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          * Revalidate the write permissions, in case security policy has
 238          * changed since the files were opened.
 239          */
 240         ret = security_file_permission(file, MAY_WRITE);
 241         if (ret)
 242                 return ret;
 243
 244         if (S_ISFIFO(inode->i_mode))
 245                 return -ESPIPE;
 246
 247         /*
 248          * Let individual file system decide if it supports preallocation
 249          * for directories or not.
 250          */
 251         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 252                 return -ENODEV;
 253
 254         /* Check for wrap through zero too */
 255         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 256                 return -EFBIG;
 257
 258         if (!file->f_op->fallocate)
 259                 return -EOPNOTSUPP;
 260
 261         return file->f_op->fallocate(file, mode, offset, len);
 262 }
 263
 264 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 265 {
 266         struct file *file;
 267         int error = -EBADF;
 268
 269         file = fget(fd);
 270         if (file) {
 271                 error = do_fallocate(file, mode, offset, len);
 272                 fput(file);
 273         }
 274
 275         return error;
 276 }
 277
 278 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 279 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 280 {
 281         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 282 }
 283 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 284 #endif
 285
 286 /*
 287  * access() needs to use the real uid/gid, not the effective uid/gid.
 288  * We do this by temporarily clearing all FS-related capabilities and
 289  * switching the fsuid/fsgid around to the real ones.
 290  */
 291 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 292 {
 293         const struct cred *old_cred;
 294         struct cred *override_cred;
 295         struct path path;
 296         struct inode *inode;
 297         int res;
 298
 299         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 300                 return -EINVAL;
 301
 302         override_cred = prepare_creds();
 303         if (!override_cred)
 304                 return -ENOMEM;
 305
 306         override_cred->fsuid = override_cred->uid;
 307         override_cred->fsgid = override_cred->gid;
 308
 309         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 310                 /* Clear the capabilities if we switch to a non-root user */
 311                 if (override_cred->uid)
 312                         cap_clear(override_cred->cap_effective);
 313                 else
 314                         override_cred->cap_effective =
 315                                 override_cred->cap_permitted;
 316         }
 317
 318         old_cred = override_creds(override_cred);
 319
 320         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 321         if (res)
 322                 goto out;
 323
 324         inode = path.dentry->d_inode;
 325
 326         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 327                 /*
 328                  * MAY_EXEC on regular files is denied if the fs is mounted
 329                  * with the "noexec" flag.
 330                  */
 331                 res = -EACCES;
 332                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 333                         goto out_path_release;
 334         }
 335
 336         res = inode_permission(inode, mode | MAY_ACCESS);
 337         /* SuS v2 requires we report a read only fs too */
 338         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 339                 goto out_path_release;
 340         /*
 341          * This is a rare case where using __mnt_is_readonly()
 342          * is OK without a mnt_want/drop_write() pair.  Since
 343          * no actual write to the fs is performed here, we do
 344          * not need to telegraph to that to anyone.
 345          *
 346          * By doing this, we accept that this access is
 347          * inherently racy and know that the fs may change
 348          * state before we even see this result.
