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, and file capabilities on truncate too */
  55         ret = dentry_needs_remove_privs(dentry);
  56         if (ret < 0)
  57                 return ret;
  58         if (ret)
  59                 newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61         mutex_lock(&dentry->d_inode->i_mutex);
  62         /* Note any delegations or leases have already been broken: */
  63         ret = notify_change(dentry, &newattrs, NULL);
  64         mutex_unlock(&dentry->d_inode->i_mutex);
  65         return ret;
  66 }
  67
  68 long vfs_truncate(struct path *path, loff_t length)
  69 {
  70         struct inode *inode;
  71         long error;
  72
  73         inode = path->dentry->d_inode;
  74
  75         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  76         if (S_ISDIR(inode->i_mode))
  77                 return -EISDIR;
  78         if (!S_ISREG(inode->i_mode))
  79                 return -EINVAL;
  80
  81         error = mnt_want_write(path->mnt);
  82         if (error)
  83                 goto out;
  84
  85         error = inode_permission(inode, MAY_WRITE);
  86         if (error)
  87                 goto mnt_drop_write_and_out;
  88
  89         error = -EPERM;
  90         if (IS_APPEND(inode))
  91                 goto mnt_drop_write_and_out;
  92
  93         error = get_write_access(inode);
  94         if (error)
  95                 goto mnt_drop_write_and_out;
  96
  97         /*
  98          * Make sure that there are no leases.  get_write_access() protects
  99          * against the truncate racing with a lease-granting setlease().
 100          */
 101         error = break_lease(inode, O_WRONLY);
 102         if (error)
 103                 goto put_write_and_out;
 104
 105         error = locks_verify_truncate(inode, NULL, length);
 106         if (!error)
 107                 error = security_path_truncate(path);
 108         if (!error)
 109                 error = do_truncate(path->dentry, length, 0, NULL);
 110
 111 put_write_and_out:
 112         put_write_access(inode);
 113 mnt_drop_write_and_out:
 114         mnt_drop_write(path->mnt);
 115 out:
 116         return error;
 117 }
 118 EXPORT_SYMBOL_GPL(vfs_truncate);
 119
 120 static long do_sys_truncate(const char __user *pathname, loff_t length)
 121 {
 122         unsigned int lookup_flags = LOOKUP_FOLLOW;
 123         struct path path;
 124         int error;
 125
 126         if (length < 0) /* sorry, but loff_t says... */
 127                 return -EINVAL;
 128
 129 retry:
 130         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 131         if (!error) {
 132                 error = vfs_truncate(&path, length);
 133                 path_put(&path);
 134         }
 135         if (retry_estale(error, lookup_flags)) {
 136                 lookup_flags |= LOOKUP_REVAL;
 137                 goto retry;
 138         }
 139         return error;
 140 }
 141
 142 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 143 {
 144         return do_sys_truncate(path, length);
 145 }
 146
 147 #ifdef CONFIG_COMPAT
 148 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 149 {
 150         return do_sys_truncate(path, length);
 151 }
 152 #endif
 153
 154 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 155 {
 156         struct inode *inode;
 157         struct dentry *dentry;
 158         struct fd f;
 159         int error;
 160
 161         error = -EINVAL;
 162         if (length < 0)
 163                 goto out;
 164         error = -EBADF;
 165         f = fdget(fd);
 166         if (!f.file)
 167                 goto out;
 168
 169         /* explicitly opened as large or we are on 64-bit box */
 170         if (f.file->f_flags & O_LARGEFILE)
 171                 small = 0;
 172
 173         dentry = f.file->f_path.dentry;
 174         inode = dentry->d_inode;
 175         error = -EINVAL;
 176         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 177                 goto out_putf;
 178
 179         error = -EINVAL;
 180         /* Cannot ftruncate over 2^31 bytes without large file support */
 181         if (small && length > MAX_NON_LFS)
 182                 goto out_putf;
 183
 184         error = -EPERM;
 185         if (IS_APPEND(inode))
 186                 goto out_putf;
 187
 188         sb_start_write(inode->i_sb);
 189         error = locks_verify_truncate(inode, f.file, length);
 190         if (!error)
 191                 error = security_path_truncate(&f.file->f_path);
 192         if (!error)
 193                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 194         sb_end_write(inode->i_sb);
 195 out_putf:
 196         fdput(f);
