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 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid/sgid on truncate too */
  55         ret = should_remove_suid(dentry);
  56         if (ret)
  57                 newattrs.ia_valid |= ret | ATTR_FORCE;
  58
  59         mutex_lock(&dentry->d_inode->i_mutex);
  60         ret = notify_change(dentry, &newattrs);
  61         mutex_unlock(&dentry->d_inode->i_mutex);
  62         return ret;
  63 }
  64
  65 long vfs_truncate(struct path *path, loff_t length)
  66 {
  67         struct inode *inode;
  68         long error;
  69
  70         inode = path->dentry->d_inode;
  71
  72         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  73         if (S_ISDIR(inode->i_mode))
  74                 return -EISDIR;
  75         if (!S_ISREG(inode->i_mode))
  76                 return -EINVAL;
  77
  78         error = mnt_want_write(path->mnt);
  79         if (error)
  80                 goto out;
  81
  82         error = inode_permission(inode, MAY_WRITE);
  83         if (error)
  84                 goto mnt_drop_write_and_out;
  85
  86         error = -EPERM;
  87         if (IS_APPEND(inode))
  88                 goto mnt_drop_write_and_out;
  89
  90         error = get_write_access(inode);
  91         if (error)
  92                 goto mnt_drop_write_and_out;
  93
  94         /*
  95          * Make sure that there are no leases.  get_write_access() protects
  96          * against the truncate racing with a lease-granting setlease().
  97          */
  98         error = break_lease(inode, O_WRONLY);
  99         if (error)
 100                 goto put_write_and_out;
 101
 102         error = locks_verify_truncate(inode, NULL, length);
 103         if (!error)
 104                 error = security_path_truncate(path);
 105         if (!error)
 106                 error = do_truncate(path->dentry, length, 0, NULL);
 107
 108 put_write_and_out:
 109         put_write_access(inode);
 110 mnt_drop_write_and_out:
 111         mnt_drop_write(path->mnt);
 112 out:
 113         return error;
 114 }
 115 EXPORT_SYMBOL_GPL(vfs_truncate);
 116
 117 static long do_sys_truncate(const char __user *pathname, loff_t length)
 118 {
 119         unsigned int lookup_flags = LOOKUP_FOLLOW;
 120         struct path path;
 121         int error;
 122
 123         if (length < 0) /* sorry, but loff_t says... */
 124                 return -EINVAL;
 125
 126 retry:
 127         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 128         if (!error) {
 129                 error = vfs_truncate(&path, length);
 130                 path_put(&path);
 131         }
 132         if (retry_estale(error, lookup_flags)) {
 133                 lookup_flags |= LOOKUP_REVAL;
 134                 goto retry;
 135         }
 136         return error;
 137 }
 138
 139 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 140 {
 141         return do_sys_truncate(path, length);
 142 }
 143
 144 #ifdef CONFIG_COMPAT
 145 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 146 {
 147         return do_sys_truncate(path, length);
 148 }
 149 #endif
 150
 151 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 152 {
 153         struct inode *inode;
 154         struct dentry *dentry;
 155         struct fd f;
 156         int error;
 157
 158         error = -EINVAL;
 159         if (length < 0)
 160                 goto out;
 161         error = -EBADF;
 162         f = fdget(fd);
 163         if (!f.file)
 164                 goto out;
 165
 166         /* explicitly opened as large or we are on 64-bit box */
 167         if (f.file->f_flags & O_LARGEFILE)
 168                 small = 0;
 169
 170         dentry = f.file->f_path.dentry;
 171         inode = dentry->d_inode;
 172         error = -EINVAL;
 173         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 174                 goto out_putf;
 175
 176         error = -EINVAL;
 177         /* Cannot ftruncate over 2^31 bytes without large file support */
 178         if (small && length > MAX_NON_LFS)
 179                 goto out_putf;
 180
 181         error = -EPERM;
 182         if (IS_APPEND(inode))
 183                 goto out_putf;
 184
 185         sb_start_write(inode->i_sb);
 186         error = locks_verify_truncate(inode, f.file, length);
 187         if (!error)
 188                 error = security_path_truncate(&f.file->f_path);
 189         if (!error)
 190                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 191         sb_end_write(inode->i_sb);
 192 out_putf:
 193         fdput(f);
