Linux Audio

Check our new training course

Loading...
v5.9
   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
  36#include "internal.h"
  37
  38int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  39	struct file *filp)
  40{
  41	int ret;
  42	struct iattr newattrs;
  43
  44	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  45	if (length < 0)
  46		return -EINVAL;
  47
  48	newattrs.ia_size = length;
  49	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  50	if (filp) {
  51		newattrs.ia_file = filp;
  52		newattrs.ia_valid |= ATTR_FILE;
  53	}
  54
  55	/* Remove suid, sgid, and file capabilities on truncate too */
  56	ret = dentry_needs_remove_privs(dentry);
  57	if (ret < 0)
  58		return ret;
  59	if (ret)
  60		newattrs.ia_valid |= ret | ATTR_FORCE;
  61
  62	inode_lock(dentry->d_inode);
  63	/* Note any delegations or leases have already been broken: */
  64	ret = notify_change(dentry, &newattrs, NULL);
  65	inode_unlock(dentry->d_inode);
  66	return ret;
  67}
  68
  69long vfs_truncate(const struct path *path, loff_t length)
  70{
 
  71	struct inode *inode;
  72	long error;
  73
  74	inode = path->dentry->d_inode;
  75
  76	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77	if (S_ISDIR(inode->i_mode))
  78		return -EISDIR;
  79	if (!S_ISREG(inode->i_mode))
  80		return -EINVAL;
  81
  82	error = mnt_want_write(path->mnt);
  83	if (error)
  84		goto out;
  85
  86	error = inode_permission(inode, MAY_WRITE);
 
  87	if (error)
  88		goto mnt_drop_write_and_out;
  89
  90	error = -EPERM;
  91	if (IS_APPEND(inode))
  92		goto mnt_drop_write_and_out;
  93
  94	error = get_write_access(inode);
  95	if (error)
  96		goto mnt_drop_write_and_out;
  97
  98	/*
  99	 * Make sure that there are no leases.  get_write_access() protects
 100	 * against the truncate racing with a lease-granting setlease().
 101	 */
 102	error = break_lease(inode, O_WRONLY);
 103	if (error)
 104		goto put_write_and_out;
 105
 106	error = locks_verify_truncate(inode, NULL, length);
 107	if (!error)
 108		error = security_path_truncate(path);
 109	if (!error)
 110		error = do_truncate(path->dentry, length, 0, NULL);
 111
 112put_write_and_out:
 113	put_write_access(inode);
 114mnt_drop_write_and_out:
 115	mnt_drop_write(path->mnt);
 116out:
 117	return error;
 118}
 119EXPORT_SYMBOL_GPL(vfs_truncate);
 120
 121long do_sys_truncate(const char __user *pathname, loff_t length)
 122{
 123	unsigned int lookup_flags = LOOKUP_FOLLOW;
 124	struct path path;
 125	int error;
 126
 127	if (length < 0)	/* sorry, but loff_t says... */
 128		return -EINVAL;
 129
 130retry:
 131	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 132	if (!error) {
 133		error = vfs_truncate(&path, length);
 134		path_put(&path);
 135	}
 136	if (retry_estale(error, lookup_flags)) {
 137		lookup_flags |= LOOKUP_REVAL;
 138		goto retry;
 139	}
 140	return error;
 141}
 142
 143SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 144{
 145	return do_sys_truncate(path, length);
 146}
 147
 148#ifdef CONFIG_COMPAT
 149COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 150{
 151	return do_sys_truncate(path, length);
 152}
 153#endif
 154
 155long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 156{
 157	struct inode *inode;
 158	struct dentry *dentry;
 159	struct fd f;
 160	int error;
 161
 162	error = -EINVAL;
 163	if (length < 0)
 164		goto out;
 165	error = -EBADF;
 166	f = fdget(fd);
 167	if (!f.file)
 168		goto out;
 169
 170	/* explicitly opened as large or we are on 64-bit box */
 171	if (f.file->f_flags & O_LARGEFILE)
 172		small = 0;
 173
 174	dentry = f.file->f_path.dentry;
 175	inode = dentry->d_inode;
 176	error = -EINVAL;
 177	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 178		goto out_putf;
 179
 180	error = -EINVAL;
 181	/* Cannot ftruncate over 2^31 bytes without large file support */
 182	if (small && length > MAX_NON_LFS)
 183		goto out_putf;
 184
 185	error = -EPERM;
 186	/* Check IS_APPEND on real upper inode */
 187	if (IS_APPEND(file_inode(f.file)))
 188		goto out_putf;
 189
 190	sb_start_write(inode->i_sb);
 191	error = locks_verify_truncate(inode, f.file, length);
 192	if (!error)
 193		error = security_path_truncate(&f.file->f_path);
 194	if (!error)
 195		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 
 196	sb_end_write(inode->i_sb);
 197out_putf:
 198	fdput(f);
 199out:
 200	return error;
 201}
 202
 203SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 204{
 205	return do_sys_ftruncate(fd, length, 1);
 206}
 207
 208#ifdef CONFIG_COMPAT
 209COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 210{
 211	return do_sys_ftruncate(fd, length, 1);
 212}
 213#endif
 214
 215/* LFS versions of truncate are only needed on 32 bit machines */
 216#if BITS_PER_LONG == 32
 217SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 218{
 219	return do_sys_truncate(path, length);
 220}
 221
 222SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 223{
 224	return do_sys_ftruncate(fd, length, 0);
 225}
 226#endif /* BITS_PER_LONG == 32 */
 227
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 228
 229int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 230{
 231	struct inode *inode = file_inode(file);
 232	long ret;
 233
 234	if (offset < 0 || len <= 0)
 235		return -EINVAL;
 236
 237	/* Return error if mode is not supported */
 238	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 239		return -EOPNOTSUPP;
 240
 241	/* Punch hole and zero range are mutually exclusive */
 242	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 243	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 244		return -EOPNOTSUPP;
 245
 246	/* Punch hole must have keep size set */
 247	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 248	    !(mode & FALLOC_FL_KEEP_SIZE))
 249		return -EOPNOTSUPP;
 250
 251	/* Collapse range should only be used exclusively. */
 252	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 253	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 254		return -EINVAL;
 255
 256	/* Insert range should only be used exclusively. */
 257	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 258	    (mode & ~FALLOC_FL_INSERT_RANGE))
 259		return -EINVAL;
 260
 261	/* Unshare range should only be used with allocate mode. */
 262	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 263	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 264		return -EINVAL;
 265
 266	if (!(file->f_mode & FMODE_WRITE))
 267		return -EBADF;
 268
 269	/*
 270	 * We can only allow pure fallocate on append only files
 271	 */
 272	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 273		return -EPERM;
 274
 275	if (IS_IMMUTABLE(inode))
 276		return -EPERM;
 277
 278	/*
 279	 * We cannot allow any fallocate operation on an active swapfile
 280	 */
 281	if (IS_SWAPFILE(inode))
 282		return -ETXTBSY;
 283
 284	/*
 285	 * Revalidate the write permissions, in case security policy has
 286	 * changed since the files were opened.
 287	 */
 288	ret = security_file_permission(file, MAY_WRITE);
 289	if (ret)
 290		return ret;
 291
 
 
 
 
 292	if (S_ISFIFO(inode->i_mode))
 293		return -ESPIPE;
 294
 295	if (S_ISDIR(inode->i_mode))
 296		return -EISDIR;
 297
 298	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 299		return -ENODEV;
 300
 301	/* Check for wrap through zero too */
 302	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 303		return -EFBIG;
 304
 305	if (!file->f_op->fallocate)
 306		return -EOPNOTSUPP;
 307
 308	file_start_write(file);
 309	ret = file->f_op->fallocate(file, mode, offset, len);
 310
 311	/*
 312	 * Create inotify and fanotify events.
 313	 *
 314	 * To keep the logic simple always create events if fallocate succeeds.
 315	 * This implies that events are even created if the file size remains
 316	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 317	 */
 318	if (ret == 0)
 319		fsnotify_modify(file);
 320
 321	file_end_write(file);
 322	return ret;
 323}
 324EXPORT_SYMBOL_GPL(vfs_fallocate);
 325
 326int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 327{
 328	struct fd f = fdget(fd);
 329	int error = -EBADF;
 330
 331	if (f.file) {
 332		error = vfs_fallocate(f.file, mode, offset, len);
 333		fdput(f);
 334	}
 335	return error;
 336}
 337
 338SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 339{
 340	return ksys_fallocate(fd, mode, offset, len);
 341}
 342
 
 
 
 
 
 
 
 
 
 343/*
 344 * access() needs to use the real uid/gid, not the effective uid/gid.
 345 * We do this by temporarily clearing all FS-related capabilities and
 346 * switching the fsuid/fsgid around to the real ones.
 
 
 
 347 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 348static const struct cred *access_override_creds(void)
 349{
 350	const struct cred *old_cred;
 351	struct cred *override_cred;
 352
 353	override_cred = prepare_creds();
 354	if (!override_cred)
 355		return NULL;
 356
 
 
 
 
 
 
 357	override_cred->fsuid = override_cred->uid;
 358	override_cred->fsgid = override_cred->gid;
 359
 360	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 361		/* Clear the capabilities if we switch to a non-root user */
 362		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 363		if (!uid_eq(override_cred->uid, root_uid))
 364			cap_clear(override_cred->cap_effective);
 365		else
 366			override_cred->cap_effective =
 367				override_cred->cap_permitted;
 368	}
 369
 370	/*
 371	 * The new set of credentials can *only* be used in
 372	 * task-synchronous circumstances, and does not need
 373	 * RCU freeing, unless somebody then takes a separate
 374	 * reference to it.
 375	 *
 376	 * NOTE! This is _only_ true because this credential
 377	 * is used purely for override_creds() that installs
 378	 * it as the subjective cred. Other threads will be
 379	 * accessing ->real_cred, not the subjective cred.
 380	 *
 381	 * If somebody _does_ make a copy of this (using the
 382	 * 'get_current_cred()' function), that will clear the
 383	 * non_rcu field, because now that other user may be
 384	 * expecting RCU freeing. But normal thread-synchronous
 385	 * cred accesses will keep things non-RCY.
 
