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/filter.h> 22 #include <linux/bpf.h> 23 24 enum bpf_type { 25 BPF_TYPE_UNSPEC = 0, 26 BPF_TYPE_PROG, 27 BPF_TYPE_MAP, 28 }; 29 bpf_any_get(void * raw,enum bpf_type type)30 static void *bpf_any_get(void *raw, enum bpf_type type) 31 { 32 switch (type) { 33 case BPF_TYPE_PROG: 34 raw = bpf_prog_inc(raw); 35 break; 36 case BPF_TYPE_MAP: 37 raw = bpf_map_inc(raw, true); 38 break; 39 default: 40 WARN_ON_ONCE(1); 41 break; 42 } 43 44 return raw; 45 } 46 bpf_any_put(void * raw,enum bpf_type type)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 bpf_fd_probe_obj(u32 ufd,enum bpf_type * type)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 bpf_get_inode(struct super_block * sb,const struct inode * dir,umode_t mode)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 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 bpf_inode_type(const struct inode * inode,enum bpf_type * type)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 bpf_mkdir(struct inode * dir,struct dentry * dentry,umode_t mode)122 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 123 { 124 struct inode *inode; 125 126 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR); 127 if (IS_ERR(inode)) 128 return PTR_ERR(inode); 129 130 inode->i_op = &bpf_dir_iops; 131 inode->i_fop = &simple_dir_operations; 132 133 inc_nlink(inode); 134 inc_nlink(dir); 135 136 d_instantiate(dentry, inode); 137 dget(dentry); 138 139 return 0; 140 } 141 bpf_mkobj_ops(struct inode * dir,struct dentry * dentry,umode_t mode,const struct inode_operations * iops)142 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry, 143 umode_t mode, const struct inode_operations *iops) 144 { 145 struct inode *inode; 146 147 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG); 148 if (IS_ERR(inode)) 149 return PTR_ERR(inode); 150 151 inode->i_op = iops; 152 inode->i_private = dentry->d_fsdata; 153 154 d_instantiate(dentry, inode); 155 dget(dentry); 156 157 return 0; 158 } 159 bpf_mkobj(struct inode * dir,struct dentry * dentry,umode_t mode,dev_t devt)160 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode, 161 dev_t devt) 162 { 163 enum bpf_type type = MINOR(devt); 164 165 if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) || 166 dentry->d_fsdata == NULL) 167 return -EPERM; 168 169 switch (type) { 170 case BPF_TYPE_PROG: 171 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops); 172 case BPF_TYPE_MAP: 173 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops); 174 default: 175 return -EPERM; 176 } 177 } 178 179 static struct dentry * bpf_lookup(struct inode * dir,struct dentry * dentry,unsigned flags)180 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags) 181 { 182 if (strchr(dentry->d_name.name, '.')) 183 return ERR_PTR(-EPERM); 184 return simple_lookup(dir, dentry, flags); 185 } 186 187 static const struct inode_operations bpf_dir_iops = { 188 .lookup = bpf_lookup, 189 .mknod = bpf_mkobj, 190 .mkdir = bpf_mkdir, 191 .rmdir = simple_rmdir, 192 .rename = simple_rename, 193 .link = simple_link, 194 .unlink = simple_unlink, 195 }; 196 bpf_obj_do_pin(const struct filename * pathname,void * raw,enum bpf_type type)197 static int bpf_obj_do_pin(const struct filename *pathname, void *raw, 198 enum bpf_type type) 199 { 200 struct dentry *dentry; 201 struct inode *dir; 202 struct path path; 203 umode_t mode; 204 dev_t devt; 205 int ret; 206 207 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0); 208 if (IS_ERR(dentry)) 209 return PTR_ERR(dentry); 210 211 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask()); 212 devt = MKDEV(UNNAMED_MAJOR, type); 213 214 ret = security_path_mknod(&path, dentry, mode, devt); 215 if (ret) 216 goto out; 217 218 dir = d_inode(path.dentry); 219 if (dir->i_op != &bpf_dir_iops) { 220 ret = -EPERM; 221 goto out; 222 } 223 224 dentry->d_fsdata = raw; 225 ret = vfs_mknod(dir, dentry, mode, devt); 226 dentry->d_fsdata = NULL; 227 out: 228 done_path_create(&path, dentry); 229 return ret; 230 } 231 bpf_obj_pin_user(u32 ufd,const char __user * pathname)232 int bpf_obj_pin_user(u32 ufd, const char __user *pathname) 233 { 234 struct filename *pname; 235 enum bpf_type type; 236 void *raw; 237 int ret; 238 239 pname = getname(pathname); 240 if (IS_ERR(pname)) 241 return PTR_ERR(pname); 242 243 raw = bpf_fd_probe_obj(ufd, &type); 244 if (IS_ERR(raw)) { 245 ret = PTR_ERR(raw); 246 goto out; 247 } 248 249 ret = bpf_obj_do_pin(pname, raw, type); 250 if (ret != 0) 251 bpf_any_put(raw, type); 252 out: 253 putname(pname); 254 return ret; 255 } 256 bpf_obj_do_get(const struct filename * pathname,enum bpf_type * type,int flags)257 static void *bpf_obj_do_get(const struct filename *pathname, 258 enum bpf_type *type, int flags) 259 { 260 struct inode *inode; 261 struct path path; 262 void *raw; 263 int ret; 264 265 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path); 266 if (ret) 267 return ERR_PTR(ret); 268 269 inode = d_backing_inode(path.dentry); 270 ret = inode_permission(inode, ACC_MODE(flags)); 271 if (ret) 272 goto out; 273 274 ret = bpf_inode_type(inode, type); 275 if (ret) 276 goto out; 277 278 raw = bpf_any_get(inode->i_private, *type); 279 if (!IS_ERR(raw)) 280 touch_atime(&path); 281 282 path_put(&path); 283 return raw; 284 out: 285 path_put(&path); 286 return ERR_PTR(ret); 287 } 288 bpf_obj_get_user(const char __user * pathname,int flags)289 int bpf_obj_get_user(const char __user *pathname, int flags) 290 { 291 enum bpf_type type = BPF_TYPE_UNSPEC; 292 struct filename *pname; 293 int ret = -ENOENT; 294 int f_flags; 295 void *raw; 296 297 f_flags = bpf_get_file_flag(flags); 298 if (f_flags < 0) 299 return f_flags; 300 301 pname = getname(pathname); 302 if (IS_ERR(pname)) 303 return PTR_ERR(pname); 304 305 raw = bpf_obj_do_get(pname, &type, f_flags); 306 if (IS_ERR(raw)) { 307 ret = PTR_ERR(raw); 308 goto out; 309 } 310 311 if (type == BPF_TYPE_PROG) 312 ret = bpf_prog_new_fd(raw); 313 else if (type == BPF_TYPE_MAP) 314 ret = bpf_map_new_fd(raw, f_flags); 315 else 316 goto out; 317 318 if (ret < 0) 319 bpf_any_put(raw, type); 320 out: 321 putname(pname); 322 return ret; 323 } 324 __get_prog_inode(struct inode * inode,enum bpf_prog_type type)325 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type) 326 { 327 struct bpf_prog *prog; 328 int ret = inode_permission(inode, MAY_READ | MAY_WRITE); 329 if (ret) 330 return ERR_PTR(ret); 331 332 if (inode->i_op == &bpf_map_iops) 333 return ERR_PTR(-EINVAL); 334 if (inode->i_op != &bpf_prog_iops) 335 return ERR_PTR(-EACCES); 336 337 prog = inode->i_private; 338 339 ret = security_bpf_prog(prog); 340 if (ret < 0) 341 return ERR_PTR(ret); 342 343 return bpf_prog_inc(prog); 344 } 345 bpf_prog_get_type_path(const char * name,enum bpf_prog_type type)346 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type) 347 { 348 struct bpf_prog *prog; 349 struct path path; 350 int ret = kern_path(name, LOOKUP_FOLLOW, &path); 351 if (ret) 352 return ERR_PTR(ret); 353 prog = __get_prog_inode(d_backing_inode(path.dentry), type); 354 if (!IS_ERR(prog)) 355 touch_atime(&path); 356 path_put(&path); 357 return prog; 358 } 359 EXPORT_SYMBOL(bpf_prog_get_type_path); 360 bpf_evict_inode(struct inode * inode)361 static void bpf_evict_inode(struct inode *inode) 362 { 363 enum bpf_type type; 364 365 truncate_inode_pages_final(&inode->i_data); 366 clear_inode(inode); 367 368 if (!bpf_inode_type(inode, &type)) 369 bpf_any_put(inode->i_private, type); 370 } 371 372 static const struct super_operations bpf_super_ops = { 373 .statfs = simple_statfs, 374 .drop_inode = generic_delete_inode, 375 .evict_inode = bpf_evict_inode, 376 }; 377 bpf_fill_super(struct super_block * sb,void * data,int silent)378 static int bpf_fill_super(struct super_block *sb, void *data, int silent) 379 { 380 static struct tree_descr bpf_rfiles[] = { { "" } }; 381 struct inode *inode; 382 int ret; 383 384 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles); 385 if (ret) 386 return ret; 387 388 sb->s_op = &bpf_super_ops; 389 390 inode = sb->s_root->d_inode; 391 inode->i_op = &bpf_dir_iops; 392 inode->i_mode &= ~S_IALLUGO; 393 inode->i_mode |= S_ISVTX | S_IRWXUGO; 394 395 return 0; 396 } 397 bpf_mount(struct file_system_type * type,int flags,const char * dev_name,void * data)398 static struct dentry *bpf_mount(struct file_system_type *type, int flags, 399 const char *dev_name, void *data) 400 { 401 return mount_nodev(type, flags, data, bpf_fill_super); 402 } 403 404 static struct file_system_type bpf_fs_type = { 405 .owner = THIS_MODULE, 406 .name = "bpf", 407 .mount = bpf_mount, 408 .kill_sb = kill_litter_super, 409 }; 410 bpf_init(void)411 static int __init bpf_init(void) 412 { 413 int ret; 414 415 ret = sysfs_create_mount_point(fs_kobj, "bpf"); 416 if (ret) 417 return ret; 418 419 ret = register_filesystem(&bpf_fs_type); 420 if (ret) 421 sysfs_remove_mount_point(fs_kobj, "bpf"); 422 423 return ret; 424 } 425 fs_initcall(bpf_init); 426