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 long vfs_truncate(struct path *path, loff_t length)
  65 {
  66         struct inode *inode;
  67         long error;
  68
  69         inode = path->dentry->d_inode;
  70
  71         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  72         if (S_ISDIR(inode->i_mode))
  73                 return -EISDIR;
  74         if (!S_ISREG(inode->i_mode))
  75                 return -EINVAL;
  76
  77         error = mnt_want_write(path->mnt);
  78         if (error)
  79                 goto out;
  80
  81         error = inode_permission(inode, MAY_WRITE);
  82         if (error)
  83                 goto mnt_drop_write_and_out;
  84
  85         error = -EPERM;
  86         if (IS_APPEND(inode))
  87                 goto mnt_drop_write_and_out;
  88
  89         error = get_write_access(inode);
  90         if (error)
  91                 goto mnt_drop_write_and_out;
  92
  93         /*
  94          * Make sure that there are no leases.  get_write_access() protects
  95          * against the truncate racing with a lease-granting setlease().
  96          */
  97         error = break_lease(inode, O_WRONLY);
  98         if (error)
  99                 goto put_write_and_out;
 100
 101         error = locks_verify_truncate(inode, NULL, length);
 102         if (!error)
 103                 error = security_path_truncate(path);
 104         if (!error)
 105                 error = do_truncate(path->dentry, length, 0, NULL);
 106
 107 put_write_and_out:
 108         put_write_access(inode);
 109 mnt_drop_write_and_out:
 110         mnt_drop_write(path->mnt);
 111 out:
 112         return error;
 113 }
 114 EXPORT_SYMBOL_GPL(vfs_truncate);
 115
 116 static long do_sys_truncate(const char __user *pathname, loff_t length)
 117 {
 118         unsigned int lookup_flags = LOOKUP_FOLLOW;
 119         struct path path;
 120         int error;
 121
 122         if (length < 0) /* sorry, but loff_t says... */
 123                 return -EINVAL;
 124
 125 retry:
 126         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 127         if (!error) {
 128                 error = vfs_truncate(&path, length);
 129                 path_put(&path);
 130         }
 131         if (retry_estale(error, lookup_flags)) {
 132                 lookup_flags |= LOOKUP_REVAL;
 133                 goto retry;
 134         }
 135         return error;
 136 }
 137
 138 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 139 {
 140         return do_sys_truncate(path, length);
 141 }
 142
 143 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 144 {
 145         struct inode *inode;
 146         struct dentry *dentry;
 147         struct fd f;
 148         int error;
 149
 150         error = -EINVAL;
 151         if (length < 0)
 152                 goto out;
 153         error = -EBADF;
 154         f = fdget(fd);
 155         if (!f.file)
 156                 goto out;
 157
 158         /* explicitly opened as large or we are on 64-bit box */
 159         if (f.file->f_flags & O_LARGEFILE)
 160                 small = 0;
 161
 162         dentry = f.file->f_path.dentry;
 163         inode = dentry->d_inode;
 164         error = -EINVAL;
 165         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 166                 goto out_putf;
 167
 168         error = -EINVAL;
 169         /* Cannot ftruncate over 2^31 bytes without large file support */
 170         if (small && length > MAX_NON_LFS)
 171                 goto out_putf;
 172
 173         error = -EPERM;
 174         if (IS_APPEND(inode))
 175                 goto out_putf;
 176
 177         sb_start_write(inode->i_sb);
 178         error = locks_verify_truncate(inode, f.file, length);
 179         if (!error)
 180                 error = security_path_truncate(&f.file->f_path);
 181         if (!error)
 182                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 183         sb_end_write(inode->i_sb);
 184 out_putf:
 185         fdput(f);
 186 out:
 187         return error;
 188 }
 189
 190 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 191 {
 192         long ret = do_sys_ftruncate(fd, length, 1);
 193         /* avoid REGPARM breakage on x86: */
 194         asmlinkage_protect(2, ret, fd, length);
 195         return ret;
 196 }
 197
 198 /* LFS versions of truncate are only needed on 32 bit machines */
 199 #if BITS_PER_LONG == 32
 200 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 201 {
 202         return do_sys_truncate(path, length);
 203 }
 204 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 205 asmlinkage long SyS_truncate64(long path, loff_t length)
 206 {
 207         return SYSC_truncate64((const char __user *) path, length);
 208 }
 209 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 210 #endif
 211
 212 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 213 {
 214         long ret = do_sys_ftruncate(fd, length, 0);
 215         /* avoid REGPARM breakage on x86: */
 216         asmlinkage_protect(2, ret, fd, length);
 217         return ret;
 218 }
 219 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 220 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 221 {
 222         return SYSC_ftruncate64((unsigned int) fd, length);
 223 }
