1/*
   2 *  linux/fs/open.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/string.h>
   8#include <linux/mm.h>
   9#include <linux/file.h>
  10#include <linux/fdtable.h>
  11#include <linux/fsnotify.h>
  12#include <linux/module.h>
  13#include <linux/tty.h>
  14#include <linux/namei.h>
  15#include <linux/backing-dev.h>
  16#include <linux/capability.h>
  17#include <linux/securebits.h>
  18#include <linux/security.h>
  19#include <linux/mount.h>
  20#include <linux/fcntl.h>
  21#include <linux/slab.h>
  22#include <asm/uaccess.h>
  23#include <linux/fs.h>
  24#include <linux/personality.h>
  25#include <linux/pagemap.h>
  26#include <linux/syscalls.h>
  27#include <linux/rcupdate.h>
  28#include <linux/audit.h>
  29#include <linux/falloc.h>
  30#include <linux/fs_struct.h>
  31#include <linux/ima.h>
  32#include <linux/dnotify.h>
 
 
 
  33
  34#include "internal.h"
  35
  36int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  37	struct file *filp)
  38{
  39	int ret;
  40	struct iattr newattrs;
  41
  42	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  43	if (length < 0)
  44		return -EINVAL;
  45
  46	newattrs.ia_size = length;
  47	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  48	if (filp) {
  49		newattrs.ia_file = filp;
  50		newattrs.ia_valid |= ATTR_FILE;
  51	}
  52
  53	/* Remove suid/sgid on truncate too */
  54	ret = should_remove_suid(dentry);
 
 
  55	if (ret)
  56		newattrs.ia_valid |= ret | ATTR_FORCE;
  57
  58	mutex_lock(&dentry->d_inode->i_mutex);
  59	ret = notify_change(dentry, &newattrs);
  60	mutex_unlock(&dentry->d_inode->i_mutex);
 
  61	return ret;
  62}
  63
  64static long do_sys_truncate(const char __user *pathname, loff_t length)
  65{
  66	struct path path;
  67	struct inode *inode;
  68	int error;
  69
  70	error = -EINVAL;
  71	if (length < 0)	/* sorry, but loff_t says... */
  72		goto out;
  73
  74	error = user_path(pathname, &path);
  75	if (error)
  76		goto out;
  77	inode = path.dentry->d_inode;
  78
  79	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  80	error = -EISDIR;
  81	if (S_ISDIR(inode->i_mode))
  82		goto dput_and_out;
  83
  84	error = -EINVAL;
  85	if (!S_ISREG(inode->i_mode))
  86		goto dput_and_out;
  87
  88	error = mnt_want_write(path.mnt);
  89	if (error)
  90		goto dput_and_out;
  91
  92	error = inode_permission(inode, MAY_WRITE);
 
  93	if (error)
  94		goto mnt_drop_write_and_out;
  95
  96	error = -EPERM;
  97	if (IS_APPEND(inode))
  98		goto mnt_drop_write_and_out;
  99
 100	error = get_write_access(inode);
 101	if (error)
 102		goto mnt_drop_write_and_out;
 103
 104	/*
 105	 * Make sure that there are no leases.  get_write_access() protects
 106	 * against the truncate racing with a lease-granting setlease().
 107	 */
 108	error = break_lease(inode, O_WRONLY);
 109	if (error)
 110		goto put_write_and_out;
 111
 112	error = locks_verify_truncate(inode, NULL, length);
 113	if (!error)
 114		error = security_path_truncate(&path);
 115	if (!error)
 116		error = do_truncate(path.dentry, length, 0, NULL);
 117
 118put_write_and_out:
 119	put_write_access(inode);
 120mnt_drop_write_and_out:
 121	mnt_drop_write(path.mnt);
 122dput_and_out:
 123	path_put(&path);
 124out:
 125	return error;
 126}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 127
 128SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 129{
 130	return do_sys_truncate(path, length);
 131}
 132
 133static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 
 134{
 135	struct inode * inode;
 
 
 
 
 
