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 lookup_flags;
 577
 578         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 579                 goto out;
 580
 581         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 582         if (flag & AT_EMPTY_PATH)
 583                 lookup_flags |= LOOKUP_EMPTY;
 584         error = user_path_at(dfd, filename, lookup_flags, &path);
 585         if (error)
 586                 goto out;
 587         error = mnt_want_write(path.mnt);
 588         if (error)
 589                 goto out_release;
 590         error = chown_common(&path, user, group);
 591         mnt_drop_write(path.mnt);
 592 out_release:
 593         path_put(&path);
 594 out:
 595         return error;
 596 }
 597
 598 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 599 {
 600         struct path path;
 601         int error;
 602
 603         error = user_lpath(filename, &path);
 604         if (error)
 605                 goto out;
 606         error = mnt_want_write(path.mnt);
 607         if (error)
 608                 goto out_release;
 609         error = chown_common(&path, user, group);
 610         mnt_drop_write(path.mnt);
 611 out_release:
 612         path_put(&path);
 613 out:
 614         return error;
 615 }
 616
 617 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 618 {
 619         struct file * file;
 620         int error = -EBADF;
 621         struct dentry * dentry;
 622
 623         file = fget(fd);
 624         if (!file)
 625                 goto out;
 626
 627         error = mnt_want_write_file(file);
 628         if (error)
 629                 goto out_fput;
 630         dentry = file->f_path.dentry;
 631         audit_inode(NULL, dentry);
 632         error = chown_common(&file->f_path, user, group);
 633         mnt_drop_write(file->f_path.mnt);
 634 out_fput:
 635         fput(file);
 636 out:
 637         return error;
 638 }
 639
 640 /*
 641  * You have to be very careful that these write
 642  * counts get cleaned up in error cases and
 643  * upon __fput().  This should probably never
 644  * be called outside of __dentry_open().
 645  */
 646 static inline int __get_file_write_access(struct inode *inode,
 647                                           struct vfsmount *mnt)
 648 {
 649         int error;
 650         error = get_write_access(inode);
 651         if (error)
 652                 return error;
 653         /*
 654          * Do not take mount writer counts on
 655          * special files since no writes to
 656          * the mount itself will occur.
 657          */
 658         if (!special_file(inode->i_mode)) {
 659                 /*
 660                  * Balanced in __fput()
 661                  */
 662                 error = mnt_want_write(mnt);
 663                 if (error)
 664                         put_write_access(inode);
 665         }
 666         return error;
 667 }
 668
 669 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 670                                         struct file *f,
 671                                         int (*open)(struct inode *, struct file *),
 672                                         const struct cred *cred)
 673 {
 674         static const struct file_operations empty_fops = {};
 675         struct inode *inode;
 676         int error;
 677
 678         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 679                                 FMODE_PREAD | FMODE_PWRITE;
 680
 681         if (unlikely(f->f_flags & O_PATH))
 682                 f->f_mode = FMODE_PATH;
 683
 684         inode = dentry->d_inode;
 685         if (f->f_mode & FMODE_WRITE) {
 686                 error = __get_file_write_access(inode, mnt);
 687                 if (error)
 688                         goto cleanup_file;
 689                 if (!special_file(inode->i_mode))
 690                         file_take_write(f);
 691         }
 692
 693         f->f_mapping = inode->i_mapping;
 694         f->f_path.dentry = dentry;
 695         f->f_path.mnt = mnt;
 696         f->f_pos = 0;
 697         file_sb_list_add(f, inode->i_sb);
 698
 699         if (unlikely(f->f_mode & FMODE_PATH)) {
 700                 f->f_op = &empty_fops;
 701                 return f;
 702         }
 703
 704         f->f_op = fops_get(inode->i_fop);
 705
 706         error = security_dentry_open(f, cred);
 707         if (error)
 708                 goto cleanup_all;
 709
 710         if (!open && f->f_op)
 711                 open = f->f_op->open;
 712         if (open) {
 713                 error = open(inode, f);
 714                 if (error)
 715                         goto cleanup_all;
 716         }
 717         ima_counts_get(f);
 718
 719         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 720
 721         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 722
 723         /* NB: we're sure to have correct a_ops only after f_op->open */
 724         if (f->f_flags & O_DIRECT) {
 725                 if (!f->f_mapping->a_ops ||
 726                     ((!f->f_mapping->a_ops->direct_IO) &&
 727                     (!f->f_mapping->a_ops->get_xip_mem))) {
 728                         fput(f);
 729                         f = ERR_PTR(-EINVAL);
 730                 }
 731         }
 732
 733         return f;
 734
 735 cleanup_all:
 736         fops_put(f->f_op);
 737         if (f->f_mode & FMODE_WRITE) {
 738                 put_write_access(inode);
 739                 if (!special_file(inode->i_mode)) {
 740                         /*
 741                          * We don't consider this a real
 742                          * mnt_want/drop_write() pair
 743                          * because it all happenend right
 744                          * here, so just reset the state.
