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 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);
 172 out_putf:
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177
 178 SYSCALL_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
 188 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 189 {
 190         return do_sys_truncate(path, length);
 191 }
 192 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 193 asmlinkage long SyS_truncate64(long path, loff_t length)
 194 {
 195         return SYSC_truncate64((const char __user *) path, length);
 196 }
 197 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 198 #endif
 199
 200 SYSCALL_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
 208 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 209 {
 210         return SYSC_ftruncate64((unsigned int) fd, length);
 211 }
 212 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 213 #endif
 214 #endif /* BITS_PER_LONG == 32 */
 215
 216
 217 int 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
 272 SYSCALL_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
 287 asmlinkage 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 }
 291 SYSCALL_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  */
 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
 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                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 320                 if (!uid_eq(override_cred->uid, root_uid))
 321                         cap_clear(override_cred->cap_effective);
 322                 else
 323                         override_cred->cap_effective =
 324                                 override_cred->cap_permitted;
 325         }
 326
 327         old_cred = override_creds(override_cred);
 328
 329         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 330         if (res)
 331                 goto out;
 332
 333         inode = path.dentry->d_inode;
 334
 335         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 336                 /*
 337                  * MAY_EXEC on regular files is denied if the fs is mounted
 338                  * with the "noexec" flag.
 339                  */
 340                 res = -EACCES;
 341                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 342                         goto out_path_release;
 343         }
 344
 345         res = inode_permission(inode, mode | MAY_ACCESS);
 346         /* SuS v2 requires we report a read only fs too */
 347         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 348                 goto out_path_release;
 349         /*
 350          * This is a rare case where using __mnt_is_readonly()
 351          * is OK without a mnt_want/drop_write() pair.  Since
 352          * no actual write to the fs is performed here, we do
 353          * not need to telegraph to that to anyone.
 354          *
 355          * By doing this, we accept that this access is
 356          * inherently racy and know that the fs may change
 357          * state before we even see this result.
 358          */
 359         if (__mnt_is_readonly(path.mnt))
 360                 res = -EROFS;
 361
 362 out_path_release:
 363         path_put(&path);
 364 out:
 365         revert_creds(old_cred);
 366         put_cred(override_cred);
 367         return res;
 368 }
 369
 370 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 371 {
 372         return sys_faccessat(AT_FDCWD, filename, mode);
 373 }
 374
 375 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 376 {
 377         struct path path;
 378         int error;
 379
 380         error = user_path_dir(filename, &path);
 381         if (error)
 382                 goto out;
 383
 384         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 385         if (error)
 386                 goto dput_and_out;
 387
 388         set_fs_pwd(current->fs, &path);
 389
 390 dput_and_out:
 391         path_put(&path);
 392 out:
 393         return error;
 394 }
 395
 396 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 397 {
 398         struct file *file;
 399         struct inode *inode;
 400         int error;
 401
 402         error = -EBADF;
 403         file = fget(fd);
 404         if (!file)
 405                 goto out;
 406
 407         inode = file->f_path.dentry->d_inode;
 408
 409         error = -ENOTDIR;
 410         if (!S_ISDIR(inode->i_mode))
 411                 goto out_putf;
 412
 413         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 414         if (!error)
 415                 set_fs_pwd(current->fs, &file->f_path);
 416 out_putf:
 417         fput(file);
 418 out:
 419         return error;
 420 }
 421
 422 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 423 {
 424         struct path path;
 425         int error;
 426
 427         error = user_path_dir(filename, &path);
 428         if (error)
 429                 goto out;
 430
 431         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 432         if (error)
 433                 goto dput_and_out;
 434
 435         error = -EPERM;
 436         if (!capable(CAP_SYS_CHROOT))
 437                 goto dput_and_out;
 438         error = security_path_chroot(&path);
 439         if (error)
 440                 goto dput_and_out;
 441
 442         set_fs_root(current->fs, &path);
 443         error = 0;
 444 dput_and_out:
 445         path_put(&path);
 446 out:
 447         return error;
 448 }
 449