 
 136	struct dentry *dentry;
 137	struct file * file;
 138	int error;
 139
 140	error = -EINVAL;
 141	if (length < 0)
 142		goto out;
 143	error = -EBADF;
 144	file = fget(fd);
 145	if (!file)
 146		goto out;
 147
 148	/* explicitly opened as large or we are on 64-bit box */
 149	if (file->f_flags & O_LARGEFILE)
 150		small = 0;
 151
 152	dentry = file->f_path.dentry;
 153	inode = dentry->d_inode;
 154	error = -EINVAL;
 155	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 156		goto out_putf;
 157
 158	error = -EINVAL;
 159	/* Cannot ftruncate over 2^31 bytes without large file support */
 160	if (small && length > MAX_NON_LFS)
 161		goto out_putf;
 162
 163	error = -EPERM;
 164	if (IS_APPEND(inode))
 
 165		goto out_putf;
 166
 167	error = locks_verify_truncate(inode, file, length);
 168	if (!error)
 169		error = security_path_truncate(&file->f_path);
 170	if (!error)
 171		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 
 
 172out_putf:
 173	fput(file);
 174out:
 175	return error;
 176}
 177
 178SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 179{
 180	long ret = do_sys_ftruncate(fd, length, 1);
 181	/* avoid REGPARM breakage on x86: */
 182	asmlinkage_protect(2, ret, fd, length);
 183	return ret;
 
 
 
 184}
 
 185
 186/* LFS versions of truncate are only needed on 32 bit machines */
 187#if BITS_PER_LONG == 32
 188SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 189{
 190	return do_sys_truncate(path, length);
 191}
 192#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 193asmlinkage long SyS_truncate64(long path, loff_t length)
 194{
 195	return SYSC_truncate64((const char __user *) path, length);
 196}
 197SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 198#endif
 199
 200SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 
 
 201{
 202	long ret = do_sys_ftruncate(fd, length, 0);
 203	/* avoid REGPARM breakage on x86: */
 204	asmlinkage_protect(2, ret, fd, length);
 205	return ret;
 206}
 207#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 208asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 
 
 
 209{
 210	return SYSC_ftruncate64((unsigned int) fd, length);
 211}
 212SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 213#endif
 214#endif /* BITS_PER_LONG == 32 */
 215
 216
 217int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 218{
 219	struct inode *inode = file->f_path.dentry->d_inode;
 220	long ret;
 221
 222	if (offset < 0 || len <= 0)
 223		return -EINVAL;
 224
 225	/* Return error if mode is not supported */
 226	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 
 
 
 
 
 227		return -EOPNOTSUPP;
 228
 229	/* Punch hole must have keep size set */
 230	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 231	    !(mode & FALLOC_FL_KEEP_SIZE))
 232		return -EOPNOTSUPP;
 233
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 234	if (!(file->f_mode & FMODE_WRITE))
 235		return -EBADF;
 236
 237	/* It's not possible punch hole on append only file */
 238	if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 
 
 239		return -EPERM;
 240
 241	if (IS_IMMUTABLE(inode))
 242		return -EPERM;
 243
 244	/*
 
 
 
 
 
 
 245	 * Revalidate the write permissions, in case security policy has
 246	 * changed since the files were opened.
 247	 */
 248	ret = security_file_permission(file, MAY_WRITE);
 249	if (ret)
 250		return ret;
 251
 
 
 
 
 252	if (S_ISFIFO(inode->i_mode))
 253		return -ESPIPE;
 254
 255	/*
 256	 * Let individual file system decide if it supports preallocation
 257	 * for directories or not.
 258	 */
 259	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 260		return -ENODEV;
 261
 262	/* Check for wrap through zero too */
 263	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 264		return -EFBIG;
 265
 266	if (!file->f_op->fallocate)
 267		return -EOPNOTSUPP;
 268
 269	return file->f_op->fallocate(file, mode, offset, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270}
 
 271
 272SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 273{
 274	struct file *file;
 275	int error = -EBADF;
 276
 277	file = fget(fd);
 278	if (file) {
 279		error = do_fallocate(file, mode, offset, len);
 280		fput(file);
 281	}
 282
 283	return error;
 284}
 285
 286#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 287asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 
 
 
 
 
 
 288{
 289	return SYSC_fallocate((int)fd, (int)mode, offset, len);
 
 290}
 291SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 292#endif
 293
 294/*
 295 * access() needs to use the real uid/gid, not the effective uid/gid.
 296 * We do this by temporarily clearing all FS-related capabilities and
 297 * switching the fsuid/fsgid around to the real ones.
 