 197 out:
 198         return error;
 199 }
 200
 201 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 202 {
 203         return do_sys_ftruncate(fd, length, 1);
 204 }
 205
 206 #ifdef CONFIG_COMPAT
 207 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 208 {
 209         return do_sys_ftruncate(fd, length, 1);
 210 }
 211 #endif
 212
 213 /* LFS versions of truncate are only needed on 32 bit machines */
 214 #if BITS_PER_LONG == 32
 215 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 216 {
 217         return do_sys_truncate(path, length);
 218 }
 219
 220 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 221 {
 222         return do_sys_ftruncate(fd, length, 0);
 223 }
 224 #endif /* BITS_PER_LONG == 32 */
 225
 226
 227 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 228 {
 229         struct inode *inode = file_inode(file);
 230         long ret;
 231
 232         if (offset < 0 || len <= 0)
 233                 return -EINVAL;
 234
 235         /* Return error if mode is not supported */
 236         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 237                 return -EOPNOTSUPP;
 238
 239         /* Punch hole and zero range are mutually exclusive */
 240         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 241             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 242                 return -EOPNOTSUPP;
 243
 244         /* Punch hole must have keep size set */
 245         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 246             !(mode & FALLOC_FL_KEEP_SIZE))
 247                 return -EOPNOTSUPP;
 248
 249         /* Collapse range should only be used exclusively. */
 250         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 251             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 252                 return -EINVAL;
 253
 254         /* Insert range should only be used exclusively. */
 255         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 256             (mode & ~FALLOC_FL_INSERT_RANGE))
 257                 return -EINVAL;
 258
 259         if (!(file->f_mode & FMODE_WRITE))
 260                 return -EBADF;
 261
 262         /*
 263          * We can only allow pure fallocate on append only files
 264          */
 265         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 266                 return -EPERM;
 267
 268         if (IS_IMMUTABLE(inode))
 269                 return -EPERM;
 270
 271         /*
 272          * We cannot allow any fallocate operation on an active swapfile
 273          */
 274         if (IS_SWAPFILE(inode))
 275                 return -ETXTBSY;
 276
 277         /*
 278          * Revalidate the write permissions, in case security policy has
 279          * changed since the files were opened.
 280          */
 281         ret = security_file_permission(file, MAY_WRITE);
 282         if (ret)
 283                 return ret;
 284
 285         if (S_ISFIFO(inode->i_mode))
 286                 return -ESPIPE;
 287
 288         /*
 289          * Let individual file system decide if it supports preallocation
 290          * for directories or not.
 291          */
 292         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 293                 return -ENODEV;
 294
 295         /* Check for wrap through zero too */
 296         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 297                 return -EFBIG;
 298
 299         if (!file->f_op->fallocate)
 300                 return -EOPNOTSUPP;
 301
 302         sb_start_write(inode->i_sb);
 303         ret = file->f_op->fallocate(file, mode, offset, len);
 304
 305         /*
 306          * Create inotify and fanotify events.
 307          *
 308          * To keep the logic simple always create events if fallocate succeeds.
 309          * This implies that events are even created if the file size remains
 310          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 311          */
 312         if (ret == 0)
 313                 fsnotify_modify(file);
 314
 315         sb_end_write(inode->i_sb);
 316         return ret;
 317 }
 318 EXPORT_SYMBOL_GPL(vfs_fallocate);
 319
 320 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 321 {
 322         struct fd f = fdget(fd);
 323         int error = -EBADF;
 324
 325         if (f.file) {
 326                 error = vfs_fallocate(f.file, mode, offset, len);
 327                 fdput(f);
 328         }
 329         return error;
 330 }
 331
 332 /*
 333  * access() needs to use the real uid/gid, not the effective uid/gid.
 334  * We do this by temporarily clearing all FS-related capabilities and
 335  * switching the fsuid/fsgid around to the real ones.
