| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/fs/open.c |
| 4 | * |
| 5 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 6 | */ |
| 7 | |
| 8 | #include <linux/string.h> |
| 9 | #include <linux/mm.h> |
| 10 | #include <linux/file.h> |
| 11 | #include <linux/fdtable.h> |
| 12 | #include <linux/fsnotify.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/tty.h> |
| 15 | #include <linux/namei.h> |
| 16 | #include <linux/backing-dev.h> |
| 17 | #include <linux/capability.h> |
| 18 | #include <linux/securebits.h> |
| 19 | #include <linux/security.h> |
| 20 | #include <linux/mount.h> |
| 21 | #include <linux/fcntl.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/uaccess.h> |
| 24 | #include <linux/fs.h> |
| 25 | #include <linux/personality.h> |
| 26 | #include <linux/pagemap.h> |
| 27 | #include <linux/syscalls.h> |
| 28 | #include <linux/rcupdate.h> |
| 29 | #include <linux/audit.h> |
| 30 | #include <linux/falloc.h> |
| 31 | #include <linux/fs_struct.h> |
| 32 | #include <linux/ima.h> |
| 33 | #include <linux/dnotify.h> |
| 34 | #include <linux/compat.h> |
| 35 | #include <linux/mnt_idmapping.h> |
| 36 | |
| 37 | #include "internal.h" |
| 38 | |
| 39 | int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry, |
| 40 | loff_t length, unsigned int time_attrs, struct file *filp) |
| 41 | { |
| 42 | int ret; |
| 43 | struct iattr newattrs; |
| 44 | |
| 45 | /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */ |
| 46 | if (length < 0) |
| 47 | return -EINVAL; |
| 48 | |
| 49 | newattrs.ia_size = length; |
| 50 | newattrs.ia_valid = ATTR_SIZE | time_attrs; |
| 51 | if (filp) { |
| 52 | newattrs.ia_file = filp; |
| 53 | newattrs.ia_valid |= ATTR_FILE; |
| 54 | } |
| 55 | |
| 56 | /* Remove suid, sgid, and file capabilities on truncate too */ |
| 57 | ret = dentry_needs_remove_privs(idmap, dentry); |
| 58 | if (ret < 0) |
| 59 | return ret; |
| 60 | if (ret) |
| 61 | newattrs.ia_valid |= ret | ATTR_FORCE; |
| 62 | |
| 63 | inode_lock(dentry->d_inode); |
| 64 | /* Note any delegations or leases have already been broken: */ |
| 65 | ret = notify_change(idmap, dentry, &newattrs, NULL); |
| 66 | inode_unlock(dentry->d_inode); |
| 67 | return ret; |
| 68 | } |
| 69 | |
| 70 | long vfs_truncate(const struct path *path, loff_t length) |
| 71 | { |
| 72 | struct mnt_idmap *idmap; |
| 73 | struct inode *inode; |
| 74 | long error; |
| 75 | |
| 76 | inode = path->dentry->d_inode; |
| 77 | |
| 78 | /* For directories it's -EISDIR, for other non-regulars - -EINVAL */ |
| 79 | if (S_ISDIR(inode->i_mode)) |
| 80 | return -EISDIR; |
| 81 | if (!S_ISREG(inode->i_mode)) |
| 82 | return -EINVAL; |
| 83 | |
| 84 | error = mnt_want_write(path->mnt); |
| 85 | if (error) |
| 86 | goto out; |
| 87 | |
| 88 | idmap = mnt_idmap(path->mnt); |
| 89 | error = inode_permission(idmap, inode, MAY_WRITE); |
| 90 | if (error) |
| 91 | goto mnt_drop_write_and_out; |
| 92 | |
| 93 | error = -EPERM; |
| 94 | if (IS_APPEND(inode)) |
| 95 | goto mnt_drop_write_and_out; |
| 96 | |
| 97 | error = get_write_access(inode); |
| 98 | if (error) |
| 99 | goto mnt_drop_write_and_out; |
| 100 | |
| 101 | /* |
| 102 | * Make sure that there are no leases. get_write_access() protects |
| 103 | * against the truncate racing with a lease-granting setlease(). |
| 104 | */ |
| 105 | error = break_lease(inode, O_WRONLY); |
| 106 | if (error) |
| 107 | goto put_write_and_out; |
| 108 | |
| 109 | error = security_path_truncate(path); |
| 110 | if (!error) |
| 111 | error = do_truncate(idmap, path->dentry, length, 0, NULL); |
| 112 | |
| 113 | put_write_and_out: |
| 114 | put_write_access(inode); |
| 115 | mnt_drop_write_and_out: |
| 116 | mnt_drop_write(path->mnt); |
| 117 | out: |
| 118 | return error; |
| 119 | } |
| 120 | EXPORT_SYMBOL_GPL(vfs_truncate); |
| 121 | |
| 122 | long do_sys_truncate(const char __user *pathname, loff_t length) |
| 123 | { |
| 124 | unsigned int lookup_flags = LOOKUP_FOLLOW; |
| 125 | struct path path; |
| 126 | int error; |
| 127 | |
| 128 | if (length < 0) /* sorry, but loff_t says... */ |
| 129 | return -EINVAL; |
| 130 | |
| 131 | retry: |
| 132 | error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path); |
| 133 | if (!error) { |
| 134 | error = vfs_truncate(&path, length); |
| 135 | path_put(&path); |
| 136 | } |
| 137 | if (retry_estale(error, lookup_flags)) { |
| 138 | lookup_flags |= LOOKUP_REVAL; |
| 139 | goto retry; |
| 140 | } |
| 141 | return error; |
| 142 | } |
| 143 | |
| 144 | SYSCALL_DEFINE2(truncate, const char __user *, path, long, length) |
| 145 | { |
| 146 | return do_sys_truncate(path, length); |
| 147 | } |
| 148 | |
| 149 | #ifdef CONFIG_COMPAT |
| 150 | COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length) |
| 151 | { |
| 152 | return do_sys_truncate(path, length); |
| 153 | } |
| 154 | #endif |
| 155 | |
| 156 | long do_sys_ftruncate(unsigned int fd, loff_t length, int small) |
| 157 | { |
| 158 | struct inode *inode; |
| 159 | struct dentry *dentry; |
| 160 | struct fd f; |
| 161 | int error; |
| 162 | |
| 163 | error = -EINVAL; |
| 164 | if (length < 0) |
| 165 | goto out; |
| 166 | error = -EBADF; |
| 167 | f = fdget(fd); |
| 168 | if (!f.file) |
| 169 | goto out; |
| 170 | |
| 171 | /* explicitly opened as large or we are on 64-bit box */ |
| 172 | if (f.file->f_flags & O_LARGEFILE) |
| 173 | small = 0; |
| 174 | |
| 175 | dentry = f.file->f_path.dentry; |
| 176 | inode = dentry->d_inode; |
| 177 | error = -EINVAL; |
| 178 | if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE)) |
| 179 | goto out_putf; |
| 180 | |
| 181 | error = -EINVAL; |
| 182 | /* Cannot ftruncate over 2^31 bytes without large file support */ |
| 183 | if (small && length > MAX_NON_LFS) |
| 184 | goto out_putf; |
| 185 | |
| 186 | error = -EPERM; |
| 187 | /* Check IS_APPEND on real upper inode */ |
| 188 | if (IS_APPEND(file_inode(f.file))) |
| 189 | goto out_putf; |
| 190 | sb_start_write(inode->i_sb); |
| 191 | error = security_file_truncate(f.file); |
| 192 | if (!error) |
| 193 | error = do_truncate(file_mnt_idmap(f.file), dentry, length, |
| 194 | ATTR_MTIME | ATTR_CTIME, f.file); |
| 195 | sb_end_write(inode->i_sb); |
| 196 | out_putf: |
| 197 | fdput(f); |
| 198 | out: |
| 199 | return error; |
| 200 | } |
| 201 | |
| 202 | SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length) |
| 203 | { |
| 204 | return do_sys_ftruncate(fd, length, 1); |
| 205 | } |
| 206 | |
| 207 | #ifdef CONFIG_COMPAT |