 194 out:
 195         return error;
 196 }
 197
 198 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 199 {
 200         return do_sys_ftruncate(fd, length, 1);
 201 }
 202
 203 #ifdef CONFIG_COMPAT
 204 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 205 {
 206         return do_sys_ftruncate(fd, length, 1);
 207 }
 208 #endif
 209
 210 /* LFS versions of truncate are only needed on 32 bit machines */
 211 #if BITS_PER_LONG == 32
 212 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 213 {
 214         return do_sys_truncate(path, length);
 215 }
 216
 217 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 218 {
 219         return do_sys_ftruncate(fd, length, 0);
 220 }
 221 #endif /* BITS_PER_LONG == 32 */
 222
 223
 224 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 225 {
 226         struct inode *inode = file_inode(file);
 227         long ret;
 228
 229         if (offset < 0 || len <= 0)
 230                 return -EINVAL;
 231
 232         /* Return error if mode is not supported */
 233         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 234                 return -EOPNOTSUPP;
 235
 236         /* Punch hole must have keep size set */
 237         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 238             !(mode & FALLOC_FL_KEEP_SIZE))
 239                 return -EOPNOTSUPP;
 240
 241         if (!(file->f_mode & FMODE_WRITE))
 242                 return -EBADF;
 243
 244         /* It's not possible punch hole on append only file */
 245         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 246                 return -EPERM;
 247
 248         if (IS_IMMUTABLE(inode))
 249                 return -EPERM;
 250
 251         /*
 252          * Revalidate the write permissions, in case security policy has
 253          * changed since the files were opened.
 254          */
 255         ret = security_file_permission(file, MAY_WRITE);
 256         if (ret)
 257                 return ret;
 258
 259         if (S_ISFIFO(inode->i_mode))
 260                 return -ESPIPE;
 261
 262         /*
 263          * Let individual file system decide if it supports preallocation
 264          * for directories or not.
 265          */
 266         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 267                 return -ENODEV;
 268
 269         /* Check for wrap through zero too */
 270         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 271                 return -EFBIG;
 272
 273         if (!file->f_op->fallocate)
 274                 return -EOPNOTSUPP;
 275
 276         sb_start_write(inode->i_sb);
 277         ret = file->f_op->fallocate(file, mode, offset, len);
 278         sb_end_write(inode->i_sb);
 279         return ret;
 280 }
 281
 282 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 283 {
 284         struct fd f = fdget(fd);
 285         int error = -EBADF;
 286
 287         if (f.file) {
 288                 error = do_fallocate(f.file, mode, offset, len);
 289                 fdput(f);
 290         }
 291         return error;
 292 }
 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  */
 299 SYSCALL_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         unsigned int lookup_flags = LOOKUP_FOLLOW;
 307
 308         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 309                 return -EINVAL;
 310
 311         override_cred = prepare_creds();
 312         if (!override_cred)
 313                 return -ENOMEM;
 314
 315         override_cred->fsuid = override_cred->uid;
 316         override_cred->fsgid = override_cred->gid;
 317
 318         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 319                 /* Clear the capabilities if we switch to a non-root user */
 320                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 321                 if (!uid_eq(override_cred->uid, root_uid))
 322                         cap_clear(override_cred->cap_effective);
 323                 else
 324                         override_cred->cap_effective =
 325                                 override_cred->cap_permitted;
 326         }
 327
 328         old_cred = override_creds(override_cred);
 329 retry:
 330         res = user_path_at(dfd, filename, lookup_flags, &path);
 331         if (res)
 332                 goto out;
 333
 334         inode = path.dentry->d_inode;
 335
 336         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 337                 /*
 338                  * MAY_EXEC on regular files is denied if the fs is mounted
 339                  * with the "noexec" flag.
