 goto out;
 346         }
 347
 348         if (type == BPF_TYPE_PROG)
 349                 ret = bpf_prog_new_fd(raw);
 350         else if (type == BPF_TYPE_MAP)
 351                 ret = bpf_map_new_fd(raw, f_flags);
 352         else
 353                 goto out;
 354
 355         if (ret < 0) {
 356                 bpf_any_put(raw, type);
 357         } else if (trace_bpf_obj_get_prog_enabled() ||
 358                    trace_bpf_obj_get_map_enabled()) {
 359                 if (type == BPF_TYPE_PROG)
 360                         trace_bpf_obj_get_prog(raw, ret, pname);
 361                 if (type == BPF_TYPE_MAP)
 362                         trace_bpf_obj_get_map(raw, ret, pname);
 363         }
 364 out:
 365         putname(pname);
 366         return ret;
 367 }
 368
 369 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
 370 {
 371         struct bpf_prog *prog;
 372         int ret = inode_permission(inode, MAY_READ | MAY_WRITE);
 373         if (ret)
 374                 return ERR_PTR(ret);
 375
 376         if (inode->i_op == &bpf_map_iops)
 377                 return ERR_PTR(-EINVAL);
 378         if (inode->i_op != &bpf_prog_iops)
 379                 return ERR_PTR(-EACCES);
 380
 381         prog = inode->i_private;
 382
 383         ret = security_bpf_prog(prog);
 384         if (ret < 0)
 385                 return ERR_PTR(ret);
 386
 387         if (!bpf_prog_get_ok(prog, &type, false))
 388                 return ERR_PTR(-EINVAL);
 389
 390         return bpf_prog_inc(prog);
 391 }
 392
 393 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
 394 {
 395         struct bpf_prog *prog;
 396         struct path path;
 397         int ret = kern_path(name, LOOKUP_FOLLOW, &path);
 398         if (ret)
 399                 return ERR_PTR(ret);
 400         prog = __get_prog_inode(d_backing_inode(path.dentry), type);
 401         if (!IS_ERR(prog))
 402                 touch_atime(&path);
 403         path_put(&path);
 404         return prog;
 405 }
 406 EXPORT_SYMBOL(bpf_prog_get_type_path);
 407
 408 static void bpf_evict_inode(struct inode *inode)
 409 {
 410         enum bpf_type type;
 411
 412         truncate_inode_pages_final(&inode->i_data);
 413         clear_inode(inode);
 414
 415         if (S_ISLNK(inode->i_mode))
 416                 kfree(inode->i_link);
 417         if (!bpf_inode_type(inode, &type))
 418                 bpf_any_put(inode->i_private, type);
 419 }
 420
 421 /*
 422  * Display the mount options in /proc/mounts.
 423  */
 424 static int bpf_show_options(struct seq_file *m, struct dentry *root)
 425 {
 426         umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
 427
 428         if (mode != S_IRWXUGO)
 429                 seq_printf(m, ",mode=%o", mode);
 430         return 0;
 431 }
 432
 433 static const struct super_operations bpf_super_ops = {
 434         .statfs         = simple_statfs,
 435         .drop_inode     = generic_delete_inode,
 436         .show_options   = bpf_show_options,
 437         .evict_inode    = bpf_evict_inode,
 438 };
 439
 440 enum {
 441         OPT_MODE,
 442         OPT_ERR,
 443 };
 444
 445 static const match_table_t bpf_mount_tokens = {
 446         { OPT_MODE, "mode=%o" },
 447         { OPT_ERR, NULL },
 448 };
 449
 450 struct bpf_mount_opts {
 451         umode_t mode;
 452 };
 453
 454 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
 455 {
 456         substring_t args[MAX_OPT_ARGS];
 457         int option, token;
 458         char *ptr;
 459
 460         opts->mode = S_IRWXUGO;
 461
 462         while ((ptr = strsep(&data, ",")) != NULL) {
 463                 if (!*ptr)
 464                         continue;
 465
 466                 token = match_token(ptr, bpf_mount_tokens, args);
 467                 switch (token) {
 468                 case OPT_MODE:
 469                         if (match_octal(&args[0], &option))
 470                                 return -EINVAL;
 471                         opts->mode = option & S_IALLUGO;
 472                         break;
 473                 /* We might like to report bad mount options here, but
 474                  * traditionally we've ignored all mount options, so we'd
 475                  * better continue to ignore non-existing options for bpf.
 476                  */
 477                 }
 478         }
 479
 480         return 0;
 481 }
 482
 483 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
 484 {
 485         static const struct tree_descr bpf_rfiles[] = { { "" } };
 486         struct bpf_mount_opts opts;
 487         struct inode *inode;
 488         int ret;
 489
 490         ret = bpf_parse_options(data, &opts);
 491         if (ret)
 492                 return ret;
 493
 494         ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
 495         if (ret)
 496                 return ret;
 497
 498         sb->s_op = &bpf_super_ops;
 499
 500         inode = sb->s_root->d_inode;
 501         inode->i_op = &bpf_dir_iops;
 502         inode->i_mode &= ~S_IALLUGO;
 503         inode->i_mode |= S_ISVTX | opts.mode;
 504
 505         return 0;
 506 }
 507
 508 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
 509                                 const char *dev_name, void *data)
 510 {
 511         return mount_nodev(type, flags, data, bpf_fill_super);
 512 }
 513
 514 static struct file_system_type bpf_fs_type = {
 515         .owner          = THIS_MODULE,
 516         .name           = "bpf",
 517         .mount          = bpf_mount,
 518         .kill_sb        = kill_litter_super,
 519 };
 520
 521 static int __init bpf_init(void)
 522 {
 523         int ret;
 524
 525         ret = sysfs_create_mount_point(fs_kobj, "bpf");
 526         if (ret)
 527                 return ret;
 528
 529         ret = register_filesystem(&bpf_fs_type);
 530         if (ret)
 531                 sysfs_remove_mount_point(fs_kobj, "bpf");
 532
 533         return ret;
 534 }
 535 fs_initcall(bpf_init);