| 208 | COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length) |
| 209 | { |
| 210 | return do_sys_ftruncate(fd, length, 1); |
| 211 | } |
| 212 | #endif |
| 213 | |
| 214 | /* LFS versions of truncate are only needed on 32 bit machines */ |
| 215 | #if BITS_PER_LONG == 32 |
| 216 | SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length) |
| 217 | { |
| 218 | return do_sys_truncate(path, length); |
| 219 | } |
| 220 | |
| 221 | SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length) |
| 222 | { |
| 223 | return do_sys_ftruncate(fd, length, 0); |
| 224 | } |
| 225 | #endif /* BITS_PER_LONG == 32 */ |
| 226 | |
| 227 | #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64) |
| 228 | COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname, |
| 229 | compat_arg_u64_dual(length)) |
| 230 | { |
| 231 | return ksys_truncate(pathname, compat_arg_u64_glue(length)); |
| 232 | } |
| 233 | #endif |
| 234 | |
| 235 | #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64) |
| 236 | COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd, |
| 237 | compat_arg_u64_dual(length)) |
| 238 | { |
| 239 | return ksys_ftruncate(fd, compat_arg_u64_glue(length)); |
| 240 | } |
| 241 | #endif |
| 242 | |
| 243 | int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
| 244 | { |
| 245 | struct inode *inode = file_inode(file); |
| 246 | long ret; |
| 247 | |
| 248 | if (offset < 0 || len <= 0) |
| 249 | return -EINVAL; |
| 250 | |
| 251 | /* Return error if mode is not supported */ |
| 252 | if (mode & ~FALLOC_FL_SUPPORTED_MASK) |
| 253 | return -EOPNOTSUPP; |
| 254 | |
| 255 | /* Punch hole and zero range are mutually exclusive */ |
| 256 | if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) == |
| 257 | (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) |
| 258 | return -EOPNOTSUPP; |
| 259 | |
| 260 | /* Punch hole must have keep size set */ |
| 261 | if ((mode & FALLOC_FL_PUNCH_HOLE) && |
| 262 | !(mode & FALLOC_FL_KEEP_SIZE)) |
| 263 | return -EOPNOTSUPP; |
| 264 | |
| 265 | /* Collapse range should only be used exclusively. */ |
| 266 | if ((mode & FALLOC_FL_COLLAPSE_RANGE) && |
| 267 | (mode & ~FALLOC_FL_COLLAPSE_RANGE)) |
| 268 | return -EINVAL; |
| 269 | |
| 270 | /* Insert range should only be used exclusively. */ |
| 271 | if ((mode & FALLOC_FL_INSERT_RANGE) && |
| 272 | (mode & ~FALLOC_FL_INSERT_RANGE)) |
| 273 | return -EINVAL; |
| 274 | |
| 275 | /* Unshare range should only be used with allocate mode. */ |
| 276 | if ((mode & FALLOC_FL_UNSHARE_RANGE) && |
| 277 | (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE))) |
| 278 | return -EINVAL; |
| 279 | |
| 280 | if (!(file->f_mode & FMODE_WRITE)) |
| 281 | return -EBADF; |
| 282 | |
| 283 | /* |
| 284 | * We can only allow pure fallocate on append only files |
| 285 | */ |
| 286 | if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode)) |
| 287 | return -EPERM; |
| 288 | |
| 289 | if (IS_IMMUTABLE(inode)) |
| 290 | return -EPERM; |
| 291 | |
| 292 | /* |
| 293 | * We cannot allow any fallocate operation on an active swapfile |
| 294 | */ |
| 295 | if (IS_SWAPFILE(inode)) |
| 296 | return -ETXTBSY; |
| 297 | |
| 298 | /* |
| 299 | * Revalidate the write permissions, in case security policy has |
| 300 | * changed since the files were opened. |
| 301 | */ |
| 302 | ret = security_file_permission(file, MAY_WRITE); |
| 303 | if (ret) |
| 304 | return ret; |
| 305 | |
| 306 | if (S_ISFIFO(inode->i_mode)) |
| 307 | return -ESPIPE; |
| 308 | |
| 309 | if (S_ISDIR(inode->i_mode)) |
| 310 | return -EISDIR; |
| 311 | |
| 312 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) |
| 313 | return -ENODEV; |
| 314 | |
| 315 | /* Check for wrap through zero too */ |
| 316 | if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0)) |
| 317 | return -EFBIG; |
| 318 | |
| 319 | if (!file->f_op->fallocate) |
| 320 | return -EOPNOTSUPP; |
| 321 | |
| 322 | file_start_write(file); |
| 323 | ret = file->f_op->fallocate(file, mode, offset, len); |
| 324 | |
| 325 | /* |
| 326 | * Create inotify and fanotify events. |
| 327 | * |
| 328 | * To keep the logic simple always create events if fallocate succeeds. |
| 329 | * This implies that events are even created if the file size remains |
| 330 | * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE. |
| 331 | */ |
| 332 | if (ret == 0) |
| 333 | fsnotify_modify(file); |
| 334 | |
| 335 | file_end_write(file); |
| 336 | return ret; |
| 337 | } |
| 338 | EXPORT_SYMBOL_GPL(vfs_fallocate); |
| 339 | |
| 340 | int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len) |
| 341 | { |
| 342 | struct fd f = fdget(fd); |
| 343 | int error = -EBADF; |
| 344 | |
| 345 | if (f.file) { |
| 346 | error = vfs_fallocate(f.file, mode, offset, len); |
| 347 | fdput(f); |
| 348 | } |
| 349 | return error; |
| 350 | } |
| 351 | |
| 352 | SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len) |
| 353 | { |
| 354 | return ksys_fallocate(fd, mode, offset, len); |
| 355 | } |
| 356 | |
| 357 | #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE) |
| 358 | COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset), |
| 359 | compat_arg_u64_dual(len)) |
| 360 | { |
| 361 | return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset), |
| 362 | compat_arg_u64_glue(len)); |
| 363 | } |
| 364 | #endif |
| 365 | |
| 366 | /* |
| 367 | * access() needs to use the real uid/gid, not the effective uid/gid. |
| 368 | * We do this by temporarily clearing all FS-related capabilities and |
| 369 | * switching the fsuid/fsgid around to the real ones. |
| 370 | */ |
| 371 | static const struct cred *access_override_creds(void) |
| 372 | { |
| 373 | const struct cred *old_cred; |
| 374 | struct cred *override_cred; |
| 375 | |
| 376 | override_cred = prepare_creds(); |
| 377 | if (!override_cred) |
| 378 | return NULL; |
| 379 | |
| 380 | override_cred->fsuid = override_cred->uid; |
| 381 | override_cred->fsgid = override_cred->gid; |
| 382 | |
| 383 | if (!issecure(SECURE_NO_SETUID_FIXUP)) { |
| 384 | /* Clear the capabilities if we switch to a non-root user */ |
| 385 | kuid_t root_uid = make_kuid(override_cred->user_ns, 0); |
| 386 | if (!uid_eq(override_cred->uid, root_uid)) |
| 387 | cap_clear(override_cred->cap_effective); |
| 388 | else |
| 389 | override_cred->cap_effective = |
| 390 | override_cred->cap_permitted; |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | * The new set of credentials can *only* be used in |