 349          */
 350         if (__mnt_is_readonly(path.mnt))
 351                 res = -EROFS;
 352
 353 out_path_release:
 354         path_put(&path);
 355 out:
 356         revert_creds(old_cred);
 357         put_cred(override_cred);
 358         return res;
 359 }
 360
 361 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 362 {
 363         return sys_faccessat(AT_FDCWD, filename, mode);
 364 }
 365
 366 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 367 {
 368         struct path path;
 369         int error;
 370
 371         error = user_path_dir(filename, &path);
 372         if (error)
 373                 goto out;
 374
 375         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 376         if (error)
 377                 goto dput_and_out;
 378
 379         set_fs_pwd(current->fs, &path);
 380
 381 dput_and_out:
 382         path_put(&path);
 383 out:
 384         return error;
 385 }
 386
 387 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 388 {
 389         struct file *file;
 390         struct inode *inode;
 391         int error;
 392
 393         error = -EBADF;
 394         file = fget(fd);
 395         if (!file)
 396                 goto out;
 397
 398         inode = file->f_path.dentry->d_inode;
 399
 400         error = -ENOTDIR;
 401         if (!S_ISDIR(inode->i_mode))
 402                 goto out_putf;
 403
 404         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 405         if (!error)
 406                 set_fs_pwd(current->fs, &file->f_path);
 407 out_putf:
 408         fput(file);
 409 out:
 410         return error;
 411 }
 412
 413 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 414 {
 415         struct path path;
 416         int error;
 417
 418         error = user_path_dir(filename, &path);
 419         if (error)
 420                 goto out;
 421
 422         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 423         if (error)
 424                 goto dput_and_out;
 425
 426         error = -EPERM;
 427         if (!capable(CAP_SYS_CHROOT))
 428                 goto dput_and_out;
 429         error = security_path_chroot(&path);
 430         if (error)
 431                 goto dput_and_out;
 432
 433         set_fs_root(current->fs, &path);
 434         error = 0;
 435 dput_and_out:
 436         path_put(&path);
 437 out:
 438         return error;
 439 }
 440
 441 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 442 {
 443         struct inode * inode;
 444         struct dentry * dentry;
 445         struct file * file;
 446         int err = -EBADF;
 447         struct iattr newattrs;
 448
 449         file = fget(fd);
 450         if (!file)
 451                 goto out;
 452
 453         dentry = file->f_path.dentry;
 454         inode = dentry->d_inode;
 455
 456         audit_inode(NULL, dentry);
 457
 458         err = mnt_want_write_file(file);
 459         if (err)
 460                 goto out_putf;
 461         mutex_lock(&inode->i_mutex);
 462         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 463         if (err)
 464                 goto out_unlock;
 465         if (mode == (mode_t) -1)
 466                 mode = inode->i_mode;
 467         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 468         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 469         err = notify_change(dentry, &newattrs);
 470 out_unlock:
 471         mutex_unlock(&inode->i_mutex);
 472         mnt_drop_write(file->f_path.mnt);
 473 out_putf:
 474         fput(file);
 475 out:
 476         return err;
 477 }
 478
 479 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 480 {
 481         struct path path;
 482         struct inode *inode;
 483         int error;
 484         struct iattr newattrs;
 485
 486         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 487         if (error)
 488                 goto out;
 489         inode = path.dentry->d_inode;
 490
 491         error = mnt_want_write(path.mnt);
 492         if (error)
 493                 goto dput_and_out;
 494         mutex_lock(&inode->i_mutex);
 495         error = security_path_chmod(path.dentry, path.mnt, mode);
 496         if (error)
 497                 goto out_unlock;
 498         if (mode == (mode_t) -1)
 499                 mode = inode->i_mode;
 500         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 501         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 502         error = notify_change(path.dentry, &newattrs);