 386	 */
 387	override_cred->non_rcu = 1;
 388
 389	old_cred = override_creds(override_cred);
 390
 391	/* override_cred() gets its own ref */
 392	put_cred(override_cred);
 393
 394	return old_cred;
 395}
 396
 397static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
 398{
 399	struct path path;
 400	struct inode *inode;
 401	int res;
 402	unsigned int lookup_flags = LOOKUP_FOLLOW;
 403	const struct cred *old_cred = NULL;
 404
 405	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 406		return -EINVAL;
 407
 408	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
 409		return -EINVAL;
 410
 411	if (flags & AT_SYMLINK_NOFOLLOW)
 412		lookup_flags &= ~LOOKUP_FOLLOW;
 413	if (flags & AT_EMPTY_PATH)
 414		lookup_flags |= LOOKUP_EMPTY;
 415
 416	if (!(flags & AT_EACCESS)) {
 417		old_cred = access_override_creds();
 418		if (!old_cred)
 419			return -ENOMEM;
 420	}
 421
 422retry:
 423	res = user_path_at(dfd, filename, lookup_flags, &path);
 424	if (res)
 425		goto out;
 426
 427	inode = d_backing_inode(path.dentry);
 428
 429	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 430		/*
 431		 * MAY_EXEC on regular files is denied if the fs is mounted
 432		 * with the "noexec" flag.
 433		 */
 434		res = -EACCES;
 435		if (path_noexec(&path))
 436			goto out_path_release;
 437	}
 438
 439	res = inode_permission(inode, mode | MAY_ACCESS);
 440	/* SuS v2 requires we report a read only fs too */
 441	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 442		goto out_path_release;
 443	/*
 444	 * This is a rare case where using __mnt_is_readonly()
 445	 * is OK without a mnt_want/drop_write() pair.  Since
 446	 * no actual write to the fs is performed here, we do
 447	 * not need to telegraph to that to anyone.
 448	 *
 449	 * By doing this, we accept that this access is
 450	 * inherently racy and know that the fs may change
 451	 * state before we even see this result.
 452	 */
 453	if (__mnt_is_readonly(path.mnt))
 454		res = -EROFS;
 455
 456out_path_release:
 457	path_put(&path);
 458	if (retry_estale(res, lookup_flags)) {
 459		lookup_flags |= LOOKUP_REVAL;
 460		goto retry;
 461	}
 462out:
 463	if (old_cred)
 464		revert_creds(old_cred);
 465
 466	return res;
 467}
 468
 469SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 470{
 471	return do_faccessat(dfd, filename, mode, 0);
 472}
 473
 474SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
 475		int, flags)
 476{
 477	return do_faccessat(dfd, filename, mode, flags);
 478}
 479
 480SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 481{
 482	return do_faccessat(AT_FDCWD, filename, mode, 0);
 483}
 484
 485SYSCALL_DEFINE1(chdir, const char __user *, filename)
 486{
 487	struct path path;
 488	int error;
 489	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 490retry:
 491	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 492	if (error)
 493		goto out;
 494
 495	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 496	if (error)
 497		goto dput_and_out;
 498
 499	set_fs_pwd(current->fs, &path);
 500
 501dput_and_out:
 502	path_put(&path);
 503	if (retry_estale(error, lookup_flags)) {
 504		lookup_flags |= LOOKUP_REVAL;
 505		goto retry;
 506	}
 507out:
 508	return error;
 509}
 510
 511SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 512{
 513	struct fd f = fdget_raw(fd);
 514	int error;
 515
 516	error = -EBADF;
 517	if (!f.file)
 518		goto out;
 519
 520	error = -ENOTDIR;
 521	if (!d_can_lookup(f.file->f_path.dentry))
 522		goto out_putf;
 523
 524	error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 525	if (!error)
 526		set_fs_pwd(current->fs, &f.file->f_path);
 527out_putf:
 528	fdput(f);
 529out:
 530	return error;
 531}
 532
 533SYSCALL_DEFINE1(chroot, const char __user *, filename)
 534{
 535	struct path path;
 536	int error;
 537	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 538retry:
 539	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 540	if (error)
 541		goto out;
 542
 543	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 544	if (error)
 545		goto dput_and_out;
 546
 547	error = -EPERM;
 548	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 549		goto dput_and_out;
 550	error = security_path_chroot(&path);
 551	if (error)
 552		goto dput_and_out;
 553
 554	set_fs_root(current->fs, &path);
 555	error = 0;
 556dput_and_out:
 557	path_put(&path);
 558	if (retry_estale(error, lookup_flags)) {
 559		lookup_flags |= LOOKUP_REVAL;
 560		goto retry;
 561	}
 562out:
 563	return error;
 564}
 565
 566int chmod_common(const struct path *path, umode_t mode)
 567{
 568	struct inode *inode = path->dentry->d_inode;
 569	struct inode *delegated_inode = NULL;
 570	struct iattr newattrs;
 571	int error;
 572
 573	error = mnt_want_write(path->mnt);
 574	if (error)
 575		return error;
 576retry_deleg:
 577	inode_lock(inode);
 578	error = security_path_chmod(path, mode);
 579	if (error)
 580		goto out_unlock;
 581	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 582	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 583	error = notify_change(path->dentry, &newattrs, &delegated_inode);
 
 584out_unlock:
 585	inode_unlock(inode);
 586	if (delegated_inode) {
 587		error = break_deleg_wait(&delegated_inode);
 588		if (!error)
 589			goto retry_deleg;
 590	}
 591	mnt_drop_write(path->mnt);
 592	return error;
 593}
 594
 595int vfs_fchmod(struct file *file, umode_t mode)
 596{
 597	audit_file(file);
 598	return chmod_common(&file->f_path, mode);
 599}
 600
 601SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 602{
 603	struct fd f = fdget(fd);
 604	int err = -EBADF;
 605
 606	if (f.file) {
 607		err = vfs_fchmod(f.file, mode);
 608		fdput(f);
 609	}
 610	return err;
 611}
 612
 613static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 
 614{
 615	struct path path;
 616	int error;
 617	unsigned int lookup_flags = LOOKUP_FOLLOW;
 
 
 
 
 
 
 
 
 618retry:
 619	error = user_path_at(dfd, filename, lookup_flags, &path);
 620	if (!error) {
 621		error = chmod_common(&path, mode);
 622		path_put(&path);
 623		if (retry_estale(error, lookup_flags)) {
 624			lookup_flags |= LOOKUP_REVAL;
 625			goto retry;
 626		}
 627	}
 628	return error;
 629}
 630
 
 
 
 
 
 
 631SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 632		umode_t, mode)
 633{
 634	return do_fchmodat(dfd, filename, mode);
 635}
 636
 637SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 638{
 639	return do_fchmodat(AT_FDCWD, filename, mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 640}
 641
 642int chown_common(const struct path *path, uid_t user, gid_t group)
 643{
 
 
 644	struct inode *inode = path->dentry->d_inode;
 645	struct inode *delegated_inode = NULL;
 646	int error;
 647	struct iattr newattrs;
 648	kuid_t uid;
 649	kgid_t gid;
 650
 651	uid = make_kuid(current_user_ns(), user);
 652	gid = make_kgid(current_user_ns(), group);
 653
 
 
 
 654retry_deleg:
 
 
 655	newattrs.ia_valid =  ATTR_CTIME;
 656	if (user != (uid_t) -1) {
 657		if (!uid_valid(uid))
 658			return -EINVAL;
 659		newattrs.ia_valid |= ATTR_UID;
 660		newattrs.ia_uid = uid;
 661	}
 662	if (group != (gid_t) -1) {
 663		if (!gid_valid(gid))
 664			return -EINVAL;
 665		newattrs.ia_valid |= ATTR_GID;
 666		newattrs.ia_gid = gid;
 667	}
 668	if (!S_ISDIR(inode->i_mode))
 669		newattrs.ia_valid |=
 670			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 671	inode_lock(inode);
 672	error = security_path_chown(path, uid, gid);
 
 
 
 
 
 
 
 673	if (!error)
 674		error = notify_change(path->dentry, &newattrs, &delegated_inode);
 
 675	inode_unlock(inode);
 676	if (delegated_inode) {
 677		error = break_deleg_wait(&delegated_inode);
 678		if (!error)
 679			goto retry_deleg;
 680	}
 681	return error;
 682}
 683
 684int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 685		int flag)
 686{
 687	struct path path;
 688	int error = -EINVAL;
 689	int lookup_flags;
 690
 691	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 692		goto out;
 693
 694	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 695	if (flag & AT_EMPTY_PATH)
 696		lookup_flags |= LOOKUP_EMPTY;
 697retry:
 698	error = user_path_at(dfd, filename, lookup_flags, &path);
 699	if (error)
 700		goto out;
 701	error = mnt_want_write(path.mnt);
 702	if (error)
 703		goto out_release;
 704	error = chown_common(&path, user, group);
 705	mnt_drop_write(path.mnt);
 706out_release:
 707	path_put(&path);
 708	if (retry_estale(error, lookup_flags)) {
 709		lookup_flags |= LOOKUP_REVAL;
 710		goto retry;
 711	}
 712out:
 713	return error;
 714}
 715
 716SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 717		gid_t, group, int, flag)
 718{
 719	return do_fchownat(dfd, filename, user, group, flag);
 720}
 721
 722SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 723{
 724	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 725}
 726
 727SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 728{
 729	return do_fchownat(AT_FDCWD, filename, user, group,
 730			   AT_SYMLINK_NOFOLLOW);
 731}
 732
 733int vfs_fchown(struct file *file, uid_t user, gid_t group)
 734{
 735	int error;
 736
 737	error = mnt_want_write_file(file);
 738	if (error)
 739		return error;
 740	audit_file(file);
 741	error = chown_common(&file->f_path, user, group);
 742	mnt_drop_write_file(file);
 743	return error;
 744}
 745
 746int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 747{
 748	struct fd f = fdget(fd);
 749	int error = -EBADF;
 750
 751	if (f.file) {
 752		error = vfs_fchown(f.file, user, group);
 753		fdput(f);
 754	}
 755	return error;
 756}
 757
 758SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 759{
 760	return ksys_fchown(fd, user, group);
 761}
 762
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 763static int do_dentry_open(struct file *f,
 764			  struct inode *inode,
 765			  int (*open)(struct inode *, struct file *))
 766{
 767	static const struct file_operations empty_fops = {};
 768	int error;
 769
 770	path_get(&f->f_path);
 771	f->f_inode = inode;
 772	f->f_mapping = inode->i_mapping;
 773	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 774	f->f_sb_err = file_sample_sb_err(f);
 775
 776	if (unlikely(f->f_flags & O_PATH)) {
 777		f->f_mode = FMODE_PATH | FMODE_OPENED;
 778		f->f_op = &empty_fops;
 779		return 0;
 780	}
 781
 782	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 783		error = get_write_access(inode);
 
