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                 if (override_cred->uid)
 320                         cap_clear(override_cred->cap_effective);
 321                 else
 322                         override_cred->cap_effective =
 323                                 override_cred->cap_permitted;
 324         }
 325
 326         old_cred = override_creds(override_cred);
 327
 328         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 329         if (res)
 330                 goto out;
 331
 332         inode = path.dentry->d_inode;
 333
 334         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 335                 /*
 336                  * MAY_EXEC on regular files is denied if the fs is mounted
 337                  * with the "noexec" flag.
 338                  */
 339                 res = -EACCES;
 340                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 341                         goto out_path_release;
 342         }
 343
 344         res = inode_permission(inode, mode | MAY_ACCESS);
 345         /* SuS v2 requires we report a read only fs too */
 346         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 347                 goto out_path_release;
 348         /*
 349          * This is a rare case where using __mnt_is_readonly()
 350          * is OK without a mnt_want/drop_write() pair.  Since
 351          * no actual write to the fs is performed here, we do
 352          * not need to telegraph to that to anyone.
 353          *
 354          * By doing this, we accept that this access is
 355          * inherently racy and know that the fs may change
 356          * state before we even see this result.
 357          */
 358         if (__mnt_is_readonly(path.mnt))
 359                 res = -EROFS;
 360
 361 out_path_release:
 362         path_put(&path);
 363 out:
 364         revert_creds(old_cred);
 365         put_cred(override_cred);
 366         return res;
 367 }
 368
 369 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 370 {
 371         return sys_faccessat(AT_FDCWD, filename, mode);
 372 }
 373
 374 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 375 {
 376         struct path path;
 377         int error;
 378
 379         error = user_path_dir(filename, &path);
 380         if (error)
 381                 goto out;
 382
 383         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 384         if (error)
 385                 goto dput_and_out;
 386
 387         set_fs_pwd(current->fs, &path);
 388
 389 dput_and_out:
 390         path_put(&path);
 391 out:
 392         return error;
 393 }
 394
 395 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 396 {
 397         struct file *file;
 398         struct inode *inode;
 399         int error;
 400
 401         error = -EBADF;
 402         file = fget(fd);
 403         if (!file)
 404                 goto out;
 405
 406         inode = file->f_path.dentry->d_inode;
 407
 408         error = -ENOTDIR;
 409         if (!S_ISDIR(inode->i_mode))
 410                 goto out_putf;
 411
 412         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 413         if (!error)
 414                 set_fs_pwd(current->fs, &file->f_path);
 415 out_putf:
 416         fput(file);
 417 out:
 418         return error;
 419 }
 420
 421 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 422 {
 423         struct path path;
 424         int error;
 425
 426         error = user_path_dir(filename, &path);
 427         if (error)
 428                 goto out;
 429
 430         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 431         if (error)
 432                 goto dput_and_out;
 433
 434         error = -EPERM;
 435         if (!capable(CAP_SYS_CHROOT))
 436                 goto dput_and_out;
 437         error = security_path_chroot(&path);
 438         if (error)
 439                 goto dput_and_out;
 440
 441         set_fs_root(current->fs, &path);
 442         error = 0;
 443 dput_and_out:
 444         path_put(&path);
 445 out:
 446         return error;
 447 }
 448
 449 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 450 {
 451         struct inode * inode;
 452         struct dentry * dentry;
 453         struct file * file;
 454         int err = -EBADF;
 455         struct iattr newattrs;
 456
 457         file = fget(fd);
 458         if (!file)
 459                 goto out;
 460
 461         dentry = file->f_path.dentry;
 462         inode = dentry->d_inode;
 463
 464         audit_inode(NULL, dentry);
 465
 466         err = mnt_want_write_file(file);
 467         if (err)
 468                 goto out_putf;
 469         mutex_lock(&inode->i_mutex);
 470         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 471         if (err)
 472                 goto out_unlock;
 473         if (mode == (mode_t) -1)
 474                 mode = inode->i_mode;
 475         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 476         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 477         err = notify_change(dentry, &newattrs);
 478 out_unlock:
 479         mutex_unlock(&inode->i_mutex);
 480         mnt_drop_write(file->f_path.mnt);
 481 out_putf:
 482         fput(file);
 483 out:
 484         return err;
 485 }
 486
 487 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 488 {
 489         struct path path;
 490         struct inode *inode;
 491         int error;
 492         struct iattr newattrs;
 493
 494         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 495         if (error)
 496                 goto out;
 497         inode = path.dentry->d_inode;
 498
 499         error = mnt_want_write(path.mnt);
 500         if (error)
 501                 goto dput_and_out;
 502         mutex_lock(&inode->i_mutex);
 503         error = security_path_chmod(path.dentry, path.mnt, mode);
 504         if (error)
 505                 goto out_unlock;
 506         if (mode == (mode_t) -1)
 507                 mode = inode->i_mode;
 508         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 509         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 510         error = notify_change(path.dentry, &newattrs);
 511 out_unlock:
 512         mutex_unlock(&inode->i_mutex);
 513         mnt_drop_write(path.mnt);
 514 dput_and_out:
 515         path_put(&path);
 516 out:
 517         return error;
 518 }
 519
 520 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 521 {
 522         return sys_fchmodat(AT_FDCWD, filename, mode);
 523 }
 524
 525 static int chown_common(struct path *path, uid_t user, gid_t group)
 526 {
 527         struct inode *inode = path->dentry->d_inode;
 528         int error;
 529         struct iattr newattrs;
 530
 531         newattrs.ia_valid =  ATTR_CTIME;
 532         if (user != (uid_t) -1) {
 533                 newattrs.ia_valid |= ATTR_UID;
 534                 newattrs.ia_uid = user;
 535         }
 536         if (group != (gid_t) -1) {
 537                 newattrs.ia_valid |= ATTR_GID;
 538                 newattrs.ia_gid = group;
 539         }
 540         if (!S_ISDIR(inode->i_mode))
 541                 newattrs.ia_valid |=
 542                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 543         mutex_lock(&inode->i_mutex);
 544         error = security_path_chown(path, user, group);
 545         if (!error)
 546                 error = notify_change(path->dentry, &newattrs);
 547         mutex_unlock(&inode->i_mutex);
 548
 549         return error;
 550 }
 551
 552 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 553 {
 554         struct path path;
 555         int error;
 556
 557         error = user_path(filename, &path);
 558         if (error)
 559                 goto out;
 560         error = mnt_want_write(path.mnt);
 561         if (error)
 562                 goto out_release;
 563         error = chown_common(&path, user, group);
 564         mnt_drop_write(path.mnt);
 565 out_release:
 566         path_put(&path);
 567 out:
 568         return error;
 569 }
 570
 571 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 572                 gid_t, group, int, flag)
 573 {
 574         struct path path;
 575         int error = -EINVAL;
 576         int follow;
 577
 578         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 579                 goto out;
 580
 581         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 582         error = user_path_at(dfd, filename, follow, &path);
 583         if (error)
 584                 goto out;
 585         error = mnt_want_write(path.mnt);
 586         if (error)
 587                 goto out_release;
 588         error = chown_common(&path, user, group);
 589         mnt_drop_write(path.mnt);
 590 out_release:
 591         path_put(&path);
 592 out:
 593         return error;
 594 }
 595
 596 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 597 {
 598         struct path path;
 599         int error;
 600
 601         error = user_lpath(filename, &path);
 602         if (error)
 603                 goto out;
 604         error = mnt_want_write(path.mnt);
 605         if (error)
 606                 goto out_release;
 607         error = chown_common(&path, user, group);
 608         mnt_drop_write(path.mnt);
 609 out_release:
 610         path_put(&path);
 611 out:
 612         return error;
 613 }
 614
 615 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 616 {
 617         struct file * file;
 618         int error = -EBADF;
 619         struct dentry * dentry;
 620
 621         file = fget(fd);
 622         if (!file)
 623                 goto out;
 624
 625         error = mnt_want_write_file(file);
 626         if (error)
 627                 goto out_fput;
 628         dentry = file->f_path.dentry;
 629         audit_inode(NULL, dentry);
 630         error = chown_common(&file->f_path, user, group);
 631         mnt_drop_write(file->f_path.mnt);
 632 out_fput:
 633         fput(file);
 634 out:
 635         return error;
 636 }
 637
 638 /*
 639  * You have to be very careful that these write
 640  * counts get cleaned up in error cases and
 641  * upon __fput().  This should probably never
 642  * be called outside of __dentry_open().