 503 out_unlock:
 504         mutex_unlock(&inode->i_mutex);
 505         mnt_drop_write(path.mnt);
 506 dput_and_out:
 507         path_put(&path);
 508 out:
 509         return error;
 510 }
 511
 512 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 513 {
 514         return sys_fchmodat(AT_FDCWD, filename, mode);
 515 }
 516
 517 static int chown_common(struct path *path, uid_t user, gid_t group)
 518 {
 519         struct inode *inode = path->dentry->d_inode;
 520         int error;
 521         struct iattr newattrs;
 522
 523         newattrs.ia_valid =  ATTR_CTIME;
 524         if (user != (uid_t) -1) {
 525                 newattrs.ia_valid |= ATTR_UID;
 526                 newattrs.ia_uid = user;
 527         }
 528         if (group != (gid_t) -1) {
 529                 newattrs.ia_valid |= ATTR_GID;
 530                 newattrs.ia_gid = group;
 531         }
 532         if (!S_ISDIR(inode->i_mode))
 533                 newattrs.ia_valid |=
 534                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 535         mutex_lock(&inode->i_mutex);
 536         error = security_path_chown(path, user, group);
 537         if (!error)
 538                 error = notify_change(path->dentry, &newattrs);
 539         mutex_unlock(&inode->i_mutex);
 540
 541         return error;
 542 }
 543
 544 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 545 {
 546         struct path path;
 547         int error;
 548
 549         error = user_path(filename, &path);
 550         if (error)
 551                 goto out;
 552         error = mnt_want_write(path.mnt);
 553         if (error)
 554                 goto out_release;
 555         error = chown_common(&path, user, group);
 556         mnt_drop_write(path.mnt);
 557 out_release:
 558         path_put(&path);
 559 out:
 560         return error;
 561 }
 562
 563 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 564                 gid_t, group, int, flag)
 565 {
 566         struct path path;
 567         int error = -EINVAL;
 568         int follow;
 569
 570         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 571                 goto out;
 572
 573         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 574         error = user_path_at(dfd, filename, follow, &path);
 575         if (error)
 576                 goto out;
 577         error = mnt_want_write(path.mnt);
 578         if (error)
 579                 goto out_release;
 580         error = chown_common(&path, user, group);
 581         mnt_drop_write(path.mnt);
 582 out_release:
 583         path_put(&path);
 584 out:
 585         return error;
 586 }
 587
 588 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 589 {
 590         struct path path;
 591         int error;
 592
 593         error = user_lpath(filename, &path);
 594         if (error)
 595                 goto out;
 596         error = mnt_want_write(path.mnt);
 597         if (error)
 598                 goto out_release;
 599         error = chown_common(&path, user, group);
 600         mnt_drop_write(path.mnt);
 601 out_release:
 602         path_put(&path);
 603 out:
 604         return error;
 605 }
 606
 607 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 608 {
 609         struct file * file;
 610         int error = -EBADF;
 611         struct dentry * dentry;
 612
 613         file = fget(fd);
 614         if (!file)
 615                 goto out;
 616
 617         error = mnt_want_write_file(file);
 618         if (error)
 619                 goto out_fput;
 620         dentry = file->f_path.dentry;
 621         audit_inode(NULL, dentry);
 622         error = chown_common(&file->f_path, user, group);
 623         mnt_drop_write(file->f_path.mnt);
 624 out_fput:
 625         fput(file);
 626 out:
 627         return error;
 628 }
 629
 630 /*
 631  * You have to be very careful that these write
 632  * counts get cleaned up in error cases and
 633  * upon __fput().  This should probably never
 634  * be called outside of __dentry_open().
 635  */
 636 static inline int __get_file_write_access(struct inode *inode,
 637                                           struct vfsmount *mnt)
 638 {
 639         int error;
 640         error = get_write_access(inode);
 641         if (error)
 642                 return error;
 643         /*
 644          * Do not take mount writer counts on
 645          * special files since no writes to
 646          * the mount itself will occur.