| 395 | * task-synchronous circumstances, and does not need |
| 396 | * RCU freeing, unless somebody then takes a separate |
| 397 | * reference to it. |
| 398 | * |
| 399 | * NOTE! This is _only_ true because this credential |
| 400 | * is used purely for override_creds() that installs |
| 401 | * it as the subjective cred. Other threads will be |
| 402 | * accessing ->real_cred, not the subjective cred. |
| 403 | * |
| 404 | * If somebody _does_ make a copy of this (using the |
| 405 | * 'get_current_cred()' function), that will clear the |
| 406 | * non_rcu field, because now that other user may be |
| 407 | * expecting RCU freeing. But normal thread-synchronous |
| 408 | * cred accesses will keep things non-RCY. |
| 409 | */ |
| 410 | override_cred->non_rcu = 1; |
| 411 | |
| 412 | old_cred = override_creds(override_cred); |
| 413 | |
| 414 | /* override_cred() gets its own ref */ |
| 415 | put_cred(override_cred); |
| 416 | |
| 417 | return old_cred; |
| 418 | } |
| 419 | |
| 420 | static long do_faccessat(int dfd, const char __user *filename, int mode, int flags) |
| 421 | { |
| 422 | struct path path; |
| 423 | struct inode *inode; |
| 424 | int res; |
| 425 | unsigned int lookup_flags = LOOKUP_FOLLOW; |
| 426 | const struct cred *old_cred = NULL; |
| 427 | |
| 428 | if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */ |
| 429 | return -EINVAL; |
| 430 | |
| 431 | if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) |
| 432 | return -EINVAL; |
| 433 | |
| 434 | if (flags & AT_SYMLINK_NOFOLLOW) |
| 435 | lookup_flags &= ~LOOKUP_FOLLOW; |
| 436 | if (flags & AT_EMPTY_PATH) |
| 437 | lookup_flags |= LOOKUP_EMPTY; |
| 438 | |
| 439 | if (!(flags & AT_EACCESS)) { |
| 440 | old_cred = access_override_creds(); |
| 441 | if (!old_cred) |
| 442 | return -ENOMEM; |
| 443 | } |
| 444 | |
| 445 | retry: |
| 446 | res = user_path_at(dfd, filename, lookup_flags, &path); |
| 447 | if (res) |
| 448 | goto out; |
| 449 | |
| 450 | inode = d_backing_inode(path.dentry); |
| 451 | |
| 452 | if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) { |
| 453 | /* |
| 454 | * MAY_EXEC on regular files is denied if the fs is mounted |
| 455 | * with the "noexec" flag. |
| 456 | */ |
| 457 | res = -EACCES; |
| 458 | if (path_noexec(&path)) |
| 459 | goto out_path_release; |
| 460 | } |
| 461 | |
| 462 | res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS); |
| 463 | /* SuS v2 requires we report a read only fs too */ |
| 464 | if (res || !(mode & S_IWOTH) || special_file(inode->i_mode)) |
| 465 | goto out_path_release; |
| 466 | /* |
| 467 | * This is a rare case where using __mnt_is_readonly() |
| 468 | * is OK without a mnt_want/drop_write() pair. Since |
| 469 | * no actual write to the fs is performed here, we do |
| 470 | * not need to telegraph to that to anyone. |
| 471 | * |
| 472 | * By doing this, we accept that this access is |
| 473 | * inherently racy and know that the fs may change |
| 474 | * state before we even see this result. |
| 475 | */ |
| 476 | if (__mnt_is_readonly(path.mnt)) |
| 477 | res = -EROFS; |
| 478 | |
| 479 | out_path_release: |
| 480 | path_put(&path); |
| 481 | if (retry_estale(res, lookup_flags)) { |
| 482 | lookup_flags |= LOOKUP_REVAL; |
| 483 | goto retry; |
| 484 | } |
| 485 | out: |
| 486 | if (old_cred) |
| 487 | revert_creds(old_cred); |
| 488 | |
| 489 | return res; |
| 490 | } |
| 491 | |
| 492 | SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode) |
| 493 | { |
| 494 | return do_faccessat(dfd, filename, mode, 0); |
| 495 | } |
| 496 | |
| 497 | SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode, |
| 498 | int, flags) |
| 499 | { |
| 500 | return do_faccessat(dfd, filename, mode, flags); |
| 501 | } |
| 502 | |
| 503 | SYSCALL_DEFINE2(access, const char __user *, filename, int, mode) |
| 504 | { |
| 505 | return do_faccessat(AT_FDCWD, filename, mode, 0); |
| 506 | } |
| 507 | |
| 508 | SYSCALL_DEFINE1(chdir, const char __user *, filename) |
| 509 | { |
| 510 | struct path path; |
| 511 | int error; |
| 512 | unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; |
| 513 | retry: |
| 514 | error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); |
| 515 | if (error) |
| 516 | goto out; |
| 517 | |
| 518 | error = path_permission(&path, MAY_EXEC | MAY_CHDIR); |
| 519 | if (error) |
| 520 | goto dput_and_out; |
| 521 | |
| 522 | set_fs_pwd(current->fs, &path); |
| 523 | |
| 524 | dput_and_out: |
| 525 | path_put(&path); |
| 526 | if (retry_estale(error, lookup_flags)) { |
| 527 | lookup_flags |= LOOKUP_REVAL; |
| 528 | goto retry; |
| 529 | } |
| 530 | out: |
| 531 | return error; |
| 532 | } |
| 533 | |
| 534 | SYSCALL_DEFINE1(fchdir, unsigned int, fd) |
| 535 | { |
| 536 | struct fd f = fdget_raw(fd); |
| 537 | int error; |
| 538 | |
| 539 | error = -EBADF; |
| 540 | if (!f.file) |
| 541 | goto out; |
| 542 | |
| 543 | error = -ENOTDIR; |
| 544 | if (!d_can_lookup(f.file->f_path.dentry)) |
| 545 | goto out_putf; |
| 546 | |
| 547 | error = file_permission(f.file, MAY_EXEC | MAY_CHDIR); |
| 548 | if (!error) |
| 549 | set_fs_pwd(current->fs, &f.file->f_path); |
| 550 | out_putf: |
| 551 | fdput(f); |
| 552 | out: |
| 553 | return error; |
| 554 | } |
| 555 | |
| 556 | SYSCALL_DEFINE1(chroot, const char __user *, filename) |
| 557 | { |
| 558 | struct path path; |
| 559 | int error; |
| 560 | unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; |
| 561 | retry: |
| 562 | error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); |
| 563 | if (error) |
| 564 | goto out; |
| 565 | |
| 566 | error = path_permission(&path, MAY_EXEC | MAY_CHDIR); |
| 567 | if (error) |
| 568 | goto dput_and_out; |
| 569 | |
| 570 | error = -EPERM; |
| 571 | if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT)) |
| 572 | goto dput_and_out; |
| 573 | error = security_path_chroot(&path); |
| 574 | if (error) |
| 575 | goto dput_and_out; |
| 576 | |
| 577 | set_fs_root(current->fs, &path); |
| 578 | error = 0; |
| 579 | dput_and_out: |
| 580 | path_put(&path); |
| 581 | if (retry_estale(error, lookup_flags)) { |
| 582 | lookup_flags |= LOOKUP_REVAL; |
| 583 | goto retry; |
| 584 | } |
| 585 | out: |
| 586 | return error; |
| 587 | } |
| 588 | |
| 589 | int chmod_common(const struct path *path, umode_t mode) |
| 590 | { |
| 591 | struct inode *inode = path->dentry->d_inode; |
| 592 | struct inode *delegated_inode = NULL; |
| 593 | struct iattr newattrs; |
| 594 | int error; |
| 595 | |
| 596 | error = mnt_want_write(path->mnt); |