 643  */
 644 static inline int __get_file_write_access(struct inode *inode,
 645                                           struct vfsmount *mnt)
 646 {
 647         int error;
 648         error = get_write_access(inode);
 649         if (error)
 650                 return error;
 651         /*
 652          * Do not take mount writer counts on
 653          * special files since no writes to
 654          * the mount itself will occur.
 655          */
 656         if (!special_file(inode->i_mode)) {
 657                 /*
 658                  * Balanced in __fput()
 659                  */
 660                 error = mnt_want_write(mnt);
 661                 if (error)
 662                         put_write_access(inode);
 663         }
 664         return error;
 665 }
 666
 667 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 668                                         struct file *f,
 669                                         int (*open)(struct inode *, struct file *),
 670                                         const struct cred *cred)
 671 {
 672         struct inode *inode;
 673         int error;
 674
 675         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 676                                 FMODE_PREAD | FMODE_PWRITE;
 677         inode = dentry->d_inode;
 678         if (f->f_mode & FMODE_WRITE) {
 679                 error = __get_file_write_access(inode, mnt);
 680                 if (error)
 681                         goto cleanup_file;
 682                 if (!special_file(inode->i_mode))
 683                         file_take_write(f);
 684         }
 685
 686         f->f_mapping = inode->i_mapping;
 687         f->f_path.dentry = dentry;
 688         f->f_path.mnt = mnt;
 689         f->f_pos = 0;
 690         f->f_op = fops_get(inode->i_fop);
 691         file_sb_list_add(f, inode->i_sb);
 692
 693         error = security_dentry_open(f, cred);
 694         if (error)
 695                 goto cleanup_all;
 696
 697         if (!open && f->f_op)
 698                 open = f->f_op->open;
 699         if (open) {
 700                 error = open(inode, f);
 701                 if (error)
 702                         goto cleanup_all;
 703         }
 704         ima_counts_get(f);
 705
 706         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 707
 708         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 709
 710         /* NB: we're sure to have correct a_ops only after f_op->open */
 711         if (f->f_flags & O_DIRECT) {
 712                 if (!f->f_mapping->a_ops ||
 713                     ((!f->f_mapping->a_ops->direct_IO) &&
 714                     (!f->f_mapping->a_ops->get_xip_mem))) {
 715                         fput(f);
 716                         f = ERR_PTR(-EINVAL);
 717                 }
 718         }
 719
 720         return f;
 721
 722 cleanup_all:
 723         fops_put(f->f_op);
 724         if (f->f_mode & FMODE_WRITE) {
 725                 put_write_access(inode);
 726                 if (!special_file(inode->i_mode)) {
 727                         /*
 728                          * We don't consider this a real
 729                          * mnt_want/drop_write() pair
 730                          * because it all happenend right
 731                          * here, so just reset the state.
 732                          */
 733                         file_reset_write(f);
 734                         mnt_drop_write(mnt);
 735                 }
 736         }
 737         file_sb_list_del(f);
 738         f->f_path.dentry = NULL;
 739         f->f_path.mnt = NULL;
 740 cleanup_file:
 741         put_filp(f);
 742         dput(dentry);
 743         mntput(mnt);
 744         return ERR_PTR(error);
 745 }
 746
 747 /**
 748  * lookup_instantiate_filp - instantiates the open intent filp
 749  * @nd: pointer to nameidata
 750  * @dentry: pointer to dentry
 751  * @open: open callback
 752  *
 753  * Helper for filesystems that want to use lookup open intents and pass back
 754  * a fully instantiated struct file to the caller.
 755  * This function is meant to be called from within a filesystem's
 756  * lookup method.
 757  * Beware of calling it for non-regular files! Those ->open methods might block
 758  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 759  * leading to a deadlock, as nobody can open that fifo anymore, because
 760  * another process to open fifo will block on locked parent when doing lookup).
 761  * Note that in case of error, nd->intent.open.file is destroyed, but the
 762  * path information remains valid.
 763  * If the open callback is set to NULL, then the standard f_op->open()
 764  * filesystem callback is substituted.