 340                  */
 341                 res = -EACCES;
 342                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 343                         goto out_path_release;
 344         }
 345
 346         res = inode_permission(inode, mode | MAY_ACCESS);
 347         /* SuS v2 requires we report a read only fs too */
 348         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 349                 goto out_path_release;
 350         /*
 351          * This is a rare case where using __mnt_is_readonly()
 352          * is OK without a mnt_want/drop_write() pair.  Since
 353          * no actual write to the fs is performed here, we do
 354          * not need to telegraph to that to anyone.
 355          *
 356          * By doing this, we accept that this access is
 357          * inherently racy and know that the fs may change
 358          * state before we even see this result.
 359          */
 360         if (__mnt_is_readonly(path.mnt))
 361                 res = -EROFS;
 362
 363 out_path_release:
 364         path_put(&path);
 365         if (retry_estale(res, lookup_flags)) {
 366                 lookup_flags |= LOOKUP_REVAL;
 367                 goto retry;
 368         }
 369 out:
 370         revert_creds(old_cred);
 371         put_cred(override_cred);
 372         return res;
 373 }
 374
 375 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 376 {
 377         return sys_faccessat(AT_FDCWD, filename, mode);
 378 }
 379
 380 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 381 {
 382         struct path path;
 383         int error;
 384         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 385 retry:
 386         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 387         if (error)
 388                 goto out;
 389
 390         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 391         if (error)
 392                 goto dput_and_out;
 393
 394         set_fs_pwd(current->fs, &path);
 395
 396 dput_and_out:
 397         path_put(&path);
 398         if (retry_estale(error, lookup_flags)) {
 399                 lookup_flags |= LOOKUP_REVAL;
 400                 goto retry;
 401         }
 402 out:
 403         return error;
 404 }
 405
 406 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 407 {
 408         struct fd f = fdget_raw(fd);
 409         struct inode *inode;
 410         int error = -EBADF;
 411
 412         error = -EBADF;
 413         if (!f.file)
 414                 goto out;
 415
 416         inode = file_inode(f.file);
 417
 418         error = -ENOTDIR;
 419         if (!S_ISDIR(inode->i_mode))
 420                 goto out_putf;
 421
 422         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 423         if (!error)
 424                 set_fs_pwd(current->fs, &f.file->f_path);
 425 out_putf:
 426         fdput(f);
 427 out:
 428         return error;
 429 }
 430
 431 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 432 {
 433         struct path path;
 434         int error;
 435         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 436 retry:
 437         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 438         if (error)
 439                 goto out;
 440
 441         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 442         if (error)
 443                 goto dput_and_out;
 444
 445         error = -EPERM;
 446         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 447                 goto dput_and_out;
 448         error = security_path_chroot(&path);
 449         if (error)
 450                 goto dput_and_out;
 451
 452         set_fs_root(current->fs, &path);
 453         error = 0;
 454 dput_and_out:
 455         path_put(&path);
 456         if (retry_estale(error, lookup_flags)) {
 457                 lookup_flags |= LOOKUP_REVAL;
 458                 goto retry;
 459         }
 460 out:
 461         return error;
 462 }
 463
 464 static int chmod_common(struct path *path, umode_t mode)
 465 {
 466         struct inode *inode = path->dentry->d_inode;
 467         struct iattr newattrs;
 468         int error;
 469
 470         error = mnt_want_write(path->mnt);
 471         if (error)
 472                 return error;
 473         mutex_lock(&inode->i_mutex);
 474         error = security_path_chmod(path, mode);
 475         if (error)
 476                 goto out_unlock;
 477         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 478         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 479         error = notify_change(path->dentry, &newattrs);
 480 out_unlock:
 481         mutex_unlock(&inode->i_mutex);
 482         mnt_drop_write(path->mnt);