 450 static int chmod_common(struct path *path, umode_t mode)
 451 {
 452         struct inode *inode = path->dentry->d_inode;
 453         struct iattr newattrs;
 454         int error;
 455
 456         error = mnt_want_write(path->mnt);
 457         if (error)
 458                 return error;
 459         mutex_lock(&inode->i_mutex);
 460         error = security_path_chmod(path, mode);
 461         if (error)
 462                 goto out_unlock;
 463         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 464         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 465         error = notify_change(path->dentry, &newattrs);
 466 out_unlock:
 467         mutex_unlock(&inode->i_mutex);
 468         mnt_drop_write(path->mnt);
 469         return error;
 470 }
 471
 472 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 473 {
 474         struct file * file;
 475         int err = -EBADF;
 476
 477         file = fget(fd);
 478         if (file) {
 479                 audit_inode(NULL, file->f_path.dentry);
 480                 err = chmod_common(&file->f_path, mode);
 481                 fput(file);
 482         }
 483         return err;
 484 }
 485
 486 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 487 {
 488         struct path path;
 489         int error;
 490
 491         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 492         if (!error) {
 493                 error = chmod_common(&path, mode);
 494                 path_put(&path);
 495         }
 496         return error;
 497 }
 498
 499 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 500 {
 501         return sys_fchmodat(AT_FDCWD, filename, mode);
 502 }
 503
 504 static int chown_common(struct path *path, uid_t user, gid_t group)
 505 {
 506         struct inode *inode = path->dentry->d_inode;
 507         int error;
 508         struct iattr newattrs;
 509         kuid_t uid;
 510         kgid_t gid;
 511
 512         uid = make_kuid(current_user_ns(), user);
 513         gid = make_kgid(current_user_ns(), group);
 514
 515         newattrs.ia_valid =  ATTR_CTIME;
 516         if (user != (uid_t) -1) {
 517                 if (!uid_valid(uid))
 518                         return -EINVAL;
 519                 newattrs.ia_valid |= ATTR_UID;
 520                 newattrs.ia_uid = uid;
 521         }
 522         if (group != (gid_t) -1) {
 523                 if (!gid_valid(gid))
 524                         return -EINVAL;
 525                 newattrs.ia_valid |= ATTR_GID;
 526                 newattrs.ia_gid = gid;
 527         }
 528         if (!S_ISDIR(inode->i_mode))
 529                 newattrs.ia_valid |=
 530                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 531         mutex_lock(&inode->i_mutex);
 532         error = security_path_chown(path, user, group);
 533         if (!error)
 534                 error = notify_change(path->dentry, &newattrs);
 535         mutex_unlock(&inode->i_mutex);
 536
 537         return error;
 538 }
 539
 540 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 541 {
 542         struct path path;
 543         int error;
 544
 545         error = user_path(filename, &path);
 546         if (error)
 547                 goto out;
 548         error = mnt_want_write(path.mnt);
 549         if (error)
 550                 goto out_release;
 551         error = chown_common(&path, user, group);
 552         mnt_drop_write(path.mnt);
 553 out_release:
 554         path_put(&path);
 555 out:
 556         return error;
 557 }
 558
 559 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 560                 gid_t, group, int, flag)
 561 {
 562         struct path path;
 563         int error = -EINVAL;
 564         int lookup_flags;
 565
 566         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 567                 goto out;
 568
 569         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 570         if (flag & AT_EMPTY_PATH)
 571                 lookup_flags |= LOOKUP_EMPTY;
 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 out:
 583         return error;
 584 }
 585
 586 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 587 {
 588         struct path path;
 589         int error;
 590
 591         error = user_lpath(filename, &path);
 592         if (error)
 593                 goto out;
 594         error = mnt_want_write(path.mnt);
 595         if (error)
 596                 goto out_release;
 597         error = chown_common(&path, user, group);
 598         mnt_drop_write(path.mnt);
 599 out_release:
 600         path_put(&path);
 601 out:
 602         return error;
 603 }
 604
 605 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 606 {
 607         struct file * file;
 608         int error = -EBADF;
 609         struct dentry * dentry;
 610
 611         file = fget(fd);
 612         if (!file)
 613                 goto out;
 614
 615         error = mnt_want_write_file(file);
 616         if (error)
 617                 goto out_fput;
 618         dentry = file->f_path.dentry;
 619         audit_inode(NULL, dentry);
 620         error = chown_common(&file->f_path, user, group);
 621         mnt_drop_write_file(file);
 622 out_fput:
 623         fput(file);
 624 out:
 625         return error;
 626 }
 627
 628 /*
 629  * You have to be very careful that these write
 630  * counts get cleaned up in error cases and
 631  * upon __fput().  This should probably never
 632  * be called outside of __dentry_open().