| 597 | if (error) |
| 598 | return error; |
| 599 | retry_deleg: |
| 600 | inode_lock(inode); |
| 601 | error = security_path_chmod(path, mode); |
| 602 | if (error) |
| 603 | goto out_unlock; |
| 604 | newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO); |
| 605 | newattrs.ia_valid = ATTR_MODE | ATTR_CTIME; |
| 606 | error = notify_change(mnt_idmap(path->mnt), path->dentry, |
| 607 | &newattrs, &delegated_inode); |
| 608 | out_unlock: |
| 609 | inode_unlock(inode); |
| 610 | if (delegated_inode) { |
| 611 | error = break_deleg_wait(&delegated_inode); |
| 612 | if (!error) |
| 613 | goto retry_deleg; |
| 614 | } |
| 615 | mnt_drop_write(path->mnt); |
| 616 | return error; |
| 617 | } |
| 618 | |
| 619 | int vfs_fchmod(struct file *file, umode_t mode) |
| 620 | { |
| 621 | audit_file(file); |
| 622 | return chmod_common(&file->f_path, mode); |
| 623 | } |
| 624 | |
| 625 | SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode) |
| 626 | { |
| 627 | struct fd f = fdget(fd); |
| 628 | int err = -EBADF; |
| 629 | |
| 630 | if (f.file) { |
| 631 | err = vfs_fchmod(f.file, mode); |
| 632 | fdput(f); |
| 633 | } |
| 634 | return err; |
| 635 | } |
| 636 | |
| 637 | static int do_fchmodat(int dfd, const char __user *filename, umode_t mode) |
| 638 | { |
| 639 | struct path path; |
| 640 | int error; |
| 641 | unsigned int lookup_flags = LOOKUP_FOLLOW; |
| 642 | retry: |
| 643 | error = user_path_at(dfd, filename, lookup_flags, &path); |
| 644 | if (!error) { |
| 645 | error = chmod_common(&path, mode); |
| 646 | path_put(&path); |
| 647 | if (retry_estale(error, lookup_flags)) { |
| 648 | lookup_flags |= LOOKUP_REVAL; |
| 649 | goto retry; |
| 650 | } |
| 651 | } |
| 652 | return error; |
| 653 | } |
| 654 | |
| 655 | SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, |
| 656 | umode_t, mode) |
| 657 | { |
| 658 | return do_fchmodat(dfd, filename, mode); |
| 659 | } |
| 660 | |
| 661 | SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode) |
| 662 | { |
| 663 | return do_fchmodat(AT_FDCWD, filename, mode); |
| 664 | } |
| 665 | |
| 666 | /** |
| 667 | * setattr_vfsuid - check and set ia_fsuid attribute |
| 668 | * @kuid: new inode owner |
| 669 | * |
| 670 | * Check whether @kuid is valid and if so generate and set vfsuid_t in |
| 671 | * ia_vfsuid. |
| 672 | * |
| 673 | * Return: true if @kuid is valid, false if not. |
| 674 | */ |
| 675 | static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid) |
| 676 | { |
| 677 | if (!uid_valid(kuid)) |
| 678 | return false; |
| 679 | attr->ia_valid |= ATTR_UID; |
| 680 | attr->ia_vfsuid = VFSUIDT_INIT(kuid); |
| 681 | return true; |
| 682 | } |
| 683 | |
| 684 | /** |
| 685 | * setattr_vfsgid - check and set ia_fsgid attribute |
| 686 | * @kgid: new inode owner |
| 687 | * |
| 688 | * Check whether @kgid is valid and if so generate and set vfsgid_t in |
| 689 | * ia_vfsgid. |
| 690 | * |
| 691 | * Return: true if @kgid is valid, false if not. |
| 692 | */ |
| 693 | static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid) |
| 694 | { |
| 695 | if (!gid_valid(kgid)) |
| 696 | return false; |
| 697 | attr->ia_valid |= ATTR_GID; |
| 698 | attr->ia_vfsgid = VFSGIDT_INIT(kgid); |
| 699 | return true; |
| 700 | } |
| 701 | |
| 702 | int chown_common(const struct path *path, uid_t user, gid_t group) |
| 703 | { |
| 704 | struct mnt_idmap *idmap; |
| 705 | struct user_namespace *mnt_userns, *fs_userns; |
| 706 | struct inode *inode = path->dentry->d_inode; |
| 707 | struct inode *delegated_inode = NULL; |
| 708 | int error; |
| 709 | struct iattr newattrs; |
| 710 | kuid_t uid; |
| 711 | kgid_t gid; |
| 712 | |
| 713 | uid = make_kuid(current_user_ns(), user); |
| 714 | gid = make_kgid(current_user_ns(), group); |
| 715 | |
| 716 | idmap = mnt_idmap(path->mnt); |
| 717 | mnt_userns = mnt_idmap_owner(idmap); |
| 718 | fs_userns = i_user_ns(inode); |
| 719 | |
| 720 | retry_deleg: |
| 721 | newattrs.ia_vfsuid = INVALID_VFSUID; |
| 722 | newattrs.ia_vfsgid = INVALID_VFSGID; |
| 723 | newattrs.ia_valid = ATTR_CTIME; |
| 724 | if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid)) |
| 725 | return -EINVAL; |
| 726 | if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid)) |
| 727 | return -EINVAL; |
| 728 | inode_lock(inode); |
| 729 | if (!S_ISDIR(inode->i_mode)) |
| 730 | newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV | |
| 731 | setattr_should_drop_sgid(idmap, inode); |
| 732 | /* Continue to send actual fs values, not the mount values. */ |
| 733 | error = security_path_chown( |
| 734 | path, |
| 735 | from_vfsuid(mnt_userns, fs_userns, newattrs.ia_vfsuid), |
| 736 | from_vfsgid(mnt_userns, fs_userns, newattrs.ia_vfsgid)); |
| 737 | if (!error) |
| 738 | error = notify_change(idmap, path->dentry, &newattrs, |
| 739 | &delegated_inode); |
| 740 | inode_unlock(inode); |
| 741 | if (delegated_inode) { |
| 742 | error = break_deleg_wait(&delegated_inode); |
| 743 | if (!error) |
| 744 | goto retry_deleg; |
| 745 | } |
| 746 | return error; |
| 747 | } |
| 748 | |
| 749 | int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group, |
| 750 | int flag) |
| 751 | { |
| 752 | struct path path; |
| 753 | int error = -EINVAL; |
| 754 | int lookup_flags; |
| 755 | |
| 756 | if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) |
| 757 | goto out; |
| 758 | |
| 759 | lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW; |
| 760 | if (flag & AT_EMPTY_PATH) |
| 761 | lookup_flags |= LOOKUP_EMPTY; |
| 762 | retry: |
| 763 | error = user_path_at(dfd, filename, lookup_flags, &path); |
| 764 | if (error) |
| 765 | goto out; |
| 766 | error = mnt_want_write(path.mnt); |
| 767 | if (error) |
| 768 | goto out_release; |
| 769 | error = chown_common(&path, user, group); |
| 770 | mnt_drop_write(path.mnt); |
| 771 | out_release: |
| 772 | path_put(&path); |
| 773 | if (retry_estale(error, lookup_flags)) { |
| 774 | lookup_flags |= LOOKUP_REVAL; |
| 775 | goto retry; |
| 776 | } |
| 777 | out: |
| 778 | return error; |
| 779 | } |
| 780 | |
| 781 | SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user, |
| 782 | gid_t, group, int, flag) |
| 783 | { |
| 784 | return do_fchownat(dfd, filename, user, group, flag); |
| 785 | } |
| 786 | |
| 787 | SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group) |
| 788 | { |
| 789 | return do_fchownat(AT_FDCWD, filename, user, group, 0); |
| 790 | } |
| 791 | |
| 792 | SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group) |