 224 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 225 #endif
 226 #endif /* BITS_PER_LONG == 32 */
 227
 228
 229 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 230 {
 231         struct inode *inode = file_inode(file);
 232         long ret;
 233
 234         if (offset < 0 || len <= 0)
 235                 return -EINVAL;
 236
 237         /* Return error if mode is not supported */
 238         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 239                 return -EOPNOTSUPP;
 240
 241         /* Punch hole must have keep size set */
 242         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 243             !(mode & FALLOC_FL_KEEP_SIZE))
 244                 return -EOPNOTSUPP;
 245
 246         if (!(file->f_mode & FMODE_WRITE))
 247                 return -EBADF;
 248
 249         /* It's not possible punch hole on append only file */
 250         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 251                 return -EPERM;
 252
 253         if (IS_IMMUTABLE(inode))
 254                 return -EPERM;
 255
 256         /*
 257          * Revalidate the write permissions, in case security policy has
 258          * changed since the files were opened.
 259          */
 260         ret = security_file_permission(file, MAY_WRITE);
 261         if (ret)
 262                 return ret;
 263
 264         if (S_ISFIFO(inode->i_mode))
 265                 return -ESPIPE;
 266
 267         /*
 268          * Let individual file system decide if it supports preallocation
 269          * for directories or not.
 270          */
 271         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 272                 return -ENODEV;
 273
 274         /* Check for wrap through zero too */
 275         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 276                 return -EFBIG;
 277
 278         if (!file->f_op->fallocate)
 279                 return -EOPNOTSUPP;
 280
 281         sb_start_write(inode->i_sb);
 282         ret = file->f_op->fallocate(file, mode, offset, len);
 283         sb_end_write(inode->i_sb);
 284         return ret;
 285 }
 286
 287 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 288 {
 289         struct fd f = fdget(fd);
 290         int error = -EBADF;
 291
 292         if (f.file) {
 293                 error = do_fallocate(f.file, mode, offset, len);
 294                 fdput(f);
 295         }
 296         return error;
 297 }
 298
 299 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 300 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 301 {
 302         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 303 }
 304 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 305 #endif
 306
 307 /*
 308  * access() needs to use the real uid/gid, not the effective uid/gid.
 309  * We do this by temporarily clearing all FS-related capabilities and
 310  * switching the fsuid/fsgid around to the real ones.
 311  */
 312 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 313 {
 314         const struct cred *old_cred;
 315         struct cred *override_cred;
 316         struct path path;
 317         struct inode *inode;
 318         int res;
 319         unsigned int lookup_flags = LOOKUP_FOLLOW;
 320
 321         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 322                 return -EINVAL;
 323
 324         override_cred = prepare_creds();
 325         if (!override_cred)
 326                 return -ENOMEM;
 327
 328         override_cred->fsuid = override_cred->uid;
 329         override_cred->fsgid = override_cred->gid;
 330
 331         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 332                 /* Clear the capabilities if we switch to a non-root user */
 333                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 334                 if (!uid_eq(override_cred->uid, root_uid))
 335                         cap_clear(override_cred->cap_effective);
 336                 else
 337                         override_cred->cap_effective =
 338                                 override_cred->cap_permitted;
 339         }
 340
 341         old_cred = override_creds(override_cred);
 342 retry:
 343         res = user_path_at(dfd, filename, lookup_flags, &path);
 344         if (res)
 345                 goto out;
 346
 347         inode = path.dentry->d_inode;
 348
 349         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 350                 /*
 351                  * MAY_EXEC on regular files is denied if the fs is mounted
 352                  * with the "noexec" flag.
 353                  */
 354                 res = -EACCES;
 355                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 356                         goto out_path_release;
 357         }
 358
 359         res = inode_permission(inode, mode | MAY_ACCESS);
 360         /* SuS v2 requires we report a read only fs too */
 361         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 362                 goto out_path_release;
 363         /*
 364          * This is a rare case where using __mnt_is_readonly()
 365          * is OK without a mnt_want/drop_write() pair.  Since
 366          * no actual write to the fs is performed here, we do
 367          * not need to telegraph to that to anyone.