 483         return error;
 484 }
 485
 486 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 487 {
 488         struct fd f = fdget(fd);
 489         int err = -EBADF;
 490
 491         if (f.file) {
 492                 audit_inode(NULL, f.file->f_path.dentry, 0);
 493                 err = chmod_common(&f.file->f_path, mode);
 494                 fdput(f);
 495         }
 496         return err;
 497 }
 498
 499 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 500 {
 501         struct path path;
 502         int error;
 503         unsigned int lookup_flags = LOOKUP_FOLLOW;
 504 retry:
 505         error = user_path_at(dfd, filename, lookup_flags, &path);
 506         if (!error) {
 507                 error = chmod_common(&path, mode);
 508                 path_put(&path);
 509                 if (retry_estale(error, lookup_flags)) {
 510                         lookup_flags |= LOOKUP_REVAL;
 511                         goto retry;
 512                 }
 513         }
 514         return error;
 515 }
 516
 517 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 518 {
 519         return sys_fchmodat(AT_FDCWD, filename, mode);
 520 }
 521
 522 static int chown_common(struct path *path, uid_t user, gid_t group)
 523 {
 524         struct inode *inode = path->dentry->d_inode;
 525         int error;
 526         struct iattr newattrs;
 527         kuid_t uid;
 528         kgid_t gid;
 529
 530         uid = make_kuid(current_user_ns(), user);
 531         gid = make_kgid(current_user_ns(), group);
 532
 533         newattrs.ia_valid =  ATTR_CTIME;
 534         if (user != (uid_t) -1) {
 535                 if (!uid_valid(uid))
 536                         return -EINVAL;
 537                 newattrs.ia_valid |= ATTR_UID;
 538                 newattrs.ia_uid = uid;
 539         }
 540         if (group != (gid_t) -1) {
 541                 if (!gid_valid(gid))
 542                         return -EINVAL;
 543                 newattrs.ia_valid |= ATTR_GID;
 544                 newattrs.ia_gid = gid;
 545         }
 546         if (!S_ISDIR(inode->i_mode))
 547                 newattrs.ia_valid |=
 548                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 549         mutex_lock(&inode->i_mutex);
 550         error = security_path_chown(path, uid, gid);
 551         if (!error)
 552                 error = notify_change(path->dentry, &newattrs);
 553         mutex_unlock(&inode->i_mutex);
 554
 555         return error;
 556 }
 557
 558 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 559                 gid_t, group, int, flag)
 560 {
 561         struct path path;
 562         int error = -EINVAL;
 563         int lookup_flags;
 564
 565         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 566                 goto out;
 567
 568         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 569         if (flag & AT_EMPTY_PATH)
 570                 lookup_flags |= LOOKUP_EMPTY;
 571 retry:
 572         error = user_path_at(dfd, filename, lookup_flags, &path);
 573         if (error)
 574                 goto out;
 575         error = mnt_want_write(path.mnt);
 576         if (error)
 577                 goto out_release;
 578         error = chown_common(&path, user, group);
 579         mnt_drop_write(path.mnt);
 580 out_release:
 581         path_put(&path);
 582         if (retry_estale(error, lookup_flags)) {
 583                 lookup_flags |= LOOKUP_REVAL;
 584                 goto retry;
 585         }
 586 out:
 587         return error;
 588 }
 589
 590 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 591 {
 592         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 593 }
 594
 595 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 596 {
 597         return sys_fchownat(AT_FDCWD, filename, user, group,
 598                             AT_SYMLINK_NOFOLLOW);
 599 }
 600
 601 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 602 {
 603         struct fd f = fdget(fd);
 604         int error = -EBADF;
 605
 606         if (!f.file)
 607                 goto out;
 608
 609         error = mnt_want_write_file(f.file);
 610         if (error)
 611                 goto out_fput;
 612         audit_inode(NULL, f.file->f_path.dentry, 0);
 613         error = chown_common(&f.file->f_path, user, group);
 614         mnt_drop_write_file(f.file);
 615 out_fput:
 616         fdput(f);
 617 out:
 618         return error;
 619 }
 620
 621 /*
 622  * You have to be very careful that these write
 623  * counts get cleaned up in error cases and
 624  * upon __fput().  This should probably never
 625  * be called outside of __dentry_open().