 336  */
 337 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 338 {
 339         const struct cred *old_cred;
 340         struct cred *override_cred;
 341         struct path path;
 342         struct inode *inode;
 343         int res;
 344         unsigned int lookup_flags = LOOKUP_FOLLOW;
 345
 346         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 347                 return -EINVAL;
 348
 349         override_cred = prepare_creds();
 350         if (!override_cred)
 351                 return -ENOMEM;
 352
 353         override_cred->fsuid = override_cred->uid;
 354         override_cred->fsgid = override_cred->gid;
 355
 356         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 357                 /* Clear the capabilities if we switch to a non-root user */
 358                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 359                 if (!uid_eq(override_cred->uid, root_uid))
 360                         cap_clear(override_cred->cap_effective);
 361                 else
 362                         override_cred->cap_effective =
 363                                 override_cred->cap_permitted;
 364         }
 365
 366         /*
 367          * The new set of credentials can *only* be used in
 368          * task-synchronous circumstances, and does not need
 369          * RCU freeing, unless somebody then takes a separate
 370          * reference to it.
 371          *
 372          * NOTE! This is _only_ true because this credential
 373          * is used purely for override_creds() that installs
 374          * it as the subjective cred. Other threads will be
 375          * accessing ->real_cred, not the subjective cred.
 376          *
 377          * If somebody _does_ make a copy of this (using the
 378          * 'get_current_cred()' function), that will clear the
 379          * non_rcu field, because now that other user may be
 380          * expecting RCU freeing. But normal thread-synchronous
 381          * cred accesses will keep things non-RCY.
 382          */
 383         override_cred->non_rcu = 1;
 384
 385         old_cred = override_creds(override_cred);
 386 retry:
 387         res = user_path_at(dfd, filename, lookup_flags, &path);
 388         if (res)
 389                 goto out;
 390
 391         inode = d_backing_inode(path.dentry);
 392
 393         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 394                 /*
 395                  * MAY_EXEC on regular files is denied if the fs is mounted
 396                  * with the "noexec" flag.
 397                  */
 398                 res = -EACCES;
 399                 if (path_noexec(&path))
 400                         goto out_path_release;
 401         }
 402
 403         res = inode_permission(inode, mode | MAY_ACCESS);
 404         /* SuS v2 requires we report a read only fs too */
 405         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 406                 goto out_path_release;
 407         /*
 408          * This is a rare case where using __mnt_is_readonly()
 409          * is OK without a mnt_want/drop_write() pair.  Since
 410          * no actual write to the fs is performed here, we do
 411          * not need to telegraph to that to anyone.
 412          *
 413          * By doing this, we accept that this access is
 414          * inherently racy and know that the fs may change
 415          * state before we even see this result.
 416          */
 417         if (__mnt_is_readonly(path.mnt))
 418                 res = -EROFS;
 419
 420 out_path_release:
 421         path_put(&path);
 422         if (retry_estale(res, lookup_flags)) {
 423                 lookup_flags |= LOOKUP_REVAL;
 424                 goto retry;
 425         }
 426 out:
 427         revert_creds(old_cred);
 428         put_cred(override_cred);
 429         return res;
 430 }
 431
 432 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 433 {
 434         return sys_faccessat(AT_FDCWD, filename, mode);
 435 }
 436
 437 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 438 {
 439         struct path path;
 440         int error;
 441         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 442 retry:
 443         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 444         if (error)
 445                 goto out;
 446
 447         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 448         if (error)
 449                 goto dput_and_out;
 450
 451         set_fs_pwd(current->fs, &path);
 452
 453 dput_and_out:
 454         path_put(&path);
 455         if (retry_estale(error, lookup_flags)) {
 456                 lookup_flags |= LOOKUP_REVAL;
 457                 goto retry;
 458         }
 459 out:
 460         return error;
 461 }
 462
 463 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 464 {
 465         struct fd f = fdget_raw(fd);
 466         struct inode *inode;
 467         int error = -EBADF;
 468
 469         error = -EBADF;
 470         if (!f.file)
 471                 goto out;
 472
 473         inode = file_inode(f.file);
 474
 475         error = -ENOTDIR;
 476         if (!S_ISDIR(inode->i_mode))
 477                 goto out_putf;
 478
 479         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 480         if (!error)
