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 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid/sgid on truncate too */
  55         ret = should_remove_suid(dentry);
  56         if (ret)
  57                 newattrs.ia_valid |= ret | ATTR_FORCE;
  58
  59         mutex_lock(&dentry->d_inode->i_mutex);
  60         /* Note any delegations or leases have already been broken: */
  61         ret = notify_change(dentry, &newattrs, NULL);
  62         mutex_unlock(&dentry->d_inode->i_mutex);
  63         return ret;
  64 }
  65
  66 long vfs_truncate(struct path *path, loff_t length)
  67 {
  68         struct inode *inode;
  69         long error;
  70
  71         inode = path->dentry->d_inode;
  72
  73         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  74         if (S_ISDIR(inode->i_mode))
  75                 return -EISDIR;
  76         if (!S_ISREG(inode->i_mode))
  77                 return -EINVAL;
  78
  79         error = mnt_want_write(path->mnt);
  80         if (error)
  81                 goto out;
  82
  83         error = inode_permission(inode, MAY_WRITE);
  84         if (error)
  85                 goto mnt_drop_write_and_out;
  86
  87         error = -EPERM;
  88         if (IS_APPEND(inode))
  89                 goto mnt_drop_write_and_out;
  90
  91         error = get_write_access(inode);
  92         if (error)
  93                 goto mnt_drop_write_and_out;
  94
  95         /*
  96          * Make sure that there are no leases.  get_write_access() protects
  97          * against the truncate racing with a lease-granting setlease().
  98          */
  99         error = break_lease(inode, O_WRONLY);
 100         if (error)
 101                 goto put_write_and_out;
 102
 103         error = locks_verify_truncate(inode, NULL, length);
 104         if (!error)
 105                 error = security_path_truncate(path);
 106         if (!error)
 107                 error = do_truncate(path->dentry, length, 0, NULL);
 108
 109 put_write_and_out:
 110         put_write_access(inode);
 111 mnt_drop_write_and_out:
 112         mnt_drop_write(path->mnt);
 113 out:
 114         return error;
 115 }
 116 EXPORT_SYMBOL_GPL(vfs_truncate);
 117
 118 static long do_sys_truncate(const char __user *pathname, loff_t length)
 119 {
 120         unsigned int lookup_flags = LOOKUP_FOLLOW;
 121         struct path path;
 122         int error;
 123
 124         if (length < 0) /* sorry, but loff_t says... */
 125                 return -EINVAL;
 126
 127 retry:
 128         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 129         if (!error) {
 130                 error = vfs_truncate(&path, length);
 131                 path_put(&path);
 132         }
 133         if (retry_estale(error, lookup_flags)) {
 134                 lookup_flags |= LOOKUP_REVAL;
 135                 goto retry;
 136         }
 137         return error;
 138 }
 139
 140 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 141 {
 142         return do_sys_truncate(path, length);
 143 }
 144
 145 #ifdef CONFIG_COMPAT
 146 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 147 {
 148         return do_sys_truncate(path, length);
 149 }
 150 #endif
 151
 152 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 153 {
 154         struct inode *inode;
 155         struct dentry *dentry;
 156         struct fd f;
 157         int error;
 158
 159         error = -EINVAL;
 160         if (length < 0)
 161                 goto out;
 162         error = -EBADF;
 163         f = fdget(fd);
 164         if (!f.file)
 165                 goto out;
 166
 167         /* explicitly opened as large or we are on 64-bit box */
 168         if (f.file->f_flags & O_LARGEFILE)
 169                 small = 0;
 170
 171         dentry = f.file->f_path.dentry;
 172         inode = dentry->d_inode;
 173         error = -EINVAL;
 174         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 175                 goto out_putf;
 176
 177         error = -EINVAL;
 178         /* Cannot ftruncate over 2^31 bytes without large file support */
 179         if (small && length > MAX_NON_LFS)
 180                 goto out_putf;
 181
 182         error = -EPERM;
 183         if (IS_APPEND(inode))
 184                 goto out_putf;
 185
 186         sb_start_write(inode->i_sb);
 187         error = locks_verify_truncate(inode, f.file, length);
 188         if (!error)
 189                 error = security_path_truncate(&f.file->f_path);
 190         if (!error)
 191                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 192         sb_end_write(inode->i_sb);
 193 out_putf:
 194         fdput(f);
 195 out:
 196         return error;
 197 }
 198
 199 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 200 {
