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