 
 
 298 */
 299SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 300{
 301	const struct cred *old_cred;
 302	struct cred *override_cred;
 303	struct path path;
 304	struct inode *inode;
 305	int res;
 306
 307	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 308		return -EINVAL;
 309
 310	override_cred = prepare_creds();
 311	if (!override_cred)
 312		return -ENOMEM;
 
 
 
 
 
 
 313
 314	override_cred->fsuid = override_cred->uid;
 315	override_cred->fsgid = override_cred->gid;
 316
 317	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 318		/* Clear the capabilities if we switch to a non-root user */
 319		if (override_cred->uid)
 
 320			cap_clear(override_cred->cap_effective);
 321		else
 322			override_cred->cap_effective =
 323				override_cred->cap_permitted;
 324	}
 325
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 326	old_cred = override_creds(override_cred);
 327
 328	res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 329	if (res)
 330		goto out;
 331
 332	inode = path.dentry->d_inode;
 333
 334	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 335		/*
 336		 * MAY_EXEC on regular files is denied if the fs is mounted
 337		 * with the "noexec" flag.
 338		 */
 339		res = -EACCES;
 340		if (path.mnt->mnt_flags & MNT_NOEXEC)
 341			goto out_path_release;
 342	}
 343
 344	res = inode_permission(inode, mode | MAY_ACCESS);
 345	/* SuS v2 requires we report a read only fs too */
 346	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 347		goto out_path_release;
 348	/*
 349	 * This is a rare case where using __mnt_is_readonly()
 350	 * is OK without a mnt_want/drop_write() pair.  Since
 351	 * no actual write to the fs is performed here, we do
 352	 * not need to telegraph to that to anyone.
 353	 *
 354	 * By doing this, we accept that this access is
 355	 * inherently racy and know that the fs may change
 356	 * state before we even see this result.
 357	 */
 358	if (__mnt_is_readonly(path.mnt))
 359		res = -EROFS;
 360
 361out_path_release:
 362	path_put(&path);
 
 
 
 
 363out:
 364	revert_creds(old_cred);
 365	put_cred(override_cred);
 
 366	return res;
 367}
 368
 
 
 
 
 
 
 
 
 
 
 
 369SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 370{
 371	return sys_faccessat(AT_FDCWD, filename, mode);
 372}
 373
 374SYSCALL_DEFINE1(chdir, const char __user *, filename)
 375{
 376	struct path path;
 377	int error;
 378
 379	error = user_path_dir(filename, &path);
 
 380	if (error)
 381		goto out;
 382
 383	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 384	if (error)
 385		goto dput_and_out;
 386
 387	set_fs_pwd(current->fs, &path);
 388
 389dput_and_out:
 390	path_put(&path);
 
 
 
 
 391out:
 392	return error;
 393}
 394
 395SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 396{
 397	struct file *file;
 398	struct inode *inode;
 399	int error;
 400
 401	error = -EBADF;
 402	file = fget(fd);
 403	if (!file)
 404		goto out;
 405
 406	inode = file->f_path.dentry->d_inode;
 407
 408	error = -ENOTDIR;
 409	if (!S_ISDIR(inode->i_mode))
 410		goto out_putf;
 411
 412	error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 413	if (!error)
 414		set_fs_pwd(current->fs, &file->f_path);
 415out_putf:
 416	fput(file);
 417out:
 418	return error;
 419}
 420
 421SYSCALL_DEFINE1(chroot, const char __user *, filename)
 422{
 423	struct path path;
 424	int error;
 425
 426	error = user_path_dir(filename, &path);
 
 427	if (error)
 428		goto out;
 429
 430	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 431	if (error)
 432		goto dput_and_out;
 433
 434	error = -EPERM;
 435	if (!capable(CAP_SYS_CHROOT))
 436		goto dput_and_out;
 437	error = security_path_chroot(&path);
 438	if (error)
 439		goto dput_and_out;
 440
 441	set_fs_root(current->fs, &path);
 442	error = 0;
 443dput_and_out:
 444	path_put(&path);
 
 
 
 
 445out:
 446	return error;
 447}
 448
 449static int chmod_common(struct path *path, umode_t mode)
 450{
 451	struct inode *inode = path->dentry->d_inode;
 
 452	struct iattr newattrs;
 453	int error;
 454
 455	error = mnt_want_write(path->mnt);
 456	if (error)
 457		return error;
 458	mutex_lock(&inode->i_mutex);
 459	error = security_path_chmod(path->dentry, path->mnt, mode);
 