 633  */
 634 static inline int __get_file_write_access(struct inode *inode,
 635                                           struct vfsmount *mnt)
 636 {
 637         int error;
 638         error = get_write_access(inode);
 639         if (error)
 640                 return error;
 641         /*
 642          * Do not take mount writer counts on
 643          * special files since no writes to
 644          * the mount itself will occur.
 645          */
 646         if (!special_file(inode->i_mode)) {
 647                 /*
 648                  * Balanced in __fput()
 649                  */
 650                 error = mnt_want_write(mnt);
 651                 if (error)
 652                         put_write_access(inode);
 653         }
 654         return error;
 655 }
 656
 657 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 658                                         struct file *f,
 659                                         int (*open)(struct inode *, struct file *),
 660                                         const struct cred *cred)
 661 {
 662         static const struct file_operations empty_fops = {};
 663         struct inode *inode;
 664         int error;
 665
 666         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 667                                 FMODE_PREAD | FMODE_PWRITE;
 668
 669         if (unlikely(f->f_flags & O_PATH))
 670                 f->f_mode = FMODE_PATH;
 671
 672         inode = dentry->d_inode;
 673         if (f->f_mode & FMODE_WRITE) {
 674                 error = __get_file_write_access(inode, mnt);
 675                 if (error)
 676                         goto cleanup_file;
 677                 if (!special_file(inode->i_mode))
 678                         file_take_write(f);
 679         }
 680
 681         f->f_mapping = inode->i_mapping;
 682         f->f_path.dentry = dentry;
 683         f->f_path.mnt = mnt;
 684         f->f_pos = 0;
 685         file_sb_list_add(f, inode->i_sb);
 686
 687         if (unlikely(f->f_mode & FMODE_PATH)) {
 688                 f->f_op = &empty_fops;
 689                 return f;
 690         }
 691
 692         f->f_op = fops_get(inode->i_fop);
 693
 694         error = security_file_open(f, cred);
 695         if (error)
 696                 goto cleanup_all;
 697
 698         error = break_lease(inode, f->f_flags);
 699         if (error)
 700                 goto cleanup_all;
 701
 702         if (!open && f->f_op)
 703                 open = f->f_op->open;
 704         if (open) {
 705                 error = open(inode, f);
 706                 if (error)
 707                         goto cleanup_all;
 708         }
 709         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 710                 i_readcount_inc(inode);
 711
 712         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 713
 714         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 715
 716         /* NB: we're sure to have correct a_ops only after f_op->open */
 717         if (f->f_flags & O_DIRECT) {
 718                 if (!f->f_mapping->a_ops ||
 719                     ((!f->f_mapping->a_ops->direct_IO) &&
 720                     (!f->f_mapping->a_ops->get_xip_mem))) {
 721                         fput(f);
 722                         f = ERR_PTR(-EINVAL);
 723                 }
 724         }
 725
 726         return f;
 727
 728 cleanup_all:
 729         fops_put(f->f_op);
 730         if (f->f_mode & FMODE_WRITE) {
 731                 put_write_access(inode);
 732                 if (!special_file(inode->i_mode)) {
 733                         /*
 734                          * We don't consider this a real
 735                          * mnt_want/drop_write() pair
 736                          * because it all happenend right
 737                          * here, so just reset the state.
 738                          */
 739                         file_reset_write(f);
 740                         mnt_drop_write(mnt);
 741                 }
 742         }
 743         file_sb_list_del(f);
 744         f->f_path.dentry = NULL;
 745         f->f_path.mnt = NULL;
 746 cleanup_file:
 747         put_filp(f);
 748         dput(dentry);
 749         mntput(mnt);
 750         return ERR_PTR(error);
 751 }
 752
 753 /**
 754  * lookup_instantiate_filp - instantiates the open intent filp
 755  * @nd: pointer to nameidata
 756  * @dentry: pointer to dentry
 757  * @open: open callback
 758  *
 759  * Helper for filesystems that want to use lookup open intents and pass back
 760  * a fully instantiated struct file to the caller.