 368          *
 369          * By doing this, we accept that this access is
 370          * inherently racy and know that the fs may change
 371          * state before we even see this result.
 372          */
 373         if (__mnt_is_readonly(path.mnt))
 374                 res = -EROFS;
 375
 376 out_path_release:
 377         path_put(&path);
 378         if (retry_estale(res, lookup_flags)) {
 379                 lookup_flags |= LOOKUP_REVAL;
 380                 goto retry;
 381         }
 382 out:
 383         revert_creds(old_cred);
 384         put_cred(override_cred);
 385         return res;
 386 }
 387
 388 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 389 {
 390         return sys_faccessat(AT_FDCWD, filename, mode);
 391 }
 392
 393 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 394 {
 395         struct path path;
 396         int error;
 397         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 398 retry:
 399         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 400         if (error)
 401                 goto out;
 402
 403         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 404         if (error)
 405                 goto dput_and_out;
 406
 407         set_fs_pwd(current->fs, &path);
 408
 409 dput_and_out:
 410         path_put(&path);
 411         if (retry_estale(error, lookup_flags)) {
 412                 lookup_flags |= LOOKUP_REVAL;
 413                 goto retry;
 414         }
 415 out:
 416         return error;
 417 }
 418
 419 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 420 {
 421         struct fd f = fdget_raw(fd);
 422         struct inode *inode;
 423         int error = -EBADF;
 424
 425         error = -EBADF;
 426         if (!f.file)
 427                 goto out;
 428
 429         inode = file_inode(f.file);
 430
 431         error = -ENOTDIR;
 432         if (!S_ISDIR(inode->i_mode))
 433                 goto out_putf;
 434
 435         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 436         if (!error)
 437                 set_fs_pwd(current->fs, &f.file->f_path);
 438 out_putf:
 439         fdput(f);
 440 out:
 441         return error;
 442 }
 443
 444 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 445 {
 446         struct path path;
 447         int error;
 448         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 449 retry:
 450         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 451         if (error)
 452                 goto out;
 453
 454         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 455         if (error)
 456                 goto dput_and_out;
 457
 458         error = -EPERM;
 459         if (!nsown_capable(CAP_SYS_CHROOT))
 460                 goto dput_and_out;
 461         error = security_path_chroot(&path);
 462         if (error)
 463                 goto dput_and_out;
 464
 465         set_fs_root(current->fs, &path);
 466         error = 0;
 467 dput_and_out:
 468         path_put(&path);
 469         if (retry_estale(error, lookup_flags)) {
 470                 lookup_flags |= LOOKUP_REVAL;
 471                 goto retry;
 472         }
 473 out:
 474         return error;
 475 }
 476
 477 static int chmod_common(struct path *path, umode_t mode)
 478 {
 479         struct inode *inode = path->dentry->d_inode;
 480         struct iattr newattrs;
 481         int error;
 482
 483         error = mnt_want_write(path->mnt);
 484         if (error)
 485                 return error;
 486         mutex_lock(&inode->i_mutex);
 487         error = security_path_chmod(path, mode);
 488         if (error)
 489                 goto out_unlock;
 490         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 491         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 492         error = notify_change(path->dentry, &newattrs);
 493 out_unlock:
 494         mutex_unlock(&inode->i_mutex);
 495         mnt_drop_write(path->mnt);
 496         return error;
 497 }
 498
 499 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 500 {
 501         struct file * file;
 502         int err = -EBADF;
 503
 504         file = fget(fd);
 505         if (file) {
 506                 audit_inode(NULL, file->f_path.dentry, 0);
 507                 err = chmod_common(&file->f_path, mode);
 508                 fput(file);
 509         }
 510         return err;
 511 }
 512
 513 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 514 {
 515         struct path path;
 516         int error;
 517         unsigned int lookup_flags = LOOKUP_FOLLOW;