 626  */
 627 static inline int __get_file_write_access(struct inode *inode,
 628                                           struct vfsmount *mnt)
 629 {
 630         int error;
 631         error = get_write_access(inode);
 632         if (error)
 633                 return error;
 634         /*
 635          * Do not take mount writer counts on
 636          * special files since no writes to
 637          * the mount itself will occur.
 638          */
 639         if (!special_file(inode->i_mode)) {
 640                 /*
 641                  * Balanced in __fput()
 642                  */
 643                 error = __mnt_want_write(mnt);
 644                 if (error)
 645                         put_write_access(inode);
 646         }
 647         return error;
 648 }
 649
 650 int open_check_o_direct(struct file *f)
 651 {
 652         /* NB: we're sure to have correct a_ops only after f_op->open */
 653         if (f->f_flags & O_DIRECT) {
 654                 if (!f->f_mapping->a_ops ||
 655                     ((!f->f_mapping->a_ops->direct_IO) &&
 656                     (!f->f_mapping->a_ops->get_xip_mem))) {
 657                         return -EINVAL;
 658                 }
 659         }
 660         return 0;
 661 }
 662
 663 static int do_dentry_open(struct file *f,
 664                           int (*open)(struct inode *, struct file *),
 665                           const struct cred *cred)
 666 {
 667         static const struct file_operations empty_fops = {};
 668         struct inode *inode;
 669         int error;
 670
 671         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 672                                 FMODE_PREAD | FMODE_PWRITE;
 673
 674         if (unlikely(f->f_flags & O_PATH))
 675                 f->f_mode = FMODE_PATH;
 676
 677         path_get(&f->f_path);
 678         inode = f->f_inode = f->f_path.dentry->d_inode;
 679         if (f->f_mode & FMODE_WRITE) {
 680                 error = __get_file_write_access(inode, f->f_path.mnt);
 681                 if (error)
 682                         goto cleanup_file;
 683                 if (!special_file(inode->i_mode))
 684                         file_take_write(f);
 685         }
 686
 687         f->f_mapping = inode->i_mapping;
 688         file_sb_list_add(f, inode->i_sb);
 689
 690         if (unlikely(f->f_mode & FMODE_PATH)) {
 691                 f->f_op = &empty_fops;
 692                 return 0;
 693         }
 694
 695         f->f_op = fops_get(inode->i_fop);
 696         if (unlikely(WARN_ON(!f->f_op))) {
 697                 error = -ENODEV;
 698                 goto cleanup_all;
 699         }
 700
 701         error = security_file_open(f, cred);
 702         if (error)
 703                 goto cleanup_all;
 704
 705         error = break_lease(inode, f->f_flags);
 706         if (error)
 707                 goto cleanup_all;
 708
 709         if (!open)
 710                 open = f->f_op->open;
 711         if (open) {
 712                 error = open(inode, f);
 713                 if (error)
 714                         goto cleanup_all;
 715         }
 716         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 717                 i_readcount_inc(inode);
 718
 719         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 720
 721         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 722
 723         return 0;
 724
 725 cleanup_all:
 726         fops_put(f->f_op);
 727         file_sb_list_del(f);
 728         if (f->f_mode & FMODE_WRITE) {
 729                 put_write_access(inode);
 730                 if (!special_file(inode->i_mode)) {
 731                         /*
 732                          * We don't consider this a real
 733                          * mnt_want/drop_write() pair
 734                          * because it all happenend right
 735                          * here, so just reset the state.