 761  * This function is meant to be called from within a filesystem's
 762  * lookup method.
 763  * Beware of calling it for non-regular files! Those ->open methods might block
 764  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 765  * leading to a deadlock, as nobody can open that fifo anymore, because
 766  * another process to open fifo will block on locked parent when doing lookup).
 767  * Note that in case of error, nd->intent.open.file is destroyed, but the
 768  * path information remains valid.
 769  * If the open callback is set to NULL, then the standard f_op->open()
 770  * filesystem callback is substituted.
 771  */
 772 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 773                 int (*open)(struct inode *, struct file *))
 774 {
 775         const struct cred *cred = current_cred();
 776
 777         if (IS_ERR(nd->intent.open.file))
 778                 goto out;
 779         if (IS_ERR(dentry))
 780                 goto out_err;
 781         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 782                                              nd->intent.open.file,
 783                                              open, cred);
 784 out:
 785         return nd->intent.open.file;
 786 out_err:
 787         release_open_intent(nd);
 788         nd->intent.open.file = ERR_CAST(dentry);
 789         goto out;
 790 }
 791 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 792
 793 /**
 794  * nameidata_to_filp - convert a nameidata to an open filp.
 795  * @nd: pointer to nameidata
 796  * @flags: open flags
 797  *
 798  * Note that this function destroys the original nameidata
 799  */
 800 struct file *nameidata_to_filp(struct nameidata *nd)
 801 {
 802         const struct cred *cred = current_cred();
 803         struct file *filp;
 804
 805         /* Pick up the filp from the open intent */
 806         filp = nd->intent.open.file;
 807         nd->intent.open.file = NULL;
 808
 809         /* Has the filesystem initialised the file for us? */
 810         if (filp->f_path.dentry == NULL) {
 811                 path_get(&nd->path);
 812                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 813                                      NULL, cred);
 814         }
 815         return filp;
 816 }
 817
 818 /*
 819  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 820  * error.
 821  */
 822 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 823                          const struct cred *cred)
 824 {
 825         int error;
 826         struct file *f;
 827
 828         validate_creds(cred);
 829
 830         /* We must always pass in a valid mount pointer. */
 831         BUG_ON(!mnt);
 832
 833         error = -ENFILE;
 834         f = get_empty_filp();
 835         if (f == NULL) {
 836                 dput(dentry);
 837                 mntput(mnt);
 838                 return ERR_PTR(error);
 839         }
 840
 841         f->f_flags = flags;
 842         return __dentry_open(dentry, mnt, f, NULL, cred);
 843 }
 844 EXPORT_SYMBOL(dentry_open);
 845
 846 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 847 {
 848         struct fdtable *fdt = files_fdtable(files);
 849         __clear_open_fd(fd, fdt);
 850         if (fd < files->next_fd)
 851                 files->next_fd = fd;
 852 }
 853
 854 void put_unused_fd(unsigned int fd)
 855 {
 856         struct files_struct *files = current->files;
 857         spin_lock(&files->file_lock);
 858         __put_unused_fd(files, fd);
 859         spin_unlock(&files->file_lock);
 860 }
 861
 862 EXPORT_SYMBOL(put_unused_fd);
 863
 864 /*
 865  * Install a file pointer in the fd array.
 866  *
 867  * The VFS is full of places where we drop the files lock between
 868  * setting the open_fds bitmap and installing the file in the file
 869  * array.  At any such point, we are vulnerable to a dup2() race
 870  * installing a file in the array before us.  We need to detect this and
 871  * fput() the struct file we are about to overwrite in this case.
 872  *
 873  * It should never happen - if we allow dup2() do it, _really_ bad things
 874  * will follow.