 518 retry:
 519         error = user_path_at(dfd, filename, lookup_flags, &path);
 520         if (!error) {
 521                 error = chmod_common(&path, mode);
 522                 path_put(&path);
 523                 if (retry_estale(error, lookup_flags)) {
 524                         lookup_flags |= LOOKUP_REVAL;
 525                         goto retry;
 526                 }
 527         }
 528         return error;
 529 }
 530
 531 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 532 {
 533         return sys_fchmodat(AT_FDCWD, filename, mode);
 534 }
 535
 536 static int chown_common(struct path *path, uid_t user, gid_t group)
 537 {
 538         struct inode *inode = path->dentry->d_inode;
 539         int error;
 540         struct iattr newattrs;
 541         kuid_t uid;
 542         kgid_t gid;
 543
 544         uid = make_kuid(current_user_ns(), user);
 545         gid = make_kgid(current_user_ns(), group);
 546
 547         newattrs.ia_valid =  ATTR_CTIME;
 548         if (user != (uid_t) -1) {
 549                 if (!uid_valid(uid))
 550                         return -EINVAL;
 551                 newattrs.ia_valid |= ATTR_UID;
 552                 newattrs.ia_uid = uid;
 553         }
 554         if (group != (gid_t) -1) {
 555                 if (!gid_valid(gid))
 556                         return -EINVAL;
 557                 newattrs.ia_valid |= ATTR_GID;
 558                 newattrs.ia_gid = gid;
 559         }
 560         if (!S_ISDIR(inode->i_mode))
 561                 newattrs.ia_valid |=
 562                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 563         mutex_lock(&inode->i_mutex);
 564         error = security_path_chown(path, uid, gid);
 565         if (!error)
 566                 error = notify_change(path->dentry, &newattrs);
 567         mutex_unlock(&inode->i_mutex);
 568
 569         return error;
 570 }
 571
 572 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 573                 gid_t, group, int, flag)
 574 {
 575         struct path path;
 576         int error = -EINVAL;
 577         int lookup_flags;
 578
 579         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 580                 goto out;
 581
 582         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 583         if (flag & AT_EMPTY_PATH)
 584                 lookup_flags |= LOOKUP_EMPTY;
 585 retry:
 586         error = user_path_at(dfd, filename, lookup_flags, &path);
 587         if (error)
 588                 goto out;
 589         error = mnt_want_write(path.mnt);
 590         if (error)
 591                 goto out_release;
 592         error = chown_common(&path, user, group);
 593         mnt_drop_write(path.mnt);
 594 out_release:
 595         path_put(&path);
 596         if (retry_estale(error, lookup_flags)) {
 597                 lookup_flags |= LOOKUP_REVAL;
 598                 goto retry;
 599         }
 600 out:
 601         return error;
 602 }
 603
 604 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 605 {
 606         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 607 }
 608
 609 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 610 {
 611         return sys_fchownat(AT_FDCWD, filename, user, group,
 612                             AT_SYMLINK_NOFOLLOW);
 613 }
 614
 615 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 616 {
 617         struct fd f = fdget(fd);
 618         int error = -EBADF;
 619
 620         if (!f.file)
 621                 goto out;
 622
 623         error = mnt_want_write_file(f.file);
 624         if (error)
 625                 goto out_fput;
 626         audit_inode(NULL, f.file->f_path.dentry, 0);
 627         error = chown_common(&f.file->f_path, user, group);
 628         mnt_drop_write_file(f.file);
 629 out_fput:
 630         fdput(f);
 631 out:
 632         return error;
 633 }
 634
 635 /*
 636  * You have to be very careful that these write
 637  * counts get cleaned up in error cases and
 638  * upon __fput().  This should probably never
 639  * be called outside of __dentry_open().
 640  */
 641 static inline int __get_file_write_access(struct inode *inode,
 642                                           struct vfsmount *mnt)
 643 {
 644         int error;
 645         error = get_write_access(inode);
 646         if (error)
 647                 return error;
 648         /*
 649          * Do not take mount writer counts on
 650          * special files since no writes to
 651          * the mount itself will occur.
