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/quotaops.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/slab.h>
  14 #include <linux/tty.h>
  15 #include <linux/namei.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/capability.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/vfs.h>
  21 #include <linux/fcntl.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
  31 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  32 {
  33         int retval = -ENODEV;
  34
  35         if (dentry) {
  36                 retval = -ENOSYS;
  37                 if (dentry->d_sb->s_op->statfs) {
  38                         memset(buf, 0, sizeof(*buf));
  39                         retval = security_sb_statfs(dentry);
  40                         if (retval)
  41                                 return retval;
  42                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  43                         if (retval == 0 && buf->f_frsize == 0)
  44                                 buf->f_frsize = buf->f_bsize;
  45                 }
  46         }
  47         return retval;
  48 }
  49
  50 EXPORT_SYMBOL(vfs_statfs);
  51
  52 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  53 {
  54         struct kstatfs st;
  55         int retval;
  56
  57         retval = vfs_statfs(dentry, &st);
  58         if (retval)
  59                 return retval;
  60
  61         if (sizeof(*buf) == sizeof(st))
  62                 memcpy(buf, &st, sizeof(st));
  63         else {
  64                 if (sizeof buf->f_blocks == 4) {
  65                         if ((st.f_blocks | st.f_bfree | st.f_bavail) &
  66                             0xffffffff00000000ULL)
  67                                 return -EOVERFLOW;
  68                         /*
  69                          * f_files and f_ffree may be -1; it's okay to stuff
  70                          * that into 32 bits
  71                          */
  72                         if (st.f_files != -1 &&
  73                             (st.f_files & 0xffffffff00000000ULL))
  74                                 return -EOVERFLOW;
  75                         if (st.f_ffree != -1 &&
  76                             (st.f_ffree & 0xffffffff00000000ULL))
  77                                 return -EOVERFLOW;
  78                 }
  79
  80                 buf->f_type = st.f_type;
  81                 buf->f_bsize = st.f_bsize;
  82                 buf->f_blocks = st.f_blocks;
  83                 buf->f_bfree = st.f_bfree;
  84                 buf->f_bavail = st.f_bavail;
  85                 buf->f_files = st.f_files;
  86                 buf->f_ffree = st.f_ffree;
  87                 buf->f_fsid = st.f_fsid;
  88                 buf->f_namelen = st.f_namelen;
  89                 buf->f_frsize = st.f_frsize;
  90                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  91         }
  92         return 0;
  93 }
  94
  95 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
  96 {
  97         struct kstatfs st;
  98         int retval;
  99
 100         retval = vfs_statfs(dentry, &st);
 101         if (retval)
 102                 return retval;
 103
 104         if (sizeof(*buf) == sizeof(st))
 105                 memcpy(buf, &st, sizeof(st));
 106         else {
 107                 buf->f_type = st.f_type;
 108                 buf->f_bsize = st.f_bsize;
 109                 buf->f_blocks = st.f_blocks;
 110                 buf->f_bfree = st.f_bfree;
 111                 buf->f_bavail = st.f_bavail;
 112                 buf->f_files = st.f_files;
 113                 buf->f_ffree = st.f_ffree;
 114                 buf->f_fsid = st.f_fsid;
 115                 buf->f_namelen = st.f_namelen;
 116                 buf->f_frsize = st.f_frsize;
 117                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 118         }
 119         return 0;
 120 }
 121
 122 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 123 {
 124         struct nameidata nd;
 125         int error;
 126
 127         error = user_path_walk(path, &nd);
 128         if (!error) {
 129                 struct statfs tmp;
 130                 error = vfs_statfs_native(nd.dentry, &tmp);