 736                          */
 737                         file_reset_write(f);
 738                         __mnt_drop_write(f->f_path.mnt);
 739                 }
 740         }
 741 cleanup_file:
 742         path_put(&f->f_path);
 743         f->f_path.mnt = NULL;
 744         f->f_path.dentry = NULL;
 745         f->f_inode = NULL;
 746         return error;
 747 }
 748
 749 /**
 750  * finish_open - finish opening a file
 751  * @file: file pointer
 752  * @dentry: pointer to dentry
 753  * @open: open callback
 754  * @opened: state of open
 755  *
 756  * This can be used to finish opening a file passed to i_op->atomic_open().
 757  *
 758  * If the open callback is set to NULL, then the standard f_op->open()
 759  * filesystem callback is substituted.
 760  *
 761  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 762  * the return value of d_splice_alias(), then the caller needs to perform dput()
 763  * on it after finish_open().
 764  *
 765  * On successful return @file is a fully instantiated open file.  After this, if
 766  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 767  *
 768  * Returns zero on success or -errno if the open failed.
 769  */
 770 int finish_open(struct file *file, struct dentry *dentry,
 771                 int (*open)(struct inode *, struct file *),
 772                 int *opened)
 773 {
 774         int error;
 775         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 776
 777         file->f_path.dentry = dentry;
 778         error = do_dentry_open(file, open, current_cred());
 779         if (!error)
 780                 *opened |= FILE_OPENED;
 781
 782         return error;
 783 }
 784 EXPORT_SYMBOL(finish_open);
 785
 786 /**
 787  * finish_no_open - finish ->atomic_open() without opening the file
 788  *
 789  * @file: file pointer
 790  * @dentry: dentry or NULL (as returned from ->lookup())
 791  *
 792  * This can be used to set the result of a successful lookup in ->atomic_open().
 793  *
 794  * NB: unlike finish_open() this function does consume the dentry reference and
 795  * the caller need not dput() it.
 796  *
 797  * Returns "1" which must be the return value of ->atomic_open() after having
 798  * called this function.
 799  */
 800 int finish_no_open(struct file *file, struct dentry *dentry)
 801 {
 802         file->f_path.dentry = dentry;
 803         return 1;
 804 }
 805 EXPORT_SYMBOL(finish_no_open);
 806
 807 struct file *dentry_open(const struct path *path, int flags,
 808                          const struct cred *cred)
 809 {
 810         int error;
 811         struct file *f;
 812
 813         validate_creds(cred);
 814
 815         /* We must always pass in a valid mount pointer. */
 816         BUG_ON(!path->mnt);
 817
 818         f = get_empty_filp();
 819         if (!IS_ERR(f)) {
 820                 f->f_flags = flags;
 821                 f->f_path = *path;
 822                 error = do_dentry_open(f, NULL, cred);
 823                 if (!error) {
 824                         /* from now on we need fput() to dispose of f */
 825                         error = open_check_o_direct(f);
 826                         if (error) {
 827                                 fput(f);
 828                                 f = ERR_PTR(error);
 829                         }
 830                 } else {
 831                         put_filp(f);
 832                         f = ERR_PTR(error);
 833                 }
 834         }
 835         return f;
 836 }
 837 EXPORT_SYMBOL(dentry_open);
 838
 839 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 840 {
 841         int lookup_flags = 0;
 842         int acc_mode;
 843
 844         if (flags & (O_CREAT | __O_TMPFILE))
 845                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 846         else
 847                 op->mode = 0;
 848
 849         /* Must never be set by userspace */
 850         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 851
 852         /*
 853          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 854          * check for O_DSYNC if the need any syncing at all we enforce it's
 855          * always set instead of having to deal with possibly weird behaviour
 856          * for malicious applications setting only __O_SYNC.