 201         return do_sys_ftruncate(fd, length, 1);
 202 }
 203
 204 #ifdef CONFIG_COMPAT
 205 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 206 {
 207         return do_sys_ftruncate(fd, length, 1);
 208 }
 209 #endif
 210
 211 /* LFS versions of truncate are only needed on 32 bit machines */
 212 #if BITS_PER_LONG == 32
 213 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 214 {
 215         return do_sys_truncate(path, length);
 216 }
 217
 218 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 219 {
 220         return do_sys_ftruncate(fd, length, 0);
 221 }
 222 #endif /* BITS_PER_LONG == 32 */
 223
 224
 225 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 226 {
 227         struct inode *inode = file_inode(file);
 228         long ret;
 229
 230         if (offset < 0 || len <= 0)
 231                 return -EINVAL;
 232
 233         /* Return error if mode is not supported */
 234         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 235                 return -EOPNOTSUPP;
 236
 237         /* Punch hole must have keep size set */
 238         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 239             !(mode & FALLOC_FL_KEEP_SIZE))
 240                 return -EOPNOTSUPP;
 241
 242         if (!(file->f_mode & FMODE_WRITE))
 243                 return -EBADF;
 244
 245         /* It's not possible punch hole on append only file */
 246         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 247                 return -EPERM;
 248
 249         if (IS_IMMUTABLE(inode))
 250                 return -EPERM;
 251
 252         /*
 253          * Revalidate the write permissions, in case security policy has
 254          * changed since the files were opened.
 255          */
 256         ret = security_file_permission(file, MAY_WRITE);
 257         if (ret)
 258                 return ret;
 259
 260         if (S_ISFIFO(inode->i_mode))
 261                 return -ESPIPE;
 262
 263         /*
 264          * Let individual file system decide if it supports preallocation
 265          * for directories or not.
 266          */
 267         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 268                 return -ENODEV;
 269
 270         /* Check for wrap through zero too */
 271         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 272                 return -EFBIG;
 273
 274         if (!file->f_op->fallocate)
 275                 return -EOPNOTSUPP;
 276
 277         sb_start_write(inode->i_sb);
 278         ret = file->f_op->fallocate(file, mode, offset, len);
 279         sb_end_write(inode->i_sb);
 280         return ret;
 281 }
 282
 283 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 284 {
 285         struct fd f = fdget(fd);
 286         int error = -EBADF;
 287
 288         if (f.file) {
 289                 error = do_fallocate(f.file, mode, offset, len);
 290                 fdput(f);
 291         }
 292         return error;
 293 }
 294
 295 /*
 296  * access() needs to use the real uid/gid, not the effective uid/gid.
 297  * We do this by temporarily clearing all FS-related capabilities and
 298  * switching the fsuid/fsgid around to the real ones.
 299  */
 300 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 301 {
 302         const struct cred *old_cred;
 303         struct cred *override_cred;
 304         struct path path;
 305         struct inode *inode;
 306         int res;
 307         unsigned int lookup_flags = LOOKUP_FOLLOW;
 308
 309         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 310                 return -EINVAL;
 311
 312         override_cred = prepare_creds();
 313         if (!override_cred)
 314                 return -ENOMEM;
 315
 316         override_cred->fsuid = override_cred->uid;
 317         override_cred->fsgid = override_cred->gid;
 318
 319         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 320                 /* Clear the capabilities if we switch to a non-root user */
 321                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 322                 if (!uid_eq(override_cred->uid, root_uid))
 323                         cap_clear(override_cred->cap_effective);
 324                 else
 325                         override_cred->cap_effective =
 326                                 override_cred->cap_permitted;
 327         }
 328
 329         old_cred = override_creds(override_cred);
 330 retry:
 331         res = user_path_at(dfd, filename, lookup_flags, &path);
 332         if (res)
 333                 goto out;
 334
 335         inode = path.dentry->d_inode;
 336
 337         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 338                 /*
 339                  * MAY_EXEC on regular files is denied if the fs is mounted
 340                  * with the "noexec" flag.