 
 784		if (unlikely(error))
 785			goto cleanup_file;
 786		error = __mnt_want_write(f->f_path.mnt);
 787		if (unlikely(error)) {
 788			put_write_access(inode);
 789			goto cleanup_file;
 790		}
 791		f->f_mode |= FMODE_WRITER;
 792	}
 793
 794	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 795	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 796		f->f_mode |= FMODE_ATOMIC_POS;
 797
 798	f->f_op = fops_get(inode->i_fop);
 799	if (WARN_ON(!f->f_op)) {
 800		error = -ENODEV;
 801		goto cleanup_all;
 802	}
 803
 804	error = security_file_open(f);
 805	if (error)
 806		goto cleanup_all;
 807
 808	error = break_lease(locks_inode(f), f->f_flags);
 809	if (error)
 810		goto cleanup_all;
 811
 812	/* normally all 3 are set; ->open() can clear them if needed */
 813	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 814	if (!open)
 815		open = f->f_op->open;
 816	if (open) {
 817		error = open(inode, f);
 818		if (error)
 819			goto cleanup_all;
 820	}
 821	f->f_mode |= FMODE_OPENED;
 822	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 823		i_readcount_inc(inode);
 824	if ((f->f_mode & FMODE_READ) &&
 825	     likely(f->f_op->read || f->f_op->read_iter))
 826		f->f_mode |= FMODE_CAN_READ;
 827	if ((f->f_mode & FMODE_WRITE) &&
 828	     likely(f->f_op->write || f->f_op->write_iter))
 829		f->f_mode |= FMODE_CAN_WRITE;
 
 
 
 
 830
 831	f->f_write_hint = WRITE_LIFE_NOT_SET;
 832	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 
 833
 834	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 835
 836	/* NB: we're sure to have correct a_ops only after f_op->open */
 837	if (f->f_flags & O_DIRECT) {
 838		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 839			return -EINVAL;
 840	}
 841
 842	/*
 843	 * XXX: Huge page cache doesn't support writing yet. Drop all page
 844	 * cache for this file before processing writes.
 845	 */
 846	if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
 847		truncate_pagecache(inode, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 848
 
 
 
 
 
 849	return 0;
 850
 851cleanup_all:
 852	if (WARN_ON_ONCE(error > 0))
 853		error = -EINVAL;
 854	fops_put(f->f_op);
 855	if (f->f_mode & FMODE_WRITER) {
 856		put_write_access(inode);
 857		__mnt_drop_write(f->f_path.mnt);
 858	}
 859cleanup_file:
 860	path_put(&f->f_path);
 861	f->f_path.mnt = NULL;
 862	f->f_path.dentry = NULL;
 863	f->f_inode = NULL;
 864	return error;
 865}
 866
 867/**
 868 * finish_open - finish opening a file
 869 * @file: file pointer
 870 * @dentry: pointer to dentry
 871 * @open: open callback
 872 * @opened: state of open
 873 *
 874 * This can be used to finish opening a file passed to i_op->atomic_open().
 875 *
 876 * If the open callback is set to NULL, then the standard f_op->open()
 877 * filesystem callback is substituted.
 878 *
 879 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 880 * the return value of d_splice_alias(), then the caller needs to perform dput()
 881 * on it after finish_open().
 882 *
 883 * Returns zero on success or -errno if the open failed.
 884 */
 885int finish_open(struct file *file, struct dentry *dentry,
 886		int (*open)(struct inode *, struct file *))
 887{
 888	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 889
 890	file->f_path.dentry = dentry;
 891	return do_dentry_open(file, d_backing_inode(dentry), open);
 892}
 893EXPORT_SYMBOL(finish_open);
 894
 895/**
 896 * finish_no_open - finish ->atomic_open() without opening the file
 897 *
 898 * @file: file pointer
 899 * @dentry: dentry or NULL (as returned from ->lookup())
 900 *
 901 * This can be used to set the result of a successful lookup in ->atomic_open().
 902 *
 903 * NB: unlike finish_open() this function does consume the dentry reference and
 904 * the caller need not dput() it.
 905 *
 906 * Returns "0" which must be the return value of ->atomic_open() after having
 907 * called this function.
 908 */
 909int finish_no_open(struct file *file, struct dentry *dentry)
 910{
 911	file->f_path.dentry = dentry;
 912	return 0;
 913}
 914EXPORT_SYMBOL(finish_no_open);
 915
 916char *file_path(struct file *filp, char *buf, int buflen)
 917{
 918	return d_path(&filp->f_path, buf, buflen);
 919}
 920EXPORT_SYMBOL(file_path);
 921
 922/**
 923 * vfs_open - open the file at the given path
 924 * @path: path to open
 925 * @file: newly allocated file with f_flag initialized
 926 * @cred: credentials to use
 927 */
 928int vfs_open(const struct path *path, struct file *file)
 929{
 930	file->f_path = *path;
 931	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
 932}
 933
 934struct file *dentry_open(const struct path *path, int flags,
 935			 const struct cred *cred)
 936{
 937	int error;
 938	struct file *f;
 939
 940	validate_creds(cred);
 941
 942	/* We must always pass in a valid mount pointer. */
 943	BUG_ON(!path->mnt);
 944
 945	f = alloc_empty_file(flags, cred);
 946	if (!IS_ERR(f)) {
 947		error = vfs_open(path, f);
 948		if (error) {
 949			fput(f);
 950			f = ERR_PTR(error);
 951		}
 952	}
 953	return f;
 954}
 955EXPORT_SYMBOL(dentry_open);
 956
 957struct file *open_with_fake_path(const struct path *path, int flags,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 958				struct inode *inode, const struct cred *cred)
 959{
 960	struct file *f = alloc_empty_file_noaccount(flags, cred);
 961	if (!IS_ERR(f)) {
 962		int error;
 963
 964		f->f_path = *path;
 965		error = do_dentry_open(f, inode, NULL);
 966		if (error) {
 967			fput(f);
 968			f = ERR_PTR(error);
 969		}
 
 
 
 970	}
 971	return f;
 972}
 973EXPORT_SYMBOL(open_with_fake_path);
 974
 975#define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
 976#define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
 977
 978inline struct open_how build_open_how(int flags, umode_t mode)
 979{
 980	struct open_how how = {
 981		.flags = flags & VALID_OPEN_FLAGS,
 982		.mode = mode & S_IALLUGO,
 983	};
 984
 985	/* O_PATH beats everything else. */
 986	if (how.flags & O_PATH)
 987		how.flags &= O_PATH_FLAGS;
 988	/* Modes should only be set for create-like flags. */
 989	if (!WILL_CREATE(how.flags))
 990		how.mode = 0;
 991	return how;
 992}
 993
 994inline int build_open_flags(const struct open_how *how, struct open_flags *op)
 995{
 996	int flags = how->flags;
 
 997	int lookup_flags = 0;
 998	int acc_mode = ACC_MODE(flags);
 999
1000	/* Must never be set by userspace */
1001	flags &= ~(FMODE_NONOTIFY | O_CLOEXEC);
 
 
 
 
 
 
 
1002
1003	/*
1004	 * Older syscalls implicitly clear all of the invalid flags or argument
1005	 * values before calling build_open_flags(), but openat2(2) checks all
1006	 * of its arguments.
1007	 */
1008	if (flags & ~VALID_OPEN_FLAGS)
1009		return -EINVAL;
1010	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1011		return -EINVAL;
1012
 
 
 
 
1013	/* Deal with the mode. */
1014	if (WILL_CREATE(flags)) {
1015		if (how->mode & ~S_IALLUGO)
1016			return -EINVAL;
1017		op->mode = how->mode | S_IFREG;
1018	} else {
1019		if (how->mode != 0)
1020			return -EINVAL;
1021		op->mode = 0;
1022	}
1023
1024	/*
1025	 * In order to ensure programs get explicit errors when trying to use
1026	 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1027	 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1028	 * have to require userspace to explicitly set it.
1029	 */
 
 
 
 
1030	if (flags & __O_TMPFILE) {
1031		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 
 
 
 
 
1032			return -EINVAL;
1033		if (!(acc_mode & MAY_WRITE))
1034			return -EINVAL;
1035	}
1036	if (flags & O_PATH) {
1037		/* O_PATH only permits certain other flags to be set. */
1038		if (flags & ~O_PATH_FLAGS)
1039			return -EINVAL;
1040		acc_mode = 0;
1041	}
1042
1043	/*
1044	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1045	 * check for O_DSYNC if the need any syncing at all we enforce it's
1046	 * always set instead of having to deal with possibly weird behaviour
1047	 * for malicious applications setting only __O_SYNC.
1048	 */
1049	if (flags & __O_SYNC)
1050		flags |= O_DSYNC;
1051
1052	op->open_flag = flags;
1053
1054	/* O_TRUNC implies we need access checks for write permissions */
1055	if (flags & O_TRUNC)
1056		acc_mode |= MAY_WRITE;
1057
1058	/* Allow the LSM permission hook to distinguish append
1059	   access from general write access. */
1060	if (flags & O_APPEND)
1061		acc_mode |= MAY_APPEND;
1062
1063	op->acc_mode = acc_mode;
1064
1065	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1066
1067	if (flags & O_CREAT) {
1068		op->intent |= LOOKUP_CREATE;
1069		if (flags & O_EXCL) {
1070			op->intent |= LOOKUP_EXCL;
1071			flags |= O_NOFOLLOW;
1072		}
1073	}
1074
1075	if (flags & O_DIRECTORY)
1076		lookup_flags |= LOOKUP_DIRECTORY;
1077	if (!(flags & O_NOFOLLOW))
1078		lookup_flags |= LOOKUP_FOLLOW;
1079
1080	if (how->resolve & RESOLVE_NO_XDEV)
1081		lookup_flags |= LOOKUP_NO_XDEV;
1082	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1083		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1084	if (how->resolve & RESOLVE_NO_SYMLINKS)
1085		lookup_flags |= LOOKUP_NO_SYMLINKS;
1086	if (how->resolve & RESOLVE_BENEATH)
1087		lookup_flags |= LOOKUP_BENEATH;
1088	if (how->resolve & RESOLVE_IN_ROOT)
1089		lookup_flags |= LOOKUP_IN_ROOT;
 
 
 
 
 
