fs/open.c

   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
  36 int 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
  64 static 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
 118 put_write_and_out:
 119         put_write_access(inode);
 120 mnt_drop_write_and_out:
 121         mnt_drop_write(path.mnt);
 122 dput_and_out:
 123         path_put(&path);
 124 out:
 125         return error;
 126 }
 127
 128 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 129 {
 130         return do_sys_truncate(path, length);
 131 }
 132
 133 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 134 {
 135         struct inode *inode;
 136         struct dentry *dentry;
 137         struct fd f;
 138         int error;
 139
 140         error = -EINVAL;
 141         if (length < 0)
 142                 goto out;
 143         error = -EBADF;
 144         f = fdget(fd);
 145         if (!f.file)
 146                 goto out;
 147
 148         /* explicitly opened as large or we are on 64-bit box */
 149         if (f.file->f_flags & O_LARGEFILE)
 150                 small = 0;
 151
 152         dentry = f.file->f_path.dentry;
 153         inode = dentry->d_inode;
 154         error = -EINVAL;
 155         if (!S_ISREG(inode->i_mode) || !(f.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         sb_start_write(inode->i_sb);
 168         error = locks_verify_truncate(inode, f.file, length);
 169         if (!error)
 170                 error = security_path_truncate(&f.file->f_path);
 171         if (!error)
 172                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 173         sb_end_write(inode->i_sb);
 174 out_putf:
 175         fdput(f);
 176 out:
 177         return error;
 178 }
 179
 180 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 181 {
 182         long ret = do_sys_ftruncate(fd, length, 1);
 183         /* avoid REGPARM breakage on x86: */
 184         asmlinkage_protect(2, ret, fd, length);
 185         return ret;
 186 }
 187
 188 /* LFS versions of truncate are only needed on 32 bit machines */
 189 #if BITS_PER_LONG == 32
 190 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 191 {
 192         return do_sys_truncate(path, length);
 193 }
 194 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 195 asmlinkage long SyS_truncate64(long path, loff_t length)
 196 {
 197         return SYSC_truncate64((const char __user *) path, length);
 198 }
 199 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 200 #endif
 201
 202 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 203 {
 204         long ret = do_sys_ftruncate(fd, length, 0);
 205         /* avoid REGPARM breakage on x86: */
 206         asmlinkage_protect(2, ret, fd, length);
 207         return ret;
 208 }
 209 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 210 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 211 {
 212         return SYSC_ftruncate64((unsigned int) fd, length);
 213 }
 214 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 215 #endif
 216 #endif /* BITS_PER_LONG == 32 */
 217
 218
 219 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 220 {
 221         struct inode *inode = file->f_path.dentry->d_inode;
 222         long ret;
 223
 224         if (offset < 0 || len <= 0)
 225                 return -EINVAL;
 226
 227         /* Return error if mode is not supported */
 228         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 229                 return -EOPNOTSUPP;
 230
 231         /* Punch hole must have keep size set */
 232         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 233             !(mode & FALLOC_FL_KEEP_SIZE))
 234                 return -EOPNOTSUPP;
 235
 236         if (!(file->f_mode & FMODE_WRITE))
 237                 return -EBADF;
 238
 239         /* It's not possible punch hole on append only file */
 240         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 241                 return -EPERM;
 242
 243         if (IS_IMMUTABLE(inode))
 244                 return -EPERM;
 245
 246         /*
 247          * Revalidate the write permissions, in case security policy has
 248          * changed since the files were opened.
 249          */
 250         ret = security_file_permission(file, MAY_WRITE);
 251         if (ret)
 252                 return ret;
 253
 254         if (S_ISFIFO(inode->i_mode))
 255                 return -ESPIPE;
 256
 257         /*
 258          * Let individual file system decide if it supports preallocation
 259          * for directories or not.
 260          */
 261         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 262                 return -ENODEV;
 263
 264         /* Check for wrap through zero too */
 265         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 266                 return -EFBIG;
 267
 268         if (!file->f_op->fallocate)
 269                 return -EOPNOTSUPP;
 270
 271         sb_start_write(inode->i_sb);
 272         ret = file->f_op->fallocate(file, mode, offset, len);
 273         sb_end_write(inode->i_sb);
 274         return ret;
 275 }
 276
 277 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 278 {
 279         struct fd f = fdget(fd);
 280         int error = -EBADF;
 281
 282         if (f.file) {
 283                 error = do_fallocate(f.file, mode, offset, len);
 284                 fdput(f);
 285         }
 286         return error;
 287 }
 288
 289 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 290 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 291 {
 292         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 293 }
 294 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 295 #endif
 296
 297 /*
 298  * access() needs to use the real uid/gid, not the effective uid/gid.