 341                  */
 342                 res = -EACCES;
 343                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 344                         goto out_path_release;
 345         }
 346
 347         res = inode_permission(inode, mode | MAY_ACCESS);
 348         /* SuS v2 requires we report a read only fs too */
 349         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 350                 goto out_path_release;
 351         /*
 352          * This is a rare case where using __mnt_is_readonly()
 353          * is OK without a mnt_want/drop_write() pair.  Since
 354          * no actual write to the fs is performed here, we do
 355          * not need to telegraph to that to anyone.
 356          *
 357          * By doing this, we accept that this access is
 358          * inherently racy and know that the fs may change
 359          * state before we even see this result.
 360          */
 361         if (__mnt_is_readonly(path.mnt))
 362                 res = -EROFS;
 363
 364 out_path_release:
 365         path_put(&path);
 366         if (retry_estale(res, lookup_flags)) {
 367                 lookup_flags |= LOOKUP_REVAL;
 368                 goto retry;
 369         }
 370 out:
 371         revert_creds(old_cred);
 372         put_cred(override_cred);
 373         return res;
 374 }
 375
 376 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 377 {
 378         return sys_faccessat(AT_FDCWD, filename, mode);
 379 }
 380
 381 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 382 {
 383         struct path path;
 384         int error;
 385         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 386 retry:
 387         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 388         if (error)
 389                 goto out;
 390
 391         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 392         if (error)
 393                 goto dput_and_out;
 394
 395         set_fs_pwd(current->fs, &path);
 396
 397 dput_and_out:
 398         path_put(&path);
 399         if (retry_estale(error, lookup_flags)) {
 400                 lookup_flags |= LOOKUP_REVAL;
 401                 goto retry;
 402         }
 403 out:
 404         return error;
 405 }
 406
 407 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 408 {
 409         struct fd f = fdget_raw(fd);
 410         struct inode *inode;
 411         int error = -EBADF;
 412
 413         error = -EBADF;
 414         if (!f.file)
 415                 goto out;
 416
 417         inode = file_inode(f.file);
 418
 419         error = -ENOTDIR;
 420         if (!S_ISDIR(inode->i_mode))
 421                 goto out_putf;
 422
 423         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 424         if (!error)
 425                 set_fs_pwd(current->fs, &f.file->f_path);
 426 out_putf:
 427         fdput(f);
 428 out:
 429         return error;
 430 }
 431
 432 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 433 {
 434         struct path path;
 435         int error;
 436         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 437 retry:
 438         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 439         if (error)
 440                 goto out;
 441
 442         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 443         if (error)
 444                 goto dput_and_out;
 445
 446         error = -EPERM;
 447         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 448                 goto dput_and_out;
 449         error = security_path_chroot(&path);
 450         if (error)
 451                 goto dput_and_out;
 452
 453         set_fs_root(current->fs, &path);
 454         error = 0;
 455 dput_and_out:
 456         path_put(&path);
 457         if (retry_estale(error, lookup_flags)) {
 458                 lookup_flags |= LOOKUP_REVAL;
 459                 goto retry;
 460         }
 461 out:
 462         return error;
 463 }
 464
 465 static int chmod_common(struct path *path, umode_t mode)
 466 {
 467         struct inode *inode = path->dentry->d_inode;
 468         struct inode *delegated_inode = NULL;
 469         struct iattr newattrs;
 470         int error;
 471
 472         error = mnt_want_write(path->mnt);
 473         if (error)
 474                 return error;
 475 retry_deleg:
 476         mutex_lock(&inode->i_mutex);
 477         error = security_path_chmod(path, mode);
 478         if (error)
 479                 goto out_unlock;
 480         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 481         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 482         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 483 out_unlock:
 484         mutex_unlock(&inode->i_mutex);
 485         if (delegated_inode) {