 
1090
1091	op->lookup_flags = lookup_flags;
1092	return 0;
1093}
1094
1095/**
1096 * file_open_name - open file and return file pointer
1097 *
1098 * @name:	struct filename containing path to open
1099 * @flags:	open flags as per the open(2) second argument
1100 * @mode:	mode for the new file if O_CREAT is set, else ignored
1101 *
1102 * This is the helper to open a file from kernelspace if you really
1103 * have to.  But in generally you should not do this, so please move
1104 * along, nothing to see here..
1105 */
1106struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1107{
1108	struct open_flags op;
1109	struct open_how how = build_open_how(flags, mode);
1110	int err = build_open_flags(&how, &op);
1111	if (err)
1112		return ERR_PTR(err);
1113	return do_filp_open(AT_FDCWD, name, &op);
1114}
1115
1116/**
1117 * filp_open - open file and return file pointer
1118 *
1119 * @filename:	path to open
1120 * @flags:	open flags as per the open(2) second argument
1121 * @mode:	mode for the new file if O_CREAT is set, else ignored
1122 *
1123 * This is the helper to open a file from kernelspace if you really
1124 * have to.  But in generally you should not do this, so please move
1125 * along, nothing to see here..
1126 */
1127struct file *filp_open(const char *filename, int flags, umode_t mode)
1128{
1129	struct filename *name = getname_kernel(filename);
1130	struct file *file = ERR_CAST(name);
1131	
1132	if (!IS_ERR(name)) {
1133		file = file_open_name(name, flags, mode);
1134		putname(name);
1135	}
1136	return file;
1137}
1138EXPORT_SYMBOL(filp_open);
1139
1140struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1141			    const char *filename, int flags, umode_t mode)
1142{
1143	struct open_flags op;
1144	struct open_how how = build_open_how(flags, mode);
1145	int err = build_open_flags(&how, &op);
1146	if (err)
1147		return ERR_PTR(err);
1148	return do_file_open_root(dentry, mnt, filename, &op);
1149}
1150EXPORT_SYMBOL(file_open_root);
1151
1152static long do_sys_openat2(int dfd, const char __user *filename,
1153			   struct open_how *how)
1154{
1155	struct open_flags op;
1156	int fd = build_open_flags(how, &op);
1157	struct filename *tmp;
1158
1159	if (fd)
1160		return fd;
1161
1162	tmp = getname(filename);
1163	if (IS_ERR(tmp))
1164		return PTR_ERR(tmp);
1165
1166	fd = get_unused_fd_flags(how->flags);
1167	if (fd >= 0) {
1168		struct file *f = do_filp_open(dfd, tmp, &op);
1169		if (IS_ERR(f)) {
1170			put_unused_fd(fd);
1171			fd = PTR_ERR(f);
1172		} else {
1173			fsnotify_open(f);
1174			fd_install(fd, f);
1175		}
1176	}
1177	putname(tmp);
1178	return fd;
1179}
1180
1181long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1182{
1183	struct open_how how = build_open_how(flags, mode);
1184	return do_sys_openat2(dfd, filename, &how);
1185}
1186
1187
1188SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1189{
1190	if (force_o_largefile())
1191		flags |= O_LARGEFILE;
1192	return do_sys_open(AT_FDCWD, filename, flags, mode);
1193}
1194
1195SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1196		umode_t, mode)
1197{
1198	if (force_o_largefile())
1199		flags |= O_LARGEFILE;
1200	return do_sys_open(dfd, filename, flags, mode);
1201}
1202
1203SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1204		struct open_how __user *, how, size_t, usize)
1205{
1206	int err;
1207	struct open_how tmp;
1208
1209	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1210	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1211
1212	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1213		return -EINVAL;
1214
1215	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1216	if (err)
1217		return err;
1218
 
 
1219	/* O_LARGEFILE is only allowed for non-O_PATH. */
1220	if (!(tmp.flags & O_PATH) && force_o_largefile())
1221		tmp.flags |= O_LARGEFILE;
1222
1223	return do_sys_openat2(dfd, filename, &tmp);
1224}
1225
1226#ifdef CONFIG_COMPAT
1227/*
1228 * Exactly like sys_open(), except that it doesn't set the
1229 * O_LARGEFILE flag.
1230 */
1231COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1232{
1233	return do_sys_open(AT_FDCWD, filename, flags, mode);
1234}
1235
1236/*
1237 * Exactly like sys_openat(), except that it doesn't set the
1238 * O_LARGEFILE flag.
1239 */
1240COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1241{
1242	return do_sys_open(dfd, filename, flags, mode);
1243}
1244#endif
1245
1246#ifndef __alpha__
1247
1248/*
1249 * For backward compatibility?  Maybe this should be moved
1250 * into arch/i386 instead?
1251 */
1252SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1253{
1254	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1255
1256	if (force_o_largefile())
1257		flags |= O_LARGEFILE;
1258	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1259}
1260#endif
1261
1262/*
1263 * "id" is the POSIX thread ID. We use the
1264 * files pointer for this..
1265 */
1266int filp_close(struct file *filp, fl_owner_t id)
1267{
1268	int retval = 0;
1269
1270	if (!file_count(filp)) {
1271		printk(KERN_ERR "VFS: Close: file count is 0\n");
 
1272		return 0;
1273	}
1274
1275	if (filp->f_op->flush)
1276		retval = filp->f_op->flush(filp, id);
1277
1278	if (likely(!(filp->f_mode & FMODE_PATH))) {
1279		dnotify_flush(filp, id);
1280		locks_remove_posix(filp, id);
1281	}
1282	fput(filp);
1283	return retval;
1284}
1285
 
 
 
 
 
 
 
 
 
1286EXPORT_SYMBOL(filp_close);
1287
1288/*
1289 * Careful here! We test whether the file pointer is NULL before
1290 * releasing the fd. This ensures that one clone task can't release
1291 * an fd while another clone is opening it.
1292 */
1293SYSCALL_DEFINE1(close, unsigned int, fd)
1294{
1295	int retval = __close_fd(current->files, fd);
 
 
 
 
 
 
 
 
 
 
 
 
 
1296
1297	/* can't restart close syscall because file table entry was cleared */
1298	if (unlikely(retval == -ERESTARTSYS ||
1299		     retval == -ERESTARTNOINTR ||
1300		     retval == -ERESTARTNOHAND ||
1301		     retval == -ERESTART_RESTARTBLOCK))
1302		retval = -EINTR;
1303
1304	return retval;
1305}
1306
1307/**
1308 * close_range() - Close all file descriptors in a given range.
1309 *
1310 * @fd:     starting file descriptor to close
1311 * @max_fd: last file descriptor to close
1312 * @flags:  reserved for future extensions
1313 *
1314 * This closes a range of file descriptors. All file descriptors
1315 * from @fd up to and including @max_fd are closed.
1316 * Currently, errors to close a given file descriptor are ignored.
1317 */
1318SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1319		unsigned int, flags)
1320{
1321	return __close_range(fd, max_fd, flags);
1322}
1323
1324/*
1325 * This routine simulates a hangup on the tty, to arrange that users
1326 * are given clean terminals at login time.
1327 */
1328SYSCALL_DEFINE0(vhangup)
1329{
1330	if (capable(CAP_SYS_TTY_CONFIG)) {
1331		tty_vhangup_self();
1332		return 0;
1333	}
1334	return -EPERM;
1335}
1336
1337/*
1338 * Called when an inode is about to be open.
1339 * We use this to disallow opening large files on 32bit systems if
1340 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1341 * on this flag in sys_open.
1342 */
1343int generic_file_open(struct inode * inode, struct file * filp)
1344{
1345	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1346		return -EOVERFLOW;
1347	return 0;
1348}
1349
1350EXPORT_SYMBOL(generic_file_open);
1351
1352/*
1353 * This is used by subsystems that don't want seekable
1354 * file descriptors. The function is not supposed to ever fail, the only
1355 * reason it returns an 'int' and not 'void' is so that it can be plugged
1356 * directly into file_operations structure.
1357 */
1358int nonseekable_open(struct inode *inode, struct file *filp)
1359{
1360	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1361	return 0;
1362}
1363
1364EXPORT_SYMBOL(nonseekable_open);
1365
1366/*
1367 * stream_open is used by subsystems that want stream-like file descriptors.
1368 * Such file descriptors are not seekable and don't have notion of position
1369 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1370 * Contrary to file descriptors of other regular files, .read() and .write()
1371 * can run simultaneously.
1372 *
1373 * stream_open never fails and is marked to return int so that it could be
1374 * directly used as file_operations.open .
1375 */
1376int stream_open(struct inode *inode, struct file *filp)
1377{
1378	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1379	filp->f_mode |= FMODE_STREAM;
1380	return 0;
1381}
1382
1383EXPORT_SYMBOL(stream_open);
v6.8
   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#include <linux/filelock.h>
  37
  38#include "internal.h"
  39
  40int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
  41		loff_t length, unsigned int time_attrs, struct file *filp)
  42{
  43	int ret;
  44	struct iattr newattrs;
  45
  46	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  47	if (length < 0)
  48		return -EINVAL;
  49
  50	newattrs.ia_size = length;
  51	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  52	if (filp) {
  53		newattrs.ia_file = filp;
  54		newattrs.ia_valid |= ATTR_FILE;
  55	}
  56
  57	/* Remove suid, sgid, and file capabilities on truncate too */
  58	ret = dentry_needs_remove_privs(idmap, dentry);
  59	if (ret < 0)
  60		return ret;
  61	if (ret)
  62		newattrs.ia_valid |= ret | ATTR_FORCE;
  63
  64	inode_lock(dentry->d_inode);
  65	/* Note any delegations or leases have already been broken: */
  66	ret = notify_change(idmap, dentry, &newattrs, NULL);
  67	inode_unlock(dentry->d_inode);
  68	return ret;
  69}
  70
  71long vfs_truncate(const struct path *path, loff_t length)
  72{
  73	struct mnt_idmap *idmap;
  74	struct inode *inode;
  75	long error;
  76
  77	inode = path->dentry->d_inode;
  78
  79	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  80	if (S_ISDIR(inode->i_mode))
  81		return -EISDIR;
  82	if (!S_ISREG(inode->i_mode))
  83		return -EINVAL;
  84
  85	error = mnt_want_write(path->mnt);
  86	if (error)
  87		goto out;
  88
  89	idmap = mnt_idmap(path->mnt);
  90	error = inode_permission(idmap, inode, MAY_WRITE);
  91	if (error)
  92		goto mnt_drop_write_and_out;
  93
  94	error = -EPERM;
  95	if (IS_APPEND(inode))
  96		goto mnt_drop_write_and_out;
  97
  98	error = get_write_access(inode);
  99	if (error)
 100		goto mnt_drop_write_and_out;
 101
 102	/*
 103	 * Make sure that there are no leases.  get_write_access() protects
 104	 * against the truncate racing with a lease-granting setlease().
 105	 */
 106	error = break_lease(inode, O_WRONLY);
 107	if (error)
 108		goto put_write_and_out;
 109
 110	error = security_path_truncate(path);
 111	if (!error)
 112		error = do_truncate(idmap, path->dentry, length, 0, NULL);
 
