fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33
  34 #include "internal.h"
  35
  36 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  37         struct file *filp)
  38 {
  39         int ret;
  40         struct iattr newattrs;
  41
  42         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  43         if (length < 0)
  44                 return -EINVAL;
  45
  46         newattrs.ia_size = length;
  47         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  48         if (filp) {
  49                 newattrs.ia_file = filp;
  50                 newattrs.ia_valid |= ATTR_FILE;
  51         }
  52
  53         /* Remove suid/sgid on truncate too */
  54         ret = should_remove_suid(dentry);
  55         if (ret)
  56                 newattrs.ia_valid |= ret | ATTR_FORCE;
  57
  58         mutex_lock(&dentry->d_inode->i_mutex);
  59         ret = notify_change(dentry, &newattrs);
  60         mutex_unlock(&dentry->d_inode->i_mutex);
  61         return ret;
  62 }
  63
  64 static long do_sys_truncate(const char __user *pathname, loff_t length)
  65 {
  66         struct path path;
  67         struct inode *inode;
  68         int error;
  69
  70         error = -EINVAL;
  71         if (length < 0) /* sorry, but loff_t says... */
  72                 goto out;
  73
  74         error = user_path(pathname, &path);
  75         if (error)
  76                 goto out;
  77         inode = path.dentry->d_inode;
  78
  79         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  80         error = -EISDIR;
  81         if (S_ISDIR(inode->i_mode))
  82                 goto dput_and_out;
  83
  84         error = -EINVAL;
  85         if (!S_ISREG(inode->i_mode))
  86                 goto dput_and_out;
  87
  88         error = mnt_want_write(path.mnt);
  89         if (error)
  90                 goto dput_and_out;
  91
  92         error = inode_permission(inode, MAY_WRITE);
  93         if (error)
  94                 goto mnt_drop_write_and_out;
  95
  96         error = -EPERM;
  97         if (IS_APPEND(inode))
  98                 goto mnt_drop_write_and_out;
  99
 100         error = get_write_access(inode);
 101         if (error)
 102                 goto mnt_drop_write_and_out;
 103
 104         /*
 105          * Make sure that there are no leases.  get_write_access() protects
 106          * against the truncate racing with a lease-granting setlease().
 107          */
 108         error = break_lease(inode, O_WRONLY);
 109         if (error)
 110                 goto put_write_and_out;
 111
 112         error = locks_verify_truncate(inode, NULL, length);
 113         if (!error)
 114                 error = security_path_truncate(&path);
 115         if (!error)
 116                 error = do_truncate(path.dentry, length, 0, NULL);
 117
 118 put_write_and_out:
 119         put_write_access(inode);
 120 mnt_drop_write_and_out:
 121         mnt_drop_write(path.mnt);
 122 dput_and_out:
 123         path_put(&path);
 124 out:
 125         return error;
 126 }
 127
 128 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 129 {
 130         return do_sys_truncate(path, length);
 131 }
 132
 133 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 134 {
 135         struct inode * inode;
 136         struct dentry *dentry;
 137         struct file * file;
 138         int error;
 139
 140         error = -EINVAL;
 141         if (length < 0)
 142                 goto out;
 143         error = -EBADF;
 144         file = fget(fd);
 145         if (!file)
 146                 goto out;
 147
 148         /* explicitly opened as large or we are on 64-bit box */
 149         if (file->f_flags & O_LARGEFILE)
 150                 small = 0;
 151
 152         dentry = file->f_path.dentry;
 153         inode = dentry->d_inode;
 154         error = -EINVAL;
 155         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 156                 goto out_putf;
 157
 158         error = -EINVAL;
 159         /* Cannot ftruncate over 2^31 bytes without large file support */
 160         if (small && length > MAX_NON_LFS)
 161                 goto out_putf;
 162
 163         error = -EPERM;
 164         if (IS_APPEND(inode))
 165                 goto out_putf;
 166
 167         error = locks_verify_truncate(inode, file, length);
 168         if (!error)
 169                 error = security_path_truncate(&file->f_path);
 170         if (!error)
 171                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 172 out_putf:
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177
 178 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 179 {
 180         long ret = do_sys_ftruncate(fd, length, 1);
 181         /* avoid REGPARM breakage on x86: */
 182         asmlinkage_protect(2, ret, fd, length);
 183         return ret;
 184 }
 185
 186 /* LFS versions of truncate are only needed on 32 bit machines */
 187 #if BITS_PER_LONG == 32
 188 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 189 {
 190         return do_sys_truncate(path, length);
 191 }