 652          */
 653         if (!special_file(inode->i_mode)) {
 654                 /*
 655                  * Balanced in __fput()
 656                  */
 657                 error = __mnt_want_write(mnt);
 658                 if (error)
 659                         put_write_access(inode);
 660         }
 661         return error;
 662 }
 663
 664 int open_check_o_direct(struct file *f)
 665 {
 666         /* NB: we're sure to have correct a_ops only after f_op->open */
 667         if (f->f_flags & O_DIRECT) {
 668                 if (!f->f_mapping->a_ops ||
 669                     ((!f->f_mapping->a_ops->direct_IO) &&
 670                     (!f->f_mapping->a_ops->get_xip_mem))) {
 671                         return -EINVAL;
 672                 }
 673         }
 674         return 0;
 675 }
 676
 677 static int do_dentry_open(struct file *f,
 678                           int (*open)(struct inode *, struct file *),
 679                           const struct cred *cred)
 680 {
 681         static const struct file_operations empty_fops = {};
 682         struct inode *inode;
 683         int error;
 684
 685         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 686                                 FMODE_PREAD | FMODE_PWRITE;
 687
 688         if (unlikely(f->f_flags & O_PATH))
 689                 f->f_mode = FMODE_PATH;
 690
 691         path_get(&f->f_path);
 692         inode = file_inode(f);
 693         if (f->f_mode & FMODE_WRITE) {
 694                 error = __get_file_write_access(inode, f->f_path.mnt);
 695                 if (error)
 696                         goto cleanup_file;
 697                 if (!special_file(inode->i_mode))
 698                         file_take_write(f);
 699         }
 700
 701         f->f_mapping = inode->i_mapping;
 702         f->f_pos = 0;
 703         file_sb_list_add(f, inode->i_sb);
 704
 705         if (unlikely(f->f_mode & FMODE_PATH)) {
 706                 f->f_op = &empty_fops;
 707                 return 0;
 708         }
 709
 710         f->f_op = fops_get(inode->i_fop);
 711
 712         error = security_file_open(f, cred);
 713         if (error)
 714                 goto cleanup_all;
 715
 716         error = break_lease(inode, f->f_flags);
 717         if (error)
 718                 goto cleanup_all;
 719
 720         if (!open && f->f_op)
 721                 open = f->f_op->open;
 722         if (open) {
 723                 error = open(inode, f);
 724                 if (error)
 725                         goto cleanup_all;
 726         }
 727         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 728                 i_readcount_inc(inode);
 729
 730         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 731
 732         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 733
 734         return 0;
 735
 736 cleanup_all:
 737         fops_put(f->f_op);
 738         file_sb_list_del(f);
 739         if (f->f_mode & FMODE_WRITE) {
 740                 put_write_access(inode);
 741                 if (!special_file(inode->i_mode)) {
 742                         /*
 743                          * We don't consider this a real
 744                          * mnt_want/drop_write() pair
 745                          * because it all happenend right
 746                          * here, so just reset the state.
 747                          */
 748                         file_reset_write(f);
 749                         __mnt_drop_write(f->f_path.mnt);
 750                 }
 751         }
 752 cleanup_file:
 753         path_put(&f->f_path);
 754         f->f_path.mnt = NULL;
 755         f->f_path.dentry = NULL;
 756         return error;
 757 }
 758
 759 /**
 760  * finish_open - finish opening a file
 761  * @od: opaque open data
 762  * @dentry: pointer to dentry
 763  * @open: open callback
 764  *
 765  * This can be used to finish opening a file passed to i_op->atomic_open().
 766  *
 767  * If the open callback is set to NULL, then the standard f_op->open()
 768  * filesystem callback is substituted.
 769  */
 770 int finish_open(struct file *file, struct dentry *dentry,
 771                 int (*open)(struct inode *, struct file *),
 772                 int *opened)
 773 {
 774         int error;
 775         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 776
 777         file->f_path.dentry = dentry;
 778         error = do_dentry_open(file, open, current_cred());
 779         if (!error)
 780                 *opened |= FILE_OPENED;
 781
 782         return error;
 783 }
 784 EXPORT_SYMBOL(finish_open);
 785
 786 /**
 787  * finish_no_open - finish ->atomic_open() without opening the file
 788  *
 789  * @od: opaque open data
 790  * @dentry: dentry or NULL (as returned from ->lookup())
 791  *
 792  * This can be used to set the result of a successful lookup in ->atomic_open().
 793  * The filesystem's atomic_open() method shall return NULL after calling this.