| 793 | { |
| 794 | return do_fchownat(AT_FDCWD, filename, user, group, |
| 795 | AT_SYMLINK_NOFOLLOW); |
| 796 | } |
| 797 | |
| 798 | int vfs_fchown(struct file *file, uid_t user, gid_t group) |
| 799 | { |
| 800 | int error; |
| 801 | |
| 802 | error = mnt_want_write_file(file); |
| 803 | if (error) |
| 804 | return error; |
| 805 | audit_file(file); |
| 806 | error = chown_common(&file->f_path, user, group); |
| 807 | mnt_drop_write_file(file); |
| 808 | return error; |
| 809 | } |
| 810 | |
| 811 | int ksys_fchown(unsigned int fd, uid_t user, gid_t group) |
| 812 | { |
| 813 | struct fd f = fdget(fd); |
| 814 | int error = -EBADF; |
| 815 | |
| 816 | if (f.file) { |
| 817 | error = vfs_fchown(f.file, user, group); |
| 818 | fdput(f); |
| 819 | } |
| 820 | return error; |
| 821 | } |
| 822 | |
| 823 | SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group) |
| 824 | { |
| 825 | return ksys_fchown(fd, user, group); |
| 826 | } |
| 827 | |
| 828 | static int do_dentry_open(struct file *f, |
| 829 | struct inode *inode, |
| 830 | int (*open)(struct inode *, struct file *)) |
| 831 | { |
| 832 | static const struct file_operations empty_fops = {}; |
| 833 | int error; |
| 834 | |
| 835 | path_get(&f->f_path); |
| 836 | f->f_inode = inode; |
| 837 | f->f_mapping = inode->i_mapping; |
| 838 | f->f_wb_err = filemap_sample_wb_err(f->f_mapping); |
| 839 | f->f_sb_err = file_sample_sb_err(f); |
| 840 | |
| 841 | if (unlikely(f->f_flags & O_PATH)) { |
| 842 | f->f_mode = FMODE_PATH | FMODE_OPENED; |
| 843 | f->f_op = &empty_fops; |
| 844 | return 0; |
| 845 | } |
| 846 | |
| 847 | if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) { |
| 848 | i_readcount_inc(inode); |
| 849 | } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) { |
| 850 | error = get_write_access(inode); |
| 851 | if (unlikely(error)) |
| 852 | goto cleanup_file; |
| 853 | error = __mnt_want_write(f->f_path.mnt); |
| 854 | if (unlikely(error)) { |
| 855 | put_write_access(inode); |
| 856 | goto cleanup_file; |
| 857 | } |
| 858 | f->f_mode |= FMODE_WRITER; |
| 859 | } |
| 860 | |
| 861 | /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */ |
| 862 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) |
| 863 | f->f_mode |= FMODE_ATOMIC_POS; |
| 864 | |
| 865 | f->f_op = fops_get(inode->i_fop); |
| 866 | if (WARN_ON(!f->f_op)) { |
| 867 | error = -ENODEV; |
| 868 | goto cleanup_all; |
| 869 | } |
| 870 | |
| 871 | error = security_file_open(f); |
| 872 | if (error) |
| 873 | goto cleanup_all; |
| 874 | |
| 875 | error = break_lease(locks_inode(f), f->f_flags); |
| 876 | if (error) |
| 877 | goto cleanup_all; |
| 878 | |
| 879 | /* normally all 3 are set; ->open() can clear them if needed */ |
| 880 | f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; |
| 881 | if (!open) |
| 882 | open = f->f_op->open; |
| 883 | if (open) { |
| 884 | error = open(inode, f); |
| 885 | if (error) |
| 886 | goto cleanup_all; |
| 887 | } |
| 888 | f->f_mode |= FMODE_OPENED; |
| 889 | if ((f->f_mode & FMODE_READ) && |
| 890 | likely(f->f_op->read || f->f_op->read_iter)) |
| 891 | f->f_mode |= FMODE_CAN_READ; |
| 892 | if ((f->f_mode & FMODE_WRITE) && |
| 893 | likely(f->f_op->write || f->f_op->write_iter)) |
| 894 | f->f_mode |= FMODE_CAN_WRITE; |
| 895 | if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek) |
| 896 | f->f_mode &= ~FMODE_LSEEK; |
| 897 | if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO) |
| 898 | f->f_mode |= FMODE_CAN_ODIRECT; |
| 899 | |
| 900 | f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); |
| 901 | f->f_iocb_flags = iocb_flags(f); |
| 902 | |
| 903 | file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping); |
| 904 | |
| 905 | if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT)) |
| 906 | return -EINVAL; |
| 907 | |
| 908 | /* |
| 909 | * XXX: Huge page cache doesn't support writing yet. Drop all page |
| 910 | * cache for this file before processing writes. |
| 911 | */ |
| 912 | if (f->f_mode & FMODE_WRITE) { |
| 913 | /* |
| 914 | * Paired with smp_mb() in collapse_file() to ensure nr_thps |
| 915 | * is up to date and the update to i_writecount by |
| 916 | * get_write_access() is visible. Ensures subsequent insertion |
| 917 | * of THPs into the page cache will fail. |
| 918 | */ |
| 919 | smp_mb(); |
| 920 | if (filemap_nr_thps(inode->i_mapping)) { |
| 921 | struct address_space *mapping = inode->i_mapping; |
| 922 | |
| 923 | filemap_invalidate_lock(inode->i_mapping); |
| 924 | /* |
| 925 | * unmap_mapping_range just need to be called once |
| 926 | * here, because the private pages is not need to be |
| 927 | * unmapped mapping (e.g. data segment of dynamic |
| 928 | * shared libraries here). |
| 929 | */ |
| 930 | unmap_mapping_range(mapping, 0, 0, 0); |
| 931 | truncate_inode_pages(mapping, 0); |
| 932 | filemap_invalidate_unlock(inode->i_mapping); |
| 933 | } |
| 934 | } |
| 935 | |
| 936 | return 0; |
| 937 | |
| 938 | cleanup_all: |
| 939 | if (WARN_ON_ONCE(error > 0)) |
| 940 | error = -EINVAL; |
| 941 | fops_put(f->f_op); |
| 942 | put_file_access(f); |
| 943 | cleanup_file: |
| 944 | path_put(&f->f_path); |
| 945 | f->f_path.mnt = NULL; |
| 946 | f->f_path.dentry = NULL; |
| 947 | f->f_inode = NULL; |
| 948 | return error; |
| 949 | } |
| 950 | |
| 951 | /** |
| 952 | * finish_open - finish opening a file |
| 953 | * @file: file pointer |
| 954 | * @dentry: pointer to dentry |
| 955 | * @open: open callback |
| 956 | * @opened: state of open |
| 957 | * |
| 958 | * This can be used to finish opening a file passed to i_op->atomic_open(). |
| 959 | * |
| 960 | * If the open callback is set to NULL, then the standard f_op->open() |
| 961 | * filesystem callback is substituted. |
| 962 | * |
| 963 | * NB: the dentry reference is _not_ consumed. If, for example, the dentry is |
| 964 | * the return value of d_splice_alias(), then the caller needs to perform dput() |
| 965 | * on it after finish_open(). |
| 966 | * |
| 967 | * Returns zero on success or -errno if the open failed. |
| 968 | */ |
| 969 | int finish_open(struct file *file, struct dentry *dentry, |
| 970 | int (*open)(struct inode *, struct file *)) |
| 971 | { |
| 972 | BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ |
| 973 | |
| 974 | file->f_path.dentry = dentry; |
| 975 | return do_dentry_open(file, d_backing_inode(dentry), open); |
| 976 | } |
| 977 | EXPORT_SYMBOL(finish_open); |
| 978 | |
| 979 | /** |