 192 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 193 asmlinkage long SyS_truncate64(long path, loff_t length)
 194 {
 195         return SYSC_truncate64((const char __user *) path, length);
 196 }
 197 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 198 #endif
 199
 200 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 201 {
 202         long ret = do_sys_ftruncate(fd, length, 0);
 203         /* avoid REGPARM breakage on x86: */
 204         asmlinkage_protect(2, ret, fd, length);
 205         return ret;
 206 }
 207 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 208 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 209 {
 210         return SYSC_ftruncate64((unsigned int) fd, length);
 211 }
 212 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 213 #endif
 214 #endif /* BITS_PER_LONG == 32 */
 215
 216
 217 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 218 {
 219         struct inode *inode = file->f_path.dentry->d_inode;
 220         long ret;
 221
 222         if (offset < 0 || len <= 0)
 223                 return -EINVAL;
 224
 225         /* Return error if mode is not supported */
 226         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 227                 return -EOPNOTSUPP;
 228
 229         /* Punch hole must have keep size set */
 230         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 231             !(mode & FALLOC_FL_KEEP_SIZE))
 232                 return -EOPNOTSUPP;
 233
 234         if (!(file->f_mode & FMODE_WRITE))
 235                 return -EBADF;
 236
 237         /* It's not possible punch hole on append only file */
 238         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 239                 return -EPERM;
 240
 241         if (IS_IMMUTABLE(inode))
 242                 return -EPERM;
 243
 244         /*
 245          * Revalidate the write permissions, in case security policy has
 246          * changed since the files were opened.
 247          */
 248         ret = security_file_permission(file, MAY_WRITE);
 249         if (ret)
 250                 return ret;
 251
 252         if (S_ISFIFO(inode->i_mode))
 253                 return -ESPIPE;
 254
 255         /*
 256          * Let individual file system decide if it supports preallocation
 257          * for directories or not.
 258          */
 259         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 260                 return -ENODEV;
 261
 262         /* Check for wrap through zero too */
 263         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 264                 return -EFBIG;
 265
 266         if (!file->f_op->fallocate)
 267                 return -EOPNOTSUPP;
 268
 269         return file->f_op->fallocate(file, mode, offset, len);
 270 }
 271
 272 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 273 {
 274         struct file *file;
 275         int error = -EBADF;
 276
 277         file = fget(fd);
 278         if (file) {
 279                 error = do_fallocate(file, mode, offset, len);
 280                 fput(file);
 281         }
 282
 283         return error;
 284 }
 285
 286 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 287 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 288 {
 289         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 290 }
 291 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 292 #endif
 293
 294 /*
 295  * access() needs to use the real uid/gid, not the effective uid/gid.
 296  * We do this by temporarily clearing all FS-related capabilities and
 297  * switching the fsuid/fsgid around to the real ones.
 298  */
 299 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 300 {
 301         const struct cred *old_cred;
 302         struct cred *override_cred;
 303         struct path path;
 304         struct inode *inode;
 305         int res;
 306
 307         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 308                 return -EINVAL;
 309
 310         override_cred = prepare_creds();
 311         if (!override_cred)
 312                 return -ENOMEM;
 313
 314         override_cred->fsuid = override_cred->uid;
 315         override_cred->fsgid = override_cred->gid;
 316
 317         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 318                 /* Clear the capabilities if we switch to a non-root user */
 319                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 320                 if (!uid_eq(override_cred->uid, root_uid))
 321                         cap_clear(override_cred->cap_effective);
 322                 else
 323                         override_cred->cap_effective =
 324                                 override_cred->cap_permitted;
 325         }
 326
 327         old_cred = override_creds(override_cred);
 328
 329         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 330         if (res)
 331                 goto out;
 332
 333         inode = path.dentry->d_inode;
 334
 335         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 336                 /*
 337                  * MAY_EXEC on regular files is denied if the fs is mounted
 338                  * with the "noexec" flag.