 486                 error = break_deleg_wait(&delegated_inode);
 487                 if (!error)
 488                         goto retry_deleg;
 489         }
 490         mnt_drop_write(path->mnt);
 491         return error;
 492 }
 493
 494 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 495 {
 496         struct fd f = fdget(fd);
 497         int err = -EBADF;
 498
 499         if (f.file) {
 500                 audit_inode(NULL, f.file->f_path.dentry, 0);
 501                 err = chmod_common(&f.file->f_path, mode);
 502                 fdput(f);
 503         }
 504         return err;
 505 }
 506
 507 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 508 {
 509         struct path path;
 510         int error;
 511         unsigned int lookup_flags = LOOKUP_FOLLOW;
 512 retry:
 513         error = user_path_at(dfd, filename, lookup_flags, &path);
 514         if (!error) {
 515                 error = chmod_common(&path, mode);
 516                 path_put(&path);
 517                 if (retry_estale(error, lookup_flags)) {
 518                         lookup_flags |= LOOKUP_REVAL;
 519                         goto retry;
 520                 }
 521         }
 522         return error;
 523 }
 524
 525 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 526 {
 527         return sys_fchmodat(AT_FDCWD, filename, mode);
 528 }
 529
 530 static int chown_common(struct path *path, uid_t user, gid_t group)
 531 {
 532         struct inode *inode = path->dentry->d_inode;
 533         struct inode *delegated_inode = NULL;
 534         int error;
 535         struct iattr newattrs;
 536         kuid_t uid;
 537         kgid_t gid;
 538
 539         uid = make_kuid(current_user_ns(), user);
 540         gid = make_kgid(current_user_ns(), group);
 541
 542         newattrs.ia_valid =  ATTR_CTIME;
 543         if (user != (uid_t) -1) {
 544                 if (!uid_valid(uid))
 545                         return -EINVAL;
 546                 newattrs.ia_valid |= ATTR_UID;
 547                 newattrs.ia_uid = uid;
 548         }
 549         if (group != (gid_t) -1) {
 550                 if (!gid_valid(gid))
 551                         return -EINVAL;
 552                 newattrs.ia_valid |= ATTR_GID;
 553                 newattrs.ia_gid = gid;
 554         }
 555         if (!S_ISDIR(inode->i_mode))
 556                 newattrs.ia_valid |=
 557                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 558 retry_deleg:
 559         mutex_lock(&inode->i_mutex);
 560         error = security_path_chown(path, uid, gid);
 561         if (!error)
 562                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 563         mutex_unlock(&inode->i_mutex);
 564         if (delegated_inode) {
 565                 error = break_deleg_wait(&delegated_inode);
 566                 if (!error)
 567                         goto retry_deleg;
 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 = f->f_inode = f->f_path.dentry->d_inode;
 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
 703         if (unlikely(f->f_mode & FMODE_PATH)) {
 704                 f->f_op = &empty_fops;
 705                 return 0;
 706         }
 707
 708         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 709         if (S_ISREG(inode->i_mode))
 710                 f->f_mode |= FMODE_ATOMIC_POS;
 711
 712         f->f_op = fops_get(inode->i_fop);
 713         if (unlikely(WARN_ON(!f->f_op))) {
 714                 error = -ENODEV;
 715                 goto cleanup_all;
 716         }
 717
 718         error = security_file_open(f, cred);
 719         if (error)
 720                 goto cleanup_all;
 721
 722         error = break_lease(inode, f->f_flags);
 723         if (error)
 724                 goto cleanup_all;
 725
 726         if (!open)
 727                 open = f->f_op->open;
 728         if (open) {
 729                 error = open(inode, f);
 730                 if (error)
 731                         goto cleanup_all;
 732         }
 733         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 734                 i_readcount_inc(inode);
 735
 736         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 737
 738         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 739
 740         return 0;
 741
 742 cleanup_all:
 743         fops_put(f->f_op);
 744         if (f->f_mode & FMODE_WRITE) {
 745                 put_write_access(inode);
 746                 if (!special_file(inode->i_mode)) {
 747                         /*
 748                          * We don't consider this a real
 749                          * mnt_want/drop_write() pair
 750                          * because it all happenend right
 751                          * here, so just reset the state.