| 980 | * finish_no_open - finish ->atomic_open() without opening the file |
| 981 | * |
| 982 | * @file: file pointer |
| 983 | * @dentry: dentry or NULL (as returned from ->lookup()) |
| 984 | * |
| 985 | * This can be used to set the result of a successful lookup in ->atomic_open(). |
| 986 | * |
| 987 | * NB: unlike finish_open() this function does consume the dentry reference and |
| 988 | * the caller need not dput() it. |
| 989 | * |
| 990 | * Returns "0" which must be the return value of ->atomic_open() after having |
| 991 | * called this function. |
| 992 | */ |
| 993 | int finish_no_open(struct file *file, struct dentry *dentry) |
| 994 | { |
| 995 | file->f_path.dentry = dentry; |
| 996 | return 0; |
| 997 | } |
| 998 | EXPORT_SYMBOL(finish_no_open); |
| 999 | |
| 1000 | char *file_path(struct file *filp, char *buf, int buflen) |
| 1001 | { |
| 1002 | return d_path(&filp->f_path, buf, buflen); |
| 1003 | } |
| 1004 | EXPORT_SYMBOL(file_path); |
| 1005 | |
| 1006 | /** |
| 1007 | * vfs_open - open the file at the given path |
| 1008 | * @path: path to open |
| 1009 | * @file: newly allocated file with f_flag initialized |
| 1010 | * @cred: credentials to use |
| 1011 | */ |
| 1012 | int vfs_open(const struct path *path, struct file *file) |
| 1013 | { |
| 1014 | file->f_path = *path; |
| 1015 | return do_dentry_open(file, d_backing_inode(path->dentry), NULL); |
| 1016 | } |
| 1017 | |
| 1018 | struct file *dentry_open(const struct path *path, int flags, |
| 1019 | const struct cred *cred) |
| 1020 | { |
| 1021 | int error; |
| 1022 | struct file *f; |
| 1023 | |
| 1024 | validate_creds(cred); |
| 1025 | |
| 1026 | /* We must always pass in a valid mount pointer. */ |
| 1027 | BUG_ON(!path->mnt); |
| 1028 | |
| 1029 | f = alloc_empty_file(flags, cred); |
| 1030 | if (!IS_ERR(f)) { |
| 1031 | error = vfs_open(path, f); |
| 1032 | if (error) { |
| 1033 | fput(f); |
| 1034 | f = ERR_PTR(error); |
| 1035 | } |
| 1036 | } |
| 1037 | return f; |
| 1038 | } |
| 1039 | EXPORT_SYMBOL(dentry_open); |
| 1040 | |
| 1041 | /** |
| 1042 | * dentry_create - Create and open a file |
| 1043 | * @path: path to create |
| 1044 | * @flags: O_ flags |
| 1045 | * @mode: mode bits for new file |
| 1046 | * @cred: credentials to use |
| 1047 | * |
| 1048 | * Caller must hold the parent directory's lock, and have prepared |
| 1049 | * a negative dentry, placed in @path->dentry, for the new file. |
| 1050 | * |
| 1051 | * Caller sets @path->mnt to the vfsmount of the filesystem where |
| 1052 | * the new file is to be created. The parent directory and the |
| 1053 | * negative dentry must reside on the same filesystem instance. |
| 1054 | * |
| 1055 | * On success, returns a "struct file *". Otherwise a ERR_PTR |
| 1056 | * is returned. |
| 1057 | */ |
| 1058 | struct file *dentry_create(const struct path *path, int flags, umode_t mode, |
| 1059 | const struct cred *cred) |
| 1060 | { |
| 1061 | struct file *f; |
| 1062 | int error; |
| 1063 | |
| 1064 | validate_creds(cred); |
| 1065 | f = alloc_empty_file(flags, cred); |
| 1066 | if (IS_ERR(f)) |
| 1067 | return f; |
| 1068 | |
| 1069 | error = vfs_create(mnt_idmap(path->mnt), |
| 1070 | d_inode(path->dentry->d_parent), |
| 1071 | path->dentry, mode, true); |
| 1072 | if (!error) |
| 1073 | error = vfs_open(path, f); |
| 1074 | |
| 1075 | if (unlikely(error)) { |
| 1076 | fput(f); |
| 1077 | return ERR_PTR(error); |
| 1078 | } |
| 1079 | return f; |
| 1080 | } |
| 1081 | EXPORT_SYMBOL(dentry_create); |
| 1082 | |
| 1083 | struct file *open_with_fake_path(const struct path *path, int flags, |
| 1084 | struct inode *inode, const struct cred *cred) |
| 1085 | { |
| 1086 | struct file *f = alloc_empty_file_noaccount(flags, cred); |
| 1087 | if (!IS_ERR(f)) { |
| 1088 | int error; |
| 1089 | |
| 1090 | f->f_path = *path; |
| 1091 | error = do_dentry_open(f, inode, NULL); |
| 1092 | if (error) { |
| 1093 | fput(f); |
| 1094 | f = ERR_PTR(error); |
| 1095 | } |
| 1096 | } |
| 1097 | return f; |
| 1098 | } |
| 1099 | EXPORT_SYMBOL(open_with_fake_path); |
| 1100 | |
| 1101 | #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE)) |
| 1102 | #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC) |
| 1103 | |
| 1104 | inline struct open_how build_open_how(int flags, umode_t mode) |
| 1105 | { |
| 1106 | struct open_how how = { |
| 1107 | .flags = flags & VALID_OPEN_FLAGS, |
| 1108 | .mode = mode & S_IALLUGO, |
| 1109 | }; |
| 1110 | |
| 1111 | /* O_PATH beats everything else. */ |
| 1112 | if (how.flags & O_PATH) |
| 1113 | how.flags &= O_PATH_FLAGS; |
| 1114 | /* Modes should only be set for create-like flags. */ |
| 1115 | if (!WILL_CREATE(how.flags)) |
| 1116 | how.mode = 0; |
| 1117 | return how; |
| 1118 | } |
| 1119 | |
| 1120 | inline int build_open_flags(const struct open_how *how, struct open_flags *op) |
| 1121 | { |
| 1122 | u64 flags = how->flags; |
| 1123 | u64 strip = FMODE_NONOTIFY | O_CLOEXEC; |
| 1124 | int lookup_flags = 0; |
| 1125 | int acc_mode = ACC_MODE(flags); |
| 1126 | |
| 1127 | BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS), |
| 1128 | "struct open_flags doesn't yet handle flags > 32 bits"); |
| 1129 | |
| 1130 | /* |
| 1131 | * Strip flags that either shouldn't be set by userspace like |
| 1132 | * FMODE_NONOTIFY or that aren't relevant in determining struct |
| 1133 | * open_flags like O_CLOEXEC. |
| 1134 | */ |
| 1135 | flags &= ~strip; |
| 1136 | |
| 1137 | /* |
| 1138 | * Older syscalls implicitly clear all of the invalid flags or argument |
| 1139 | * values before calling build_open_flags(), but openat2(2) checks all |
| 1140 | * of its arguments. |
| 1141 | */ |
| 1142 | if (flags & ~VALID_OPEN_FLAGS) |
| 1143 | return -EINVAL; |
| 1144 | if (how->resolve & ~VALID_RESOLVE_FLAGS) |
| 1145 | return -EINVAL; |
| 1146 | |
| 1147 | /* Scoping flags are mutually exclusive. */ |
| 1148 | if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT)) |
| 1149 | return -EINVAL; |
| 1150 | |
| 1151 | /* Deal with the mode. */ |
| 1152 | if (WILL_CREATE(flags)) { |
| 1153 | if (how->mode & ~S_IALLUGO) |
| 1154 | return -EINVAL; |
| 1155 | op->mode = how->mode | S_IFREG; |
| 1156 | } else { |
| 1157 | if (how->mode != 0) |
| 1158 | return -EINVAL; |
| 1159 | op->mode = 0; |
| 1160 | } |
| 1161 | |
| 1162 | /* |
| 1163 | * In order to ensure programs get explicit errors when trying to use |
| 1164 | * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it |