 
 113
 114put_write_and_out:
 115	put_write_access(inode);
 116mnt_drop_write_and_out:
 117	mnt_drop_write(path->mnt);
 118out:
 119	return error;
 120}
 121EXPORT_SYMBOL_GPL(vfs_truncate);
 122
 123long do_sys_truncate(const char __user *pathname, loff_t length)
 124{
 125	unsigned int lookup_flags = LOOKUP_FOLLOW;
 126	struct path path;
 127	int error;
 128
 129	if (length < 0)	/* sorry, but loff_t says... */
 130		return -EINVAL;
 131
 132retry:
 133	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 134	if (!error) {
 135		error = vfs_truncate(&path, length);
 136		path_put(&path);
 137	}
 138	if (retry_estale(error, lookup_flags)) {
 139		lookup_flags |= LOOKUP_REVAL;
 140		goto retry;
 141	}
 142	return error;
 143}
 144
 145SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 146{
 147	return do_sys_truncate(path, length);
 148}
 149
 150#ifdef CONFIG_COMPAT
 151COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 152{
 153	return do_sys_truncate(path, length);
 154}
 155#endif
 156
 157long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 158{
 159	struct inode *inode;
 160	struct dentry *dentry;
 161	struct fd f;
 162	int error;
 163
 164	error = -EINVAL;
 165	if (length < 0)
 166		goto out;
 167	error = -EBADF;
 168	f = fdget(fd);
 169	if (!f.file)
 170		goto out;
 171
 172	/* explicitly opened as large or we are on 64-bit box */
 173	if (f.file->f_flags & O_LARGEFILE)
 174		small = 0;
 175
 176	dentry = f.file->f_path.dentry;
 177	inode = dentry->d_inode;
 178	error = -EINVAL;
 179	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 180		goto out_putf;
 181
 182	error = -EINVAL;
 183	/* Cannot ftruncate over 2^31 bytes without large file support */
 184	if (small && length > MAX_NON_LFS)
 185		goto out_putf;
 186
 187	error = -EPERM;
 188	/* Check IS_APPEND on real upper inode */
 189	if (IS_APPEND(file_inode(f.file)))
 190		goto out_putf;
 
 191	sb_start_write(inode->i_sb);
 192	error = security_file_truncate(f.file);
 