 131                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 132                         error = -EFAULT;
 133                 path_release(&nd);
 134         }
 135         return error;
 136 }
 137
 138
 139 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 140 {
 141         struct nameidata nd;
 142         long error;
 143
 144         if (sz != sizeof(*buf))
 145                 return -EINVAL;
 146         error = user_path_walk(path, &nd);
 147         if (!error) {
 148                 struct statfs64 tmp;
 149                 error = vfs_statfs64(nd.dentry, &tmp);
 150                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 151                         error = -EFAULT;
 152                 path_release(&nd);
 153         }
 154         return error;
 155 }
 156
 157
 158 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 159 {
 160         struct file * file;
 161         struct statfs tmp;
 162         int error;
 163
 164         error = -EBADF;
 165         file = fget(fd);
 166         if (!file)
 167                 goto out;
 168         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 169         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 170                 error = -EFAULT;
 171         fput(file);
 172 out:
 173         return error;
 174 }
 175
 176 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 177 {
 178         struct file * file;
 179         struct statfs64 tmp;
 180         int error;
 181
 182         if (sz != sizeof(*buf))
 183                 return -EINVAL;
 184
 185         error = -EBADF;
 186         file = fget(fd);
 187         if (!file)
 188                 goto out;
 189         error = vfs_statfs64(file->f_path.dentry, &tmp);
 190         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 191                 error = -EFAULT;
 192         fput(file);
 193 out:
 194         return error;
 195 }
 196
 197 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 198         struct file *filp)
 199 {
 200         int err;
 201         struct iattr newattrs;
 202
 203         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 204         if (length < 0)
 205                 return -EINVAL;
 206
 207         newattrs.ia_size = length;
 208         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 209         if (filp) {
 210                 newattrs.ia_file = filp;
 211                 newattrs.ia_valid |= ATTR_FILE;
 212         }
 213
 214         /* Remove suid/sgid on truncate too */
 215         newattrs.ia_valid |= should_remove_suid(dentry);
 216
 217         mutex_lock(&dentry->d_inode->i_mutex);
 218         err = notify_change(dentry, &newattrs);
 219         mutex_unlock(&dentry->d_inode->i_mutex);
 220         return err;
 221 }
 222
 223 static long do_sys_truncate(const char __user * path, loff_t length)
 224 {
 225         struct nameidata nd;
 226         struct inode * inode;
 227         int error;
 228
 229         error = -EINVAL;
 230         if (length < 0) /* sorry, but loff_t says... */
 231                 goto out;
 232
 233         error = user_path_walk(path, &nd);
 234         if (error)
 235                 goto out;
 236         inode = nd.dentry->d_inode;
 237
 238         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 239         error = -EISDIR;
 240         if (S_ISDIR(inode->i_mode))
 241                 goto dput_and_out;
 242
 243         error = -EINVAL;
 244         if (!S_ISREG(inode->i_mode))
 245                 goto dput_and_out;
 246
 247         error = vfs_permission(&nd, MAY_WRITE);
 248         if (error)
 249                 goto dput_and_out;
 250
 251         error = -EROFS;
 252         if (IS_RDONLY(inode))
 253                 goto dput_and_out;
 254
 255         error = -EPERM;
 256         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 257                 goto dput_and_out;
 258
 259         error = get_write_access(inode);
 260         if (error)
 261                 goto dput_and_out;
 262
 263         /*
 264          * Make sure that there are no leases.  get_write_access() protects
 265          * against the truncate racing with a lease-granting setlease().