 339                  */
 340                 res = -EACCES;
 341                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 342                         goto out_path_release;
 343         }
 344
 345         res = inode_permission(inode, mode | MAY_ACCESS);
 346         /* SuS v2 requires we report a read only fs too */
 347         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 348                 goto out_path_release;
 349         /*
 350          * This is a rare case where using __mnt_is_readonly()
 351          * is OK without a mnt_want/drop_write() pair.  Since
 352          * no actual write to the fs is performed here, we do
 353          * not need to telegraph to that to anyone.
 354          *
 355          * By doing this, we accept that this access is
 356          * inherently racy and know that the fs may change
 357          * state before we even see this result.
 358          */
 359         if (__mnt_is_readonly(path.mnt))
 360                 res = -EROFS;
 361
 362 out_path_release:
 363         path_put(&path);
 364 out:
 365         revert_creds(old_cred);
 366         put_cred(override_cred);
 367         return res;
 368 }
 369
 370 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 371 {
 372         return sys_faccessat(AT_FDCWD, filename, mode);
 373 }
 374
 375 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 376 {
 377         struct path path;
 378         int error;
 379
 380         error = user_path_dir(filename, &path);
 381         if (error)
 382                 goto out;
 383
 384         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 385         if (error)
 386                 goto dput_and_out;
 387
 388         set_fs_pwd(current->fs, &path);
 389
 390 dput_and_out:
 391         path_put(&path);
 392 out:
 393         return error;
 394 }
 395
 396 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 397 {
 398         struct file *file;
 399         struct inode *inode;
 400         int error, fput_needed;
 401
 402         error = -EBADF;
 403         file = fget_raw_light(fd, &fput_needed);
 404         if (!file)
 405                 goto out;
 406
 407         inode = file->f_path.dentry->d_inode;
 408
 409         error = -ENOTDIR;
 410         if (!S_ISDIR(inode->i_mode))
 411                 goto out_putf;
 412
 413         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 414         if (!error)
 415                 set_fs_pwd(current->fs, &file->f_path);
 416 out_putf:
 417         fput_light(file, fput_needed);
 418 out:
 419         return error;
 420 }
 421
 422 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 423 {
 424         struct path path;
 425         int error;
 426
 427         error = user_path_dir(filename, &path);
 428         if (error)
 429                 goto out;
 430
 431         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 432         if (error)
 433                 goto dput_and_out;
 434
 435         error = -EPERM;
 436         if (!capable(CAP_SYS_CHROOT))
 437                 goto dput_and_out;
 438         error = security_path_chroot(&path);
 439         if (error)
 440                 goto dput_and_out;
 441
 442         set_fs_root(current->fs, &path);
 443         error = 0;
 444 dput_and_out:
 445         path_put(&path);
 446 out:
 447         return error;
 448 }
 449
 450 static int chmod_common(struct path *path, umode_t mode)
 451 {
 452         struct inode *inode = path->dentry->d_inode;
 453         struct iattr newattrs;
 454         int error;
 455
 456         error = mnt_want_write(path->mnt);
 457         if (error)
 458                 return error;
 459         mutex_lock(&inode->i_mutex);
 460         error = security_path_chmod(path, mode);
 461         if (error)
 462                 goto out_unlock;
 463         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 464         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 465         error = notify_change(path->dentry, &newattrs);
 466 out_unlock:
 467         mutex_unlock(&inode->i_mutex);
 468         mnt_drop_write(path->mnt);
 469         return error;
 470 }
 471
 472 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 473 {
 474         struct file * file;