 647          */
 648         if (!special_file(inode->i_mode)) {
 649                 /*
 650                  * Balanced in __fput()
 651                  */
 652                 error = mnt_want_write(mnt);
 653                 if (error)
 654                         put_write_access(inode);
 655         }
 656         return error;
 657 }
 658
 659 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 660                                         struct file *f,
 661                                         int (*open)(struct inode *, struct file *),
 662                                         const struct cred *cred)
 663 {
 664         struct inode *inode;
 665         int error;
 666
 667         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 668                                 FMODE_PREAD | FMODE_PWRITE;
 669         inode = dentry->d_inode;
 670         if (f->f_mode & FMODE_WRITE) {
 671                 error = __get_file_write_access(inode, mnt);
 672                 if (error)
 673                         goto cleanup_file;
 674                 if (!special_file(inode->i_mode))
 675                         file_take_write(f);
 676         }
 677
 678         f->f_mapping = inode->i_mapping;
 679         f->f_path.dentry = dentry;
 680         f->f_path.mnt = mnt;
 681         f->f_pos = 0;
 682         f->f_op = fops_get(inode->i_fop);
 683         file_sb_list_add(f, inode->i_sb);
 684
 685         error = security_dentry_open(f, cred);
 686         if (error)
 687                 goto cleanup_all;
 688
 689         if (!open && f->f_op)
 690                 open = f->f_op->open;
 691         if (open) {
 692                 error = open(inode, f);
 693                 if (error)
 694                         goto cleanup_all;
 695         }
 696         ima_counts_get(f);
 697
 698         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 699
 700         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 701
 702         /* NB: we're sure to have correct a_ops only after f_op->open */
 703         if (f->f_flags & O_DIRECT) {
 704                 if (!f->f_mapping->a_ops ||
 705                     ((!f->f_mapping->a_ops->direct_IO) &&
 706                     (!f->f_mapping->a_ops->get_xip_mem))) {
 707                         fput(f);
 708                         f = ERR_PTR(-EINVAL);
 709                 }
 710         }
 711
 712         return f;
 713
 714 cleanup_all:
 715         fops_put(f->f_op);
 716         if (f->f_mode & FMODE_WRITE) {
 717                 put_write_access(inode);
 718                 if (!special_file(inode->i_mode)) {
 719                         /*
 720                          * We don't consider this a real
 721                          * mnt_want/drop_write() pair
 722                          * because it all happenend right
 723                          * here, so just reset the state.
 724                          */
 725                         file_reset_write(f);
 726                         mnt_drop_write(mnt);
 727                 }
 728         }
 729         file_sb_list_del(f);
 730         f->f_path.dentry = NULL;
 731         f->f_path.mnt = NULL;
 732 cleanup_file:
 733         put_filp(f);
 734         dput(dentry);
 735         mntput(mnt);
 736         return ERR_PTR(error);
 737 }
 738
 739 /**
 740  * lookup_instantiate_filp - instantiates the open intent filp
 741  * @nd: pointer to nameidata
 742  * @dentry: pointer to dentry
 743  * @open: open callback
 744  *
 745  * Helper for filesystems that want to use lookup open intents and pass back
 746  * a fully instantiated struct file to the caller.
 747  * This function is meant to be called from within a filesystem's
 748  * lookup method.
 749  * Beware of calling it for non-regular files! Those ->open methods might block
 750  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 751  * leading to a deadlock, as nobody can open that fifo anymore, because
 752  * another process to open fifo will block on locked parent when doing lookup).
 753  * Note that in case of error, nd->intent.open.file is destroyed, but the
 754  * path information remains valid.
 755  * If the open callback is set to NULL, then the standard f_op->open()
 756  * filesystem callback is substituted.
 757  */
 758 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 759                 int (*open)(struct inode *, struct file *))
 760 {
 761         const struct cred *cred = current_cred();
 762
 763         if (IS_ERR(nd->intent.open.file))
 764                 goto out;
 765         if (IS_ERR(dentry))
 766                 goto out_err;
 767         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 768                                              nd->intent.open.file,
 769                                              open, cred);
 770 out:
 771         return nd->intent.open.file;
 772 out_err:
 773         release_open_intent(nd);
 774         nd->intent.open.file = (struct file *)dentry;
 775         goto out;
 776 }
 777 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 778
 779 /**
 780  * nameidata_to_filp - convert a nameidata to an open filp.
 781  * @nd: pointer to nameidata
 782  * @flags: open flags
 783  *
 784  * Note that this function destroys the original nameidata
 785  */
 786 struct file *nameidata_to_filp(struct nameidata *nd)
 787 {
 788         const struct cred *cred = current_cred();
 789         struct file *filp;
 790
 791         /* Pick up the filp from the open intent */
 792         filp = nd->intent.open.file;
 793         /* Has the filesystem initialised the file for us? */
 794         if (filp->f_path.dentry == NULL) {
 795                 path_get(&nd->path);
 796                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 797                                      NULL, cred);
 798         }
 799         return filp;
 800 }
 801
 802 /*
 803  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 804  * error.