 266          */
 267         error = break_lease(inode, FMODE_WRITE);
 268         if (error)
 269                 goto put_write_and_out;
 270
 271         error = locks_verify_truncate(inode, NULL, length);
 272         if (!error) {
 273                 DQUOT_INIT(inode);
 274                 error = do_truncate(nd.dentry, length, 0, NULL);
 275         }
 276
 277 put_write_and_out:
 278         put_write_access(inode);
 279 dput_and_out:
 280         path_release(&nd);
 281 out:
 282         return error;
 283 }
 284
 285 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 286 {
 287         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 288         return do_sys_truncate(path, (long)length);
 289 }
 290
 291 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 292 {
 293         struct inode * inode;
 294         struct dentry *dentry;
 295         struct file * file;
 296         int error;
 297
 298         error = -EINVAL;
 299         if (length < 0)
 300                 goto out;
 301         error = -EBADF;
 302         file = fget(fd);
 303         if (!file)
 304                 goto out;
 305
 306         /* explicitly opened as large or we are on 64-bit box */
 307         if (file->f_flags & O_LARGEFILE)
 308                 small = 0;
 309
 310         dentry = file->f_path.dentry;
 311         inode = dentry->d_inode;
 312         error = -EINVAL;
 313         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 314                 goto out_putf;
 315
 316         error = -EINVAL;
 317         /* Cannot ftruncate over 2^31 bytes without large file support */
 318         if (small && length > MAX_NON_LFS)
 319                 goto out_putf;
 320
 321         error = -EPERM;
 322         if (IS_APPEND(inode))
 323                 goto out_putf;
 324
 325         error = locks_verify_truncate(inode, file, length);
 326         if (!error)
 327                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 328 out_putf:
 329         fput(file);
 330 out:
 331         return error;
 332 }
 333
 334 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 335 {
 336         long ret = do_sys_ftruncate(fd, length, 1);
 337         /* avoid REGPARM breakage on x86: */
 338         prevent_tail_call(ret);
 339         return ret;
 340 }
 341
 342 /* LFS versions of truncate are only needed on 32 bit machines */
 343 #if BITS_PER_LONG == 32
 344 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 345 {
 346         return do_sys_truncate(path, length);
 347 }
 348
 349 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 350 {
 351         long ret = do_sys_ftruncate(fd, length, 0);
 352         /* avoid REGPARM breakage on x86: */
 353         prevent_tail_call(ret);
 354         return ret;
 355 }
 356 #endif
 357
 358 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 359 {
 360         struct file *file;
 361         struct inode *inode;
 362         long ret = -EINVAL;
 363
 364         if (offset < 0 || len <= 0)
 365                 goto out;
 366
 367         /* Return error if mode is not supported */
 368         ret = -EOPNOTSUPP;
 369         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 370                 goto out;
 371
 372         ret = -EBADF;
 373         file = fget(fd);
 374         if (!file)
 375                 goto out;
 376         if (!(file->f_mode & FMODE_WRITE))
 377                 goto out_fput;
 378         /*
 379          * Revalidate the write permissions, in case security policy has
 380          * changed since the files were opened.
 381          */
 382         ret = security_file_permission(file, MAY_WRITE);
 383         if (ret)
 384                 goto out_fput;
 385
 386         inode = file->f_path.dentry->d_inode;
 387
 388         ret = -ESPIPE;
 389         if (S_ISFIFO(inode->i_mode))
 390                 goto out_fput;
 391
 392         ret = -ENODEV;
 393         /*
 394          * Let individual file system decide if it supports preallocation
 395          * for directories or not.
 396          */
 397         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 398                 goto out_fput;
 399
 400         ret = -EFBIG;
 401         /* Check for wrap through zero too */
 402         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 403                 goto out_fput;
 404
 405         if (inode->i_op && inode->i_op->fallocate)
 406                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 407         else
 408                 ret = -EOPNOTSUPP;
 409
 410 out_fput:
 411         fput(file);
 412 out:
 413         return ret;
 414 }
 415
 416 /*
 417  * access() needs to use the real uid/gid, not the effective uid/gid.
 418  * We do this by temporarily clearing all FS-related capabilities and
 419  * switching the fsuid/fsgid around to the real ones.
 420  */
 421 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 422 {
 423         struct nameidata nd;
 424         int old_fsuid, old_fsgid;
 425         kernel_cap_t old_cap;
 426         int res;
 427
 428         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 429                 return -EINVAL;
 430
 431         old_fsuid = current->fsuid;
 432         old_fsgid = current->fsgid;
 433         old_cap = current->cap_effective;
 434
 435         current->fsuid = current->uid;
 436         current->fsgid = current->gid;
 437
 438         /*
 439          * Clear the capabilities if we switch to a non-root user
 440          *
 441          * FIXME: There is a race here against sys_capset.  The
 442          * capabilities can change yet we will restore the old
 443          * value below.  We should hold task_capabilities_lock,
 444          * but we cannot because user_path_walk can sleep.
