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 struct file *dentry_open(const struct path *path, int flags,
 770                          const struct cred *cred)
 771 {
 772         int error;
 773         struct file *f;
 774
 775         validate_creds(cred);
 776
 777         /* We must always pass in a valid mount pointer. */
 778         BUG_ON(!path->mnt);
 779
 780         error = -ENFILE;
 781         f = get_empty_filp();
 782         if (f == NULL)
 783                 return ERR_PTR(error);
 784
 785         f->f_flags = flags;
 786         f->f_path = *path;
 787         path_get(&f->f_path);
 788         error = do_dentry_open(f, NULL, cred);
 789         if (!error) {
 790                 error = open_check_o_direct(f);
 791                 if (error) {
 792                         fput(f);
 793                         f = ERR_PTR(error);
 794                 }
 795         } else {
 796                 put_filp(f);
 797                 f = ERR_PTR(error);
 798         }
 799         return f;
 800 }
 801 EXPORT_SYMBOL(dentry_open);
 802
 803 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 804 {
 805         struct fdtable *fdt = files_fdtable(files);
 806         __clear_open_fd(fd, fdt);
 807         if (fd < files->next_fd)
 808                 files->next_fd = fd;
 809 }
 810
 811 void put_unused_fd(unsigned int fd)
 812 {
 813         struct files_struct *files = current->files;
 814         spin_lock(&files->file_lock);
 815         __put_unused_fd(files, fd);
 816         spin_unlock(&files->file_lock);
 817 }
 818
 819 EXPORT_SYMBOL(put_unused_fd);
 820
 821 /*
 822  * Install a file pointer in the fd array.
 823  *
 824  * The VFS is full of places where we drop the files lock between
 825  * setting the open_fds bitmap and installing the file in the file
 826  * array.  At any such point, we are vulnerable to a dup2() race
 827  * installing a file in the array before us.  We need to detect this and
 828  * fput() the struct file we are about to overwrite in this case.
 829  *
 830  * It should never happen - if we allow dup2() do it, _really_ bad things
 831  * will follow.
 832  */
 833
 834 void fd_install(unsigned int fd, struct file *file)
 835 {
 836         struct files_struct *files = current->files;
 837         struct fdtable *fdt;
 838         spin_lock(&files->file_lock);
 839         fdt = files_fdtable(files);
 840         BUG_ON(fdt->fd[fd] != NULL);
 841         rcu_assign_pointer(fdt->fd[fd], file);
 842         spin_unlock(&files->file_lock);
 843 }
 844
 845 EXPORT_SYMBOL(fd_install);
 846
 847 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 848 {
 849         int lookup_flags = 0;
 850         int acc_mode;
 851
 852         if (!(flags & O_CREAT))
 853                 mode = 0;
 854         op->mode = mode;
 855
 856         /* Must never be set by userspace */
 857         flags &= ~FMODE_NONOTIFY;
 858
 859         /*
 860          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 861          * check for O_DSYNC if the need any syncing at all we enforce it's
 862          * always set instead of having to deal with possibly weird behaviour
 863          * for malicious applications setting only __O_SYNC.
 864          */
 865         if (flags & __O_SYNC)
 866                 flags |= O_DSYNC;
 867
 868         /*
 869          * If we have O_PATH in the open flag. Then we
 870          * cannot have anything other than the below set of flags
 871          */
 872         if (flags & O_PATH) {
 873                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 874                 acc_mode = 0;
 875         } else {
 876                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 877         }
 878
 879         op->open_flag = flags;
 880
 881         /* O_TRUNC implies we need access checks for write permissions */
 882         if (flags & O_TRUNC)
 883                 acc_mode |= MAY_WRITE;
 884
 885         /* Allow the LSM permission hook to distinguish append
 886            access from general write access. */
 887         if (flags & O_APPEND)
 888                 acc_mode |= MAY_APPEND;
 889
 890         op->acc_mode = acc_mode;
 891
 892         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 893
 894         if (flags & O_CREAT) {
 895                 op->intent |= LOOKUP_CREATE;
 896                 if (flags & O_EXCL)
 897                         op->intent |= LOOKUP_EXCL;
 898         }
 899
 900         if (flags & O_DIRECTORY)
 901                 lookup_flags |= LOOKUP_DIRECTORY;
 902         if (!(flags & O_NOFOLLOW))
 903                 lookup_flags |= LOOKUP_FOLLOW;
 904         return lookup_flags;
 905 }
 906
 907 /**
 908  * filp_open - open file and return file pointer
 909  *
 910  * @filename:   path to open
 911  * @flags:      open flags as per the open(2) second argument
 912  * @mode:       mode for the new file if O_CREAT is set, else ignored
 913  *
 914  * This is the helper to open a file from kernelspace if you really
 915  * have to.  But in generally you should not do this, so please move
 916  * along, nothing to see here..