 745                          */
 746                         file_reset_write(f);
 747                         mnt_drop_write(mnt);
 748                 }
 749         }
 750         file_sb_list_del(f);
 751         f->f_path.dentry = NULL;
 752         f->f_path.mnt = NULL;
 753 cleanup_file:
 754         put_filp(f);
 755         dput(dentry);
 756         mntput(mnt);
 757         return ERR_PTR(error);
 758 }
 759
 760 /**
 761  * lookup_instantiate_filp - instantiates the open intent filp
 762  * @nd: pointer to nameidata
 763  * @dentry: pointer to dentry
 764  * @open: open callback
 765  *
 766  * Helper for filesystems that want to use lookup open intents and pass back
 767  * a fully instantiated struct file to the caller.
 768  * This function is meant to be called from within a filesystem's
 769  * lookup method.
 770  * Beware of calling it for non-regular files! Those ->open methods might block
 771  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 772  * leading to a deadlock, as nobody can open that fifo anymore, because
 773  * another process to open fifo will block on locked parent when doing lookup).
 774  * Note that in case of error, nd->intent.open.file is destroyed, but the
 775  * path information remains valid.
 776  * If the open callback is set to NULL, then the standard f_op->open()
 777  * filesystem callback is substituted.
 778  */
 779 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 780                 int (*open)(struct inode *, struct file *))
 781 {
 782         const struct cred *cred = current_cred();
 783
 784         if (IS_ERR(nd->intent.open.file))
 785                 goto out;
 786         if (IS_ERR(dentry))
 787                 goto out_err;
 788         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 789                                              nd->intent.open.file,
 790                                              open, cred);
 791 out:
 792         return nd->intent.open.file;
 793 out_err:
 794         release_open_intent(nd);
 795         nd->intent.open.file = (struct file *)dentry;
 796         goto out;
 797 }
 798 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 799
 800 /**
 801  * nameidata_to_filp - convert a nameidata to an open filp.
 802  * @nd: pointer to nameidata
 803  * @flags: open flags
 804  *
 805  * Note that this function destroys the original nameidata
 806  */
 807 struct file *nameidata_to_filp(struct nameidata *nd)
 808 {
 809         const struct cred *cred = current_cred();
 810         struct file *filp;
 811
 812         /* Pick up the filp from the open intent */
 813         filp = nd->intent.open.file;
 814         nd->intent.open.file = NULL;
 815
 816         /* Has the filesystem initialised the file for us? */
 817         if (filp->f_path.dentry == NULL) {
 818                 path_get(&nd->path);
 819                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 820                                      NULL, cred);
 821         }
 822         return filp;
 823 }
 824
 825 /*
 826  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 827  * error.
 828  */
 829 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 830                          const struct cred *cred)
 831 {
 832         int error;
 833         struct file *f;
 834
 835         validate_creds(cred);
 836
 837         /*
 838          * We must always pass in a valid mount pointer.   Historically
 839          * callers got away with not passing it, but we must enforce this at
 840          * the earliest possible point now to avoid strange problems deep in the
 841          * filesystem stack.
 842          */
 843         if (!mnt) {
 844                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 845                 dump_stack();
 846                 return ERR_PTR(-EINVAL);
 847         }
 848
 849         error = -ENFILE;
 850         f = get_empty_filp();
 851         if (f == NULL) {
 852                 dput(dentry);
 853                 mntput(mnt);
 854                 return ERR_PTR(error);
 855         }
 856
 857         f->f_flags = flags;
 858         return __dentry_open(dentry, mnt, f, NULL, cred);
 859 }
 860 EXPORT_SYMBOL(dentry_open);
 861
 862 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 863 {
 864         struct fdtable *fdt = files_fdtable(files);
 865         __FD_CLR(fd, fdt->open_fds);
 866         if (fd < files->next_fd)
 867                 files->next_fd = fd;
 868 }
 869
 870 void put_unused_fd(unsigned int fd)
 871 {
 872         struct files_struct *files = current->files;
 873         spin_lock(&files->file_lock);
 874         __put_unused_fd(files, fd);
 875         spin_unlock(&files->file_lock);
 876 }
 877
 878 EXPORT_SYMBOL(put_unused_fd);
 879
 880 /*
 881  * Install a file pointer in the fd array.