 857          */
 858         if (flags & __O_SYNC)
 859                 flags |= O_DSYNC;
 860
 861         if (flags & __O_TMPFILE) {
 862                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 863                         return -EINVAL;
 864                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 865                 if (!(acc_mode & MAY_WRITE))
 866                         return -EINVAL;
 867         } else if (flags & O_PATH) {
 868                 /*
 869                  * If we have O_PATH in the open flag. Then we
 870                  * cannot have anything other than the below set of flags
 871                  */
 872                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 873                 acc_mode = 0;
 874         } else {
 875                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 876         }
 877
 878         op->open_flag = flags;
 879
 880         /* O_TRUNC implies we need access checks for write permissions */
 881         if (flags & O_TRUNC)
 882                 acc_mode |= MAY_WRITE;
 883
 884         /* Allow the LSM permission hook to distinguish append
 885            access from general write access. */
 886         if (flags & O_APPEND)
 887                 acc_mode |= MAY_APPEND;
 888
 889         op->acc_mode = acc_mode;
 890
 891         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 892
 893         if (flags & O_CREAT) {
 894                 op->intent |= LOOKUP_CREATE;
 895                 if (flags & O_EXCL)
 896                         op->intent |= LOOKUP_EXCL;
 897         }
 898
 899         if (flags & O_DIRECTORY)
 900                 lookup_flags |= LOOKUP_DIRECTORY;
 901         if (!(flags & O_NOFOLLOW))
 902                 lookup_flags |= LOOKUP_FOLLOW;
 903         op->lookup_flags = lookup_flags;
 904         return 0;
 905 }
 906
 907 /**
 908  * file_open_name - open file and return file pointer
 909  *
 910  * @name:       struct filename containing path to open
 911  * @flags:      open flags as per the open(2) second argument
 912  * @mode:       mode for the new file if O_CREAT is set, else ignored
 913  *
 914  * This is the helper to open a file from kernelspace if you really
 915  * have to.  But in generally you should not do this, so please move
 916  * along, nothing to see here..
 917  */
 918 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 919 {
 920         struct open_flags op;
 921         int err = build_open_flags(flags, mode, &op);
 922         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 923 }
 924
 925 /**
 926  * filp_open - open file and return file pointer
 927  *
 928  * @filename:   path to open
 929  * @flags:      open flags as per the open(2) second argument
 930  * @mode:       mode for the new file if O_CREAT is set, else ignored
 931  *
 932  * This is the helper to open a file from kernelspace if you really
 933  * have to.  But in generally you should not do this, so please move
 934  * along, nothing to see here..
 935  */
 936 struct file *filp_open(const char *filename, int flags, umode_t mode)
 937 {
 938         struct filename name = {.name = filename};
 939         return file_open_name(&name, flags, mode);
 940 }
 941 EXPORT_SYMBOL(filp_open);
 942
 943 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 944                             const char *filename, int flags)
 945 {
 946         struct open_flags op;
 947         int err = build_open_flags(flags, 0, &op);
 948         if (err)
 949                 return ERR_PTR(err);
 950         if (flags & O_CREAT)
 951                 return ERR_PTR(-EINVAL);
 952         if (!filename && (flags & O_DIRECTORY))
 953                 if (!dentry->d_inode->i_op->lookup)
 954                         return ERR_PTR(-ENOTDIR);
 955         return do_file_open_root(dentry, mnt, filename, &op);
 956 }
 957 EXPORT_SYMBOL(file_open_root);
 958
 959 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 960 {
 961         struct open_flags op;
 962         int fd = build_open_flags(flags, mode, &op);
 963         struct filename *tmp;
 964
 965         if (fd)
 966                 return fd;
 967
 968         tmp = getname(filename);
 969         if (IS_ERR(tmp))
 970                 return PTR_ERR(tmp);
 971
 972         fd = get_unused_fd_flags(flags);
 973         if (fd >= 0) {
 974                 struct file *f = do_filp_open(dfd, tmp, &op);
 975                 if (IS_ERR(f)) {
 976                         put_unused_fd(fd);
 977                         fd = PTR_ERR(f);
 978                 } else {
 979                         fsnotify_open(f);
 980                         fd_install(fd, f);
 981                 }
 982         }
 983         putname(tmp);
 984         return fd;
 985 }
 986
 987 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 988 {
 989         if (force_o_largefile())
 990                 flags |= O_LARGEFILE;
 991
 992         return do_sys_open(AT_FDCWD, filename, flags, mode);
 993 }
 994
 995 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 996                 umode_t, mode)
 997 {
 998         if (force_o_largefile())
 999                 flags |= O_LARGEFILE;
1000
1001         return do_sys_open(dfd, filename, flags, mode);
1002 }
1003
1004 #ifndef __alpha__
1005
1006 /*
1007  * For backward compatibility?  Maybe this should be moved
1008  * into arch/i386 instead?