| 1165 | * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we |
| 1166 | * have to require userspace to explicitly set it. |
| 1167 | */ |
| 1168 | if (flags & __O_TMPFILE) { |
| 1169 | if ((flags & O_TMPFILE_MASK) != O_TMPFILE) |
| 1170 | return -EINVAL; |
| 1171 | if (!(acc_mode & MAY_WRITE)) |
| 1172 | return -EINVAL; |
| 1173 | } |
| 1174 | if (flags & O_PATH) { |
| 1175 | /* O_PATH only permits certain other flags to be set. */ |
| 1176 | if (flags & ~O_PATH_FLAGS) |
| 1177 | return -EINVAL; |
| 1178 | acc_mode = 0; |
| 1179 | } |
| 1180 | |
| 1181 | /* |
| 1182 | * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only |
| 1183 | * check for O_DSYNC if the need any syncing at all we enforce it's |
| 1184 | * always set instead of having to deal with possibly weird behaviour |
| 1185 | * for malicious applications setting only __O_SYNC. |
| 1186 | */ |
| 1187 | if (flags & __O_SYNC) |
| 1188 | flags |= O_DSYNC; |
| 1189 | |
| 1190 | op->open_flag = flags; |
| 1191 | |
| 1192 | /* O_TRUNC implies we need access checks for write permissions */ |
| 1193 | if (flags & O_TRUNC) |
| 1194 | acc_mode |= MAY_WRITE; |
| 1195 | |
| 1196 | /* Allow the LSM permission hook to distinguish append |
| 1197 | access from general write access. */ |
| 1198 | if (flags & O_APPEND) |
| 1199 | acc_mode |= MAY_APPEND; |
| 1200 | |
| 1201 | op->acc_mode = acc_mode; |
| 1202 | |
| 1203 | op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN; |
| 1204 | |
| 1205 | if (flags & O_CREAT) { |
| 1206 | op->intent |= LOOKUP_CREATE; |
| 1207 | if (flags & O_EXCL) { |
| 1208 | op->intent |= LOOKUP_EXCL; |
| 1209 | flags |= O_NOFOLLOW; |
| 1210 | } |
| 1211 | } |
| 1212 | |
| 1213 | if (flags & O_DIRECTORY) |
| 1214 | lookup_flags |= LOOKUP_DIRECTORY; |
| 1215 | if (!(flags & O_NOFOLLOW)) |
| 1216 | lookup_flags |= LOOKUP_FOLLOW; |
| 1217 | |
| 1218 | if (how->resolve & RESOLVE_NO_XDEV) |
| 1219 | lookup_flags |= LOOKUP_NO_XDEV; |
| 1220 | if (how->resolve & RESOLVE_NO_MAGICLINKS) |
| 1221 | lookup_flags |= LOOKUP_NO_MAGICLINKS; |
| 1222 | if (how->resolve & RESOLVE_NO_SYMLINKS) |
| 1223 | lookup_flags |= LOOKUP_NO_SYMLINKS; |
| 1224 | if (how->resolve & RESOLVE_BENEATH) |
| 1225 | lookup_flags |= LOOKUP_BENEATH; |
| 1226 | if (how->resolve & RESOLVE_IN_ROOT) |
| 1227 | lookup_flags |= LOOKUP_IN_ROOT; |
| 1228 | if (how->resolve & RESOLVE_CACHED) { |
| 1229 | /* Don't bother even trying for create/truncate/tmpfile open */ |
| 1230 | if (flags & (O_TRUNC | O_CREAT | O_TMPFILE)) |
| 1231 | return -EAGAIN; |
| 1232 | lookup_flags |= LOOKUP_CACHED; |
| 1233 | } |
| 1234 | |
| 1235 | op->lookup_flags = lookup_flags; |
| 1236 | return 0; |
| 1237 | } |
| 1238 | |
| 1239 | /** |
| 1240 | * file_open_name - open file and return file pointer |
| 1241 | * |
| 1242 | * @name: struct filename containing path to open |
| 1243 | * @flags: open flags as per the open(2) second argument |
| 1244 | * @mode: mode for the new file if O_CREAT is set, else ignored |
| 1245 | * |
| 1246 | * This is the helper to open a file from kernelspace if you really |
| 1247 | * have to. But in generally you should not do this, so please move |
| 1248 | * along, nothing to see here.. |
| 1249 | */ |
| 1250 | struct file *file_open_name(struct filename *name, int flags, umode_t mode) |
| 1251 | { |
| 1252 | struct open_flags op; |
| 1253 | struct open_how how = build_open_how(flags, mode); |
| 1254 | int err = build_open_flags(&how, &op); |
| 1255 | if (err) |
| 1256 | return ERR_PTR(err); |
| 1257 | return do_filp_open(AT_FDCWD, name, &op); |
| 1258 | } |
| 1259 | |
| 1260 | /** |
| 1261 | * filp_open - open file and return file pointer |
| 1262 | * |
| 1263 | * @filename: path to open |
| 1264 | * @flags: open flags as per the open(2) second argument |
| 1265 | * @mode: mode for the new file if O_CREAT is set, else ignored |
| 1266 | * |
| 1267 | * This is the helper to open a file from kernelspace if you really |
| 1268 | * have to. But in generally you should not do this, so please move |
| 1269 | * along, nothing to see here.. |
| 1270 | */ |
| 1271 | struct file *filp_open(const char *filename, int flags, umode_t mode) |
| 1272 | { |
| 1273 | struct filename *name = getname_kernel(filename); |
| 1274 | struct file *file = ERR_CAST(name); |
| 1275 | |
| 1276 | if (!IS_ERR(name)) { |
| 1277 | file = file_open_name(name, flags, mode); |
| 1278 | putname(name); |
| 1279 | } |
| 1280 | return file; |
| 1281 | } |
| 1282 | EXPORT_SYMBOL(filp_open); |
| 1283 | |
| 1284 | struct file *file_open_root(const struct path *root, |
| 1285 | const char *filename, int flags, umode_t mode) |
| 1286 | { |
| 1287 | struct open_flags op; |
| 1288 | struct open_how how = build_open_how(flags, mode); |
| 1289 | int err = build_open_flags(&how, &op); |
| 1290 | if (err) |
| 1291 | return ERR_PTR(err); |
| 1292 | return do_file_open_root(root, filename, &op); |
| 1293 | } |
| 1294 | EXPORT_SYMBOL(file_open_root); |
| 1295 | |
| 1296 | static long do_sys_openat2(int dfd, const char __user *filename, |
| 1297 | struct open_how *how) |
| 1298 | { |
| 1299 | struct open_flags op; |
| 1300 | int fd = build_open_flags(how, &op); |
| 1301 | struct filename *tmp; |
| 1302 | |
| 1303 | if (fd) |
| 1304 | return fd; |
| 1305 | |
| 1306 | tmp = getname(filename); |
| 1307 | if (IS_ERR(tmp)) |
| 1308 | return PTR_ERR(tmp); |
| 1309 | |
| 1310 | fd = get_unused_fd_flags(how->flags); |
| 1311 | if (fd >= 0) { |
| 1312 | struct file *f = do_filp_open(dfd, tmp, &op); |
| 1313 | if (IS_ERR(f)) { |
| 1314 | put_unused_fd(fd); |
| 1315 | fd = PTR_ERR(f); |
| 1316 | } else { |
| 1317 | fsnotify_open(f); |
| 1318 | fd_install(fd, f); |
| 1319 | } |
| 1320 | } |
| 1321 | putname(tmp); |
| 1322 | return fd; |
| 1323 | } |
| 1324 | |
| 1325 | long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode) |
| 1326 | { |
| 1327 | struct open_how how = build_open_how(flags, mode); |
| 1328 | return do_sys_openat2(dfd, filename, &how); |
| 1329 | } |
| 1330 | |
| 1331 | |
| 1332 | SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) |
| 1333 | { |
| 1334 | if (force_o_largefile()) |
| 1335 | flags |= O_LARGEFILE; |
| 1336 | return do_sys_open(AT_FDCWD, filename, flags, mode); |
| 1337 | } |
| 1338 | |
| 1339 | SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, |
| 1340 | umode_t, mode) |
| 1341 | { |
| 1342 | if (force_o_largefile()) |
| 1343 | flags |= O_LARGEFILE; |
| 1344 | return do_sys_open(dfd, filename, flags, mode); |
| 1345 | } |