 445          */
 446         if (current->uid)
 447                 cap_clear(current->cap_effective);
 448         else
 449                 current->cap_effective = current->cap_permitted;
 450
 451         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 452         if (res)
 453                 goto out;
 454
 455         res = vfs_permission(&nd, mode);
 456         /* SuS v2 requires we report a read only fs too */
 457         if(res || !(mode & S_IWOTH) ||
 458            special_file(nd.dentry->d_inode->i_mode))
 459                 goto out_path_release;
 460
 461         if(IS_RDONLY(nd.dentry->d_inode))
 462                 res = -EROFS;
 463
 464 out_path_release:
 465         path_release(&nd);
 466 out:
 467         current->fsuid = old_fsuid;
 468         current->fsgid = old_fsgid;
 469         current->cap_effective = old_cap;
 470
 471         return res;
 472 }
 473
 474 asmlinkage long sys_access(const char __user *filename, int mode)
 475 {
 476         return sys_faccessat(AT_FDCWD, filename, mode);
 477 }
 478
 479 asmlinkage long sys_chdir(const char __user * filename)
 480 {
 481         struct nameidata nd;
 482         int error;
 483
 484         error = __user_walk(filename,
 485                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 486         if (error)
 487                 goto out;
 488
 489         error = vfs_permission(&nd, MAY_EXEC);
 490         if (error)
 491                 goto dput_and_out;
 492
 493         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 494
 495 dput_and_out:
 496         path_release(&nd);
 497 out:
 498         return error;
 499 }
 500
 501 asmlinkage long sys_fchdir(unsigned int fd)
 502 {
 503         struct file *file;
 504         struct dentry *dentry;
 505         struct inode *inode;
 506         struct vfsmount *mnt;
 507         int error;
 508
 509         error = -EBADF;
 510         file = fget(fd);
 511         if (!file)
 512                 goto out;
 513
 514         dentry = file->f_path.dentry;
 515         mnt = file->f_path.mnt;
 516         inode = dentry->d_inode;
 517
 518         error = -ENOTDIR;
 519         if (!S_ISDIR(inode->i_mode))
 520                 goto out_putf;
 521
 522         error = file_permission(file, MAY_EXEC);
 523         if (!error)
 524                 set_fs_pwd(current->fs, mnt, dentry);
 525 out_putf:
 526         fput(file);
 527 out:
 528         return error;
 529 }
 530
 531 asmlinkage long sys_chroot(const char __user * filename)
 532 {
 533         struct nameidata nd;
 534         int error;
 535
 536         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 537         if (error)
 538                 goto out;
 539
 540         error = vfs_permission(&nd, MAY_EXEC);
 541         if (error)
 542                 goto dput_and_out;
 543
 544         error = -EPERM;
 545         if (!capable(CAP_SYS_CHROOT))
 546                 goto dput_and_out;
 547
 548         set_fs_root(current->fs, nd.mnt, nd.dentry);
 549         set_fs_altroot();
 550         error = 0;
 551 dput_and_out:
 552         path_release(&nd);
 553 out:
 554         return error;
 555 }
 556
 557 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 558 {
 559         struct inode * inode;
 560         struct dentry * dentry;
 561         struct file * file;
 562         int err = -EBADF;
 563         struct iattr newattrs;
 564
 565         file = fget(fd);
 566         if (!file)
 567                 goto out;
 568
 569         dentry = file->f_path.dentry;
 570         inode = dentry->d_inode;
 571
 572         audit_inode(NULL, inode);
 573
 574         err = -EROFS;
 575         if (IS_RDONLY(inode))
 576                 goto out_putf;
 577         err = -EPERM;
 578         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 579                 goto out_putf;
 580         mutex_lock(&inode->i_mutex);
 581         if (mode == (mode_t) -1)
 582                 mode = inode->i_mode;
 583         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 584         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 585         err = notify_change(dentry, &newattrs);
 586         mutex_unlock(&inode->i_mutex);
 587
 588 out_putf:
 589         fput(file);
 590 out:
 591         return err;
 592 }
 593
 594 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 595                              mode_t mode)
 596 {
 597         struct nameidata nd;
 598         struct inode * inode;
 599         int error;
 600         struct iattr newattrs;
 601