 794  */
 795 int finish_no_open(struct file *file, struct dentry *dentry)
 796 {
 797         file->f_path.dentry = dentry;
 798         return 1;
 799 }
 800 EXPORT_SYMBOL(finish_no_open);
 801
 802 struct file *dentry_open(const struct path *path, int flags,
 803                          const struct cred *cred)
 804 {
 805         int error;
 806         struct file *f;
 807
 808         validate_creds(cred);
 809
 810         /* We must always pass in a valid mount pointer. */
 811         BUG_ON(!path->mnt);
 812
 813         error = -ENFILE;
 814         f = get_empty_filp();
 815         if (f == NULL)
 816                 return ERR_PTR(error);
 817
 818         f->f_flags = flags;
 819         f->f_path = *path;
 820         error = do_dentry_open(f, NULL, cred);
 821         if (!error) {
 822                 error = open_check_o_direct(f);
 823                 if (error) {
 824                         fput(f);
 825                         f = ERR_PTR(error);
 826                 }
 827         } else {
 828                 put_filp(f);
 829                 f = ERR_PTR(error);
 830         }
 831         return f;
 832 }
 833 EXPORT_SYMBOL(dentry_open);
 834
 835 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 836 {
 837         int lookup_flags = 0;
 838         int acc_mode;
 839
 840         if (flags & O_CREAT)
 841                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 842         else
 843                 op->mode = 0;
 844
 845         /* Must never be set by userspace */
 846         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 847
 848         /*
 849          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 850          * check for O_DSYNC if the need any syncing at all we enforce it's
 851          * always set instead of having to deal with possibly weird behaviour
 852          * for malicious applications setting only __O_SYNC.
 853          */
 854         if (flags & __O_SYNC)
 855                 flags |= O_DSYNC;
 856
 857         /*
 858          * If we have O_PATH in the open flag. Then we
 859          * cannot have anything other than the below set of flags
 860          */
 861         if (flags & O_PATH) {
 862                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 863                 acc_mode = 0;
 864         } else {
 865                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 866         }
 867
 868         op->open_flag = flags;
 869
 870         /* O_TRUNC implies we need access checks for write permissions */
 871         if (flags & O_TRUNC)
 872                 acc_mode |= MAY_WRITE;
 873
 874         /* Allow the LSM permission hook to distinguish append
 875            access from general write access. */
 876         if (flags & O_APPEND)
 877                 acc_mode |= MAY_APPEND;
 878
 879         op->acc_mode = acc_mode;
 880
 881         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 882
 883         if (flags & O_CREAT) {
 884                 op->intent |= LOOKUP_CREATE;
 885                 if (flags & O_EXCL)
 886                         op->intent |= LOOKUP_EXCL;
 887         }
 888
 889         if (flags & O_DIRECTORY)
 890                 lookup_flags |= LOOKUP_DIRECTORY;
 891         if (!(flags & O_NOFOLLOW))
 892                 lookup_flags |= LOOKUP_FOLLOW;
 893         return lookup_flags;
 894 }
 895
 896 /**
 897  * file_open_name - open file and return file pointer
 898  *
 899  * @name:       struct filename containing path to open
 900  * @flags:      open flags as per the open(2) second argument
 901  * @mode:       mode for the new file if O_CREAT is set, else ignored
 902  *
 903  * This is the helper to open a file from kernelspace if you really
 904  * have to.  But in generally you should not do this, so please move
 905  * along, nothing to see here..
 906  */
 907 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 908 {
 909         struct open_flags op;
 910         int lookup = build_open_flags(flags, mode, &op);
 911         return do_filp_open(AT_FDCWD, name, &op, lookup);
 912 }
 913
 914 /**
 915  * filp_open - open file and return file pointer
 916  *
 917  * @filename:   path to open
 918  * @flags:      open flags as per the open(2) second argument
 919  * @mode:       mode for the new file if O_CREAT is set, else ignored
 920  *
 921  * This is the helper to open a file from kernelspace if you really
 922  * have to.  But in generally you should not do this, so please move
 923  * along, nothing to see here..