 475         int err = -EBADF;
 476
 477         file = fget(fd);
 478         if (file) {
 479                 audit_inode(NULL, file->f_path.dentry);
 480                 err = chmod_common(&file->f_path, mode);
 481                 fput(file);
 482         }
 483         return err;
 484 }
 485
 486 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 487 {
 488         struct path path;
 489         int error;
 490
 491         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 492         if (!error) {
 493                 error = chmod_common(&path, mode);
 494                 path_put(&path);
 495         }
 496         return error;
 497 }
 498
 499 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 500 {
 501         return sys_fchmodat(AT_FDCWD, filename, mode);
 502 }
 503
 504 static int chown_common(struct path *path, uid_t user, gid_t group)
 505 {
 506         struct inode *inode = path->dentry->d_inode;
 507         int error;
 508         struct iattr newattrs;
 509         kuid_t uid;
 510         kgid_t gid;
 511
 512         uid = make_kuid(current_user_ns(), user);
 513         gid = make_kgid(current_user_ns(), group);
 514
 515         newattrs.ia_valid =  ATTR_CTIME;
 516         if (user != (uid_t) -1) {
 517                 if (!uid_valid(uid))
 518                         return -EINVAL;
 519                 newattrs.ia_valid |= ATTR_UID;
 520                 newattrs.ia_uid = uid;
 521         }
 522         if (group != (gid_t) -1) {
 523                 if (!gid_valid(gid))
 524                         return -EINVAL;
 525                 newattrs.ia_valid |= ATTR_GID;
 526                 newattrs.ia_gid = gid;
 527         }
 528         if (!S_ISDIR(inode->i_mode))
 529                 newattrs.ia_valid |=
 530                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 531         mutex_lock(&inode->i_mutex);
 532         error = security_path_chown(path, user, group);
 533         if (!error)
 534                 error = notify_change(path->dentry, &newattrs);
 535         mutex_unlock(&inode->i_mutex);
 536
 537         return error;
 538 }
 539
 540 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 541                 gid_t, group, int, flag)
 542 {
 543         struct path path;
 544         int error = -EINVAL;
 545         int lookup_flags;
 546
 547         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 548                 goto out;
 549
 550         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 551         if (flag & AT_EMPTY_PATH)
 552                 lookup_flags |= LOOKUP_EMPTY;
 553         error = user_path_at(dfd, filename, lookup_flags, &path);
 554         if (error)
 555                 goto out;
 556         error = mnt_want_write(path.mnt);
 557         if (error)
 558                 goto out_release;
 559         error = chown_common(&path, user, group);
 560         mnt_drop_write(path.mnt);
 561 out_release:
 562         path_put(&path);
 563 out:
 564         return error;
 565 }
 566
 567 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 568 {
 569         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 570 }
 571
 572 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 573 {
 574         return sys_fchownat(AT_FDCWD, filename, user, group,
 575                             AT_SYMLINK_NOFOLLOW);
 576 }
 577
 578 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 579 {
 580         struct file * file;
 581         int error = -EBADF;
 582         struct dentry * dentry;
 583
 584         file = fget(fd);
 585         if (!file)
 586                 goto out;
 587
 588         error = mnt_want_write_file(file);
 589         if (error)
 590                 goto out_fput;
 591         dentry = file->f_path.dentry;
 592         audit_inode(NULL, dentry);
 593         error = chown_common(&file->f_path, user, group);
 594         mnt_drop_write_file(file);
 595 out_fput:
 596         fput(file);
 597 out:
 598         return error;
 599 }
 600
 601 /*
 602  * You have to be very careful that these write
 603  * counts get cleaned up in error cases and
 604  * upon __fput().  This should probably never
 605  * be called outside of __dentry_open().