 882  *
 883  * The VFS is full of places where we drop the files lock between
 884  * setting the open_fds bitmap and installing the file in the file
 885  * array.  At any such point, we are vulnerable to a dup2() race
 886  * installing a file in the array before us.  We need to detect this and
 887  * fput() the struct file we are about to overwrite in this case.
 888  *
 889  * It should never happen - if we allow dup2() do it, _really_ bad things
 890  * will follow.
 891  */
 892
 893 void fd_install(unsigned int fd, struct file *file)
 894 {
 895         struct files_struct *files = current->files;
 896         struct fdtable *fdt;
 897         spin_lock(&files->file_lock);
 898         fdt = files_fdtable(files);
 899         BUG_ON(fdt->fd[fd] != NULL);
 900         rcu_assign_pointer(fdt->fd[fd], file);
 901         spin_unlock(&files->file_lock);
 902 }
 903
 904 EXPORT_SYMBOL(fd_install);
 905
 906 static inline int build_open_flags(int flags, int mode, struct open_flags *op)
 907 {
 908         int lookup_flags = 0;
 909         int acc_mode;
 910
 911         if (!(flags & O_CREAT))
 912                 mode = 0;
 913         op->mode = mode;
 914
 915         /* Must never be set by userspace */
 916         flags &= ~FMODE_NONOTIFY;
 917
 918         /*
 919          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 920          * check for O_DSYNC if the need any syncing at all we enforce it's
 921          * always set instead of having to deal with possibly weird behaviour
 922          * for malicious applications setting only __O_SYNC.
 923          */
 924         if (flags & __O_SYNC)
 925                 flags |= O_DSYNC;
 926
 927         /*
 928          * If we have O_PATH in the open flag. Then we
 929          * cannot have anything other than the below set of flags
 930          */
 931         if (flags & O_PATH) {
 932                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 933                 acc_mode = 0;
 934         } else {
 935                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 936         }
 937
 938         op->open_flag = flags;
 939
 940         /* O_TRUNC implies we need access checks for write permissions */
 941         if (flags & O_TRUNC)
 942                 acc_mode |= MAY_WRITE;
 943
 944         /* Allow the LSM permission hook to distinguish append
 945            access from general write access. */
 946         if (flags & O_APPEND)
 947                 acc_mode |= MAY_APPEND;
 948
 949         op->acc_mode = acc_mode;
 950
 951         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 952
 953         if (flags & O_CREAT) {
 954                 op->intent |= LOOKUP_CREATE;
 955                 if (flags & O_EXCL)
 956                         op->intent |= LOOKUP_EXCL;
 957         }
 958
 959         if (flags & O_DIRECTORY)
 960                 lookup_flags |= LOOKUP_DIRECTORY;
 961         if (!(flags & O_NOFOLLOW))
 962                 lookup_flags |= LOOKUP_FOLLOW;
 963         return lookup_flags;
 964 }
 965
 966 /**
 967  * filp_open - open file and return file pointer
 968  *
 969  * @filename:   path to open
 970  * @flags:      open flags as per the open(2) second argument
 971  * @mode:       mode for the new file if O_CREAT is set, else ignored
 972  *
 973  * This is the helper to open a file from kernelspace if you really
 974  * have to.  But in generally you should not do this, so please move
 975  * along, nothing to see here..
 976  */
 977 struct file *filp_open(const char *filename, int flags, int mode)
 978 {
 979         struct open_flags op;
 980         int lookup = build_open_flags(flags, mode, &op);
 981         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 982 }
 983 EXPORT_SYMBOL(filp_open);
 984
 985 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 986                             const char *filename, int flags)
 987 {
 988         struct open_flags op;
 989         int lookup = build_open_flags(flags, 0, &op);
 990         if (flags & O_CREAT)
 991                 return ERR_PTR(-EINVAL);
 992         if (!filename && (flags & O_DIRECTORY))
 993                 if (!dentry->d_inode->i_op->lookup)
 994                         return ERR_PTR(-ENOTDIR);
 995         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 996 }
 997 EXPORT_SYMBOL(file_open_root);
 998
 999 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1000 {
1001         struct open_flags op;
1002         int lookup = build_open_flags(flags, mode, &op);
1003         char *tmp = getname(filename);
1004         int fd = PTR_ERR(tmp);
1005
1006         if (!IS_ERR(tmp)) {
1007                 fd = get_unused_fd_flags(flags);
1008                 if (fd >= 0) {
1009                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
1010                         if (IS_ERR(f)) {
1011                                 put_unused_fd(fd);
1012                                 fd = PTR_ERR(f);
1013                         } else {
1014                                 fsnotify_open(f);
1015                                 fd_install(fd, f);
1016                         }
1017                 }
1018                 putname(tmp);
1019         }
1020         return fd;
1021 }
1022
1023 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1024 {
1025         long ret;
1026
1027         if (force_o_largefile())
1028                 flags |= O_LARGEFILE;
1029
1030         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1031         /* avoid REGPARM breakage on x86: */
1032         asmlinkage_protect(3, ret, filename, flags, mode);
1033         return ret;
1034 }
1035
1036 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1037                 int, mode)
1038 {
1039         long ret;
1040
1041         if (force_o_largefile())
1042                 flags |= O_LARGEFILE;
1043
1044         ret = do_sys_open(dfd, filename, flags, mode);
1045         /* avoid REGPARM breakage on x86: */
1046         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1047         return ret;
1048 }
1049
1050 #ifndef __alpha__
1051
1052 /*
1053  * For backward compatibility?  Maybe this should be moved
1054  * into arch/i386 instead?