 875  */
 876
 877 void fd_install(unsigned int fd, struct file *file)
 878 {
 879         struct files_struct *files = current->files;
 880         struct fdtable *fdt;
 881         spin_lock(&files->file_lock);
 882         fdt = files_fdtable(files);
 883         BUG_ON(fdt->fd[fd] != NULL);
 884         rcu_assign_pointer(fdt->fd[fd], file);
 885         spin_unlock(&files->file_lock);
 886 }
 887
 888 EXPORT_SYMBOL(fd_install);
 889
 890 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 891 {
 892         int lookup_flags = 0;
 893         int acc_mode;
 894
 895         if (!(flags & O_CREAT))
 896                 mode = 0;
 897         op->mode = mode;
 898
 899         /* Must never be set by userspace */
 900         flags &= ~FMODE_NONOTIFY;
 901
 902         /*
 903          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 904          * check for O_DSYNC if the need any syncing at all we enforce it's
 905          * always set instead of having to deal with possibly weird behaviour
 906          * for malicious applications setting only __O_SYNC.
 907          */
 908         if (flags & __O_SYNC)
 909                 flags |= O_DSYNC;
 910
 911         /*
 912          * If we have O_PATH in the open flag. Then we
 913          * cannot have anything other than the below set of flags
 914          */
 915         if (flags & O_PATH) {
 916                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 917                 acc_mode = 0;
 918         } else {
 919                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 920         }
 921
 922         op->open_flag = flags;
 923
 924         /* O_TRUNC implies we need access checks for write permissions */
 925         if (flags & O_TRUNC)
 926                 acc_mode |= MAY_WRITE;
 927
 928         /* Allow the LSM permission hook to distinguish append
 929            access from general write access. */
 930         if (flags & O_APPEND)
 931                 acc_mode |= MAY_APPEND;
 932
 933         op->acc_mode = acc_mode;
 934
 935         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 936
 937         if (flags & O_CREAT) {
 938                 op->intent |= LOOKUP_CREATE;
 939                 if (flags & O_EXCL)
 940                         op->intent |= LOOKUP_EXCL;
 941         }
 942
 943         if (flags & O_DIRECTORY)
 944                 lookup_flags |= LOOKUP_DIRECTORY;
 945         if (!(flags & O_NOFOLLOW))
 946                 lookup_flags |= LOOKUP_FOLLOW;
 947         return lookup_flags;
 948 }
 949
 950 /**
 951  * filp_open - open file and return file pointer
 952  *
 953  * @filename:   path to open
 954  * @flags:      open flags as per the open(2) second argument
 955  * @mode:       mode for the new file if O_CREAT is set, else ignored
 956  *
 957  * This is the helper to open a file from kernelspace if you really
 958  * have to.  But in generally you should not do this, so please move
 959  * along, nothing to see here..
 960  */
 961 struct file *filp_open(const char *filename, int flags, umode_t mode)
 962 {
 963         struct open_flags op;
 964         int lookup = build_open_flags(flags, mode, &op);
 965         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 966 }
 967 EXPORT_SYMBOL(filp_open);
 968
 969 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 970                             const char *filename, int flags)
 971 {
 972         struct open_flags op;
 973         int lookup = build_open_flags(flags, 0, &op);
 974         if (flags & O_CREAT)
 975                 return ERR_PTR(-EINVAL);
 976         if (!filename && (flags & O_DIRECTORY))
 977                 if (!dentry->d_inode->i_op->lookup)
 978                         return ERR_PTR(-ENOTDIR);
 979         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 980 }
 981 EXPORT_SYMBOL(file_open_root);
 982
 983 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 984 {
 985         struct open_flags op;
 986         int lookup = build_open_flags(flags, mode, &op);
 987         char *tmp = getname(filename);
 988         int fd = PTR_ERR(tmp);
 989
 990         if (!IS_ERR(tmp)) {
 991                 fd = get_unused_fd_flags(flags);
 992                 if (fd >= 0) {
 993                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 994                         if (IS_ERR(f)) {
 995                                 put_unused_fd(fd);
 996                                 fd = PTR_ERR(f);
 997                         } else {
 998                                 fsnotify_open(f);
 999                                 fd_install(fd, f);
1000                         }
1001                 }
1002                 putname(tmp);
1003         }
1004         return fd;
1005 }
1006
1007 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1008 {
1009         long ret;
1010
1011         if (force_o_largefile())
1012                 flags |= O_LARGEFILE;
1013
1014         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1015         /* avoid REGPARM breakage on x86: */
1016         asmlinkage_protect(3, ret, filename, flags, mode);
1017         return ret;
1018 }
1019
1020 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1021                 umode_t, mode)
1022 {
1023         long ret;
1024
1025         if (force_o_largefile())
1026                 flags |= O_LARGEFILE;
1027
1028         ret = do_sys_open(dfd, filename, flags, mode);
1029         /* avoid REGPARM breakage on x86: */
1030         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1031         return ret;
1032 }
1033
1034 #ifndef __alpha__
1035
1036 /*
1037  * For backward compatibility?  Maybe this should be moved
1038  * into arch/i386 instead?