 917  */
 918 struct file *filp_open(const char *filename, int flags, umode_t mode)
 919 {
 920         struct open_flags op;
 921         int lookup = build_open_flags(flags, mode, &op);
 922         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 923 }
 924 EXPORT_SYMBOL(filp_open);
 925
 926 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 927                             const char *filename, int flags)
 928 {
 929         struct open_flags op;
 930         int lookup = build_open_flags(flags, 0, &op);
 931         if (flags & O_CREAT)
 932                 return ERR_PTR(-EINVAL);
 933         if (!filename && (flags & O_DIRECTORY))
 934                 if (!dentry->d_inode->i_op->lookup)
 935                         return ERR_PTR(-ENOTDIR);
 936         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 937 }
 938 EXPORT_SYMBOL(file_open_root);
 939
 940 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 941 {
 942         struct open_flags op;
 943         int lookup = build_open_flags(flags, mode, &op);
 944         char *tmp = getname(filename);
 945         int fd = PTR_ERR(tmp);
 946
 947         if (!IS_ERR(tmp)) {
 948                 fd = get_unused_fd_flags(flags);
 949                 if (fd >= 0) {
 950                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 951                         if (IS_ERR(f)) {
 952                                 put_unused_fd(fd);
 953                                 fd = PTR_ERR(f);
 954                         } else {
 955                                 fsnotify_open(f);
 956                                 fd_install(fd, f);
 957                         }
 958                 }
 959                 putname(tmp);
 960         }
 961         return fd;
 962 }
 963
 964 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 965 {
 966         long ret;
 967
 968         if (force_o_largefile())
 969                 flags |= O_LARGEFILE;
 970
 971         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 972         /* avoid REGPARM breakage on x86: */
 973         asmlinkage_protect(3, ret, filename, flags, mode);
 974         return ret;
 975 }
 976
 977 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 978                 umode_t, mode)
 979 {
 980         long ret;
 981
 982         if (force_o_largefile())
 983                 flags |= O_LARGEFILE;
 984
 985         ret = do_sys_open(dfd, filename, flags, mode);
 986         /* avoid REGPARM breakage on x86: */
 987         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 988         return ret;
 989 }
 990
 991 #ifndef __alpha__
 992
 993 /*
 994  * For backward compatibility?  Maybe this should be moved
 995  * into arch/i386 instead?
 996  */
 997 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
 998 {
 999         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1000 }
1001
1002 #endif
1003
1004 /*
1005  * "id" is the POSIX thread ID. We use the
1006  * files pointer for this..
1007  */
1008 int filp_close(struct file *filp, fl_owner_t id)
1009 {
1010         int retval = 0;
1011
1012         if (!file_count(filp)) {
1013                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1014                 return 0;
1015         }
1016
1017         if (filp->f_op && filp->f_op->flush)
1018                 retval = filp->f_op->flush(filp, id);
1019
1020         if (likely(!(filp->f_mode & FMODE_PATH))) {
1021                 dnotify_flush(filp, id);
1022                 locks_remove_posix(filp, id);
1023         }
1024         fput(filp);
1025         return retval;
1026 }
1027
1028 EXPORT_SYMBOL(filp_close);
1029
1030 /*
1031  * Careful here! We test whether the file pointer is NULL before
1032  * releasing the fd. This ensures that one clone task can't release
1033  * an fd while another clone is opening it.
1034  */
1035 SYSCALL_DEFINE1(close, unsigned int, fd)
1036 {
1037         struct file * filp;
1038         struct files_struct *files = current->files;
1039         struct fdtable *fdt;
1040         int retval;
1041
1042         spin_lock(&files->file_lock);
1043         fdt = files_fdtable(files);
1044         if (fd >= fdt->max_fds)
1045                 goto out_unlock;
1046         filp = fdt->fd[fd];
1047         if (!filp)
1048                 goto out_unlock;
1049         rcu_assign_pointer(fdt->fd[fd], NULL);
1050         __clear_close_on_exec(fd, fdt);
1051         __put_unused_fd(files, fd);
1052         spin_unlock(&files->file_lock);
1053         retval = filp_close(filp, files);
1054
1055         /* can't restart close syscall because file table entry was cleared */
1056         if (unlikely(retval == -ERESTARTSYS ||
1057                      retval == -ERESTARTNOINTR ||
1058                      retval == -ERESTARTNOHAND ||
1059                      retval == -ERESTART_RESTARTBLOCK))
1060                 retval = -EINTR;
1061
1062         return retval;
1063
1064 out_unlock:
1065         spin_unlock(&files->file_lock);
1066         return -EBADF;
1067 }
1068 EXPORT_SYMBOL(sys_close);
1069
1070 /*
1071  * This routine simulates a hangup on the tty, to arrange that users
1072  * are given clean terminals at login time.
1073  */
1074 SYSCALL_DEFINE0(vhangup)
1075 {
1076         if (capable(CAP_SYS_TTY_CONFIG)) {
1077                 tty_vhangup_self();
1078                 return 0;
1079         }
1080         return -EPERM;
1081 }
1082
1083 /*
1084  * Called when an inode is about to be open.
1085  * We use this to disallow opening large files on 32bit systems if
1086  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1087  * on this flag in sys_open.
1088  */
1089 int generic_file_open(struct inode * inode, struct file * filp)
1090 {
1091         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1092                 return -EOVERFLOW;
1093         return 0;
1094 }
1095
1096 EXPORT_SYMBOL(generic_file_open);
1097
1098 /*
1099  * This is used by subsystems that don't want seekable
1100  * file descriptors. The function is not supposed to ever fail, the only
1101  * reason it returns an 'int' and not 'void' is so that it can be plugged
1102  * directly into file_operations structure.
1103  */
1104 int nonseekable_open(struct inode *inode, struct file *filp)
1105 {
1106         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1107         return 0;
1108 }
1109
1110 EXPORT_SYMBOL(nonseekable_open);