| 1346 | |
| 1347 | SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename, |
| 1348 | struct open_how __user *, how, size_t, usize) |
| 1349 | { |
| 1350 | int err; |
| 1351 | struct open_how tmp; |
| 1352 | |
| 1353 | BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0); |
| 1354 | BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST); |
| 1355 | |
| 1356 | if (unlikely(usize < OPEN_HOW_SIZE_VER0)) |
| 1357 | return -EINVAL; |
| 1358 | |
| 1359 | err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize); |
| 1360 | if (err) |
| 1361 | return err; |
| 1362 | |
| 1363 | audit_openat2_how(&tmp); |
| 1364 | |
| 1365 | /* O_LARGEFILE is only allowed for non-O_PATH. */ |
| 1366 | if (!(tmp.flags & O_PATH) && force_o_largefile()) |
| 1367 | tmp.flags |= O_LARGEFILE; |
| 1368 | |
| 1369 | return do_sys_openat2(dfd, filename, &tmp); |
| 1370 | } |
| 1371 | |
| 1372 | #ifdef CONFIG_COMPAT |
| 1373 | /* |
| 1374 | * Exactly like sys_open(), except that it doesn't set the |
| 1375 | * O_LARGEFILE flag. |
| 1376 | */ |
| 1377 | COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) |
| 1378 | { |
| 1379 | return do_sys_open(AT_FDCWD, filename, flags, mode); |
| 1380 | } |
| 1381 | |
| 1382 | /* |
| 1383 | * Exactly like sys_openat(), except that it doesn't set the |
| 1384 | * O_LARGEFILE flag. |
| 1385 | */ |
| 1386 | COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode) |
| 1387 | { |
| 1388 | return do_sys_open(dfd, filename, flags, mode); |
| 1389 | } |
| 1390 | #endif |
| 1391 | |
| 1392 | #ifndef __alpha__ |
| 1393 | |
| 1394 | /* |
| 1395 | * For backward compatibility? Maybe this should be moved |
| 1396 | * into arch/i386 instead? |
| 1397 | */ |
| 1398 | SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode) |
| 1399 | { |
| 1400 | int flags = O_CREAT | O_WRONLY | O_TRUNC; |
| 1401 | |
| 1402 | if (force_o_largefile()) |
| 1403 | flags |= O_LARGEFILE; |
| 1404 | return do_sys_open(AT_FDCWD, pathname, flags, mode); |
| 1405 | } |
| 1406 | #endif |
| 1407 | |
| 1408 | /* |
| 1409 | * "id" is the POSIX thread ID. We use the |
| 1410 | * files pointer for this.. |
| 1411 | */ |
| 1412 | int filp_close(struct file *filp, fl_owner_t id) |
| 1413 | { |
| 1414 | int retval = 0; |
| 1415 | |
| 1416 | if (!file_count(filp)) { |
| 1417 | printk(KERN_ERR "VFS: Close: file count is 0\n"); |
| 1418 | return 0; |
| 1419 | } |
| 1420 | |
| 1421 | if (filp->f_op->flush) |
| 1422 | retval = filp->f_op->flush(filp, id); |
| 1423 | |
| 1424 | if (likely(!(filp->f_mode & FMODE_PATH))) { |
| 1425 | dnotify_flush(filp, id); |
| 1426 | locks_remove_posix(filp, id); |
| 1427 | } |
| 1428 | fput(filp); |
| 1429 | return retval; |
| 1430 | } |
| 1431 | |
| 1432 | EXPORT_SYMBOL(filp_close); |
| 1433 | |
| 1434 | /* |
| 1435 | * Careful here! We test whether the file pointer is NULL before |
| 1436 | * releasing the fd. This ensures that one clone task can't release |
| 1437 | * an fd while another clone is opening it. |
| 1438 | */ |
| 1439 | SYSCALL_DEFINE1(close, unsigned int, fd) |
| 1440 | { |
| 1441 | int retval = close_fd(fd); |
| 1442 | |
| 1443 | /* can't restart close syscall because file table entry was cleared */ |
| 1444 | if (unlikely(retval == -ERESTARTSYS || |
| 1445 | retval == -ERESTARTNOINTR || |
| 1446 | retval == -ERESTARTNOHAND || |
| 1447 | retval == -ERESTART_RESTARTBLOCK)) |
| 1448 | retval = -EINTR; |
| 1449 | |
| 1450 | return retval; |
| 1451 | } |
| 1452 | |
| 1453 | /** |
| 1454 | * close_range() - Close all file descriptors in a given range. |
| 1455 | * |
| 1456 | * @fd: starting file descriptor to close |
| 1457 | * @max_fd: last file descriptor to close |
| 1458 | * @flags: reserved for future extensions |
| 1459 | * |
| 1460 | * This closes a range of file descriptors. All file descriptors |
| 1461 | * from @fd up to and including @max_fd are closed. |
| 1462 | * Currently, errors to close a given file descriptor are ignored. |
| 1463 | */ |
| 1464 | SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd, |
| 1465 | unsigned int, flags) |
| 1466 | { |
| 1467 | return __close_range(fd, max_fd, flags); |
| 1468 | } |
| 1469 | |
| 1470 | /* |
| 1471 | * This routine simulates a hangup on the tty, to arrange that users |
| 1472 | * are given clean terminals at login time. |
| 1473 | */ |
| 1474 | SYSCALL_DEFINE0(vhangup) |
| 1475 | { |
| 1476 | if (capable(CAP_SYS_TTY_CONFIG)) { |
| 1477 | tty_vhangup_self(); |
| 1478 | return 0; |
| 1479 | } |
| 1480 | return -EPERM; |
| 1481 | } |
| 1482 | |
| 1483 | /* |
| 1484 | * Called when an inode is about to be open. |
| 1485 | * We use this to disallow opening large files on 32bit systems if |
| 1486 | * the caller didn't specify O_LARGEFILE. On 64bit systems we force |
| 1487 | * on this flag in sys_open. |
| 1488 | */ |
| 1489 | int generic_file_open(struct inode * inode, struct file * filp) |
| 1490 | { |
| 1491 | if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
| 1492 | return -EOVERFLOW; |
| 1493 | return 0; |
| 1494 | } |
| 1495 | |
| 1496 | EXPORT_SYMBOL(generic_file_open); |
| 1497 | |
| 1498 | /* |
| 1499 | * This is used by subsystems that don't want seekable |
| 1500 | * file descriptors. The function is not supposed to ever fail, the only |
| 1501 | * reason it returns an 'int' and not 'void' is so that it can be plugged |
| 1502 | * directly into file_operations structure. |
| 1503 | */ |
| 1504 | int nonseekable_open(struct inode *inode, struct file *filp) |
| 1505 | { |
| 1506 | filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE); |
| 1507 | return 0; |
| 1508 | } |
| 1509 | |
| 1510 | EXPORT_SYMBOL(nonseekable_open); |
| 1511 | |
| 1512 | /* |
| 1513 | * stream_open is used by subsystems that want stream-like file descriptors. |
| 1514 | * Such file descriptors are not seekable and don't have notion of position |
| 1515 | * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL). |
| 1516 | * Contrary to file descriptors of other regular files, .read() and .write() |
| 1517 | * can run simultaneously. |
| 1518 | * |
| 1519 | * stream_open never fails and is marked to return int so that it could be |
| 1520 | * directly used as file_operations.open . |
| 1521 | */ |
| 1522 | int stream_open(struct inode *inode, struct file *filp) |
| 1523 | { |
| 1524 | filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS); |
| 1525 | filp->f_mode |= FMODE_STREAM; |
| 1526 | return 0; |
| 1527 | } |
| 1528 | |
| 1529 | EXPORT_SYMBOL(stream_open); |