 602         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 603         if (error)
 604                 goto out;
 605         inode = nd.dentry->d_inode;
 606
 607         error = -EROFS;
 608         if (IS_RDONLY(inode))
 609                 goto dput_and_out;
 610
 611         error = -EPERM;
 612         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 613                 goto dput_and_out;
 614
 615         mutex_lock(&inode->i_mutex);
 616         if (mode == (mode_t) -1)
 617                 mode = inode->i_mode;
 618         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 619         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 620         error = notify_change(nd.dentry, &newattrs);
 621         mutex_unlock(&inode->i_mutex);
 622
 623 dput_and_out:
 624         path_release(&nd);
 625 out:
 626         return error;
 627 }
 628
 629 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 630 {
 631         return sys_fchmodat(AT_FDCWD, filename, mode);
 632 }
 633
 634 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 635 {
 636         struct inode * inode;
 637         int error;
 638         struct iattr newattrs;
 639
 640         error = -ENOENT;
 641         if (!(inode = dentry->d_inode)) {
 642                 printk(KERN_ERR "chown_common: NULL inode\n");
 643                 goto out;
 644         }
 645         error = -EROFS;
 646         if (IS_RDONLY(inode))
 647                 goto out;
 648         error = -EPERM;
 649         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 650                 goto out;
 651         newattrs.ia_valid =  ATTR_CTIME;
 652         if (user != (uid_t) -1) {
 653                 newattrs.ia_valid |= ATTR_UID;
 654                 newattrs.ia_uid = user;
 655         }
 656         if (group != (gid_t) -1) {
 657                 newattrs.ia_valid |= ATTR_GID;
 658                 newattrs.ia_gid = group;
 659         }
 660         if (!S_ISDIR(inode->i_mode))
 661                 newattrs.ia_valid |=
 662                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 663         mutex_lock(&inode->i_mutex);
 664         error = notify_change(dentry, &newattrs);
 665         mutex_unlock(&inode->i_mutex);
 666 out:
 667         return error;
 668 }
 669
 670 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 671 {
 672         struct nameidata nd;
 673         int error;
 674
 675         error = user_path_walk(filename, &nd);
 676         if (error)
 677                 goto out;
 678         error = chown_common(nd.dentry, user, group);
 679         path_release(&nd);
 680 out:
 681         return error;
 682 }
 683
 684 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 685                              gid_t group, int flag)
 686 {
 687         struct nameidata nd;
 688         int error = -EINVAL;
 689         int follow;
 690
 691         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 692                 goto out;
 693
 694         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 695         error = __user_walk_fd(dfd, filename, follow, &nd);
 696         if (error)
 697                 goto out;
 698         error = chown_common(nd.dentry, user, group);
 699         path_release(&nd);
 700 out:
 701         return error;
 702 }
 703
 704 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 705 {
 706         struct nameidata nd;
 707         int error;
 708
 709         error = user_path_walk_link(filename, &nd);
 710         if (error)
 711                 goto out;
 712         error = chown_common(nd.dentry, user, group);
 713         path_release(&nd);
 714 out:
 715         return error;
 716 }
 717
 718
 719 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 720 {
 721         struct file * file;
 722         int error = -EBADF;
 723         struct dentry * dentry;
 724
 725         file = fget(fd);
 726         if (!file)
 727                 goto out;
 728
 729         dentry = file->f_path.dentry;
 730         audit_inode(NULL, dentry->d_inode);
 731         error = chown_common(dentry, user, group);
 732         fput(file);
 733 out:
 734         return error;
 735 }
 736
 737 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 738                                         int flags, struct file *f,
 739                                         int (*open)(struct inode *, struct file *))
 740 {
 741         struct inode *inode;
 742         int error;
 743
 744         f->f_flags = flags;
 745         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 746                                 FMODE_PREAD | FMODE_PWRITE;