 
 193	if (!error)
 194		error = do_truncate(file_mnt_idmap(f.file), dentry, length,
 195				    ATTR_MTIME | ATTR_CTIME, f.file);
 196	sb_end_write(inode->i_sb);
 197out_putf:
 198	fdput(f);
 199out:
 200	return error;
 201}
 202
 203SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 204{
 205	return do_sys_ftruncate(fd, length, 1);
 206}
 207
 208#ifdef CONFIG_COMPAT
 209COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 210{
 211	return do_sys_ftruncate(fd, length, 1);
 212}
 213#endif
 214
 215/* LFS versions of truncate are only needed on 32 bit machines */
 216#if BITS_PER_LONG == 32
 217SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 218{
 219	return do_sys_truncate(path, length);
 220}
 221
 222SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 223{
 224	return do_sys_ftruncate(fd, length, 0);
 225}
 226#endif /* BITS_PER_LONG == 32 */
 227
 228#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
 229COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
 230		       compat_arg_u64_dual(length))
 231{
 232	return ksys_truncate(pathname, compat_arg_u64_glue(length));
 233}
 234#endif
 235
 236#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
 237COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
 238		       compat_arg_u64_dual(length))
 239{
 240	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
 241}
 242#endif
 243
 244int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 245{
 246	struct inode *inode = file_inode(file);
 247	long ret;
 248
 249	if (offset < 0 || len <= 0)
 250		return -EINVAL;
 251
 252	/* Return error if mode is not supported */
 253	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 254		return -EOPNOTSUPP;
 255
 256	/* Punch hole and zero range are mutually exclusive */
 257	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 258	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 259		return -EOPNOTSUPP;
 260
 261	/* Punch hole must have keep size set */
 262	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 263	    !(mode & FALLOC_FL_KEEP_SIZE))
 264		return -EOPNOTSUPP;
 265
 266	/* Collapse range should only be used exclusively. */
 267	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 268	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 269		return -EINVAL;
 270
 271	/* Insert range should only be used exclusively. */
 272	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 273	    (mode & ~FALLOC_FL_INSERT_RANGE))
 274		return -EINVAL;
 275
 276	/* Unshare range should only be used with allocate mode. */
 277	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 278	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 279		return -EINVAL;
 280
 281	if (!(file->f_mode & FMODE_WRITE))
 282		return -EBADF;
 283
 284	/*
 285	 * We can only allow pure fallocate on append only files
 286	 */
 287	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 288		return -EPERM;
 289
 290	if (IS_IMMUTABLE(inode))
 291		return -EPERM;
 292
 293	/*
 294	 * We cannot allow any fallocate operation on an active swapfile
 295	 */
 296	if (IS_SWAPFILE(inode))
 297		return -ETXTBSY;
 298
 299	/*
 300	 * Revalidate the write permissions, in case security policy has
 301	 * changed since the files were opened.
 302	 */
 303	ret = security_file_permission(file, MAY_WRITE);
 304	if (ret)
 305		return ret;
 306
 307	ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
 308	if (ret)
 309		return ret;
 310
 311	if (S_ISFIFO(inode->i_mode))
 312		return -ESPIPE;
 313
 314	if (S_ISDIR(inode->i_mode))
 315		return -EISDIR;
 316
 317	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 318		return -ENODEV;
 319
 320	/* Check for wrap through zero too */
 321	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 322		return -EFBIG;
 323
 324	if (!file->f_op->fallocate)
 325		return -EOPNOTSUPP;
 326
 327	file_start_write(file);
 328	ret = file->f_op->fallocate(file, mode, offset, len);
 329
 330	/*
 331	 * Create inotify and fanotify events.
 332	 *
 333	 * To keep the logic simple always create events if fallocate succeeds.
 334	 * This implies that events are even created if the file size remains
 335	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 336	 */
 337	if (ret == 0)
 338		fsnotify_modify(file);
 339
 340	file_end_write(file);
 341	return ret;
 342}
 343EXPORT_SYMBOL_GPL(vfs_fallocate);
 344
 345int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 346{
 347	struct fd f = fdget(fd);
 348	int error = -EBADF;
 349
 350	if (f.file) {
 351		error = vfs_fallocate(f.file, mode, offset, len);
 352		fdput(f);
 353	}
 354	return error;
 355}
 356
 357SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 358{
 359	return ksys_fallocate(fd, mode, offset, len);
 360}
 361
 362#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
 363COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
 364		       compat_arg_u64_dual(len))
 365{
 366	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
 367			      compat_arg_u64_glue(len));
 368}
 369#endif
 370
 371/*
 372 * access() needs to use the real uid/gid, not the effective uid/gid.
 373 * We do this by temporarily clearing all FS-related capabilities and
 374 * switching the fsuid/fsgid around to the real ones.
 375 *
 376 * Creating new credentials is expensive, so we try to skip doing it,
 377 * which we can if the result would match what we already got.
 378 */
 379static bool access_need_override_creds(int flags)
 380{
 381	const struct cred *cred;
 382
 383	if (flags & AT_EACCESS)
 384		return false;
 385
 386	cred = current_cred();
 387	if (!uid_eq(cred->fsuid, cred->uid) ||
 388	    !gid_eq(cred->fsgid, cred->gid))
 389		return true;
 390
 391	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 392		kuid_t root_uid = make_kuid(cred->user_ns, 0);
 393		if (!uid_eq(cred->uid, root_uid)) {
 394			if (!cap_isclear(cred->cap_effective))
 395				return true;
 396		} else {
 397			if (!cap_isidentical(cred->cap_effective,
 398			    cred->cap_permitted))
 399				return true;
 400		}
 401	}
 402
 403	return false;
 404}
 405
 406static const struct cred *access_override_creds(void)
 407{
 408	const struct cred *old_cred;
 409	struct cred *override_cred;
 410
 411	override_cred = prepare_creds();
 412	if (!override_cred)
 413		return NULL;
 414
 415	/*
 416	 * XXX access_need_override_creds performs checks in hopes of skipping
 417	 * this work. Make sure it stays in sync if making any changes in this
 418	 * routine.
 419	 */
 420
 421	override_cred->fsuid = override_cred->uid;
 422	override_cred->fsgid = override_cred->gid;
 423
 424	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 425		/* Clear the capabilities if we switch to a non-root user */
 426		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 427		if (!uid_eq(override_cred->uid, root_uid))
 428			cap_clear(override_cred->cap_effective);
 429		else
 430			override_cred->cap_effective =
 431				override_cred->cap_permitted;
 432	}
 433
 434	/*
 435	 * The new set of credentials can *only* be used in
 436	 * task-synchronous circumstances, and does not need
 437	 * RCU freeing, unless somebody then takes a separate
 438	 * reference to it.
 439	 *
 440	 * NOTE! This is _only_ true because this credential
 441	 * is used purely for override_creds() that installs
 442	 * it as the subjective cred. Other threads will be
 443	 * accessing ->real_cred, not the subjective cred.
 444	 *
 445	 * If somebody _does_ make a copy of this (using the
 446	 * 'get_current_cred()' function), that will clear the
 447	 * non_rcu field, because now that other user may be
 448	 * expecting RCU freeing. But normal thread-synchronous
 449	 * cred accesses will keep things non-racy to avoid RCU
 450	 * freeing.
 451	 */
 452	override_cred->non_rcu = 1;
 453
 454	old_cred = override_creds(override_cred);
 455
 456	/* override_cred() gets its own ref */
 457	put_cred(override_cred);
 458
 459	return old_cred;
 460}
 461
 462static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
 463{
 464	struct path path;
 465	struct inode *inode;
 466	int res;
 467	unsigned int lookup_flags = LOOKUP_FOLLOW;
 468	const struct cred *old_cred = NULL;
 469
 470	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 471		return -EINVAL;
 472
 473	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
 474		return -EINVAL;
 475
 476	if (flags & AT_SYMLINK_NOFOLLOW)
 477		lookup_flags &= ~LOOKUP_FOLLOW;
 478	if (flags & AT_EMPTY_PATH)
 479		lookup_flags |= LOOKUP_EMPTY;
 480
 481	if (access_need_override_creds(flags)) {
 482		old_cred = access_override_creds();
 483		if (!old_cred)
 484			return -ENOMEM;
 485	}
 486
 487retry:
 488	res = user_path_at(dfd, filename, lookup_flags, &path);
 489	if (res)
 490		goto out;
 491
 492	inode = d_backing_inode(path.dentry);
 493
 494	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 495		/*
 496		 * MAY_EXEC on regular files is denied if the fs is mounted
 497		 * with the "noexec" flag.
 498		 */
 499		res = -EACCES;
 500		if (path_noexec(&path))
 501			goto out_path_release;
 502	}
 503
 504	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
 505	/* SuS v2 requires we report a read only fs too */
 506	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 507		goto out_path_release;
 508	/*
 509	 * This is a rare case where using __mnt_is_readonly()
 510	 * is OK without a mnt_want/drop_write() pair.  Since
 511	 * no actual write to the fs is performed here, we do
 512	 * not need to telegraph to that to anyone.
 513	 *
 514	 * By doing this, we accept that this access is
 515	 * inherently racy and know that the fs may change
 516	 * state before we even see this result.
 517	 */
 518	if (__mnt_is_readonly(path.mnt))
 519		res = -EROFS;
 520
 521out_path_release:
 522	path_put(&path);
 523	if (retry_estale(res, lookup_flags)) {
 524		lookup_flags |= LOOKUP_REVAL;
 525		goto retry;
 526	}
 527out:
 528	if (old_cred)
 529		revert_creds(old_cred);
 530
 531	return res;
 532}
 533
 534SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 535{
 536	return do_faccessat(dfd, filename, mode, 0);
 537}
 538
 539SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
 540		int, flags)
 541{
 542	return do_faccessat(dfd, filename, mode, flags);
 543}
 544
 545SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 546{
 547	return do_faccessat(AT_FDCWD, filename, mode, 0);
 548}
 549
 550SYSCALL_DEFINE1(chdir, const char __user *, filename)
 551{
 552	struct path path;
 553	int error;
 554	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 555retry:
 556	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 557	if (error)
 558		goto out;
 559
 560	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 561	if (error)
 562		goto dput_and_out;
 563
 564	set_fs_pwd(current->fs, &path);
 565
 566dput_and_out:
 567	path_put(&path);
 568	if (retry_estale(error, lookup_flags)) {
 569		lookup_flags |= LOOKUP_REVAL;
 570		goto retry;
 571	}
 572out:
 573	return error;
 574}
 575
 576SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 577{
 578	struct fd f = fdget_raw(fd);
 579	int error;
 580
 581	error = -EBADF;
 582	if (!f.file)
 583		goto out;
 584
 585	error = -ENOTDIR;
 586	if (!d_can_lookup(f.file->f_path.dentry))
 587		goto out_putf;
 588
 589	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
 590	if (!error)
 591		set_fs_pwd(current->fs, &f.file->f_path);
 592out_putf:
 593	fdput(f);
 594out:
 595	return error;
 596}
 597
 598SYSCALL_DEFINE1(chroot, const char __user *, filename)
 599{
 600	struct path path;
 601	int error;
 602	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 603retry:
 604	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 605	if (error)
 606		goto out;
 607
 608	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 609	if (error)
 610		goto dput_and_out;
 611
 612	error = -EPERM;
 613	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 614		goto dput_and_out;
 615	error = security_path_chroot(&path);
 616	if (error)
 617		goto dput_and_out;
 618
 619	set_fs_root(current->fs, &path);
 620	error = 0;
 621dput_and_out:
 622	path_put(&path);
 623	if (retry_estale(error, lookup_flags)) {
 624		lookup_flags |= LOOKUP_REVAL;
 625		goto retry;
 626	}
 627out:
 628	return error;
 629}
 630
 631int chmod_common(const struct path *path, umode_t mode)
 632{
 633	struct inode *inode = path->dentry->d_inode;
 634	struct inode *delegated_inode = NULL;
 635	struct iattr newattrs;
 636	int error;
 637
 638	error = mnt_want_write(path->mnt);
 639	if (error)
 640		return error;
 641retry_deleg:
 642	inode_lock(inode);
 643	error = security_path_chmod(path, mode);
 644	if (error)
 645		goto out_unlock;
 646	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 647	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 648	error = notify_change(mnt_idmap(path->mnt), path->dentry,
 649			      &newattrs, &delegated_inode);
 650out_unlock:
 651	inode_unlock(inode);
 652	if (delegated_inode) {
 653		error = break_deleg_wait(&delegated_inode);
 654		if (!error)
 655			goto retry_deleg;
 656	}
 657	mnt_drop_write(path->mnt);
 658	return error;
 659}
 660
 661int vfs_fchmod(struct file *file, umode_t mode)
 662{
 663	audit_file(file);
 664	return chmod_common(&file->f_path, mode);
 665}
 666
 667SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 668{
 669	struct fd f = fdget(fd);
 670	int err = -EBADF;
 671
 672	if (f.file) {
 673		err = vfs_fchmod(f.file, mode);
 674		fdput(f);
 675	}
 676	return err;
 677}
 678
 679static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
 680		       unsigned int flags)
 681{
 682	struct path path;
 683	int error;
 684	unsigned int lookup_flags;
 685
 686	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
 687		return -EINVAL;
 688
 689	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 690	if (flags & AT_EMPTY_PATH)
 691		lookup_flags |= LOOKUP_EMPTY;
 692
 693retry:
 694	error = user_path_at(dfd, filename, lookup_flags, &path);
 695	if (!error) {
 696		error = chmod_common(&path, mode);
 697		path_put(&path);
 698		if (retry_estale(error, lookup_flags)) {
 699			lookup_flags |= LOOKUP_REVAL;
 700			goto retry;
 701		}
 702	}
 703	return error;
 704}
 705
 706SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
 707		umode_t, mode, unsigned int, flags)
 708{
 709	return do_fchmodat(dfd, filename, mode, flags);
 710}
 711
 712SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 713		umode_t, mode)
 714{
 715	return do_fchmodat(dfd, filename, mode, 0);
 716}
 717
 718SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 719{
 720	return do_fchmodat(AT_FDCWD, filename, mode, 0);
 721}
 722
 723/*
 724 * Check whether @kuid is valid and if so generate and set vfsuid_t in
 725 * ia_vfsuid.
 726 *
 727 * Return: true if @kuid is valid, false if not.
 728 */
 729static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
 730{
 731	if (!uid_valid(kuid))
 732		return false;
 733	attr->ia_valid |= ATTR_UID;
 734	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
 735	return true;
 736}
 737
 738/*
 739 * Check whether @kgid is valid and if so generate and set vfsgid_t in
 740 * ia_vfsgid.
 741 *
 742 * Return: true if @kgid is valid, false if not.
 743 */
 744static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
 745{
 746	if (!gid_valid(kgid))
 747		return false;
 748	attr->ia_valid |= ATTR_GID;
 749	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
 750	return true;
 751}
 752
 753int chown_common(const struct path *path, uid_t user, gid_t group)
 754{
 755	struct mnt_idmap *idmap;
 756	struct user_namespace *fs_userns;
 757	struct inode *inode = path->dentry->d_inode;
 758	struct inode *delegated_inode = NULL;
 759	int error;
 760	struct iattr newattrs;
 761	kuid_t uid;
 762	kgid_t gid;
 763
 764	uid = make_kuid(current_user_ns(), user);
 765	gid = make_kgid(current_user_ns(), group);
 766
 767	idmap = mnt_idmap(path->mnt);
 768	fs_userns = i_user_ns(inode);
 769
 770retry_deleg:
 771	newattrs.ia_vfsuid = INVALID_VFSUID;
 772	newattrs.ia_vfsgid = INVALID_VFSGID;
 773	newattrs.ia_valid =  ATTR_CTIME;
 774	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
 775		return -EINVAL;
 776	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
 777		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 778	inode_lock(inode);
 779	if (!S_ISDIR(inode->i_mode))
 780		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
 781				     setattr_should_drop_sgid(idmap, inode);
 782	/* Continue to send actual fs values, not the mount values. */
 783	error = security_path_chown(
 784		path,
 785		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
 786		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
 787	if (!error)
 788		error = notify_change(idmap, path->dentry, &newattrs,
 789				      &delegated_inode);
 790	inode_unlock(inode);
 791	if (delegated_inode) {
 792		error = break_deleg_wait(&delegated_inode);
 793		if (!error)
 794			goto retry_deleg;
 795	}
 796	return error;
 797}
 798
 799int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 800		int flag)
 801{
 802	struct path path;
 803	int error = -EINVAL;
 804	int lookup_flags;
 805
 806	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 807		goto out;
 808
 809	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 810	if (flag & AT_EMPTY_PATH)
 811		lookup_flags |= LOOKUP_EMPTY;
 812retry:
 813	error = user_path_at(dfd, filename, lookup_flags, &path);
 814	if (error)
 815		goto out;
 816	error = mnt_want_write(path.mnt);
 817	if (error)
 818		goto out_release;
 819	error = chown_common(&path, user, group);
 820	mnt_drop_write(path.mnt);
 821out_release:
 822	path_put(&path);
 823	if (retry_estale(error, lookup_flags)) {
 824		lookup_flags |= LOOKUP_REVAL;
 825		goto retry;
 826	}
 827out:
 828	return error;
 829}
 830
 831SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 832		gid_t, group, int, flag)
 833{
 834	return do_fchownat(dfd, filename, user, group, flag);
 835}
 836
 837SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 838{
 839	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 840}
 841
 842SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 843{
 844	return do_fchownat(AT_FDCWD, filename, user, group,
 845			   AT_SYMLINK_NOFOLLOW);
 846}
 847
 848int vfs_fchown(struct file *file, uid_t user, gid_t group)
 849{
 850	int error;
 851
 852	error = mnt_want_write_file(file);
 853	if (error)
 854		return error;
 855	audit_file(file);
 856	error = chown_common(&file->f_path, user, group);
 857	mnt_drop_write_file(file);
 858	return error;
 859}
 860
 861int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 862{
 863	struct fd f = fdget(fd);
 864	int error = -EBADF;
 865
 866	if (f.file) {
 867		error = vfs_fchown(f.file, user, group);
 868		fdput(f);
 869	}
 870	return error;
 871}
 872
 873SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 874{
 875	return ksys_fchown(fd, user, group);
 876}
 877
 878static inline int file_get_write_access(struct file *f)
 879{
 880	int error;
 881
 882	error = get_write_access(f->f_inode);
 883	if (unlikely(error))
 884		return error;
 885	error = mnt_get_write_access(f->f_path.mnt);
 886	if (unlikely(error))
 887		goto cleanup_inode;
 888	if (unlikely(f->f_mode & FMODE_BACKING)) {
 889		error = mnt_get_write_access(backing_file_user_path(f)->mnt);
 890		if (unlikely(error))
 891			goto cleanup_mnt;
 892	}
 893	return 0;
 894
 895cleanup_mnt:
 896	mnt_put_write_access(f->f_path.mnt);
 897cleanup_inode:
 898	put_write_access(f->f_inode);
 899	return error;
 900}
 901
 902static int do_dentry_open(struct file *f,
 903			  struct inode *inode,
 904			  int (*open)(struct inode *, struct file *))
 905{
 906	static const struct file_operations empty_fops = {};
 907	int error;
 908
 909	path_get(&f->f_path);
 910	f->f_inode = inode;
 911	f->f_mapping = inode->i_mapping;
 912	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 913	f->f_sb_err = file_sample_sb_err(f);
 914
 915	if (unlikely(f->f_flags & O_PATH)) {
 916		f->f_mode = FMODE_PATH | FMODE_OPENED;
 917		f->f_op = &empty_fops;
 918		return 0;
 919	}
 920
 921	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
 922		i_readcount_inc(inode);
 923	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 924		error = file_get_write_access(f);
 925		if (unlikely(error))
 926			goto cleanup_file;
 
 
 
 
 
 927		f->f_mode |= FMODE_WRITER;
 928	}
 929
 930	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 931	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 932		f->f_mode |= FMODE_ATOMIC_POS;
 933
 934	f->f_op = fops_get(inode->i_fop);
 935	if (WARN_ON(!f->f_op)) {
 936		error = -ENODEV;
 937		goto cleanup_all;
 938	}
 939
 940	error = security_file_open(f);
 941	if (error)
 942		goto cleanup_all;
 943
 944	error = break_lease(file_inode(f), f->f_flags);
 945	if (error)
 946		goto cleanup_all;
 947
 948	/* normally all 3 are set; ->open() can clear them if needed */
 949	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 950	if (!open)
 951		open = f->f_op->open;
 952	if (open) {
 953		error = open(inode, f);
 954		if (error)
 955			goto cleanup_all;
 956	}
 957	f->f_mode |= FMODE_OPENED;
 