 765  */
 766 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 767                 int (*open)(struct inode *, struct file *))
 768 {
 769         const struct cred *cred = current_cred();
 770
 771         if (IS_ERR(nd->intent.open.file))
 772                 goto out;
 773         if (IS_ERR(dentry))
 774                 goto out_err;
 775         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 776                                              nd->intent.open.file,
 777                                              open, cred);
 778 out:
 779         return nd->intent.open.file;
 780 out_err:
 781         release_open_intent(nd);
 782         nd->intent.open.file = (struct file *)dentry;
 783         goto out;
 784 }
 785 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 786
 787 /**
 788  * nameidata_to_filp - convert a nameidata to an open filp.
 789  * @nd: pointer to nameidata
 790  * @flags: open flags
 791  *
 792  * Note that this function destroys the original nameidata
 793  */
 794 struct file *nameidata_to_filp(struct nameidata *nd)
 795 {
 796         const struct cred *cred = current_cred();
 797         struct file *filp;
 798
 799         /* Pick up the filp from the open intent */
 800         filp = nd->intent.open.file;
 801         nd->intent.open.file = NULL;
 802
 803         /* Has the filesystem initialised the file for us? */
 804         if (filp->f_path.dentry == NULL) {
 805                 path_get(&nd->path);
 806                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 807                                      NULL, cred);
 808         }
 809         return filp;
 810 }
 811
 812 /*
 813  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 814  * error.
 815  */
 816 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 817                          const struct cred *cred)
 818 {
 819         int error;
 820         struct file *f;
 821
 822         validate_creds(cred);
 823
 824         /*
 825          * We must always pass in a valid mount pointer.   Historically
 826          * callers got away with not passing it, but we must enforce this at
 827          * the earliest possible point now to avoid strange problems deep in the
 828          * filesystem stack.
 829          */
 830         if (!mnt) {
 831                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 832                 dump_stack();
 833                 return ERR_PTR(-EINVAL);
 834         }
 835
 836         error = -ENFILE;
 837         f = get_empty_filp();
 838         if (f == NULL) {
 839                 dput(dentry);
 840                 mntput(mnt);
 841                 return ERR_PTR(error);
 842         }
 843
 844         f->f_flags = flags;
 845         return __dentry_open(dentry, mnt, f, NULL, cred);
 846 }
 847 EXPORT_SYMBOL(dentry_open);
 848
 849 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 850 {
 851         struct fdtable *fdt = files_fdtable(files);
 852         __FD_CLR(fd, fdt->open_fds);
 853         if (fd < files->next_fd)
 854                 files->next_fd = fd;
 855 }
 856
 857 void put_unused_fd(unsigned int fd)
 858 {
 859         struct files_struct *files = current->files;
 860         spin_lock(&files->file_lock);
 861         __put_unused_fd(files, fd);
 862         spin_unlock(&files->file_lock);
 863 }
 864
 865 EXPORT_SYMBOL(put_unused_fd);
 866
 867 /*
 868  * Install a file pointer in the fd array.
 869  *
 870  * The VFS is full of places where we drop the files lock between
 871  * setting the open_fds bitmap and installing the file in the file
 872  * array.  At any such point, we are vulnerable to a dup2() race
 873  * installing a file in the array before us.  We need to detect this and
 874  * fput() the struct file we are about to overwrite in this case.
 875  *
 876  * It should never happen - if we allow dup2() do it, _really_ bad things
 877  * will follow.
 878  */
 879
 880 void fd_install(unsigned int fd, struct file *file)
 881 {
 882         struct files_struct *files = current->files;
 883         struct fdtable *fdt;
 884         spin_lock(&files->file_lock);
 885         fdt = files_fdtable(files);
 886         BUG_ON(fdt->fd[fd] != NULL);
 887         rcu_assign_pointer(fdt->fd[fd], file);
 888         spin_unlock(&files->file_lock);
 889 }
 890
 891 EXPORT_SYMBOL(fd_install);
 892
 893 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 894 {
 895         char *tmp = getname(filename);
 896         int fd = PTR_ERR(tmp);
 897
 898         if (!IS_ERR(tmp)) {
 899                 fd = get_unused_fd_flags(flags);
 900                 if (fd >= 0) {
 901                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
 902                         if (IS_ERR(f)) {
 903                                 put_unused_fd(fd);
 904                                 fd = PTR_ERR(f);
 905                         } else {
 906                                 fsnotify_open(f);
 907                                 fd_install(fd, f);
 908                         }
 909                 }
 910                 putname(tmp);
 911         }
 912         return fd;
 913 }
 914
 915 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 916 {
 917         long ret;
 918
 919         if (force_o_largefile())
 920                 flags |= O_LARGEFILE;
 921
 922         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 923         /* avoid REGPARM breakage on x86: */
 924         asmlinkage_protect(3, ret, filename, flags, mode);
 925         return ret;
 926 }
 927
 928 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 929                 int, mode)
 930 {
 931         long ret;
 932
 933         if (force_o_largefile())
 934                 flags |= O_LARGEFILE;
 935
 936         ret = do_sys_open(dfd, filename, flags, mode);
 937         /* avoid REGPARM breakage on x86: */
 938         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 939         return ret;
 940 }
 941
 942 #ifndef __alpha__
 943
 944 /*
 945  * For backward compatibility?  Maybe this should be moved
 946  * into arch/i386 instead?