 481                 set_fs_pwd(current->fs, &f.file->f_path);
 482 out_putf:
 483         fdput(f);
 484 out:
 485         return error;
 486 }
 487
 488 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 489 {
 490         struct path path;
 491         int error;
 492         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 493 retry:
 494         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 495         if (error)
 496                 goto out;
 497
 498         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 499         if (error)
 500                 goto dput_and_out;
 501
 502         error = -EPERM;
 503         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 504                 goto dput_and_out;
 505         error = security_path_chroot(&path);
 506         if (error)
 507                 goto dput_and_out;
 508
 509         set_fs_root(current->fs, &path);
 510         error = 0;
 511 dput_and_out:
 512         path_put(&path);
 513         if (retry_estale(error, lookup_flags)) {
 514                 lookup_flags |= LOOKUP_REVAL;
 515                 goto retry;
 516         }
 517 out:
 518         return error;
 519 }
 520
 521 static int chmod_common(struct path *path, umode_t mode)
 522 {
 523         struct inode *inode = path->dentry->d_inode;
 524         struct inode *delegated_inode = NULL;
 525         struct iattr newattrs;
 526         int error;
 527
 528         error = mnt_want_write(path->mnt);
 529         if (error)
 530                 return error;
 531 retry_deleg:
 532         mutex_lock(&inode->i_mutex);
 533         error = security_path_chmod(path, mode);
 534         if (error)
 535                 goto out_unlock;
 536         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 537         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 538         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 539 out_unlock:
 540         mutex_unlock(&inode->i_mutex);
 541         if (delegated_inode) {
 542                 error = break_deleg_wait(&delegated_inode);
 543                 if (!error)
 544                         goto retry_deleg;
 545         }
 546         mnt_drop_write(path->mnt);
 547         return error;
 548 }
 549
 550 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 551 {
 552         struct fd f = fdget(fd);
 553         int err = -EBADF;
 554
 555         if (f.file) {
 556                 audit_file(f.file);
 557                 err = chmod_common(&f.file->f_path, mode);
 558                 fdput(f);
 559         }
 560         return err;
 561 }
 562
 563 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 564 {
 565         struct path path;
 566         int error;
 567         unsigned int lookup_flags = LOOKUP_FOLLOW;
 568 retry:
 569         error = user_path_at(dfd, filename, lookup_flags, &path);
 570         if (!error) {
 571                 error = chmod_common(&path, mode);
 572                 path_put(&path);
 573                 if (retry_estale(error, lookup_flags)) {
 574                         lookup_flags |= LOOKUP_REVAL;
 575                         goto retry;
 576                 }
 577         }
 578         return error;
 579 }
 580
 581 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 582 {
 583         return sys_fchmodat(AT_FDCWD, filename, mode);
 584 }
 585
 586 static int chown_common(struct path *path, uid_t user, gid_t group)
 587 {
 588         struct inode *inode = path->dentry->d_inode;
 589         struct inode *delegated_inode = NULL;
 590         int error;
 591         struct iattr newattrs;
 592         kuid_t uid;
 593         kgid_t gid;
 594
 595         uid = make_kuid(current_user_ns(), user);
 596         gid = make_kgid(current_user_ns(), group);
 597
 598 retry_deleg:
 599         newattrs.ia_valid =  ATTR_CTIME;
 600         if (user != (uid_t) -1) {
 601                 if (!uid_valid(uid))
 602                         return -EINVAL;
 603                 newattrs.ia_valid |= ATTR_UID;
 604                 newattrs.ia_uid = uid;
 605         }
 606         if (group != (gid_t) -1) {
 607                 if (!gid_valid(gid))
 608                         return -EINVAL;
 609                 newattrs.ia_valid |= ATTR_GID;
 610                 newattrs.ia_gid = gid;
 611         }
 612         if (!S_ISDIR(inode->i_mode))
 613                 newattrs.ia_valid |=
 614                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 615         mutex_lock(&inode->i_mutex);
 616         error = security_path_chown(path, uid, gid);
 617         if (!error)
 618                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 619         mutex_unlock(&inode->i_mutex);
 620         if (delegated_inode) {
 621                 error = break_deleg_wait(&delegated_inode);
 622                 if (!error)
 623                         goto retry_deleg;
 624         }
 625         return error;
 626 }
 627
 628 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 629                 gid_t, group, int, flag)
 630 {
 631         struct path path;
 632         int error = -EINVAL;
 633         int lookup_flags;
 634
 635         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 636                 goto out;
 637