1055  */
1056 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1057 {
1058         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1059 }
1060
1061 #endif
1062
1063 /*
1064  * "id" is the POSIX thread ID. We use the
1065  * files pointer for this..
1066  */
1067 int filp_close(struct file *filp, fl_owner_t id)
1068 {
1069         int retval = 0;
1070
1071         if (!file_count(filp)) {
1072                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1073                 return 0;
1074         }
1075
1076         if (filp->f_op && filp->f_op->flush)
1077                 retval = filp->f_op->flush(filp, id);
1078
1079         if (likely(!(filp->f_mode & FMODE_PATH))) {
1080                 dnotify_flush(filp, id);
1081                 locks_remove_posix(filp, id);
1082         }
1083         fput(filp);
1084         return retval;
1085 }
1086
1087 EXPORT_SYMBOL(filp_close);
1088
1089 /*
1090  * Careful here! We test whether the file pointer is NULL before
1091  * releasing the fd. This ensures that one clone task can't release
1092  * an fd while another clone is opening it.
1093  */
1094 SYSCALL_DEFINE1(close, unsigned int, fd)
1095 {
1096         struct file * filp;
1097         struct files_struct *files = current->files;
1098         struct fdtable *fdt;
1099         int retval;
1100
1101         spin_lock(&files->file_lock);
1102         fdt = files_fdtable(files);
1103         if (fd >= fdt->max_fds)
1104                 goto out_unlock;
1105         filp = fdt->fd[fd];
1106         if (!filp)
1107                 goto out_unlock;
1108         rcu_assign_pointer(fdt->fd[fd], NULL);
1109         FD_CLR(fd, fdt->close_on_exec);
1110         __put_unused_fd(files, fd);
1111         spin_unlock(&files->file_lock);
1112         retval = filp_close(filp, files);
1113
1114         /* can't restart close syscall because file table entry was cleared */
1115         if (unlikely(retval == -ERESTARTSYS ||
1116                      retval == -ERESTARTNOINTR ||
1117                      retval == -ERESTARTNOHAND ||
1118                      retval == -ERESTART_RESTARTBLOCK))
1119                 retval = -EINTR;
1120
1121         return retval;
1122
1123 out_unlock:
1124         spin_unlock(&files->file_lock);
1125         return -EBADF;
1126 }
1127 EXPORT_SYMBOL(sys_close);
1128
1129 /*
1130  * This routine simulates a hangup on the tty, to arrange that users
1131  * are given clean terminals at login time.
1132  */
1133 SYSCALL_DEFINE0(vhangup)
1134 {
1135         if (capable(CAP_SYS_TTY_CONFIG)) {
1136                 tty_vhangup_self();
1137                 return 0;
1138         }
1139         return -EPERM;
1140 }
1141
1142 /*
1143  * Called when an inode is about to be open.
1144  * We use this to disallow opening large files on 32bit systems if
1145  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1146  * on this flag in sys_open.
1147  */
1148 int generic_file_open(struct inode * inode, struct file * filp)
1149 {
1150         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1151                 return -EOVERFLOW;
1152         return 0;
1153 }
1154
1155 EXPORT_SYMBOL(generic_file_open);
1156
1157 /*
1158  * This is used by subsystems that don't want seekable
1159  * file descriptors. The function is not supposed to ever fail, the only
1160  * reason it returns an 'int' and not 'void' is so that it can be plugged
1161  * directly into file_operations structure.
1162  */
1163 int nonseekable_open(struct inode *inode, struct file *filp)
1164 {
1165         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1166         return 0;
1167 }
1168
1169 EXPORT_SYMBOL(nonseekable_open);