 299  * We do this by temporarily clearing all FS-related capabilities and
 300  * switching the fsuid/fsgid around to the real ones.
 301  */
 302 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 303 {
 304         const struct cred *old_cred;
 305         struct cred *override_cred;
 306         struct path path;
 307         struct inode *inode;
 308         int res;
 309
 310         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 311                 return -EINVAL;
 312
 313         override_cred = prepare_creds();
 314         if (!override_cred)
 315                 return -ENOMEM;
 316
 317         override_cred->fsuid = override_cred->uid;
 318         override_cred->fsgid = override_cred->gid;
 319
 320         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 321                 /* Clear the capabilities if we switch to a non-root user */
 322                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 323                 if (!uid_eq(override_cred->uid, root_uid))
 324                         cap_clear(override_cred->cap_effective);
 325                 else
 326                         override_cred->cap_effective =
 327                                 override_cred->cap_permitted;
 328         }
 329
 330         old_cred = override_creds(override_cred);
 331
 332         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 333         if (res)
 334                 goto out;
 335
 336         inode = path.dentry->d_inode;
 337
 338         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 339                 /*
 340                  * MAY_EXEC on regular files is denied if the fs is mounted
 341                  * with the "noexec" flag.
 342                  */
 343                 res = -EACCES;
 344                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 345                         goto out_path_release;
 346         }
 347
 348         res = inode_permission(inode, mode | MAY_ACCESS);
 349         /* SuS v2 requires we report a read only fs too */
 350         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 351                 goto out_path_release;
 352         /*
 353          * This is a rare case where using __mnt_is_readonly()
 354          * is OK without a mnt_want/drop_write() pair.  Since
 355          * no actual write to the fs is performed here, we do
 356          * not need to telegraph to that to anyone.
 357          *
 358          * By doing this, we accept that this access is
 359          * inherently racy and know that the fs may change
 360          * state before we even see this result.
 361          */
 362         if (__mnt_is_readonly(path.mnt))
 363                 res = -EROFS;
 364
 365 out_path_release:
 366         path_put(&path);
 367 out:
 368         revert_creds(old_cred);
 369         put_cred(override_cred);
 370         return res;
 371 }
 372
 373 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 374 {
 375         return sys_faccessat(AT_FDCWD, filename, mode);
 376 }
 377
 378 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 379 {
 380         struct path path;
 381         int error;
 382
 383         error = user_path_dir(filename, &path);
 384         if (error)
 385                 goto out;
 386
 387         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 388         if (error)
 389                 goto dput_and_out;
 390
 391         set_fs_pwd(current->fs, &path);
 392
 393 dput_and_out:
 394         path_put(&path);
 395 out:
 396         return error;
 397 }
 398
 399 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 400 {
 401         struct fd f = fdget_raw(fd);
 402         struct inode *inode;
 403         int error = -EBADF;
 404
 405         error = -EBADF;
 406         if (!f.file)
 407                 goto out;
 408
 409         inode = f.file->f_path.dentry->d_inode;
 410
 411         error = -ENOTDIR;
 412         if (!S_ISDIR(inode->i_mode))
 413                 goto out_putf;
 414
 415         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 416         if (!error)
 417                 set_fs_pwd(current->fs, &f.file->f_path);
 418 out_putf:
 419         fdput(f);
 420 out:
 421         return error;
 422 }
 423
 424 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 425 {
 426         struct path path;
 427         int error;
 428
 429         error = user_path_dir(filename, &path);
 430         if (error)
 431                 goto out;
 432
 433         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 434         if (error)
 435                 goto dput_and_out;
 436
 437         error = -EPERM;
 438         if (!capable(CAP_SYS_CHROOT))
 439                 goto dput_and_out;
 440         error = security_path_chroot(&path);
 441         if (error)
 442                 goto dput_and_out;
 443
 444         set_fs_root(current->fs, &path);
 445         error = 0;
 446 dput_and_out:
 447         path_put(&path);
 448 out:
 449         return error;
 450 }
 451
 452 static int chmod_common(struct path *path, umode_t mode)
 453 {
 454         struct inode *inode = path->dentry->d_inode;
 455         struct iattr newattrs;
 456         int error;
 457
 458         error = mnt_want_write(path->mnt);