 638         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 639         if (flag & AT_EMPTY_PATH)
 640                 lookup_flags |= LOOKUP_EMPTY;
 641 retry:
 642         error = user_path_at(dfd, filename, lookup_flags, &path);
 643         if (error)
 644                 goto out;
 645         error = mnt_want_write(path.mnt);
 646         if (error)
 647                 goto out_release;
 648         error = chown_common(&path, user, group);
 649         mnt_drop_write(path.mnt);
 650 out_release:
 651         path_put(&path);
 652         if (retry_estale(error, lookup_flags)) {
 653                 lookup_flags |= LOOKUP_REVAL;
 654                 goto retry;
 655         }
 656 out:
 657         return error;
 658 }
 659
 660 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 661 {
 662         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 663 }
 664
 665 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 666 {
 667         return sys_fchownat(AT_FDCWD, filename, user, group,
 668                             AT_SYMLINK_NOFOLLOW);
 669 }
 670
 671 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 672 {
 673         struct fd f = fdget(fd);
 674         int error = -EBADF;
 675
 676         if (!f.file)
 677                 goto out;
 678
 679         error = mnt_want_write_file(f.file);
 680         if (error)
 681                 goto out_fput;
 682         audit_file(f.file);
 683         error = chown_common(&f.file->f_path, user, group);
 684         mnt_drop_write_file(f.file);
 685 out_fput:
 686         fdput(f);
 687 out:
 688         return error;
 689 }
 690
 691 int open_check_o_direct(struct file *f)
 692 {
 693         /* NB: we're sure to have correct a_ops only after f_op->open */
 694         if (f->f_flags & O_DIRECT) {
 695                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 696                         return -EINVAL;
 697         }
 698         return 0;
 699 }
 700
 701 static int do_dentry_open(struct file *f,
 702                           struct inode *inode,
 703                           int (*open)(struct inode *, struct file *),
 704                           const struct cred *cred)
 705 {
 706         static const struct file_operations empty_fops = {};
 707         int error;
 708
 709         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 710                                 FMODE_PREAD | FMODE_PWRITE;
 711
 712         path_get(&f->f_path);
 713         f->f_inode = inode;
 714         f->f_mapping = inode->i_mapping;
 715
 716         if (unlikely(f->f_flags & O_PATH)) {
 717                 f->f_mode = FMODE_PATH;
 718                 f->f_op = &empty_fops;
 719                 return 0;
 720         }
 721
 722         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 723                 error = get_write_access(inode);
 724                 if (unlikely(error))
 725                         goto cleanup_file;
 726                 error = __mnt_want_write(f->f_path.mnt);
 727                 if (unlikely(error)) {
 728                         put_write_access(inode);
 729                         goto cleanup_file;
 730                 }
 731                 f->f_mode |= FMODE_WRITER;
 732         }
 733
 734         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 735         if (S_ISREG(inode->i_mode))
 736                 f->f_mode |= FMODE_ATOMIC_POS;
 737
 738         f->f_op = fops_get(inode->i_fop);
 739         if (unlikely(WARN_ON(!f->f_op))) {
 740                 error = -ENODEV;
 741                 goto cleanup_all;
 742         }
 743
 744         error = security_file_open(f, cred);
 745         if (error)
 746                 goto cleanup_all;
 747
 748         error = break_lease(inode, f->f_flags);
 749         if (error)
 750                 goto cleanup_all;
 751
 752         if (!open)
 753                 open = f->f_op->open;
 754         if (open) {
 755                 error = open(inode, f);
 756                 if (error)
 757                         goto cleanup_all;
 758         }
 759         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 760                 i_readcount_inc(inode);
 761         if ((f->f_mode & FMODE_READ) &&
 762              likely(f->f_op->read || f->f_op->read_iter))
 763                 f->f_mode |= FMODE_CAN_READ;
 764         if ((f->f_mode & FMODE_WRITE) &&
 765              likely(f->f_op->write || f->f_op->write_iter))
 766                 f->f_mode |= FMODE_CAN_WRITE;
 767
 768         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 769
 770         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 771
 772         return 0;
 773
 774 cleanup_all:
 775         fops_put(f->f_op);
 776         if (f->f_mode & FMODE_WRITER) {
 777                 put_write_access(inode);
 778                 __mnt_drop_write(f->f_path.mnt);
 779         }
 780 cleanup_file:
 781         path_put(&f->f_path);
 782         f->f_path.mnt = NULL;
 783         f->f_path.dentry = NULL;
 784         f->f_inode = NULL;
 785         return error;
 786 }
 787
 788 /**
 789  * finish_open - finish opening a file
 790  * @file: file pointer
 791  * @dentry: pointer to dentry
 792  * @open: open callback
 793  * @opened: state of open
 794  *
 795  * This can be used to finish opening a file passed to i_op->atomic_open().
