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 inode *dir, struct dentry *dentry,
 154                          umode_t mode, const struct inode_operations *iops)
 155 {
 156         struct inode *inode;
 157
 158         inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
 159         if (IS_ERR(inode))
 160                 return PTR_ERR(inode);
 161
 162         inode->i_op = iops;
 163         inode->i_private = dentry->d_fsdata;
 164
 165         bpf_dentry_finalize(dentry, inode, dir);
 166         return 0;
 167 }
 168
 169 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
 170                      dev_t devt)
 171 {
 172         enum bpf_type type = MINOR(devt);
 173
 174         if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
 175             dentry->d_fsdata == NULL)
 176                 return -EPERM;
 177
 178         switch (type) {
 179         case BPF_TYPE_PROG:
 180                 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
 181         case BPF_TYPE_MAP:
 182                 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
 183         default:
 184                 return -EPERM;
 185         }
 186 }
 187
 188 static struct dentry *
 189 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
 190 {
 191         if (strchr(dentry->d_name.name, '.'))
 192                 return ERR_PTR(-EPERM);
 193
 194         return simple_lookup(dir, dentry, flags);
 195 }
 196
 197 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
 198                        const char *target)
 199 {
 200         char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
 201         struct inode *inode;
 202
 203         if (!link)
 204                 return -ENOMEM;
 205
 206         inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
 207         if (IS_ERR(inode)) {
 208                 kfree(link);
 209                 return PTR_ERR(inode);
 210         }
 211
 212         inode->i_op = &simple_symlink_inode_operations;
 213         inode->i_link = link;
 214
 215         bpf_dentry_finalize(dentry, inode, dir);
 216         return 0;
 217 }
 218
 219 static const struct inode_operations bpf_dir_iops = {
 220         .lookup         = bpf_lookup,
 221         .mknod          = bpf_mkobj,
 222         .mkdir          = bpf_mkdir,
 223         .symlink        = bpf_symlink,
 224         .rmdir          = simple_rmdir,
 225         .rename         = simple_rename,
 226         .link           = simple_link,
 227         .unlink         = simple_unlink,
 228 };
 229
 230 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
 231                           enum bpf_type type)
 232 {
 233         struct dentry *dentry;
 234         struct inode *dir;
 235         struct path path;
 236         umode_t mode;
 237         dev_t devt;
 238         int ret;
 239
 240         dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
 241         if (IS_ERR(dentry))
 242                 return PTR_ERR(dentry);
 243
 244         mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
 245         devt = MKDEV(UNNAMED_MAJOR, type);
 246
 247         ret = security_path_mknod(&path, dentry, mode, devt);
 248         if (ret)
 249                 goto out;
 250
 251         dir = d_inode(path.dentry);
 252         if (dir->i_op != &bpf_dir_iops) {
 253                 ret = -EPERM;
 254                 goto out;
 255         }
 256
 257         dentry->d_fsdata = raw;
 258         ret = vfs_mknod(dir, dentry, mode, devt);
 259         dentry->d_fsdata = NULL;
 260 out:
 261         done_path_create(&path, dentry);
 262         return ret;
 263 }
 264
 265 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
 266 {
 267         struct filename *pname;
 268         enum bpf_type type;
 269         void *raw;
 270         int ret;
 271
 272         pname = getname(pathname);
 273         if (IS_ERR(pname))
 274                 return PTR_ERR(pname);
 275
 276         raw = bpf_fd_probe_obj(ufd, &type);
 277         if (IS_ERR(raw)) {
 278                 ret = PTR_ERR(raw);
 279                 goto out;
 280         }
 281
 282         ret = bpf_obj_do_pin(pname, raw, type);
 283         if (ret != 0)
 284                 bpf_any_put(raw, type);
 285         if ((trace_bpf_obj_pin_prog_enabled() ||
 286              trace_bpf_obj_pin_map_enabled()) && !ret) {
 287                 if (type == BPF_TYPE_PROG)
 288                         trace_bpf_obj_pin_prog(raw, ufd, pname);
 289                 if (type == BPF_TYPE_MAP)
 290                         trace_bpf_obj_pin_map(raw, ufd, pname);
 291         }
 292 out:
 293         putname(pname);
 294         return ret;
 295 }
 296
 297 static void *bpf_obj_do_get(const struct filename *pathname,
 298                             enum bpf_type *type, int flags)
 299 {
 300         struct inode *inode;
 301         struct path path;
 302         void *raw;
 303         int ret;
 304
 305         ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
 306         if (ret)
 307                 return ERR_PTR(ret);
 308
 309         inode = d_backing_inode(path.dentry);
 310         ret = inode_permission(inode, ACC_MODE(flags));
 311         if (ret)
 312                 goto out;
 313
 314         ret = bpf_inode_type(inode, type);
 315         if (ret)
 316                 goto out;
 317
 318         raw = bpf_any_get(inode->i_private, *type);
 319         if (!IS_ERR(raw))
 320                 touch_atime(&path);
 321
 322         path_put(&path);