 460	if (error)
 461		goto out_unlock;
 462	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 463	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 464	error = notify_change(path->dentry, &newattrs);
 
 465out_unlock:
 466	mutex_unlock(&inode->i_mutex);
 
 
 
 
 
 467	mnt_drop_write(path->mnt);
 468	return error;
 469}
 470
 471SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 
 
 
 
 
 
 472{
 473	struct file * file;
 474	int err = -EBADF;
 475
 476	file = fget(fd);
 477	if (file) {
 478		audit_inode(NULL, file->f_path.dentry);
 479		err = chmod_common(&file->f_path, mode);
 480		fput(file);
 481	}
 482	return err;
 483}
 484
 485SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 
 486{
 487	struct path path;
 488	int error;
 
 489
 490	error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 
 
 
 
 
 
 
 
 491	if (!error) {
 492		error = chmod_common(&path, mode);
 493		path_put(&path);
 
 
 
 
 494	}
 495	return error;
 496}
 497
 498SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 
 499{
 500	return sys_fchmodat(AT_FDCWD, filename, mode);
 501}
 502
 503static int chown_common(struct path *path, uid_t user, gid_t group)
 
 504{
 505	struct inode *inode = path->dentry->d_inode;
 506	int error;
 507	struct iattr newattrs;
 508
 509	newattrs.ia_valid =  ATTR_CTIME;
 510	if (user != (uid_t) -1) {
 511		newattrs.ia_valid |= ATTR_UID;
 512		newattrs.ia_uid = user;
 513	}
 514	if (group != (gid_t) -1) {
 515		newattrs.ia_valid |= ATTR_GID;
 516		newattrs.ia_gid = group;
 517	}
 518	if (!S_ISDIR(inode->i_mode))
 519		newattrs.ia_valid |=
 520			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 521	mutex_lock(&inode->i_mutex);
 522	error = security_path_chown(path, user, group);
 523	if (!error)
 524		error = notify_change(path->dentry, &newattrs);
 525	mutex_unlock(&inode->i_mutex);
 526
 527	return error;
 
 
 
 
 
 
 
 
 
 
 
 
 528}
 529
 530SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 
 
 
 
 
 
 531{
 532	struct path path;
 
 
 
 
 
 
 
 
 
 
 
 
 533	int error;
 
 
 
 534
 535	error = user_path(filename, &path);
 536	if (error)
 537		goto out;
 538	error = mnt_want_write(path.mnt);
 539	if (error)
 540		goto out_release;
 541	error = chown_common(&path, user, group);
 542	mnt_drop_write(path.mnt);
 543out_release:
 544	path_put(&path);
 545out:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 546	return error;
 547}
 548
 549SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 550		gid_t, group, int, flag)
 551{
 552	struct path path;
 553	int error = -EINVAL;
 554	int lookup_flags;
 555
 556	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 557		goto out;
 558
 559	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 560	if (flag & AT_EMPTY_PATH)
 561		lookup_flags |= LOOKUP_EMPTY;
 
 562	error = user_path_at(dfd, filename, lookup_flags, &path);
 563	if (error)
 564		goto out;
 565	error = mnt_want_write(path.mnt);
 566	if (error)
 567		goto out_release;
 568	error = chown_common(&path, user, group);
 569	mnt_drop_write(path.mnt);
 570out_release:
 571	path_put(&path);
 
 
 
 
 572out:
 573	return error;
 574}
 575
 
 
 
 
 
 
 
 
 
 
 
 576SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 577{
 578	struct path path;
 
 
 
 
 
 579	int error;
 580
 581	error = user_lpath(filename, &path);
 582	if (error)
 583		goto out;
 584	error = mnt_want_write(path.mnt);
 585	if (error)
 586		goto out_release;
 587	error = chown_common(&path, user, group);
 588	mnt_drop_write(path.mnt);
 589out_release:
 590	path_put(&path);
 591out:
 592	return error;
 593}
 594
 595SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 596{
 597	struct file * file;
 598	int error = -EBADF;
 599	struct dentry * dentry;
 600
 601	file = fget(fd);
 602	if (!file)
 603		goto out;
 604
 605	error = mnt_want_write_file(file);
 606	if (error)
 607		goto out_fput;
 608	dentry = file->f_path.dentry;
 609	audit_inode(NULL, dentry);
 610	error = chown_common(&file->f_path, user, group);
 611	mnt_drop_write(file->f_path.mnt);
 612out_fput:
 613	fput(file);
 614out:
 615	return error;
 616}
 617
 618/*
 619 * You have to be very careful that these write
 620 * counts get cleaned up in error cases and
 621 * upon __fput().  This should probably never
 622 * be called outside of __dentry_open().
 623 */
 624static inline int __get_file_write_access(struct inode *inode,
 625					  struct vfsmount *mnt)
 626{
 627	int error;
 628	error = get_write_access(inode);
 629	if (error)
 