 459         if (error)
 460                 return error;
 461         mutex_lock(&inode->i_mutex);
 462         error = security_path_chmod(path, mode);
 463         if (error)
 464                 goto out_unlock;
 465         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 466         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 467         error = notify_change(path->dentry, &newattrs);
 468 out_unlock:
 469         mutex_unlock(&inode->i_mutex);
 470         mnt_drop_write(path->mnt);
 471         return error;
 472 }
 473
 474 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 475 {
 476         struct file * file;
 477         int err = -EBADF;
 478
 479         file = fget(fd);
 480         if (file) {
 481                 audit_inode(NULL, file->f_path.dentry);
 482                 err = chmod_common(&file->f_path, mode);
 483                 fput(file);
 484         }
 485         return err;
 486 }
 487
 488 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 489 {
 490         struct path path;
 491         int error;
 492
 493         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 494         if (!error) {
 495                 error = chmod_common(&path, mode);
 496                 path_put(&path);
 497         }
 498         return error;
 499 }
 500
 501 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 502 {
 503         return sys_fchmodat(AT_FDCWD, filename, mode);
 504 }
 505
 506 static int chown_common(struct path *path, uid_t user, gid_t group)
 507 {
 508         struct inode *inode = path->dentry->d_inode;
 509         int error;
 510         struct iattr newattrs;
 511         kuid_t uid;
 512         kgid_t gid;
 513
 514         uid = make_kuid(current_user_ns(), user);
 515         gid = make_kgid(current_user_ns(), group);
 516
 517         newattrs.ia_valid =  ATTR_CTIME;
 518         if (user != (uid_t) -1) {
 519                 if (!uid_valid(uid))
 520                         return -EINVAL;
 521                 newattrs.ia_valid |= ATTR_UID;
 522                 newattrs.ia_uid = uid;
 523         }
 524         if (group != (gid_t) -1) {
 525                 if (!gid_valid(gid))
 526                         return -EINVAL;
 527                 newattrs.ia_valid |= ATTR_GID;
 528                 newattrs.ia_gid = gid;
 529         }
 530         if (!S_ISDIR(inode->i_mode))
 531                 newattrs.ia_valid |=
 532                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 533         mutex_lock(&inode->i_mutex);
 534         error = security_path_chown(path, uid, gid);
 535         if (!error)
 536                 error = notify_change(path->dentry, &newattrs);
 537         mutex_unlock(&inode->i_mutex);
 538
 539         return error;
 540 }
 541
 542 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 543                 gid_t, group, int, flag)
 544 {
 545         struct path path;
 546         int error = -EINVAL;
 547         int lookup_flags;
 548
 549         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 550                 goto out;
 551
 552         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 553         if (flag & AT_EMPTY_PATH)
 554                 lookup_flags |= LOOKUP_EMPTY;
 555         error = user_path_at(dfd, filename, lookup_flags, &path);
 556         if (error)
 557                 goto out;
 558         error = mnt_want_write(path.mnt);
 559         if (error)
 560                 goto out_release;
 561         error = chown_common(&path, user, group);
 562         mnt_drop_write(path.mnt);
 563 out_release:
 564         path_put(&path);
 565 out:
 566         return error;
 567 }
 568
 569 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 570 {
 571         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 572 }
 573
 574 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 575 {
 576         return sys_fchownat(AT_FDCWD, filename, user, group,
 577                             AT_SYMLINK_NOFOLLOW);
 578 }
 579
 580 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 581 {
 582         struct fd f = fdget(fd);
 583         int error = -EBADF;
 584
 585         if (!f.file)
 586                 goto out;
 587
 588         error = mnt_want_write_file(f.file);
 589         if (error)
 590                 goto out_fput;
 591         audit_inode(NULL, f.file->f_path.dentry);
 592         error = chown_common(&f.file->f_path, user, group);
 593         mnt_drop_write_file(f.file);
 594 out_fput:
 595         fdput(f);
 596 out:
 597         return error;
 598 }
 599
 600 /*
 601  * You have to be very careful that these write
 602  * counts get cleaned up in error cases and
 603  * upon __fput().  This should probably never
 604  * be called outside of __dentry_open().
 605  */
 606 static inline int __get_file_write_access(struct inode *inode,
 607                                           struct vfsmount *mnt)
 608 {
 609         int error;
 610         error = get_write_access(inode);
 611         if (error)
 612                 return error;
 613         /*
 614          * Do not take mount writer counts on
 615          * special files since no writes to
 616          * the mount itself will occur.