 606  */
 607 static inline int __get_file_write_access(struct inode *inode,
 608                                           struct vfsmount *mnt)
 609 {
 610         int error;
 611         error = get_write_access(inode);
 612         if (error)
 613                 return error;
 614         /*
 615          * Do not take mount writer counts on
 616          * special files since no writes to
 617          * the mount itself will occur.
 618          */
 619         if (!special_file(inode->i_mode)) {
 620                 /*
 621                  * Balanced in __fput()
 622                  */
 623                 error = mnt_want_write(mnt);
 624                 if (error)
 625                         put_write_access(inode);
 626         }
 627         return error;
 628 }
 629
 630 int open_check_o_direct(struct file *f)
 631 {
 632         /* NB: we're sure to have correct a_ops only after f_op->open */
 633         if (f->f_flags & O_DIRECT) {
 634                 if (!f->f_mapping->a_ops ||
 635                     ((!f->f_mapping->a_ops->direct_IO) &&
 636                     (!f->f_mapping->a_ops->get_xip_mem))) {
 637                         return -EINVAL;
 638                 }
 639         }
 640         return 0;
 641 }
 642
 643 static int do_dentry_open(struct file *f,
 644                           int (*open)(struct inode *, struct file *),
 645                           const struct cred *cred)
 646 {
 647         static const struct file_operations empty_fops = {};
 648         struct inode *inode;
 649         int error;
 650
 651         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 652                                 FMODE_PREAD | FMODE_PWRITE;
 653
 654         if (unlikely(f->f_flags & O_PATH))
 655                 f->f_mode = FMODE_PATH;
 656
 657         inode = f->f_path.dentry->d_inode;
 658         if (f->f_mode & FMODE_WRITE) {
 659                 error = __get_file_write_access(inode, f->f_path.mnt);
 660                 if (error)
 661                         goto cleanup_file;
 662                 if (!special_file(inode->i_mode))
 663                         file_take_write(f);
 664         }
 665
 666         f->f_mapping = inode->i_mapping;
 667         f->f_pos = 0;
 668         file_sb_list_add(f, inode->i_sb);
 669
 670         if (unlikely(f->f_mode & FMODE_PATH)) {
 671                 f->f_op = &empty_fops;
 672                 return 0;
 673         }
 674
 675         f->f_op = fops_get(inode->i_fop);
 676
 677         error = security_file_open(f, cred);
 678         if (error)
 679                 goto cleanup_all;
 680
 681         error = break_lease(inode, f->f_flags);
 682         if (error)
 683                 goto cleanup_all;
 684
 685         if (!open && f->f_op)
 686                 open = f->f_op->open;
 687         if (open) {
 688                 error = open(inode, f);
 689                 if (error)
 690                         goto cleanup_all;
 691         }
 692         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 693                 i_readcount_inc(inode);
 694
 695         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 696
 697         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 698
 699         return 0;
 700
 701 cleanup_all:
 702         fops_put(f->f_op);
 703         file_sb_list_del(f);
 704         if (f->f_mode & FMODE_WRITE) {
 705                 put_write_access(inode);
 706                 if (!special_file(inode->i_mode)) {
 707                         /*
 708                          * We don't consider this a real
 709                          * mnt_want/drop_write() pair
 710                          * because it all happenend right
 711                          * here, so just reset the state.
 712                          */
 713                         file_reset_write(f);
 714                         mnt_drop_write(f->f_path.mnt);
 715                 }
 716         }
 717 cleanup_file:
 718         path_put(&f->f_path);
 719         f->f_path.mnt = NULL;
 720         f->f_path.dentry = NULL;
 721         return error;
 722 }
 723
 724 /**
 725  * finish_open - finish opening a file
 726  * @od: opaque open data
 727  * @dentry: pointer to dentry
 728  * @open: open callback
 729  *
 730  * This can be used to finish opening a file passed to i_op->atomic_open().
 731  *
 732  * If the open callback is set to NULL, then the standard f_op->open()
 733  * filesystem callback is substituted.