 805  */
 806 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 807                          const struct cred *cred)
 808 {
 809         int error;
 810         struct file *f;
 811
 812         validate_creds(cred);
 813
 814         /*
 815          * We must always pass in a valid mount pointer.   Historically
 816          * callers got away with not passing it, but we must enforce this at
 817          * the earliest possible point now to avoid strange problems deep in the
 818          * filesystem stack.
 819          */
 820         if (!mnt) {
 821                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 822                 dump_stack();
 823                 return ERR_PTR(-EINVAL);
 824         }
 825
 826         error = -ENFILE;
 827         f = get_empty_filp();
 828         if (f == NULL) {
 829                 dput(dentry);
 830                 mntput(mnt);
 831                 return ERR_PTR(error);
 832         }
 833
 834         f->f_flags = flags;
 835         return __dentry_open(dentry, mnt, f, NULL, cred);
 836 }
 837 EXPORT_SYMBOL(dentry_open);
 838
 839 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 840 {
 841         struct fdtable *fdt = files_fdtable(files);
 842         __FD_CLR(fd, fdt->open_fds);
 843         if (fd < files->next_fd)
 844                 files->next_fd = fd;
 845 }
 846
 847 void put_unused_fd(unsigned int fd)
 848 {
 849         struct files_struct *files = current->files;
 850         spin_lock(&files->file_lock);
 851         __put_unused_fd(files, fd);
 852         spin_unlock(&files->file_lock);
 853 }
 854
 855 EXPORT_SYMBOL(put_unused_fd);
 856
 857 /*
 858  * Install a file pointer in the fd array.
 859  *
 860  * The VFS is full of places where we drop the files lock between
 861  * setting the open_fds bitmap and installing the file in the file
 862  * array.  At any such point, we are vulnerable to a dup2() race
 863  * installing a file in the array before us.  We need to detect this and
 864  * fput() the struct file we are about to overwrite in this case.
 865  *
 866  * It should never happen - if we allow dup2() do it, _really_ bad things
 867  * will follow.
 868  */
 869
 870 void fd_install(unsigned int fd, struct file *file)
 871 {
 872         struct files_struct *files = current->files;
 873         struct fdtable *fdt;
 874         spin_lock(&files->file_lock);
 875         fdt = files_fdtable(files);
 876         BUG_ON(fdt->fd[fd] != NULL);
 877         rcu_assign_pointer(fdt->fd[fd], file);
 878         spin_unlock(&files->file_lock);
 879 }
 880
 881 EXPORT_SYMBOL(fd_install);
 882
 883 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 884 {
 885         char *tmp = getname(filename);
 886         int fd = PTR_ERR(tmp);
 887
 888         if (!IS_ERR(tmp)) {
 889                 fd = get_unused_fd_flags(flags);
 890                 if (fd >= 0) {
 891                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
 892                         if (IS_ERR(f)) {
 893                                 put_unused_fd(fd);
 894                                 fd = PTR_ERR(f);
 895                         } else {
 896                                 fsnotify_open(f);
 897                                 fd_install(fd, f);
 898                         }
 899                 }
 900                 putname(tmp);
 901         }
 902         return fd;
 903 }
 904
 905 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 906 {
 907         long ret;
 908
 909         if (force_o_largefile())
 910                 flags |= O_LARGEFILE;
 911
 912         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 913         /* avoid REGPARM breakage on x86: */
 914         asmlinkage_protect(3, ret, filename, flags, mode);
 915         return ret;
 916 }
 917
 918 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 919                 int, mode)
 920 {
 921         long ret;
 922
 923         if (force_o_largefile())
 924                 flags |= O_LARGEFILE;
 925
 926         ret = do_sys_open(dfd, filename, flags, mode);
 927         /* avoid REGPARM breakage on x86: */
 928         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 929         return ret;
 930 }
 931
 932 #ifndef __alpha__
 933
 934 /*
 935  * For backward compatibility?  Maybe this should be moved
 936  * into arch/i386 instead?