 617          */
 618         if (!special_file(inode->i_mode)) {
 619                 /*
 620                  * Balanced in __fput()
 621                  */
 622                 error = __mnt_want_write(mnt);
 623                 if (error)
 624                         put_write_access(inode);
 625         }
 626         return error;
 627 }
 628
 629 int open_check_o_direct(struct file *f)
 630 {
 631         /* NB: we're sure to have correct a_ops only after f_op->open */
 632         if (f->f_flags & O_DIRECT) {
 633                 if (!f->f_mapping->a_ops ||
 634                     ((!f->f_mapping->a_ops->direct_IO) &&
 635                     (!f->f_mapping->a_ops->get_xip_mem))) {
 636                         return -EINVAL;
 637                 }
 638         }
 639         return 0;
 640 }
 641
 642 static int do_dentry_open(struct file *f,
 643                           int (*open)(struct inode *, struct file *),
 644                           const struct cred *cred)
 645 {
 646         static const struct file_operations empty_fops = {};
 647         struct inode *inode;
 648         int error;
 649
 650         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 651                                 FMODE_PREAD | FMODE_PWRITE;
 652
 653         if (unlikely(f->f_flags & O_PATH))
 654                 f->f_mode = FMODE_PATH;
 655
 656         path_get(&f->f_path);
 657         inode = f->f_path.dentry->d_inode;
 658         if (f->f_mode & FMODE_WRITE) {
 659                 error = __get_file_write_access(inode, f->f_path.mnt);
 660                 if (error)
 661                         goto cleanup_file;
 662                 if (!special_file(inode->i_mode))
 663                         file_take_write(f);
 664         }
 665
 666         f->f_mapping = inode->i_mapping;
 667         f->f_pos = 0;
 668         file_sb_list_add(f, inode->i_sb);
 669
 670         if (unlikely(f->f_mode & FMODE_PATH)) {
 671                 f->f_op = &empty_fops;
 672                 return 0;
 673         }
 674
 675         f->f_op = fops_get(inode->i_fop);
 676
 677         error = security_file_open(f, cred);
 678         if (error)
 679                 goto cleanup_all;
 680
 681         error = break_lease(inode, f->f_flags);
 682         if (error)
 683                 goto cleanup_all;
 684
 685         if (!open && f->f_op)
 686                 open = f->f_op->open;
 687         if (open) {
 688                 error = open(inode, f);
 689                 if (error)
 690                         goto cleanup_all;
 691         }
 692         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 693                 i_readcount_inc(inode);
 694
 695         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 696
 697         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 698
 699         return 0;
 700
 701 cleanup_all:
 702         fops_put(f->f_op);
 703         file_sb_list_del(f);
 704         if (f->f_mode & FMODE_WRITE) {
 705                 put_write_access(inode);
 706                 if (!special_file(inode->i_mode)) {
 707                         /*
 708                          * We don't consider this a real
 709                          * mnt_want/drop_write() pair
 710                          * because it all happenend right
 711                          * here, so just reset the state.
 712                          */
 713                         file_reset_write(f);
 714                         __mnt_drop_write(f->f_path.mnt);
 715                 }
 716         }
 717 cleanup_file:
 718         path_put(&f->f_path);
 719         f->f_path.mnt = NULL;
 720         f->f_path.dentry = NULL;
 721         return error;
 722 }
 723
 724 /**
 725  * finish_open - finish opening a file
 726  * @od: opaque open data
 727  * @dentry: pointer to dentry
 728  * @open: open callback
 729  *
 730  * This can be used to finish opening a file passed to i_op->atomic_open().
 731  *
 732  * If the open callback is set to NULL, then the standard f_op->open()
 733  * filesystem callback is substituted.
 734  */
 735 int finish_open(struct file *file, struct dentry *dentry,
 736                 int (*open)(struct inode *, struct file *),
 737                 int *opened)
 738 {
 739         int error;
 740         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 741
 742         file->f_path.dentry = dentry;
 743         error = do_dentry_open(file, open, current_cred());
 744         if (!error)
 745                 *opened |= FILE_OPENED;
 746
 747         return error;
 748 }
 749 EXPORT_SYMBOL(finish_open);
 750
 751 /**
 752  * finish_no_open - finish ->atomic_open() without opening the file
 753  *
 754  * @od: opaque open data
 755  * @dentry: dentry or NULL (as returned from ->lookup())
 756  *
 757  * This can be used to set the result of a successful lookup in ->atomic_open().