 752                          */
 753                         file_reset_write(f);
 754                         __mnt_drop_write(f->f_path.mnt);
 755                 }
 756         }
 757 cleanup_file:
 758         path_put(&f->f_path);
 759         f->f_path.mnt = NULL;
 760         f->f_path.dentry = NULL;
 761         f->f_inode = NULL;
 762         return error;
 763 }
 764
 765 /**
 766  * finish_open - finish opening a file
 767  * @file: file pointer
 768  * @dentry: pointer to dentry
 769  * @open: open callback
 770  * @opened: state of open
 771  *
 772  * This can be used to finish opening a file passed to i_op->atomic_open().
 773  *
 774  * If the open callback is set to NULL, then the standard f_op->open()
 775  * filesystem callback is substituted.
 776  *
 777  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 778  * the return value of d_splice_alias(), then the caller needs to perform dput()
 779  * on it after finish_open().
 780  *
 781  * On successful return @file is a fully instantiated open file.  After this, if
 782  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 783  *
 784  * Returns zero on success or -errno if the open failed.
 785  */
 786 int finish_open(struct file *file, struct dentry *dentry,
 787                 int (*open)(struct inode *, struct file *),
 788                 int *opened)
 789 {
 790         int error;
 791         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 792
 793         file->f_path.dentry = dentry;
 794         error = do_dentry_open(file, open, current_cred());
 795         if (!error)
 796                 *opened |= FILE_OPENED;
 797
 798         return error;
 799 }
 800 EXPORT_SYMBOL(finish_open);
 801
 802 /**
 803  * finish_no_open - finish ->atomic_open() without opening the file
 804  *
 805  * @file: file pointer
 806  * @dentry: dentry or NULL (as returned from ->lookup())
 807  *
 808  * This can be used to set the result of a successful lookup in ->atomic_open().
 809  *
 810  * NB: unlike finish_open() this function does consume the dentry reference and
 811  * the caller need not dput() it.
 812  *
 813  * Returns "1" which must be the return value of ->atomic_open() after having
 814  * called this function.
 815  */
 816 int finish_no_open(struct file *file, struct dentry *dentry)
 817 {
 818         file->f_path.dentry = dentry;
 819         return 1;
 820 }
 821 EXPORT_SYMBOL(finish_no_open);
 822
 823 struct file *dentry_open(const struct path *path, int flags,
 824                          const struct cred *cred)
 825 {
 826         int error;
 827         struct file *f;
 828
 829         validate_creds(cred);
 830
 831         /* We must always pass in a valid mount pointer. */
 832         BUG_ON(!path->mnt);
 833
 834         f = get_empty_filp();
 835         if (!IS_ERR(f)) {
 836                 f->f_flags = flags;
 837                 f->f_path = *path;
 838                 error = do_dentry_open(f, NULL, cred);
 839                 if (!error) {
 840                         /* from now on we need fput() to dispose of f */
 841                         error = open_check_o_direct(f);
 842                         if (error) {
 843                                 fput(f);
 844                                 f = ERR_PTR(error);
 845                         }
 846                 } else {
 847                         put_filp(f);
 848                         f = ERR_PTR(error);
 849                 }
 850         }
 851         return f;
 852 }
 853 EXPORT_SYMBOL(dentry_open);
 854
 855 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 856 {
 857         int lookup_flags = 0;
 858         int acc_mode;
 859
 860         if (flags & (O_CREAT | __O_TMPFILE))
 861                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 862         else
 863                 op->mode = 0;
 864
 865         /* Must never be set by userspace */
 866         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 867
 868         /*
 869          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 870          * check for O_DSYNC if the need any syncing at all we enforce it's
 871          * always set instead of having to deal with possibly weird behaviour
 872          * for malicious applications setting only __O_SYNC.