 734  */
 735 int finish_open(struct file *file, struct dentry *dentry,
 736                 int (*open)(struct inode *, struct file *),
 737                 int *opened)
 738 {
 739         int error;
 740         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 741
 742         mntget(file->f_path.mnt);
 743         file->f_path.dentry = dget(dentry);
 744
 745         error = do_dentry_open(file, open, current_cred());
 746         if (!error)
 747                 *opened |= FILE_OPENED;
 748
 749         return error;
 750 }
 751 EXPORT_SYMBOL(finish_open);
 752
 753 /**
 754  * finish_no_open - finish ->atomic_open() without opening the file
 755  *
 756  * @od: opaque open data
 757  * @dentry: dentry or NULL (as returned from ->lookup())
 758  *
 759  * This can be used to set the result of a successful lookup in ->atomic_open().
 760  * The filesystem's atomic_open() method shall return NULL after calling this.
 761  */
 762 int finish_no_open(struct file *file, struct dentry *dentry)
 763 {
 764         file->f_path.dentry = dentry;
 765         return 1;
 766 }
 767 EXPORT_SYMBOL(finish_no_open);
 768
 769 /*
 770  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 771  * error.
 772  */
 773 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 774                          const struct cred *cred)
 775 {
 776         int error;
 777         struct file *f;
 778
 779         validate_creds(cred);
 780
 781         /* We must always pass in a valid mount pointer. */
 782         BUG_ON(!mnt);
 783
 784         error = -ENFILE;
 785         f = get_empty_filp();
 786         if (f == NULL) {
 787                 dput(dentry);
 788                 mntput(mnt);
 789                 return ERR_PTR(error);
 790         }
 791
 792         f->f_flags = flags;
 793         f->f_path.mnt = mnt;
 794         f->f_path.dentry = dentry;
 795         error = do_dentry_open(f, NULL, cred);
 796         if (!error) {
 797                 error = open_check_o_direct(f);
 798                 if (error) {
 799                         fput(f);
 800                         f = ERR_PTR(error);
 801                 }
 802         } else {
 803                 put_filp(f);
 804                 f = ERR_PTR(error);
 805         }
 806         return f;
 807 }
 808 EXPORT_SYMBOL(dentry_open);
 809
 810 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 811 {
 812         struct fdtable *fdt = files_fdtable(files);
 813         __clear_open_fd(fd, fdt);
 814         if (fd < files->next_fd)
 815                 files->next_fd = fd;
 816 }
 817
 818 void put_unused_fd(unsigned int fd)
 819 {
 820         struct files_struct *files = current->files;
 821         spin_lock(&files->file_lock);
 822         __put_unused_fd(files, fd);
 823         spin_unlock(&files->file_lock);
 824 }
 825
 826 EXPORT_SYMBOL(put_unused_fd);
 827
 828 /*
 829  * Install a file pointer in the fd array.
 830  *
 831  * The VFS is full of places where we drop the files lock between
 832  * setting the open_fds bitmap and installing the file in the file
 833  * array.  At any such point, we are vulnerable to a dup2() race
 834  * installing a file in the array before us.  We need to detect this and
 835  * fput() the struct file we are about to overwrite in this case.
 836  *
 837  * It should never happen - if we allow dup2() do it, _really_ bad things
 838  * will follow.
 839  */
 840
 841 void fd_install(unsigned int fd, struct file *file)
 842 {
 843         struct files_struct *files = current->files;
 844         struct fdtable *fdt;
 845         spin_lock(&files->file_lock);
 846         fdt = files_fdtable(files);
 847         BUG_ON(fdt->fd[fd] != NULL);
 848         rcu_assign_pointer(fdt->fd[fd], file);
 849         spin_unlock(&files->file_lock);
 850 }
 851
 852 EXPORT_SYMBOL(fd_install);
 853
 854 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 855 {
 856         int lookup_flags = 0;
 857         int acc_mode;
 858
 859         if (!(flags & O_CREAT))
 860                 mode = 0;
 861         op->mode = mode;
 862
 863         /* Must never be set by userspace */
 864         flags &= ~FMODE_NONOTIFY;
 865
 866         /*
 867          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 868          * check for O_DSYNC if the need any syncing at all we enforce it's
 869          * always set instead of having to deal with possibly weird behaviour
 870          * for malicious applications setting only __O_SYNC.