 630		return error;
 631	/*
 632	 * Do not take mount writer counts on
 633	 * special files since no writes to
 634	 * the mount itself will occur.
 635	 */
 636	if (!special_file(inode->i_mode)) {
 637		/*
 638		 * Balanced in __fput()
 639		 */
 640		error = mnt_want_write(mnt);
 641		if (error)
 642			put_write_access(inode);
 643	}
 
 
 
 
 
 
 644	return error;
 645}
 646
 647static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 648					struct file *f,
 649					int (*open)(struct inode *, struct file *),
 650					const struct cred *cred)
 651{
 652	static const struct file_operations empty_fops = {};
 653	struct inode *inode;
 654	int error;
 655
 656	f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 657				FMODE_PREAD | FMODE_PWRITE;
 
 
 
 658
 659	if (unlikely(f->f_flags & O_PATH))
 660		f->f_mode = FMODE_PATH;
 
 
 
 661
 662	inode = dentry->d_inode;
 663	if (f->f_mode & FMODE_WRITE) {
 664		error = __get_file_write_access(inode, mnt);
 665		if (error)
 
 666			goto cleanup_file;
 667		if (!special_file(inode->i_mode))
 668			file_take_write(f);
 669	}
 670
 671	f->f_mapping = inode->i_mapping;
 672	f->f_path.dentry = dentry;
 673	f->f_path.mnt = mnt;
 674	f->f_pos = 0;
 675	file_sb_list_add(f, inode->i_sb);
 676
 677	if (unlikely(f->f_mode & FMODE_PATH)) {
 678		f->f_op = &empty_fops;
 679		return f;
 
 680	}
 681
 682	f->f_op = fops_get(inode->i_fop);
 
 
 683
 684	error = security_dentry_open(f, cred);
 685	if (error)
 686		goto cleanup_all;
 687
 688	if (!open && f->f_op)
 
 
 689		open = f->f_op->open;
 690	if (open) {
 691		error = open(inode, f);
 692		if (error)
 693			goto cleanup_all;
 694	}
 695	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 696		i_readcount_inc(inode);
 
 
 
 
 
 
 
 
 
 697
 698	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 
 699
 700	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 701
 702	/* NB: we're sure to have correct a_ops only after f_op->open */
 703	if (f->f_flags & O_DIRECT) {
 704		if (!f->f_mapping->a_ops ||
 705		    ((!f->f_mapping->a_ops->direct_IO) &&
 706		    (!f->f_mapping->a_ops->get_xip_mem))) {
 707			fput(f);
 708			f = ERR_PTR(-EINVAL);
 709		}
 710	}
 711
 712	return f;
 713
 714cleanup_all:
 715	fops_put(f->f_op);
 
 
 716	if (f->f_mode & FMODE_WRITE) {
 717		put_write_access(inode);
 718		if (!special_file(inode->i_mode)) {
 
 
 
 
 
 
 
 
 
 719			/*
 720			 * We don't consider this a real
 721			 * mnt_want/drop_write() pair
 722			 * because it all happenend right
 723			 * here, so just reset the state.
 724			 */
 725			file_reset_write(f);
 726			mnt_drop_write(mnt);
 
 727		}
 728	}
 729	file_sb_list_del(f);
 730	f->f_path.dentry = NULL;
 731	f->f_path.mnt = NULL;
 
 
 
 
 
 
 
 
 
 
 732cleanup_file:
 733	put_filp(f);
 734	dput(dentry);
 735	mntput(mnt);
 736	return ERR_PTR(error);
 
 737}
 738
 739/**
 740 * lookup_instantiate_filp - instantiates the open intent filp
 741 * @nd: pointer to nameidata
 742 * @dentry: pointer to dentry
 743 * @open: open callback
 744 *
 745 * Helper for filesystems that want to use lookup open intents and pass back
 746 * a fully instantiated struct file to the caller.
 747 * This function is meant to be called from within a filesystem's
 748 * lookup method.
 749 * Beware of calling it for non-regular files! Those ->open methods might block
 750 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 751 * leading to a deadlock, as nobody can open that fifo anymore, because
 752 * another process to open fifo will block on locked parent when doing lookup).
 753 * Note that in case of error, nd->intent.open.file is destroyed, but the
 754 * path information remains valid.
 755 * If the open callback is set to NULL, then the standard f_op->open()
 756 * filesystem callback is substituted.
 