 796  *
 797  * If the open callback is set to NULL, then the standard f_op->open()
 798  * filesystem callback is substituted.
 799  *
 800  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 801  * the return value of d_splice_alias(), then the caller needs to perform dput()
 802  * on it after finish_open().
 803  *
 804  * On successful return @file is a fully instantiated open file.  After this, if
 805  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 806  *
 807  * Returns zero on success or -errno if the open failed.
 808  */
 809 int finish_open(struct file *file, struct dentry *dentry,
 810                 int (*open)(struct inode *, struct file *),
 811                 int *opened)
 812 {
 813         int error;
 814         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 815
 816         file->f_path.dentry = dentry;
 817         error = do_dentry_open(file, d_backing_inode(dentry), open,
 818                                current_cred());
 819         if (!error)
 820                 *opened |= FILE_OPENED;
 821
 822         return error;
 823 }
 824 EXPORT_SYMBOL(finish_open);
 825
 826 /**
 827  * finish_no_open - finish ->atomic_open() without opening the file
 828  *
 829  * @file: file pointer
 830  * @dentry: dentry or NULL (as returned from ->lookup())
 831  *
 832  * This can be used to set the result of a successful lookup in ->atomic_open().
 833  *
 834  * NB: unlike finish_open() this function does consume the dentry reference and
 835  * the caller need not dput() it.
 836  *
 837  * Returns "1" which must be the return value of ->atomic_open() after having
 838  * called this function.
 839  */
 840 int finish_no_open(struct file *file, struct dentry *dentry)
 841 {
 842         file->f_path.dentry = dentry;
 843         return 1;
 844 }
 845 EXPORT_SYMBOL(finish_no_open);
 846
 847 char *file_path(struct file *filp, char *buf, int buflen)
 848 {
 849         return d_path(&filp->f_path, buf, buflen);
 850 }
 851 EXPORT_SYMBOL(file_path);
 852
 853 /**
 854  * vfs_open - open the file at the given path
 855  * @path: path to open
 856  * @file: newly allocated file with f_flag initialized
 857  * @cred: credentials to use
 858  */
 859 int vfs_open(const struct path *path, struct file *file,
 860              const struct cred *cred)
 861 {
 862         struct inode *inode = vfs_select_inode(path->dentry, file->f_flags);
 863
 864         if (IS_ERR(inode))
 865                 return PTR_ERR(inode);
 866
 867         file->f_path = *path;
 868         return do_dentry_open(file, inode, NULL, cred);
 869 }
 870
 871 struct file *dentry_open(const struct path *path, int flags,
 872                          const struct cred *cred)
 873 {
 874         int error;
 875         struct file *f;
 876
 877         validate_creds(cred);
 878
 879         /* We must always pass in a valid mount pointer. */
 880         BUG_ON(!path->mnt);
 881
 882         f = get_empty_filp();
 883         if (!IS_ERR(f)) {
 884                 f->f_flags = flags;
 885                 error = vfs_open(path, f, cred);
 886                 if (!error) {
 887                         /* from now on we need fput() to dispose of f */
 888                         error = open_check_o_direct(f);
 889                         if (error) {
 890                                 fput(f);
 891                                 f = ERR_PTR(error);
 892                         }
 893                 } else {
 894                         put_filp(f);
 895                         f = ERR_PTR(error);
 896                 }
 897         }
 898         return f;
 899 }
 900 EXPORT_SYMBOL(dentry_open);
 901
 902 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 903 {
 904         int lookup_flags = 0;
 905         int acc_mode;
 906
 907         /*
 908          * Clear out all open flags we don't know about so that we don't report
 909          * them in fcntl(F_GETFD) or similar interfaces.