 758  * The filesystem's atomic_open() method shall return NULL after calling this.
 759  */
 760 int finish_no_open(struct file *file, struct dentry *dentry)
 761 {
 762         file->f_path.dentry = dentry;
 763         return 1;
 764 }
 765 EXPORT_SYMBOL(finish_no_open);
 766
 767 struct file *dentry_open(const struct path *path, int flags,
 768                          const struct cred *cred)
 769 {
 770         int error;
 771         struct file *f;
 772
 773         validate_creds(cred);
 774
 775         /* We must always pass in a valid mount pointer. */
 776         BUG_ON(!path->mnt);
 777
 778         error = -ENFILE;
 779         f = get_empty_filp();
 780         if (f == NULL)
 781                 return ERR_PTR(error);
 782
 783         f->f_flags = flags;
 784         f->f_path = *path;
 785         error = do_dentry_open(f, NULL, cred);
 786         if (!error) {
 787                 error = open_check_o_direct(f);
 788                 if (error) {
 789                         fput(f);
 790                         f = ERR_PTR(error);
 791                 }
 792         } else {
 793                 put_filp(f);
 794                 f = ERR_PTR(error);
 795         }
 796         return f;
 797 }
 798 EXPORT_SYMBOL(dentry_open);
 799
 800 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 801 {
 802         int lookup_flags = 0;
 803         int acc_mode;
 804
 805         if (flags & O_CREAT)
 806                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 807         else
 808                 op->mode = 0;
 809
 810         /* Must never be set by userspace */
 811         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 812
 813         /*
 814          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 815          * check for O_DSYNC if the need any syncing at all we enforce it's
 816          * always set instead of having to deal with possibly weird behaviour
 817          * for malicious applications setting only __O_SYNC.
 818          */
 819         if (flags & __O_SYNC)
 820                 flags |= O_DSYNC;
 821
 822         /*
 823          * If we have O_PATH in the open flag. Then we
 824          * cannot have anything other than the below set of flags
 825          */
 826         if (flags & O_PATH) {
 827                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 828                 acc_mode = 0;
 829         } else {
 830                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 831         }
 832
 833         op->open_flag = flags;
 834
 835         /* O_TRUNC implies we need access checks for write permissions */
 836         if (flags & O_TRUNC)
 837                 acc_mode |= MAY_WRITE;
 838
 839         /* Allow the LSM permission hook to distinguish append
 840            access from general write access. */
 841         if (flags & O_APPEND)
 842                 acc_mode |= MAY_APPEND;
 843
 844         op->acc_mode = acc_mode;
 845
 846         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 847
 848         if (flags & O_CREAT) {
 849                 op->intent |= LOOKUP_CREATE;
 850                 if (flags & O_EXCL)
 851                         op->intent |= LOOKUP_EXCL;
 852         }
 853
 854         if (flags & O_DIRECTORY)
 855                 lookup_flags |= LOOKUP_DIRECTORY;
 856         if (!(flags & O_NOFOLLOW))
 857                 lookup_flags |= LOOKUP_FOLLOW;
 858         return lookup_flags;
 859 }
 860
 861 /**
 862  * filp_open - open file and return file pointer
 863  *
 864  * @filename:   path to open
 865  * @flags:      open flags as per the open(2) second argument
 866  * @mode:       mode for the new file if O_CREAT is set, else ignored
 867  *
 868  * This is the helper to open a file from kernelspace if you really
 869  * have to.  But in generally you should not do this, so please move
 870  * along, nothing to see here..