 
 
 
 
 
 757 */
 758struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 759		int (*open)(struct inode *, struct file *))
 760{
 761	const struct cred *cred = current_cred();
 762
 763	if (IS_ERR(nd->intent.open.file))
 764		goto out;
 765	if (IS_ERR(dentry))
 766		goto out_err;
 767	nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 768					     nd->intent.open.file,
 769					     open, cred);
 770out:
 771	return nd->intent.open.file;
 772out_err:
 773	release_open_intent(nd);
 774	nd->intent.open.file = ERR_CAST(dentry);
 775	goto out;
 776}
 777EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 778
 779/**
 780 * nameidata_to_filp - convert a nameidata to an open filp.
 781 * @nd: pointer to nameidata
 782 * @flags: open flags
 
 
 
 
 
 
 783 *
 784 * Note that this function destroys the original nameidata
 
 785 */
 786struct file *nameidata_to_filp(struct nameidata *nd)
 787{
 788	const struct cred *cred = current_cred();
 789	struct file *filp;
 790
 791	/* Pick up the filp from the open intent */
 792	filp = nd->intent.open.file;
 793	nd->intent.open.file = NULL;
 794
 795	/* Has the filesystem initialised the file for us? */
 796	if (filp->f_path.dentry == NULL) {
 797		path_get(&nd->path);
 798		filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 799				     NULL, cred);
 800	}
 801	return filp;
 802}
 
 803
 804/*
 805 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 806 * error.
 
 807 */
 808struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 
 
 
 
 
 
 809			 const struct cred *cred)
 810{
 811	int error;
 812	struct file *f;
 813
 814	validate_creds(cred);
 815
 816	/* We must always pass in a valid mount pointer. */
 817	BUG_ON(!mnt);
 818
 819	error = -ENFILE;
 820	f = get_empty_filp();
 821	if (f == NULL) {
 822		dput(dentry);
 823		mntput(mnt);
 824		return ERR_PTR(error);
 
 825	}
 826
 827	f->f_flags = flags;
 828	return __dentry_open(dentry, mnt, f, NULL, cred);
 829}
 830EXPORT_SYMBOL(dentry_open);
 831
 832static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 833{
 834	struct fdtable *fdt = files_fdtable(files);
 835	__FD_CLR(fd, fdt->open_fds);
 836	if (fd < files->next_fd)
 837		files->next_fd = fd;
 838}
 839
 840void put_unused_fd(unsigned int fd)
 841{
 842	struct files_struct *files = current->files;
 843	spin_lock(&files->file_lock);
 844	__put_unused_fd(files, fd);
 845	spin_unlock(&files->file_lock);
 846}
 847
 848EXPORT_SYMBOL(put_unused_fd);
 
 
 
 
 849
 850/*
 851 * Install a file pointer in the fd array.
 
 
 
 
 
 
 
 
 
 
 
 
 852 *
 853 * The VFS is full of places where we drop the files lock between
 854 * setting the open_fds bitmap and installing the file in the file
 855 * array.  At any such point, we are vulnerable to a dup2() race
 856 * installing a file in the array before us.  We need to detect this and
 857 * fput() the struct file we are about to overwrite in this case.
 858 *
 859 * It should never happen - if we allow dup2() do it, _really_ bad things
 860 * will follow.
 861 */
 862
 863void fd_install(unsigned int fd, struct file *file)
 864{
 865	struct files_struct *files = current->files;
 866	struct fdtable *fdt;
 867	spin_lock(&files->file_lock);
 868	fdt = files_fdtable(files);
 869	BUG_ON(fdt->fd[fd] != NULL);
 870	rcu_assign_pointer(fdt->fd[fd], file);
 871	spin_unlock(&files->file_lock);
 
 
 
 
 
 
 
 872}
 
 
 