 873          */
 874         if (flags & __O_SYNC)
 875                 flags |= O_DSYNC;
 876
 877         if (flags & __O_TMPFILE) {
 878                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 879                         return -EINVAL;
 880                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 881                 if (!(acc_mode & MAY_WRITE))
 882                         return -EINVAL;
 883         } else if (flags & O_PATH) {
 884                 /*
 885                  * If we have O_PATH in the open flag. Then we
 886                  * cannot have anything other than the below set of flags
 887                  */
 888                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 889                 acc_mode = 0;
 890         } else {
 891                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 892         }
 893
 894         op->open_flag = flags;
 895
 896         /* O_TRUNC implies we need access checks for write permissions */
 897         if (flags & O_TRUNC)
 898                 acc_mode |= MAY_WRITE;
 899
 900         /* Allow the LSM permission hook to distinguish append
 901            access from general write access. */
 902         if (flags & O_APPEND)
 903                 acc_mode |= MAY_APPEND;
 904
 905         op->acc_mode = acc_mode;
 906
 907         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 908
 909         if (flags & O_CREAT) {
 910                 op->intent |= LOOKUP_CREATE;
 911                 if (flags & O_EXCL)
 912                         op->intent |= LOOKUP_EXCL;
 913         }
 914
 915         if (flags & O_DIRECTORY)
 916                 lookup_flags |= LOOKUP_DIRECTORY;
 917         if (!(flags & O_NOFOLLOW))
 918                 lookup_flags |= LOOKUP_FOLLOW;
 919         op->lookup_flags = lookup_flags;
 920         return 0;
 921 }
 922
 923 /**
 924  * file_open_name - open file and return file pointer
 925  *
 926  * @name:       struct filename containing path to open
 927  * @flags:      open flags as per the open(2) second argument
 928  * @mode:       mode for the new file if O_CREAT is set, else ignored
 929  *
 930  * This is the helper to open a file from kernelspace if you really
 931  * have to.  But in generally you should not do this, so please move
 932  * along, nothing to see here..
 933  */
 934 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 935 {
 936         struct open_flags op;
 937         int err = build_open_flags(flags, mode, &op);
 938         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 939 }
 940
 941 /**
 942  * filp_open - open file and return file pointer
 943  *
 944  * @filename:   path to open
 945  * @flags:      open flags as per the open(2) second argument
 946  * @mode:       mode for the new file if O_CREAT is set, else ignored
 947  *
 948  * This is the helper to open a file from kernelspace if you really
 949  * have to.  But in generally you should not do this, so please move
 950  * along, nothing to see here..
 951  */
 952 struct file *filp_open(const char *filename, int flags, umode_t mode)
 953 {
 954         struct filename name = {.name = filename};
 955         return file_open_name(&name, flags, mode);
 956 }
 957 EXPORT_SYMBOL(filp_open);
 958
 959 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 960                             const char *filename, int flags)
 961 {
 962         struct open_flags op;
 963         int err = build_open_flags(flags, 0, &op);
 964         if (err)
 965                 return ERR_PTR(err);
 966         if (flags & O_CREAT)
 967                 return ERR_PTR(-EINVAL);
 968         if (!filename && (flags & O_DIRECTORY))
 969                 if (!dentry->d_inode->i_op->lookup)
 970                         return ERR_PTR(-ENOTDIR);
 971         return do_file_open_root(dentry, mnt, filename, &op);
 972 }
 973 EXPORT_SYMBOL(file_open_root);
 974
 975 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 976 {
 977         struct open_flags op;
 978         int fd = build_open_flags(flags, mode, &op);
 979         struct filename *tmp;
 980
 981         if (fd)
 982                 return fd;
 983
 984         tmp = getname(filename);
 985         if (IS_ERR(tmp))
 986                 return PTR_ERR(tmp);
 987
 988         fd = get_unused_fd_flags(flags);
 989         if (fd >= 0) {
 990                 struct file *f = do_filp_open(dfd, tmp, &op);
 991                 if (IS_ERR(f)) {
 992                         put_unused_fd(fd);
 993                         fd = PTR_ERR(f);
 994                 } else {
 995                         fsnotify_open(f);
 996                         fd_install(fd, f);
 997                 }
 998         }
 999         putname(tmp);
1000         return fd;
1001 }
1002
1003 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1004 {
1005         if (force_o_largefile())
1006                 flags |= O_LARGEFILE;
1007
1008         return do_sys_open(AT_FDCWD, filename, flags, mode);
1009 }
1010
1011 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1012                 umode_t, mode)
1013 {
1014         if (force_o_largefile())
1015                 flags |= O_LARGEFILE;
1016
1017         return do_sys_open(dfd, filename, flags, mode);
1018 }
1019
1020 #ifndef __alpha__
1021
1022 /*
1023  * For backward compatibility?  Maybe this should be moved
1024  * into arch/i386 instead?