 947  */
 948 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
 949 {
 950         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 951 }
 952
 953 #endif
 954
 955 /*
 956  * "id" is the POSIX thread ID. We use the
 957  * files pointer for this..
 958  */
 959 int filp_close(struct file *filp, fl_owner_t id)
 960 {
 961         int retval = 0;
 962
 963         if (!file_count(filp)) {
 964                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 965                 return 0;
 966         }
 967
 968         if (filp->f_op && filp->f_op->flush)
 969                 retval = filp->f_op->flush(filp, id);
 970
 971         dnotify_flush(filp, id);
 972         locks_remove_posix(filp, id);
 973         fput(filp);
 974         return retval;
 975 }
 976
 977 EXPORT_SYMBOL(filp_close);
 978
 979 /*
 980  * Careful here! We test whether the file pointer is NULL before
 981  * releasing the fd. This ensures that one clone task can't release
 982  * an fd while another clone is opening it.
 983  */
 984 SYSCALL_DEFINE1(close, unsigned int, fd)
 985 {
 986         struct file * filp;
 987         struct files_struct *files = current->files;
 988         struct fdtable *fdt;
 989         int retval;
 990
 991         spin_lock(&files->file_lock);
 992         fdt = files_fdtable(files);
 993         if (fd >= fdt->max_fds)
 994                 goto out_unlock;
 995         filp = fdt->fd[fd];
 996         if (!filp)
 997                 goto out_unlock;
 998         rcu_assign_pointer(fdt->fd[fd], NULL);
 999         FD_CLR(fd, fdt->close_on_exec);
1000         __put_unused_fd(files, fd);
1001         spin_unlock(&files->file_lock);
1002         retval = filp_close(filp, files);
1003
1004         /* can't restart close syscall because file table entry was cleared */
1005         if (unlikely(retval == -ERESTARTSYS ||
1006                      retval == -ERESTARTNOINTR ||
1007                      retval == -ERESTARTNOHAND ||
1008                      retval == -ERESTART_RESTARTBLOCK))
1009                 retval = -EINTR;
1010
1011         return retval;
1012
1013 out_unlock:
1014         spin_unlock(&files->file_lock);
1015         return -EBADF;
1016 }
1017 EXPORT_SYMBOL(sys_close);
1018
1019 /*
1020  * This routine simulates a hangup on the tty, to arrange that users
1021  * are given clean terminals at login time.
1022  */
1023 SYSCALL_DEFINE0(vhangup)
1024 {
1025         if (capable(CAP_SYS_TTY_CONFIG)) {
1026                 tty_vhangup_self();
1027                 return 0;
1028         }
1029         return -EPERM;
1030 }
1031
1032 /*
1033  * Called when an inode is about to be open.
1034  * We use this to disallow opening large files on 32bit systems if
1035  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1036  * on this flag in sys_open.
1037  */
1038 int generic_file_open(struct inode * inode, struct file * filp)
1039 {
1040         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1041                 return -EOVERFLOW;
1042         return 0;
1043 }
1044
1045 EXPORT_SYMBOL(generic_file_open);
1046
1047 /*
1048  * This is used by subsystems that don't want seekable
1049  * file descriptors. The function is not supposed to ever fail, the only
1050  * reason it returns an 'int' and not 'void' is so that it can be plugged
1051  * directly into file_operations structure.
1052  */
1053 int nonseekable_open(struct inode *inode, struct file *filp)
1054 {
1055         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1056         return 0;
1057 }
1058
1059 EXPORT_SYMBOL(nonseekable_open);