 
 873
 874EXPORT_SYMBOL(fd_install);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 875
 876static inline int build_open_flags(int flags, int mode, struct open_flags *op)
 877{
 
 
 878	int lookup_flags = 0;
 879	int acc_mode;
 880
 881	if (!(flags & O_CREAT))
 882		mode = 0;
 883	op->mode = mode;
 884
 885	/* Must never be set by userspace */
 886	flags &= ~FMODE_NONOTIFY;
 
 
 
 
 887
 888	/*
 889	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 890	 * check for O_DSYNC if the need any syncing at all we enforce it's
 891	 * always set instead of having to deal with possibly weird behaviour
 892	 * for malicious applications setting only __O_SYNC.
 893	 */
 894	if (flags & __O_SYNC)
 895		flags |= O_DSYNC;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 896
 897	/*
 898	 * If we have O_PATH in the open flag. Then we
 899	 * cannot have anything other than the below set of flags
 
 900	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 901	if (flags & O_PATH) {
 902		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 
 
 903		acc_mode = 0;
 904	} else {
 905		acc_mode = MAY_OPEN | ACC_MODE(flags);
 906	}
 907
 
 
 
 
 
 
 
 
 
 908	op->open_flag = flags;
 909
 910	/* O_TRUNC implies we need access checks for write permissions */
 911	if (flags & O_TRUNC)
 912		acc_mode |= MAY_WRITE;
 913
 914	/* Allow the LSM permission hook to distinguish append
 915	   access from general write access. */
 916	if (flags & O_APPEND)
 917		acc_mode |= MAY_APPEND;
 918
 919	op->acc_mode = acc_mode;
 920
 921	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 922
 923	if (flags & O_CREAT) {
 924		op->intent |= LOOKUP_CREATE;
 925		if (flags & O_EXCL)
 926			op->intent |= LOOKUP_EXCL;
 
 
 927	}
 928
 929	if (flags & O_DIRECTORY)
 930		lookup_flags |= LOOKUP_DIRECTORY;
 931	if (!(flags & O_NOFOLLOW))
 932		lookup_flags |= LOOKUP_FOLLOW;
 933	return lookup_flags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 934}
 935
 936/**
 937 * filp_open - open file and return file pointer
 938 *
 939 * @filename:	path to open
 940 * @flags:	open flags as per the open(2) second argument
 941 * @mode:	mode for the new file if O_CREAT is set, else ignored
 942 *
 943 * This is the helper to open a file from kernelspace if you really
 944 * have to.  But in generally you should not do this, so please move
 945 * along, nothing to see here..
 946 */
 947struct file *filp_open(const char *filename, int flags, int mode)
 948{
 949	struct open_flags op;
 950	int lookup = build_open_flags(flags, mode, &op);
 951	return do_filp_open(AT_FDCWD, filename, &op, lookup);
 
 
 
 
 
 952}
 953EXPORT_SYMBOL(filp_open);
 954
 955struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 956			    const char *filename, int flags)
 957{
 958	struct open_flags op;
 959	int lookup = build_open_flags(flags, 0, &op);
 960	if (flags & O_CREAT)
 961		return ERR_PTR(-EINVAL);
 962	if (!filename && (flags & O_DIRECTORY))
 963		if (!dentry->d_inode->i_op->lookup)
 964			return ERR_PTR(-ENOTDIR);
 965	return do_file_open_root(dentry, mnt, filename, &op, lookup);
 966}
 967EXPORT_SYMBOL(file_open_root);
 968
 969long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 
 970{
 971	struct open_flags op;
 972	int lookup = build_open_flags(flags, mode, &op);
 973	char *tmp = getname(filename);
 974	int fd = PTR_ERR(tmp);
 975
 976	if (!IS_ERR(tmp)) {
 977		fd = get_unused_fd_flags(flags);
 978		if (fd >= 0) {
 979			struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 980			if (IS_ERR(f)) {
 981				put_unused_fd(fd);
 982				fd = PTR_ERR(f);
 983			} else {
 984				fsnotify_open(f);
 985				fd_install(fd, f);
 986			}
 
 
 
 987		}
 988		putname(tmp);
 989	}
 
 990	return fd;
 991}
 992
 993SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 994{
 995	long ret;
 
 
 