1009  */
1010 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1011 {
1012         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1013 }
1014
1015 #endif
1016
1017 /*
1018  * "id" is the POSIX thread ID. We use the
1019  * files pointer for this..
1020  */
1021 int filp_close(struct file *filp, fl_owner_t id)
1022 {
1023         int retval = 0;
1024
1025         if (!file_count(filp)) {
1026                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1027                 return 0;
1028         }
1029
1030         if (filp->f_op->flush)
1031                 retval = filp->f_op->flush(filp, id);
1032
1033         if (likely(!(filp->f_mode & FMODE_PATH))) {
1034                 dnotify_flush(filp, id);
1035                 locks_remove_posix(filp, id);
1036         }
1037         fput(filp);
1038         return retval;
1039 }
1040
1041 EXPORT_SYMBOL(filp_close);
1042
1043 /*
1044  * Careful here! We test whether the file pointer is NULL before
1045  * releasing the fd. This ensures that one clone task can't release
1046  * an fd while another clone is opening it.
1047  */
1048 SYSCALL_DEFINE1(close, unsigned int, fd)
1049 {
1050         int retval = __close_fd(current->files, fd);
1051
1052         /* can't restart close syscall because file table entry was cleared */
1053         if (unlikely(retval == -ERESTARTSYS ||
1054                      retval == -ERESTARTNOINTR ||
1055                      retval == -ERESTARTNOHAND ||
1056                      retval == -ERESTART_RESTARTBLOCK))
1057                 retval = -EINTR;
1058
1059         return retval;
1060 }
1061 EXPORT_SYMBOL(sys_close);
1062
1063 /*
1064  * This routine simulates a hangup on the tty, to arrange that users
1065  * are given clean terminals at login time.
1066  */
1067 SYSCALL_DEFINE0(vhangup)
1068 {
1069         if (capable(CAP_SYS_TTY_CONFIG)) {
1070                 tty_vhangup_self();
1071                 return 0;
1072         }
1073         return -EPERM;
1074 }
1075
1076 /*
1077  * Called when an inode is about to be open.
1078  * We use this to disallow opening large files on 32bit systems if
1079  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1080  * on this flag in sys_open.
1081  */
1082 int generic_file_open(struct inode * inode, struct file * filp)
1083 {
1084         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1085                 return -EOVERFLOW;
1086         return 0;
1087 }
1088
1089 EXPORT_SYMBOL(generic_file_open);
1090
1091 /*
1092  * This is used by subsystems that don't want seekable
1093  * file descriptors. The function is not supposed to ever fail, the only
1094  * reason it returns an 'int' and not 'void' is so that it can be plugged
1095  * directly into file_operations structure.
1096  */
1097 int nonseekable_open(struct inode *inode, struct file *filp)
1098 {
1099         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1100         return 0;
1101 }
1102
1103 EXPORT_SYMBOL(nonseekable_open);