1025  */
1026 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1027 {
1028         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1029 }
1030
1031 #endif
1032
1033 /*
1034  * "id" is the POSIX thread ID. We use the
1035  * files pointer for this..
1036  */
1037 int filp_close(struct file *filp, fl_owner_t id)
1038 {
1039         int retval = 0;
1040
1041         if (!file_count(filp)) {
1042                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1043                 return 0;
1044         }
1045
1046         if (filp->f_op->flush)
1047                 retval = filp->f_op->flush(filp, id);
1048
1049         if (likely(!(filp->f_mode & FMODE_PATH))) {
1050                 dnotify_flush(filp, id);
1051                 locks_remove_posix(filp, id);
1052         }
1053         fput(filp);
1054         return retval;
1055 }
1056
1057 EXPORT_SYMBOL(filp_close);
1058
1059 /*
1060  * Careful here! We test whether the file pointer is NULL before
1061  * releasing the fd. This ensures that one clone task can't release
1062  * an fd while another clone is opening it.
1063  */
1064 SYSCALL_DEFINE1(close, unsigned int, fd)
1065 {
1066         int retval = __close_fd(current->files, fd);
1067
1068         /* can't restart close syscall because file table entry was cleared */
1069         if (unlikely(retval == -ERESTARTSYS ||
1070                      retval == -ERESTARTNOINTR ||
1071                      retval == -ERESTARTNOHAND ||
1072                      retval == -ERESTART_RESTARTBLOCK))
1073                 retval = -EINTR;
1074
1075         return retval;
1076 }
1077 EXPORT_SYMBOL(sys_close);
1078
1079 /*
1080  * This routine simulates a hangup on the tty, to arrange that users
1081  * are given clean terminals at login time.
1082  */
1083 SYSCALL_DEFINE0(vhangup)
1084 {
1085         if (capable(CAP_SYS_TTY_CONFIG)) {
1086                 tty_vhangup_self();
1087                 return 0;
1088         }
1089         return -EPERM;
1090 }
1091
1092 /*
1093  * Called when an inode is about to be open.
1094  * We use this to disallow opening large files on 32bit systems if
1095  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1096  * on this flag in sys_open.
1097  */
1098 int generic_file_open(struct inode * inode, struct file * filp)
1099 {
1100         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1101                 return -EOVERFLOW;
1102         return 0;
1103 }
1104
1105 EXPORT_SYMBOL(generic_file_open);
1106
1107 /*
1108  * This is used by subsystems that don't want seekable
1109  * file descriptors. The function is not supposed to ever fail, the only
1110  * reason it returns an 'int' and not 'void' is so that it can be plugged
1111  * directly into file_operations structure.
1112  */
1113 int nonseekable_open(struct inode *inode, struct file *filp)
1114 {
1115         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1116         return 0;
1117 }
1118
1119 EXPORT_SYMBOL(nonseekable_open);