 
 958	if ((f->f_mode & FMODE_READ) &&
 959	     likely(f->f_op->read || f->f_op->read_iter))
 960		f->f_mode |= FMODE_CAN_READ;
 961	if ((f->f_mode & FMODE_WRITE) &&
 962	     likely(f->f_op->write || f->f_op->write_iter))
 963		f->f_mode |= FMODE_CAN_WRITE;
 964	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
 965		f->f_mode &= ~FMODE_LSEEK;
 966	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
 967		f->f_mode |= FMODE_CAN_ODIRECT;
 968
 
 969	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 970	f->f_iocb_flags = iocb_flags(f);
 971
 972	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 973
 974	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
 975		return -EINVAL;
 
 
 
 976
 977	/*
 978	 * XXX: Huge page cache doesn't support writing yet. Drop all page
 979	 * cache for this file before processing writes.
 980	 */
 981	if (f->f_mode & FMODE_WRITE) {
 982		/*
 983		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
 984		 * is up to date and the update to i_writecount by
 985		 * get_write_access() is visible. Ensures subsequent insertion
 986		 * of THPs into the page cache will fail.
 987		 */
 988		smp_mb();
 989		if (filemap_nr_thps(inode->i_mapping)) {
 990			struct address_space *mapping = inode->i_mapping;
 991
 992			filemap_invalidate_lock(inode->i_mapping);
 993			/*
 994			 * unmap_mapping_range just need to be called once
 995			 * here, because the private pages is not need to be
 996			 * unmapped mapping (e.g. data segment of dynamic
 997			 * shared libraries here).
 998			 */
 999			unmap_mapping_range(mapping, 0, 0, 0);
1000			truncate_inode_pages(mapping, 0);
1001			filemap_invalidate_unlock(inode->i_mapping);
1002		}
1003	}
1004
1005	/*
1006	 * Once we return a file with FMODE_OPENED, __fput() will call
1007	 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1008	 */
1009	fsnotify_open(f);
1010	return 0;
1011
1012cleanup_all:
1013	if (WARN_ON_ONCE(error > 0))
1014		error = -EINVAL;
1015	fops_put(f->f_op);
1016	put_file_access(f);
 
 
 
1017cleanup_file:
1018	path_put(&f->f_path);
1019	f->f_path.mnt = NULL;
1020	f->f_path.dentry = NULL;
1021	f->f_inode = NULL;
1022	return error;
1023}
1024
1025/**
1026 * finish_open - finish opening a file
1027 * @file: file pointer
1028 * @dentry: pointer to dentry
1029 * @open: open callback
 
1030 *
1031 * This can be used to finish opening a file passed to i_op->atomic_open().
1032 *
1033 * If the open callback is set to NULL, then the standard f_op->open()
1034 * filesystem callback is substituted.
1035 *
1036 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1037 * the return value of d_splice_alias(), then the caller needs to perform dput()
1038 * on it after finish_open().
1039 *
1040 * Returns zero on success or -errno if the open failed.
1041 */
1042int finish_open(struct file *file, struct dentry *dentry,
1043		int (*open)(struct inode *, struct file *))
1044{
1045	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1046
1047	file->f_path.dentry = dentry;
1048	return do_dentry_open(file, d_backing_inode(dentry), open);
1049}
1050EXPORT_SYMBOL(finish_open);
1051
1052/**
1053 * finish_no_open - finish ->atomic_open() without opening the file
1054 *
1055 * @file: file pointer
1056 * @dentry: dentry or NULL (as returned from ->lookup())
1057 *
1058 * This can be used to set the result of a successful lookup in ->atomic_open().
1059 *
1060 * NB: unlike finish_open() this function does consume the dentry reference and
1061 * the caller need not dput() it.
1062 *
1063 * Returns "0" which must be the return value of ->atomic_open() after having
1064 * called this function.
1065 */
1066int finish_no_open(struct file *file, struct dentry *dentry)
1067{
1068	file->f_path.dentry = dentry;
1069	return 0;
1070}
1071EXPORT_SYMBOL(finish_no_open);
1072
1073char *file_path(struct file *filp, char *buf, int buflen)
1074{
1075	return d_path(&filp->f_path, buf, buflen);
1076}
1077EXPORT_SYMBOL(file_path);
1078
1079/**
1080 * vfs_open - open the file at the given path
1081 * @path: path to open
1082 * @file: newly allocated file with f_flag initialized
 
1083 */
1084int vfs_open(const struct path *path, struct file *file)
1085{
1086	file->f_path = *path;
1087	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1088}
1089
1090struct file *dentry_open(const struct path *path, int flags,
1091			 const struct cred *cred)
1092{
1093	int error;
1094	struct file *f;
1095
 
 
1096	/* We must always pass in a valid mount pointer. */
1097	BUG_ON(!path->mnt);
1098
1099	f = alloc_empty_file(flags, cred);
1100	if (!IS_ERR(f)) {
1101		error = vfs_open(path, f);
1102		if (error) {
1103			fput(f);
1104			f = ERR_PTR(error);
1105		}
1106	}
1107	return f;
1108}
1109EXPORT_SYMBOL(dentry_open);
1110
1111/**
1112 * dentry_create - Create and open a file
1113 * @path: path to create
1114 * @flags: O_ flags
1115 * @mode: mode bits for new file
1116 * @cred: credentials to use
1117 *
1118 * Caller must hold the parent directory's lock, and have prepared
1119 * a negative dentry, placed in @path->dentry, for the new file.
1120 *
1121 * Caller sets @path->mnt to the vfsmount of the filesystem where
1122 * the new file is to be created. The parent directory and the
1123 * negative dentry must reside on the same filesystem instance.
1124 *
1125 * On success, returns a "struct file *". Otherwise a ERR_PTR
1126 * is returned.
1127 */
1128struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1129			   const struct cred *cred)
1130{
1131	struct file *f;
1132	int error;
1133
1134	f = alloc_empty_file(flags, cred);
1135	if (IS_ERR(f))
1136		return f;
1137
1138	error = vfs_create(mnt_idmap(path->mnt),
1139			   d_inode(path->dentry->d_parent),
1140			   path->dentry, mode, true);
1141	if (!error)
1142		error = vfs_open(path, f);
1143
1144	if (unlikely(error)) {
1145		fput(f);
1146		return ERR_PTR(error);
1147	}
1148	return f;
1149}
1150EXPORT_SYMBOL(dentry_create);
1151
1152/**
1153 * kernel_file_open - open a file for kernel internal use
1154 * @path:	path of the file to open
1155 * @flags:	open flags
1156 * @inode:	the inode
1157 * @cred:	credentials for open
1158 *
1159 * Open a file for use by in-kernel consumers. The file is not accounted
1160 * against nr_files and must not be installed into the file descriptor
1161 * table.
1162 *
1163 * Return: Opened file on success, an error pointer on failure.
1164 */
1165struct file *kernel_file_open(const struct path *path, int flags,
1166				struct inode *inode, const struct cred *cred)
1167{
1168	struct file *f;
1169	int error;
 
1170
1171	f = alloc_empty_file_noaccount(flags, cred);
1172	if (IS_ERR(f))
1173		return f;
1174
1175	f->f_path = *path;
1176	error = do_dentry_open(f, inode, NULL);
1177	if (error) {
1178		fput(f);
1179		f = ERR_PTR(error);
1180	}
1181	return f;
1182}
1183EXPORT_SYMBOL_GPL(kernel_file_open);
1184
1185#define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1186#define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1187
1188inline struct open_how build_open_how(int flags, umode_t mode)
1189{
1190	struct open_how how = {
1191		.flags = flags & VALID_OPEN_FLAGS,
1192		.mode = mode & S_IALLUGO,
1193	};
1194
1195	/* O_PATH beats everything else. */
1196	if (how.flags & O_PATH)
1197		how.flags &= O_PATH_FLAGS;
1198	/* Modes should only be set for create-like flags. */
1199	if (!WILL_CREATE(how.flags))
1200		how.mode = 0;
1201	return how;
1202}
1203
1204inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1205{
1206	u64 flags = how->flags;
1207	u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1208	int lookup_flags = 0;
1209	int acc_mode = ACC_MODE(flags);
1210
1211	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1212			 "struct open_flags doesn't yet handle flags > 32 bits");
1213
1214	/*
1215	 * Strip flags that either shouldn't be set by userspace like
1216	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1217	 * open_flags like O_CLOEXEC.
1218	 */
1219	flags &= ~strip;
1220
1221	/*
1222	 * Older syscalls implicitly clear all of the invalid flags or argument
1223	 * values before calling build_open_flags(), but openat2(2) checks all
1224	 * of its arguments.
1225	 */
1226	if (flags & ~VALID_OPEN_FLAGS)
1227		return -EINVAL;
1228	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1229		return -EINVAL;
1230
1231	/* Scoping flags are mutually exclusive. */
1232	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1233		return -EINVAL;
1234
1235	/* Deal with the mode. */
1236	if (WILL_CREATE(flags)) {
1237		if (how->mode & ~S_IALLUGO)
1238			return -EINVAL;
1239		op->mode = how->mode | S_IFREG;
1240	} else {
1241		if (how->mode != 0)
1242			return -EINVAL;
1243		op->mode = 0;
1244	}
1245
1246	/*
1247	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1248	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1249	 * O_TMPFILE below which requires O_DIRECTORY being raised.
 