 996
 
 
 997	if (force_o_largefile())
 998		flags |= O_LARGEFILE;
 999
1000	ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1001	/* avoid REGPARM breakage on x86: */
1002	asmlinkage_protect(3, ret, filename, flags, mode);
1003	return ret;
1004}
1005
1006SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1007		int, mode)
1008{
1009	long ret;
1010
1011	if (force_o_largefile())
1012		flags |= O_LARGEFILE;
 
 
1013
1014	ret = do_sys_open(dfd, filename, flags, mode);
1015	/* avoid REGPARM breakage on x86: */
1016	asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1017	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1018}
1019
 
 
 
 
 
 
 
 
 
 
1020#ifndef __alpha__
1021
1022/*
1023 * For backward compatibility?  Maybe this should be moved
1024 * into arch/i386 instead?
1025 */
1026SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1027{
1028	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1029}
1030
 
 
 
 
1031#endif
1032
1033/*
1034 * "id" is the POSIX thread ID. We use the
1035 * files pointer for this..
1036 */
1037int filp_close(struct file *filp, fl_owner_t id)
1038{
1039	int retval = 0;
1040
1041	if (!file_count(filp)) {
1042		printk(KERN_ERR "VFS: Close: file count is 0\n");
 
1043		return 0;
1044	}
1045
1046	if (filp->f_op && filp->f_op->flush)
1047		retval = filp->f_op->flush(filp, id);
1048
1049	if (likely(!(filp->f_mode & FMODE_PATH))) {
1050		dnotify_flush(filp, id);
1051		locks_remove_posix(filp, id);
1052	}
1053	fput(filp);
1054	return retval;
1055}
1056
 
 
 
 
 
 
 
 
 
1057EXPORT_SYMBOL(filp_close);
1058
1059/*
1060 * Careful here! We test whether the file pointer is NULL before
1061 * releasing the fd. This ensures that one clone task can't release
1062 * an fd while another clone is opening it.
1063 */
1064SYSCALL_DEFINE1(close, unsigned int, fd)
1065{
1066	struct file * filp;
1067	struct files_struct *files = current->files;
1068	struct fdtable *fdt;
1069	int retval;
 
1070
1071	spin_lock(&files->file_lock);
1072	fdt = files_fdtable(files);
1073	if (fd >= fdt->max_fds)
1074		goto out_unlock;
1075	filp = fdt->fd[fd];
1076	if (!filp)
1077		goto out_unlock;
1078	rcu_assign_pointer(fdt->fd[fd], NULL);
1079	FD_CLR(fd, fdt->close_on_exec);
1080	__put_unused_fd(files, fd);
1081	spin_unlock(&files->file_lock);
1082	retval = filp_close(filp, files);
1083
1084	/* can't restart close syscall because file table entry was cleared */
1085	if (unlikely(retval == -ERESTARTSYS ||
1086		     retval == -ERESTARTNOINTR ||
1087		     retval == -ERESTARTNOHAND ||
1088		     retval == -ERESTART_RESTARTBLOCK))
1089		retval = -EINTR;
1090
1091	return retval;
 
1092
1093out_unlock:
1094	spin_unlock(&files->file_lock);
1095	return -EBADF;
 
 
 
 
 
 
 
 
 
 
 
 
1096}
1097EXPORT_SYMBOL(sys_close);
1098
1099/*
1100 * This routine simulates a hangup on the tty, to arrange that users
1101 * are given clean terminals at login time.
1102 */
1103SYSCALL_DEFINE0(vhangup)
1104{
1105	if (capable(CAP_SYS_TTY_CONFIG)) {
1106		tty_vhangup_self();
1107		return 0;
1108	}
1109	return -EPERM;
1110}
1111
1112/*
1113 * Called when an inode is about to be open.
1114 * We use this to disallow opening large files on 32bit systems if
1115 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1116 * on this flag in sys_open.
1117 */
1118int generic_file_open(struct inode * inode, struct file * filp)
1119{
1120	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1121		return -EOVERFLOW;
1122	return 0;
1123}
1124
1125EXPORT_SYMBOL(generic_file_open);
1126
1127/*
1128 * This is used by subsystems that don't want seekable
1129 * file descriptors. The function is not supposed to ever fail, the only
1130 * reason it returns an 'int' and not 'void' is so that it can be plugged
1131 * directly into file_operations structure.
1132 */
1133int nonseekable_open(struct inode *inode, struct file *filp)
1134{
1135	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1136	return 0;
1137}
1138
1139EXPORT_SYMBOL(nonseekable_open);