 871          */
 872         if (flags & __O_SYNC)
 873                 flags |= O_DSYNC;
 874
 875         /*
 876          * If we have O_PATH in the open flag. Then we
 877          * cannot have anything other than the below set of flags
 878          */
 879         if (flags & O_PATH) {
 880                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 881                 acc_mode = 0;
 882         } else {
 883                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 884         }
 885
 886         op->open_flag = flags;
 887
 888         /* O_TRUNC implies we need access checks for write permissions */
 889         if (flags & O_TRUNC)
 890                 acc_mode |= MAY_WRITE;
 891
 892         /* Allow the LSM permission hook to distinguish append
 893            access from general write access. */
 894         if (flags & O_APPEND)
 895                 acc_mode |= MAY_APPEND;
 896
 897         op->acc_mode = acc_mode;
 898
 899         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 900
 901         if (flags & O_CREAT) {
 902                 op->intent |= LOOKUP_CREATE;
 903                 if (flags & O_EXCL)
 904                         op->intent |= LOOKUP_EXCL;
 905         }
 906
 907         if (flags & O_DIRECTORY)
 908                 lookup_flags |= LOOKUP_DIRECTORY;
 909         if (!(flags & O_NOFOLLOW))
 910                 lookup_flags |= LOOKUP_FOLLOW;
 911         return lookup_flags;
 912 }
 913
 914 /**
 915  * filp_open - open file and return file pointer
 916  *
 917  * @filename:   path to open
 918  * @flags:      open flags as per the open(2) second argument
 919  * @mode:       mode for the new file if O_CREAT is set, else ignored
 920  *
 921  * This is the helper to open a file from kernelspace if you really
 922  * have to.  But in generally you should not do this, so please move
 923  * along, nothing to see here..
 924  */
 925 struct file *filp_open(const char *filename, int flags, umode_t mode)
 926 {
 927         struct open_flags op;
 928         int lookup = build_open_flags(flags, mode, &op);
 929         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 930 }
 931 EXPORT_SYMBOL(filp_open);
 932
 933 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 934                             const char *filename, int flags)
 935 {
 936         struct open_flags op;
 937         int lookup = build_open_flags(flags, 0, &op);
 938         if (flags & O_CREAT)
 939                 return ERR_PTR(-EINVAL);
 940         if (!filename && (flags & O_DIRECTORY))
 941                 if (!dentry->d_inode->i_op->lookup)
 942                         return ERR_PTR(-ENOTDIR);
 943         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 944 }
 945 EXPORT_SYMBOL(file_open_root);
 946
 947 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 948 {
 949         struct open_flags op;
 950         int lookup = build_open_flags(flags, mode, &op);
 951         char *tmp = getname(filename);
 952         int fd = PTR_ERR(tmp);
 953
 954         if (!IS_ERR(tmp)) {
 955                 fd = get_unused_fd_flags(flags);
 956                 if (fd >= 0) {
 957                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 958                         if (IS_ERR(f)) {
 959                                 put_unused_fd(fd);
 960                                 fd = PTR_ERR(f);
 961                         } else {
 962                                 fsnotify_open(f);
 963                                 fd_install(fd, f);
 964                         }
 965                 }
 966                 putname(tmp);
 967         }
 968         return fd;
 969 }
 970
 971 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 972 {
 973         long ret;
 974
 975         if (force_o_largefile())
 976                 flags |= O_LARGEFILE;
 977
 978         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 979         /* avoid REGPARM breakage on x86: */
 980         asmlinkage_protect(3, ret, filename, flags, mode);
 981         return ret;
 982 }
 983
 984 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 985                 umode_t, mode)
 986 {
 987         long ret;
 988
 989         if (force_o_largefile())
 990                 flags |= O_LARGEFILE;
 991
 992         ret = do_sys_open(dfd, filename, flags, mode);
 993         /* avoid REGPARM breakage on x86: */
 994         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 995         return ret;
 996 }
 997
 998 #ifndef __alpha__
 999
1000 /*
1001  * For backward compatibility?  Maybe this should be moved
1002  * into arch/i386 instead?