 871  */
 872 struct file *filp_open(const char *filename, int flags, umode_t mode)
 873 {
 874         struct open_flags op;
 875         int lookup = build_open_flags(flags, mode, &op);
 876         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 877 }
 878 EXPORT_SYMBOL(filp_open);
 879
 880 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 881                             const char *filename, int flags)
 882 {
 883         struct open_flags op;
 884         int lookup = build_open_flags(flags, 0, &op);
 885         if (flags & O_CREAT)
 886                 return ERR_PTR(-EINVAL);
 887         if (!filename && (flags & O_DIRECTORY))
 888                 if (!dentry->d_inode->i_op->lookup)
 889                         return ERR_PTR(-ENOTDIR);
 890         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 891 }
 892 EXPORT_SYMBOL(file_open_root);
 893
 894 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 895 {
 896         struct open_flags op;
 897         int lookup = build_open_flags(flags, mode, &op);
 898         char *tmp = getname(filename);
 899         int fd = PTR_ERR(tmp);
 900
 901         if (!IS_ERR(tmp)) {
 902                 fd = get_unused_fd_flags(flags);
 903                 if (fd >= 0) {
 904                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 905                         if (IS_ERR(f)) {
 906                                 put_unused_fd(fd);
 907                                 fd = PTR_ERR(f);
 908                         } else {
 909                                 fsnotify_open(f);
 910                                 fd_install(fd, f);
 911                         }
 912                 }
 913                 putname(tmp);
 914         }
 915         return fd;
 916 }
 917
 918 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 919 {
 920         long ret;
 921
 922         if (force_o_largefile())
 923                 flags |= O_LARGEFILE;
 924
 925         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 926         /* avoid REGPARM breakage on x86: */
 927         asmlinkage_protect(3, ret, filename, flags, mode);
 928         return ret;
 929 }
 930
 931 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 932                 umode_t, mode)
 933 {
 934         long ret;
 935
 936         if (force_o_largefile())
 937                 flags |= O_LARGEFILE;
 938
 939         ret = do_sys_open(dfd, filename, flags, mode);
 940         /* avoid REGPARM breakage on x86: */
 941         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 942         return ret;
 943 }
 944
 945 #ifndef __alpha__
 946
 947 /*
 948  * For backward compatibility?  Maybe this should be moved
 949  * into arch/i386 instead?
 950  */
 951 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
 952 {
 953         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 954 }
 955
 956 #endif
 957
 958 /*
 959  * "id" is the POSIX thread ID. We use the
 960  * files pointer for this..
 961  */
 962 int filp_close(struct file *filp, fl_owner_t id)
 963 {
 964         int retval = 0;
 965
 966         if (!file_count(filp)) {
 967                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 968                 return 0;
 969         }
 970
 971         if (filp->f_op && filp->f_op->flush)
 972                 retval = filp->f_op->flush(filp, id);
 973
 974         if (likely(!(filp->f_mode & FMODE_PATH))) {
 975                 dnotify_flush(filp, id);
 976                 locks_remove_posix(filp, id);
 977         }
 978         fput(filp);
 979         return retval;
 980 }
 981
 982 EXPORT_SYMBOL(filp_close);
 983
 984 /*
 985  * Careful here! We test whether the file pointer is NULL before
 986  * releasing the fd. This ensures that one clone task can't release
 987  * an fd while another clone is opening it.
 988  */
 989 SYSCALL_DEFINE1(close, unsigned int, fd)
 990 {
 991         int retval = __close_fd(current->files, fd);
 992
 993         /* can't restart close syscall because file table entry was cleared */
 994         if (unlikely(retval == -ERESTARTSYS ||
 995                      retval == -ERESTARTNOINTR ||
 996                      retval == -ERESTARTNOHAND ||
 997                      retval == -ERESTART_RESTARTBLOCK))
 998                 retval = -EINTR;
 999
1000         return retval;
1001 }
1002 EXPORT_SYMBOL(sys_close);
1003
1004 /*
1005  * This routine simulates a hangup on the tty, to arrange that users
1006  * are given clean terminals at login time.
1007  */
1008 SYSCALL_DEFINE0(vhangup)
1009 {
1010         if (capable(CAP_SYS_TTY_CONFIG)) {
1011                 tty_vhangup_self();
1012                 return 0;
1013         }
1014         return -EPERM;
1015 }
1016
1017 /*
1018  * Called when an inode is about to be open.
1019  * We use this to disallow opening large files on 32bit systems if
1020  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1021  * on this flag in sys_open.
1022  */
1023 int generic_file_open(struct inode * inode, struct file * filp)
1024 {
1025         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1026                 return -EOVERFLOW;
1027         return 0;
1028 }
1029
1030 EXPORT_SYMBOL(generic_file_open);
1031
1032 /*
1033  * This is used by subsystems that don't want seekable
1034  * file descriptors. The function is not supposed to ever fail, the only
1035  * reason it returns an 'int' and not 'void' is so that it can be plugged
1036  * directly into file_operations structure.
1037  */
1038 int nonseekable_open(struct inode *inode, struct file *filp)
1039 {
1040         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1041         return 0;
1042 }
1043
1044 EXPORT_SYMBOL(nonseekable_open);