 910          */
 911         flags &= VALID_OPEN_FLAGS;
 912
 913         if (flags & (O_CREAT | __O_TMPFILE))
 914                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 915         else
 916                 op->mode = 0;
 917
 918         /* Must never be set by userspace */
 919         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 920
 921         /*
 922          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 923          * check for O_DSYNC if the need any syncing at all we enforce it's
 924          * always set instead of having to deal with possibly weird behaviour
 925          * for malicious applications setting only __O_SYNC.
 926          */
 927         if (flags & __O_SYNC)
 928                 flags |= O_DSYNC;
 929
 930         if (flags & __O_TMPFILE) {
 931                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 932                         return -EINVAL;
 933                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 934                 if (!(acc_mode & MAY_WRITE))
 935                         return -EINVAL;
 936         } else if (flags & O_PATH) {
 937                 /*
 938                  * If we have O_PATH in the open flag. Then we
 939                  * cannot have anything other than the below set of flags
 940                  */
 941                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 942                 acc_mode = 0;
 943         } else {
 944                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 945         }
 946
 947         op->open_flag = flags;
 948
 949         /* O_TRUNC implies we need access checks for write permissions */
 950         if (flags & O_TRUNC)
 951                 acc_mode |= MAY_WRITE;
 952
 953         /* Allow the LSM permission hook to distinguish append
 954            access from general write access. */
 955         if (flags & O_APPEND)
 956                 acc_mode |= MAY_APPEND;
 957
 958         op->acc_mode = acc_mode;
 959
 960         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 961
 962         if (flags & O_CREAT) {
 963                 op->intent |= LOOKUP_CREATE;
 964                 if (flags & O_EXCL)
 965                         op->intent |= LOOKUP_EXCL;
 966         }
 967
 968         if (flags & O_DIRECTORY)
 969                 lookup_flags |= LOOKUP_DIRECTORY;
 970         if (!(flags & O_NOFOLLOW))
 971                 lookup_flags |= LOOKUP_FOLLOW;
 972         op->lookup_flags = lookup_flags;
 973         return 0;
 974 }
 975
 976 /**
 977  * file_open_name - open file and return file pointer
 978  *
 979  * @name:       struct filename containing path to open
 980  * @flags:      open flags as per the open(2) second argument
 981  * @mode:       mode for the new file if O_CREAT is set, else ignored
 982  *
 983  * This is the helper to open a file from kernelspace if you really
 984  * have to.  But in generally you should not do this, so please move
 985  * along, nothing to see here..
 986  */
 987 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 988 {
 989         struct open_flags op;
 990         int err = build_open_flags(flags, mode, &op);
 991         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 992 }
 993
 994 /**
 995  * filp_open - open file and return file pointer
 996  *
 997  * @filename:   path to open
 998  * @flags:      open flags as per the open(2) second argument
 999  * @mode:       mode for the new file if O_CREAT is set, else ignored
1000  *
1001  * This is the helper to open a file from kernelspace if you really
1002  * have to.  But in generally you should not do this, so please move
1003  * along, nothing to see here..
1004  */
1005 struct file *filp_open(const char *filename, int flags, umode_t mode)
1006 {
1007         struct filename *name = getname_kernel(filename);
1008         struct file *file = ERR_CAST(name);
1009
1010         if (!IS_ERR(name)) {
1011                 file = file_open_name(name, flags, mode);
1012                 putname(name);
1013         }
1014         return file;
1015 }
1016 EXPORT_SYMBOL(filp_open);
1017
1018 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1019                             const char *filename, int flags, umode_t mode)
1020 {
1021         struct open_flags op;
1022         int err = build_open_flags(flags, mode, &op);
1023         if (err)
1024                 return ERR_PTR(err);
1025         return do_file_open_root(dentry, mnt, filename, &op);
1026 }
1027 EXPORT_SYMBOL(file_open_root);
1028
1029 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1030 {
1031         struct open_flags op;
1032         int fd = build_open_flags(flags, mode, &op);
1033         struct filename *tmp;
1034
1035         if (fd)
1036                 return fd;
1037
1038         tmp = getname(filename);
1039         if (IS_ERR(tmp))
1040                 return PTR_ERR(tmp);
1041
1042         fd = get_unused_fd_flags(flags);
1043         if (fd >= 0) {
1044                 struct file *f = do_filp_open(dfd, tmp, &op);
1045                 if (IS_ERR(f)) {
1046                         put_unused_fd(fd);
1047                         fd = PTR_ERR(f);
1048                 } else {
1049                         fsnotify_open(f);
1050                         fd_install(fd, f);
1051                 }
1052         }
1053         putname(tmp);
1054         return fd;
1055 }
1056
1057 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1058 {
1059         if (force_o_largefile())
1060                 flags |= O_LARGEFILE;
1061
1062         return do_sys_open(AT_FDCWD, filename, flags, mode);
1063 }
1064
1065 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1066                 umode_t, mode)
1067 {
1068         if (force_o_largefile())
1069                 flags |= O_LARGEFILE;
1070
1071         return do_sys_open(dfd, filename, flags, mode);
1072 }
1073
1074 #ifndef __alpha__
1075
1076 /*
1077  * For backward compatibility?  Maybe this should be moved
1078  * into arch/i386 instead?