1003  */
1004 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1005 {
1006         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1007 }
1008
1009 #endif
1010
1011 /*
1012  * "id" is the POSIX thread ID. We use the
1013  * files pointer for this..
1014  */
1015 int filp_close(struct file *filp, fl_owner_t id)
1016 {
1017         int retval = 0;
1018
1019         if (!file_count(filp)) {
1020                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1021                 return 0;
1022         }
1023
1024         if (filp->f_op && filp->f_op->flush)
1025                 retval = filp->f_op->flush(filp, id);
1026
1027         if (likely(!(filp->f_mode & FMODE_PATH))) {
1028                 dnotify_flush(filp, id);
1029                 locks_remove_posix(filp, id);
1030         }
1031         fput(filp);
1032         return retval;
1033 }
1034
1035 EXPORT_SYMBOL(filp_close);
1036
1037 /*
1038  * Careful here! We test whether the file pointer is NULL before
1039  * releasing the fd. This ensures that one clone task can't release
1040  * an fd while another clone is opening it.
1041  */
1042 SYSCALL_DEFINE1(close, unsigned int, fd)
1043 {
1044         struct file * filp;
1045         struct files_struct *files = current->files;
1046         struct fdtable *fdt;
1047         int retval;
1048
1049         spin_lock(&files->file_lock);
1050         fdt = files_fdtable(files);
1051         if (fd >= fdt->max_fds)
1052                 goto out_unlock;
1053         filp = fdt->fd[fd];
1054         if (!filp)
1055                 goto out_unlock;
1056         rcu_assign_pointer(fdt->fd[fd], NULL);
1057         __clear_close_on_exec(fd, fdt);
1058         __put_unused_fd(files, fd);
1059         spin_unlock(&files->file_lock);
1060         retval = filp_close(filp, files);
1061
1062         /* can't restart close syscall because file table entry was cleared */
1063         if (unlikely(retval == -ERESTARTSYS ||
1064                      retval == -ERESTARTNOINTR ||
1065                      retval == -ERESTARTNOHAND ||
1066                      retval == -ERESTART_RESTARTBLOCK))
1067                 retval = -EINTR;
1068
1069         return retval;
1070
1071 out_unlock:
1072         spin_unlock(&files->file_lock);
1073         return -EBADF;
1074 }
1075 EXPORT_SYMBOL(sys_close);
1076
1077 /*
1078  * This routine simulates a hangup on the tty, to arrange that users
1079  * are given clean terminals at login time.
1080  */
1081 SYSCALL_DEFINE0(vhangup)
1082 {
1083         if (capable(CAP_SYS_TTY_CONFIG)) {
1084                 tty_vhangup_self();
1085                 return 0;
1086         }
1087         return -EPERM;
1088 }
1089
1090 /*
1091  * Called when an inode is about to be open.
1092  * We use this to disallow opening large files on 32bit systems if
1093  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1094  * on this flag in sys_open.
1095  */
1096 int generic_file_open(struct inode * inode, struct file * filp)
1097 {
1098         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1099                 return -EOVERFLOW;
1100         return 0;
1101 }
1102
1103 EXPORT_SYMBOL(generic_file_open);
1104
1105 /*
1106  * This is used by subsystems that don't want seekable
1107  * file descriptors. The function is not supposed to ever fail, the only
1108  * reason it returns an 'int' and not 'void' is so that it can be plugged
1109  * directly into file_operations structure.
1110  */
1111 int nonseekable_open(struct inode *inode, struct file *filp)
1112 {
1113         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1114         return 0;
1115 }
1116
1117 EXPORT_SYMBOL(nonseekable_open);