1250	 */
1251	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1252		return -EINVAL;
1253
1254	/* Now handle the creative implementation of O_TMPFILE. */
1255	if (flags & __O_TMPFILE) {
1256		/*
1257		 * In order to ensure programs get explicit errors when trying
1258		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1259		 * is raised alongside __O_TMPFILE.
1260		 */
1261		if (!(flags & O_DIRECTORY))
1262			return -EINVAL;
1263		if (!(acc_mode & MAY_WRITE))
1264			return -EINVAL;
1265	}
1266	if (flags & O_PATH) {
1267		/* O_PATH only permits certain other flags to be set. */
1268		if (flags & ~O_PATH_FLAGS)
1269			return -EINVAL;
1270		acc_mode = 0;
1271	}
1272
1273	/*
1274	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1275	 * check for O_DSYNC if the need any syncing at all we enforce it's
1276	 * always set instead of having to deal with possibly weird behaviour
1277	 * for malicious applications setting only __O_SYNC.
1278	 */
1279	if (flags & __O_SYNC)
1280		flags |= O_DSYNC;
1281
1282	op->open_flag = flags;
1283
1284	/* O_TRUNC implies we need access checks for write permissions */
1285	if (flags & O_TRUNC)
1286		acc_mode |= MAY_WRITE;
1287
1288	/* Allow the LSM permission hook to distinguish append
1289	   access from general write access. */
1290	if (flags & O_APPEND)
1291		acc_mode |= MAY_APPEND;
1292
1293	op->acc_mode = acc_mode;
1294
1295	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1296
1297	if (flags & O_CREAT) {
1298		op->intent |= LOOKUP_CREATE;
1299		if (flags & O_EXCL) {
1300			op->intent |= LOOKUP_EXCL;
1301			flags |= O_NOFOLLOW;
1302		}
1303	}
1304
1305	if (flags & O_DIRECTORY)
1306		lookup_flags |= LOOKUP_DIRECTORY;
1307	if (!(flags & O_NOFOLLOW))
1308		lookup_flags |= LOOKUP_FOLLOW;
1309
1310	if (how->resolve & RESOLVE_NO_XDEV)
1311		lookup_flags |= LOOKUP_NO_XDEV;
1312	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1313		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1314	if (how->resolve & RESOLVE_NO_SYMLINKS)
1315		lookup_flags |= LOOKUP_NO_SYMLINKS;
1316	if (how->resolve & RESOLVE_BENEATH)
1317		lookup_flags |= LOOKUP_BENEATH;
1318	if (how->resolve & RESOLVE_IN_ROOT)
1319		lookup_flags |= LOOKUP_IN_ROOT;
1320	if (how->resolve & RESOLVE_CACHED) {
1321		/* Don't bother even trying for create/truncate/tmpfile open */
1322		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1323			return -EAGAIN;
1324		lookup_flags |= LOOKUP_CACHED;
1325	}
1326
1327	op->lookup_flags = lookup_flags;
1328	return 0;
1329}
1330
1331/**
1332 * file_open_name - open file and return file pointer
1333 *
1334 * @name:	struct filename containing path to open
1335 * @flags:	open flags as per the open(2) second argument
1336 * @mode:	mode for the new file if O_CREAT is set, else ignored
1337 *
1338 * This is the helper to open a file from kernelspace if you really
1339 * have to.  But in generally you should not do this, so please move
1340 * along, nothing to see here..
1341 */
1342struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1343{
1344	struct open_flags op;
1345	struct open_how how = build_open_how(flags, mode);
1346	int err = build_open_flags(&how, &op);
1347	if (err)
1348		return ERR_PTR(err);
1349	return do_filp_open(AT_FDCWD, name, &op);
1350}
1351
1352/**
1353 * filp_open - open file and return file pointer
1354 *
1355 * @filename:	path to open
1356 * @flags:	open flags as per the open(2) second argument
1357 * @mode:	mode for the new file if O_CREAT is set, else ignored
1358 *
1359 * This is the helper to open a file from kernelspace if you really
1360 * have to.  But in generally you should not do this, so please move
1361 * along, nothing to see here..
1362 */
1363struct file *filp_open(const char *filename, int flags, umode_t mode)
1364{
1365	struct filename *name = getname_kernel(filename);
1366	struct file *file = ERR_CAST(name);
1367	
1368	if (!IS_ERR(name)) {
1369		file = file_open_name(name, flags, mode);
1370		putname(name);
1371	}
1372	return file;
1373}
1374EXPORT_SYMBOL(filp_open);
1375
1376struct file *file_open_root(const struct path *root,
1377			    const char *filename, int flags, umode_t mode)
1378{
1379	struct open_flags op;
1380	struct open_how how = build_open_how(flags, mode);
1381	int err = build_open_flags(&how, &op);
1382	if (err)
1383		return ERR_PTR(err);
1384	return do_file_open_root(root, filename, &op);
1385}
1386EXPORT_SYMBOL(file_open_root);
1387
1388static long do_sys_openat2(int dfd, const char __user *filename,
1389			   struct open_how *how)
1390{
1391	struct open_flags op;
1392	int fd = build_open_flags(how, &op);
1393	struct filename *tmp;
1394
1395	if (fd)
1396		return fd;
1397
1398	tmp = getname(filename);
1399	if (IS_ERR(tmp))
1400		return PTR_ERR(tmp);
1401
1402	fd = get_unused_fd_flags(how->flags);
1403	if (fd >= 0) {
1404		struct file *f = do_filp_open(dfd, tmp, &op);
1405		if (IS_ERR(f)) {
1406			put_unused_fd(fd);
1407			fd = PTR_ERR(f);
1408		} else {
 
1409			fd_install(fd, f);
1410		}
1411	}
1412	putname(tmp);
1413	return fd;
1414}
1415
1416long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1417{
1418	struct open_how how = build_open_how(flags, mode);
1419	return do_sys_openat2(dfd, filename, &how);
1420}
1421
1422
1423SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1424{
1425	if (force_o_largefile())
1426		flags |= O_LARGEFILE;
1427	return do_sys_open(AT_FDCWD, filename, flags, mode);
1428}
1429
1430SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1431		umode_t, mode)
1432{
1433	if (force_o_largefile())
1434		flags |= O_LARGEFILE;
1435	return do_sys_open(dfd, filename, flags, mode);
1436}
1437
1438SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1439		struct open_how __user *, how, size_t, usize)
1440{
1441	int err;
1442	struct open_how tmp;
1443
1444	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1445	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1446
1447	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1448		return -EINVAL;
1449
1450	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1451	if (err)
1452		return err;
1453
1454	audit_openat2_how(&tmp);
1455
1456	/* O_LARGEFILE is only allowed for non-O_PATH. */
1457	if (!(tmp.flags & O_PATH) && force_o_largefile())
1458		tmp.flags |= O_LARGEFILE;
1459
1460	return do_sys_openat2(dfd, filename, &tmp);
1461}
1462
1463#ifdef CONFIG_COMPAT
1464/*
1465 * Exactly like sys_open(), except that it doesn't set the
1466 * O_LARGEFILE flag.
1467 */
1468COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1469{
1470	return do_sys_open(AT_FDCWD, filename, flags, mode);
1471}
1472
1473/*
1474 * Exactly like sys_openat(), except that it doesn't set the
1475 * O_LARGEFILE flag.
1476 */
1477COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1478{
1479	return do_sys_open(dfd, filename, flags, mode);
1480}
1481#endif
1482
1483#ifndef __alpha__
1484
1485/*
1486 * For backward compatibility?  Maybe this should be moved
1487 * into arch/i386 instead?
1488 */
1489SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1490{
1491	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1492
1493	if (force_o_largefile())
1494		flags |= O_LARGEFILE;
1495	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1496}
1497#endif
1498
1499/*
1500 * "id" is the POSIX thread ID. We use the
1501 * files pointer for this..
1502 */
1503static int filp_flush(struct file *filp, fl_owner_t id)
1504{
1505	int retval = 0;
1506
1507	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1508			"VFS: Close: file count is 0 (f_op=%ps)",
1509			filp->f_op)) {
1510		return 0;
1511	}
1512
1513	if (filp->f_op->flush)
1514		retval = filp->f_op->flush(filp, id);
1515
1516	if (likely(!(filp->f_mode & FMODE_PATH))) {
1517		dnotify_flush(filp, id);
1518		locks_remove_posix(filp, id);
1519	}
 
1520	return retval;
1521}
1522
1523int filp_close(struct file *filp, fl_owner_t id)
1524{
1525	int retval;
1526
1527	retval = filp_flush(filp, id);
1528	fput(filp);
1529
1530	return retval;
1531}
1532EXPORT_SYMBOL(filp_close);
1533
1534/*
1535 * Careful here! We test whether the file pointer is NULL before
1536 * releasing the fd. This ensures that one clone task can't release
1537 * an fd while another clone is opening it.
1538 */
1539SYSCALL_DEFINE1(close, unsigned int, fd)
1540{
1541	int retval;
1542	struct file *file;
1543
1544	file = file_close_fd(fd);
1545	if (!file)
1546		return -EBADF;
1547
1548	retval = filp_flush(file, current->files);
1549
1550	/*
1551	 * We're returning to user space. Don't bother
1552	 * with any delayed fput() cases.
1553	 */
1554	__fput_sync(file);
1555
1556	/* can't restart close syscall because file table entry was cleared */
1557	if (unlikely(retval == -ERESTARTSYS ||
1558		     retval == -ERESTARTNOINTR ||
1559		     retval == -ERESTARTNOHAND ||
1560		     retval == -ERESTART_RESTARTBLOCK))
1561		retval = -EINTR;
1562
1563	return retval;
1564}
1565
1566/**
1567 * sys_close_range() - Close all file descriptors in a given range.
1568 *
1569 * @fd:     starting file descriptor to close
1570 * @max_fd: last file descriptor to close
1571 * @flags:  reserved for future extensions
1572 *
1573 * This closes a range of file descriptors. All file descriptors
1574 * from @fd up to and including @max_fd are closed.
1575 * Currently, errors to close a given file descriptor are ignored.
1576 */
1577SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1578		unsigned int, flags)
1579{
1580	return __close_range(fd, max_fd, flags);
1581}
1582
1583/*
1584 * This routine simulates a hangup on the tty, to arrange that users
1585 * are given clean terminals at login time.
1586 */
1587SYSCALL_DEFINE0(vhangup)
1588{
1589	if (capable(CAP_SYS_TTY_CONFIG)) {
1590		tty_vhangup_self();
1591		return 0;
1592	}
1593	return -EPERM;
1594}
1595
1596/*
1597 * Called when an inode is about to be open.
1598 * We use this to disallow opening large files on 32bit systems if
1599 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1600 * on this flag in sys_open.
1601 */
1602int generic_file_open(struct inode * inode, struct file * filp)
1603{
1604	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1605		return -EOVERFLOW;
1606	return 0;
1607}
1608
1609EXPORT_SYMBOL(generic_file_open);
1610
1611/*
1612 * This is used by subsystems that don't want seekable
1613 * file descriptors. The function is not supposed to ever fail, the only
1614 * reason it returns an 'int' and not 'void' is so that it can be plugged
1615 * directly into file_operations structure.
1616 */
1617int nonseekable_open(struct inode *inode, struct file *filp)
1618{
1619	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1620	return 0;
1621}
1622
1623EXPORT_SYMBOL(nonseekable_open);
1624
1625/*
1626 * stream_open is used by subsystems that want stream-like file descriptors.
1627 * Such file descriptors are not seekable and don't have notion of position
1628 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1629 * Contrary to file descriptors of other regular files, .read() and .write()
1630 * can run simultaneously.
1631 *
1632 * stream_open never fails and is marked to return int so that it could be
1633 * directly used as file_operations.open .
1634 */
1635int stream_open(struct inode *inode, struct file *filp)
1636{
1637	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1638	filp->f_mode |= FMODE_STREAM;
1639	return 0;
1640}
1641
1642EXPORT_SYMBOL(stream_open);