1079  */
1080 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1081 {
1082         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1083 }
1084
1085 #endif
1086
1087 /*
1088  * "id" is the POSIX thread ID. We use the
1089  * files pointer for this..
1090  */
1091 int filp_close(struct file *filp, fl_owner_t id)
1092 {
1093         int retval = 0;
1094
1095         if (!file_count(filp)) {
1096                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1097                 return 0;
1098         }
1099
1100         if (filp->f_op->flush)
1101                 retval = filp->f_op->flush(filp, id);
1102
1103         if (likely(!(filp->f_mode & FMODE_PATH))) {
1104                 dnotify_flush(filp, id);
1105                 locks_remove_posix(filp, id);
1106         }
1107         fput(filp);
1108         return retval;
1109 }
1110
1111 EXPORT_SYMBOL(filp_close);
1112
1113 /*
1114  * Careful here! We test whether the file pointer is NULL before
1115  * releasing the fd. This ensures that one clone task can't release
1116  * an fd while another clone is opening it.
1117  */
1118 SYSCALL_DEFINE1(close, unsigned int, fd)
1119 {
1120         int retval = __close_fd(current->files, fd);
1121
1122         /* can't restart close syscall because file table entry was cleared */
1123         if (unlikely(retval == -ERESTARTSYS ||
1124                      retval == -ERESTARTNOINTR ||
1125                      retval == -ERESTARTNOHAND ||
1126                      retval == -ERESTART_RESTARTBLOCK))
1127                 retval = -EINTR;
1128
1129         return retval;
1130 }
1131 EXPORT_SYMBOL(sys_close);
1132
1133 /*
1134  * This routine simulates a hangup on the tty, to arrange that users
1135  * are given clean terminals at login time.
1136  */
1137 SYSCALL_DEFINE0(vhangup)
1138 {
1139         if (capable(CAP_SYS_TTY_CONFIG)) {
1140                 tty_vhangup_self();
1141                 return 0;
1142         }
1143         return -EPERM;
1144 }
1145
1146 /*
1147  * Called when an inode is about to be open.
1148  * We use this to disallow opening large files on 32bit systems if
1149  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1150  * on this flag in sys_open.
1151  */
1152 int generic_file_open(struct inode * inode, struct file * filp)
1153 {
1154         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1155                 return -EOVERFLOW;
1156         return 0;
1157 }
1158
1159 EXPORT_SYMBOL(generic_file_open);
1160
1161 /*
1162  * This is used by subsystems that don't want seekable
1163  * file descriptors. The function is not supposed to ever fail, the only
1164  * reason it returns an 'int' and not 'void' is so that it can be plugged
1165  * directly into file_operations structure.
1166  */
1167 int nonseekable_open(struct inode *inode, struct file *filp)
1168 {
1169         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1170         return 0;
1171 }
1172
1173 EXPORT_SYMBOL(nonseekable_open);
1174
1175 /*
1176  * stream_open is used by subsystems that want stream-like file descriptors.
1177  * Such file descriptors are not seekable and don't have notion of position
1178  * (file.f_pos is always 0). Contrary to file descriptors of other regular
1179  * files, .read() and .write() can run simultaneously.
1180  *
1181  * stream_open never fails and is marked to return int so that it could be
1182  * directly used as file_operations.open .
1183  */
1184 int stream_open(struct inode *inode, struct file *filp)
1185 {
1186         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1187         filp->f_mode |= FMODE_STREAM;
1188         return 0;
1189 }
1190
1191 EXPORT_SYMBOL(stream_open);