 924  */
 925 struct file *filp_open(const char *filename, int flags, umode_t mode)
 926 {
 927         struct filename name = {.name = filename};
 928         return file_open_name(&name, flags, mode);
 929 }
 930 EXPORT_SYMBOL(filp_open);
 931
 932 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 933                             const char *filename, int flags)
 934 {
 935         struct open_flags op;
 936         int lookup = build_open_flags(flags, 0, &op);
 937         if (flags & O_CREAT)
 938                 return ERR_PTR(-EINVAL);
 939         if (!filename && (flags & O_DIRECTORY))
 940                 if (!dentry->d_inode->i_op->lookup)
 941                         return ERR_PTR(-ENOTDIR);
 942         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 943 }
 944 EXPORT_SYMBOL(file_open_root);
 945
 946 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 947 {
 948         struct open_flags op;
 949         int lookup = build_open_flags(flags, mode, &op);
 950         struct filename *tmp = getname(filename);
 951         int fd = PTR_ERR(tmp);
 952
 953         if (!IS_ERR(tmp)) {
 954                 fd = get_unused_fd_flags(flags);
 955                 if (fd >= 0) {
 956                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 957                         if (IS_ERR(f)) {
 958                                 put_unused_fd(fd);
 959                                 fd = PTR_ERR(f);
 960                         } else {
 961                                 fsnotify_open(f);
 962                                 fd_install(fd, f);
 963                         }
 964                 }
 965                 putname(tmp);
 966         }
 967         return fd;
 968 }
 969
 970 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 971 {
 972         long ret;
 973
 974         if (force_o_largefile())
 975                 flags |= O_LARGEFILE;
 976
 977         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 978         /* avoid REGPARM breakage on x86: */
 979         asmlinkage_protect(3, ret, filename, flags, mode);
 980         return ret;
 981 }
 982
 983 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 984                 umode_t, mode)
 985 {
 986         long ret;
 987
 988         if (force_o_largefile())
 989                 flags |= O_LARGEFILE;
 990
 991         ret = do_sys_open(dfd, filename, flags, mode);
 992         /* avoid REGPARM breakage on x86: */
 993         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 994         return ret;
 995 }
 996
 997 #ifndef __alpha__
 998
 999 /*
1000  * For backward compatibility?  Maybe this should be moved
1001  * into arch/i386 instead?
1002  */
1003 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1004 {
1005         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1006 }
1007
1008 #endif
1009
1010 /*
1011  * "id" is the POSIX thread ID. We use the
1012  * files pointer for this..
1013  */
1014 int filp_close(struct file *filp, fl_owner_t id)
1015 {
1016         int retval = 0;
1017
1018         if (!file_count(filp)) {
1019                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1020                 return 0;
1021         }
1022
1023         if (filp->f_op && filp->f_op->flush)
1024                 retval = filp->f_op->flush(filp, id);
1025
1026         if (likely(!(filp->f_mode & FMODE_PATH))) {
1027                 dnotify_flush(filp, id);
1028                 locks_remove_posix(filp, id);
1029         }
1030         fput(filp);
1031         return retval;
1032 }
1033
1034 EXPORT_SYMBOL(filp_close);
1035
1036 /*
1037  * Careful here! We test whether the file pointer is NULL before
1038  * releasing the fd. This ensures that one clone task can't release
1039  * an fd while another clone is opening it.
1040  */
1041 SYSCALL_DEFINE1(close, unsigned int, fd)
1042 {
1043         int retval = __close_fd(current->files, fd);
1044
1045         /* can't restart close syscall because file table entry was cleared */
1046         if (unlikely(retval == -ERESTARTSYS ||
1047                      retval == -ERESTARTNOINTR ||
1048                      retval == -ERESTARTNOHAND ||
1049                      retval == -ERESTART_RESTARTBLOCK))
1050                 retval = -EINTR;
1051
1052         return retval;
1053 }
1054 EXPORT_SYMBOL(sys_close);
1055
1056 /*
1057  * This routine simulates a hangup on the tty, to arrange that users
1058  * are given clean terminals at login time.
1059  */
1060 SYSCALL_DEFINE0(vhangup)
1061 {
1062         if (capable(CAP_SYS_TTY_CONFIG)) {
1063                 tty_vhangup_self();
1064                 return 0;
1065         }
1066         return -EPERM;
1067 }
1068
1069 /*
1070  * Called when an inode is about to be open.
1071  * We use this to disallow opening large files on 32bit systems if
1072  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1073  * on this flag in sys_open.
1074  */
1075 int generic_file_open(struct inode * inode, struct file * filp)
1076 {
1077         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1078                 return -EOVERFLOW;
1079         return 0;
1080 }
1081
1082 EXPORT_SYMBOL(generic_file_open);
1083
1084 /*
1085  * This is used by subsystems that don't want seekable
1086  * file descriptors. The function is not supposed to ever fail, the only
1087  * reason it returns an 'int' and not 'void' is so that it can be plugged
1088  * directly into file_operations structure.
1089  */
1090 int nonseekable_open(struct inode *inode, struct file *filp)
1091 {
1092         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1093         return 0;
1094 }
1095
1096 EXPORT_SYMBOL(nonseekable_open);