 937  */
 938 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
 939 {
 940         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 941 }
 942
 943 #endif
 944
 945 /*
 946  * "id" is the POSIX thread ID. We use the
 947  * files pointer for this..
 948  */
 949 int filp_close(struct file *filp, fl_owner_t id)
 950 {
 951         int retval = 0;
 952
 953         if (!file_count(filp)) {
 954                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 955                 return 0;
 956         }
 957
 958         if (filp->f_op && filp->f_op->flush)
 959                 retval = filp->f_op->flush(filp, id);
 960
 961         dnotify_flush(filp, id);
 962         locks_remove_posix(filp, id);
 963         fput(filp);
 964         return retval;
 965 }
 966
 967 EXPORT_SYMBOL(filp_close);
 968
 969 /*
 970  * Careful here! We test whether the file pointer is NULL before
 971  * releasing the fd. This ensures that one clone task can't release
 972  * an fd while another clone is opening it.
 973  */
 974 SYSCALL_DEFINE1(close, unsigned int, fd)
 975 {
 976         struct file * filp;
 977         struct files_struct *files = current->files;
 978         struct fdtable *fdt;
 979         int retval;
 980
 981         spin_lock(&files->file_lock);
 982         fdt = files_fdtable(files);
 983         if (fd >= fdt->max_fds)
 984                 goto out_unlock;
 985         filp = fdt->fd[fd];
 986         if (!filp)
 987                 goto out_unlock;
 988         rcu_assign_pointer(fdt->fd[fd], NULL);
 989         FD_CLR(fd, fdt->close_on_exec);
 990         __put_unused_fd(files, fd);
 991         spin_unlock(&files->file_lock);
 992         retval = filp_close(filp, files);
 993
 994         /* can't restart close syscall because file table entry was cleared */
 995         if (unlikely(retval == -ERESTARTSYS ||
 996                      retval == -ERESTARTNOINTR ||
 997                      retval == -ERESTARTNOHAND ||
 998                      retval == -ERESTART_RESTARTBLOCK))
 999                 retval = -EINTR;
1000
1001         return retval;
1002
1003 out_unlock:
1004         spin_unlock(&files->file_lock);
1005         return -EBADF;
1006 }
1007 EXPORT_SYMBOL(sys_close);
1008
1009 /*
1010  * This routine simulates a hangup on the tty, to arrange that users
1011  * are given clean terminals at login time.
1012  */
1013 SYSCALL_DEFINE0(vhangup)
1014 {
1015         if (capable(CAP_SYS_TTY_CONFIG)) {
1016                 tty_vhangup_self();
1017                 return 0;
1018         }
1019         return -EPERM;
1020 }
1021
1022 /*
1023  * Called when an inode is about to be open.
1024  * We use this to disallow opening large files on 32bit systems if
1025  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1026  * on this flag in sys_open.
1027  */
1028 int generic_file_open(struct inode * inode, struct file * filp)
1029 {
1030         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1031                 return -EOVERFLOW;
1032         return 0;
1033 }
1034
1035 EXPORT_SYMBOL(generic_file_open);
1036
1037 /*
1038  * This is used by subsystems that don't want seekable
1039  * file descriptors. The function is not supposed to ever fail, the only
1040  * reason it returns an 'int' and not 'void' is so that it can be plugged
1041  * directly into file_operations structure.
1042  */
1043 int nonseekable_open(struct inode *inode, struct file *filp)
1044 {
1045         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1046         return 0;
1047 }
1048
1049 EXPORT_SYMBOL(nonseekable_open);