1039  */
1040 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1041 {
1042         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1043 }
1044
1045 #endif
1046
1047 /*
1048  * "id" is the POSIX thread ID. We use the
1049  * files pointer for this..
1050  */
1051 int filp_close(struct file *filp, fl_owner_t id)
1052 {
1053         int retval = 0;
1054
1055         if (!file_count(filp)) {
1056                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1057                 return 0;
1058         }
1059
1060         if (filp->f_op && filp->f_op->flush)
1061                 retval = filp->f_op->flush(filp, id);
1062
1063         if (likely(!(filp->f_mode & FMODE_PATH))) {
1064                 dnotify_flush(filp, id);
1065                 locks_remove_posix(filp, id);
1066         }
1067         fput(filp);
1068         return retval;
1069 }
1070
1071 EXPORT_SYMBOL(filp_close);
1072
1073 /*
1074  * Careful here! We test whether the file pointer is NULL before
1075  * releasing the fd. This ensures that one clone task can't release
1076  * an fd while another clone is opening it.
1077  */
1078 SYSCALL_DEFINE1(close, unsigned int, fd)
1079 {
1080         struct file * filp;
1081         struct files_struct *files = current->files;
1082         struct fdtable *fdt;
1083         int retval;
1084
1085         spin_lock(&files->file_lock);
1086         fdt = files_fdtable(files);
1087         if (fd >= fdt->max_fds)
1088                 goto out_unlock;
1089         filp = fdt->fd[fd];
1090         if (!filp)
1091                 goto out_unlock;
1092         rcu_assign_pointer(fdt->fd[fd], NULL);
1093         __clear_close_on_exec(fd, fdt);
1094         __put_unused_fd(files, fd);
1095         spin_unlock(&files->file_lock);
1096         retval = filp_close(filp, files);
1097
1098         /* can't restart close syscall because file table entry was cleared */
1099         if (unlikely(retval == -ERESTARTSYS ||
1100                      retval == -ERESTARTNOINTR ||
1101                      retval == -ERESTARTNOHAND ||
1102                      retval == -ERESTART_RESTARTBLOCK))
1103                 retval = -EINTR;
1104
1105         return retval;
1106
1107 out_unlock:
1108         spin_unlock(&files->file_lock);
1109         return -EBADF;
1110 }
1111 EXPORT_SYMBOL(sys_close);
1112
1113 /*
1114  * This routine simulates a hangup on the tty, to arrange that users
1115  * are given clean terminals at login time.
1116  */
1117 SYSCALL_DEFINE0(vhangup)
1118 {
1119         if (capable(CAP_SYS_TTY_CONFIG)) {
1120                 tty_vhangup_self();
1121                 return 0;
1122         }
1123         return -EPERM;
1124 }
1125
1126 /*
1127  * Called when an inode is about to be open.
1128  * We use this to disallow opening large files on 32bit systems if
1129  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1130  * on this flag in sys_open.
1131  */
1132 int generic_file_open(struct inode * inode, struct file * filp)
1133 {
1134         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1135                 return -EOVERFLOW;
1136         return 0;
1137 }
1138
1139 EXPORT_SYMBOL(generic_file_open);
1140
1141 /*
1142  * This is used by subsystems that don't want seekable
1143  * file descriptors. The function is not supposed to ever fail, the only
1144  * reason it returns an 'int' and not 'void' is so that it can be plugged
1145  * directly into file_operations structure.
1146  */
1147 int nonseekable_open(struct inode *inode, struct file *filp)
1148 {
1149         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1150         return 0;
1151 }
1152
1153 EXPORT_SYMBOL(nonseekable_open);