 323         return raw;
 324 out:
 325         path_put(&path);
 326         return ERR_PTR(ret);
 327 }
 328
 329 int bpf_obj_get_user(const char __user *pathname, int flags)
 330 {
 331         enum bpf_type type = BPF_TYPE_UNSPEC;
 332         struct filename *pname;
 333         int ret = -ENOENT;
 334         int f_flags;
 335         void *raw;
 336
 337         f_flags = bpf_get_file_flag(flags);
 338         if (f_flags < 0)
 339                 return f_flags;
 340
 341         pname = getname(pathname);
 342         if (IS_ERR(pname))
 343                 return PTR_ERR(pname);
 344
 345         raw = bpf_obj_do_get(pname, &type, f_flags);
 346         if (IS_ERR(raw)) {
 347                 ret = PTR_ERR(raw);
 348                 goto out;
 349         }
 350
 351         if (type == BPF_TYPE_PROG)
 352                 ret = bpf_prog_new_fd(raw);
 353         else if (type == BPF_TYPE_MAP)
 354                 ret = bpf_map_new_fd(raw, f_flags);
 355         else
 356                 goto out;
 357
 358         if (ret < 0) {
 359                 bpf_any_put(raw, type);
 360         } else if (trace_bpf_obj_get_prog_enabled() ||
 361                    trace_bpf_obj_get_map_enabled()) {
 362                 if (type == BPF_TYPE_PROG)
 363                         trace_bpf_obj_get_prog(raw, ret, pname);
 364                 if (type == BPF_TYPE_MAP)
 365                         trace_bpf_obj_get_map(raw, ret, pname);
 366         }
 367 out:
 368         putname(pname);
 369         return ret;
 370 }
 371 EXPORT_SYMBOL_GPL(bpf_obj_get_user);
 372
 373 static void bpf_evict_inode(struct inode *inode)
 374 {
 375         enum bpf_type type;
 376
 377         truncate_inode_pages_final(&inode->i_data);
 378         clear_inode(inode);
 379
 380         if (S_ISLNK(inode->i_mode))
 381                 kfree(inode->i_link);
 382         if (!bpf_inode_type(inode, &type))
 383                 bpf_any_put(inode->i_private, type);
 384 }
 385
 386 /*
 387  * Display the mount options in /proc/mounts.
 388  */
 389 static int bpf_show_options(struct seq_file *m, struct dentry *root)
 390 {
 391         umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
 392
 393         if (mode != S_IRWXUGO)
 394                 seq_printf(m, ",mode=%o", mode);
 395         return 0;
 396 }
 397
 398 static const struct super_operations bpf_super_ops = {
 399         .statfs         = simple_statfs,
 400         .drop_inode     = generic_delete_inode,
 401         .show_options   = bpf_show_options,
 402         .evict_inode    = bpf_evict_inode,
 403 };
 404
 405 enum {
 406         OPT_MODE,
 407         OPT_ERR,
 408 };
 409
 410 static const match_table_t bpf_mount_tokens = {
 411         { OPT_MODE, "mode=%o" },
 412         { OPT_ERR, NULL },
 413 };
 414
 415 struct bpf_mount_opts {
 416         umode_t mode;
 417 };
 418
 419 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
 420 {
 421         substring_t args[MAX_OPT_ARGS];
 422         int option, token;
 423         char *ptr;
 424
 425         opts->mode = S_IRWXUGO;
 426
 427         while ((ptr = strsep(&data, ",")) != NULL) {
 428                 if (!*ptr)
 429                         continue;
 430
 431                 token = match_token(ptr, bpf_mount_tokens, args);
 432                 switch (token) {
 433                 case OPT_MODE:
 434                         if (match_octal(&args[0], &option))
 435                                 return -EINVAL;
 436                         opts->mode = option & S_IALLUGO;
 437                         break;
 438                 /* We might like to report bad mount options here, but
 439                  * traditionally we've ignored all mount options, so we'd
 440                  * better continue to ignore non-existing options for bpf.
 441                  */
 442                 }
 443         }
 444
 445         return 0;
 446 }
 447
 448 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
 449 {
 450         static const struct tree_descr bpf_rfiles[] = { { "" } };
 451         struct bpf_mount_opts opts;
 452         struct inode *inode;
 453         int ret;
 454
 455         ret = bpf_parse_options(data, &opts);
 456         if (ret)
 457                 return ret;
 458
 459         ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
 460         if (ret)
 461                 return ret;
 462
 463         sb->s_op = &bpf_super_ops;
 464
 465         inode = sb->s_root->d_inode;
 466         inode->i_op = &bpf_dir_iops;
 467         inode->i_mode &= ~S_IALLUGO;
 468         inode->i_mode |= S_ISVTX | opts.mode;
 469
 470         return 0;
 471 }
 472
 473 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
 474                                 const char *dev_name, void *data)
 475 {
 476         return mount_nodev(type, flags, data, bpf_fill_super);
 477 }
 478
 479 static struct file_system_type bpf_fs_type = {
 480         .owner          = THIS_MODULE,
 481         .name           = "bpf",
 482         .mount          = bpf_mount,
 483         .kill_sb        = kill_litter_super,
 484 };
 485
 486 static int __init bpf_init(void)
 487 {
 488         int ret;
 489
 490         ret = sysfs_create_mount_point(fs_kobj, "bpf");
 491         if (ret)
 492                 return ret;
 493
 494         ret = register_filesystem(&bpf_fs_type);
 495         if (ret)
 496                 sysfs_remove_mount_point(fs_kobj, "bpf");
 497
 498         return ret;
 499 }
 500 fs_initcall(bpf_init);