 747         inode = dentry->d_inode;
 748         if (f->f_mode & FMODE_WRITE) {
 749                 error = get_write_access(inode);
 750                 if (error)
 751                         goto cleanup_file;
 752         }
 753
 754         f->f_mapping = inode->i_mapping;
 755         f->f_path.dentry = dentry;
 756         f->f_path.mnt = mnt;
 757         f->f_pos = 0;
 758         f->f_op = fops_get(inode->i_fop);
 759         file_move(f, &inode->i_sb->s_files);
 760
 761         error = security_dentry_open(f);
 762         if (error)
 763                 goto cleanup_all;
 764
 765         if (!open && f->f_op)
 766                 open = f->f_op->open;
 767         if (open) {
 768                 error = open(inode, f);
 769                 if (error)
 770                         goto cleanup_all;
 771         }
 772
 773         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 774
 775         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 776
 777         /* NB: we're sure to have correct a_ops only after f_op->open */
 778         if (f->f_flags & O_DIRECT) {
 779                 if (!f->f_mapping->a_ops ||
 780                     ((!f->f_mapping->a_ops->direct_IO) &&
 781                     (!f->f_mapping->a_ops->get_xip_page))) {
 782                         fput(f);
 783                         f = ERR_PTR(-EINVAL);
 784                 }
 785         }
 786
 787         return f;
 788
 789 cleanup_all:
 790         fops_put(f->f_op);
 791         if (f->f_mode & FMODE_WRITE)
 792                 put_write_access(inode);
 793         file_kill(f);
 794         f->f_path.dentry = NULL;
 795         f->f_path.mnt = NULL;
 796 cleanup_file:
 797         put_filp(f);
 798         dput(dentry);
 799         mntput(mnt);
 800         return ERR_PTR(error);
 801 }
 802
 803 /*
 804  * Note that while the flag value (low two bits) for sys_open means:
 805  *      00 - read-only
 806  *      01 - write-only
 807  *      10 - read-write
 808  *      11 - special
 809  * it is changed into
 810  *      00 - no permissions needed
 811  *      01 - read-permission
 812  *      10 - write-permission
 813  *      11 - read-write
 814  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 815  * used by symlinks.
 816  */
 817 static struct file *do_filp_open(int dfd, const char *filename, int flags,
 818                                  int mode)
 819 {
 820         int namei_flags, error;
 821         struct nameidata nd;
 822
 823         namei_flags = flags;
 824         if ((namei_flags+1) & O_ACCMODE)
 825                 namei_flags++;
 826
 827         error = open_namei(dfd, filename, namei_flags, mode, &nd);
 828         if (!error)
 829                 return nameidata_to_filp(&nd, flags);
 830
 831         return ERR_PTR(error);
 832 }
 833
 834 struct file *filp_open(const char *filename, int flags, int mode)
 835 {
 836         return do_filp_open(AT_FDCWD, filename, flags, mode);
 837 }
 838 EXPORT_SYMBOL(filp_open);
 839
 840 /**
 841  * lookup_instantiate_filp - instantiates the open intent filp
 842  * @nd: pointer to nameidata
 843  * @dentry: pointer to dentry
 844  * @open: open callback
 845  *
 846  * Helper for filesystems that want to use lookup open intents and pass back
 847  * a fully instantiated struct file to the caller.
 848  * This function is meant to be called from within a filesystem's
 849  * lookup method.
 850  * Beware of calling it for non-regular files! Those ->open methods might block
 851  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 852  * leading to a deadlock, as nobody can open that fifo anymore, because
 853  * another process to open fifo will block on locked parent when doing lookup).
 854  * Note that in case of error, nd->intent.open.file is destroyed, but the
 855  * path information remains valid.
 856  * If the open callback is set to NULL, then the standard f_op->open()
 857  * filesystem callback is substituted.
 858  */
 859 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 860                 int (*open)(struct inode *, struct file *))
 861 {
 862         if (IS_ERR(nd->intent.open.file))
 863                 goto out;
 864         if (IS_ERR(dentry))
 865                 goto out_err;
 866         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
 867                                              nd->intent.open.flags - 1,
 868                                              nd->intent.open.file,
 869                                              open);
 870 out:
 871         return nd->intent.open.file;
 872 out_err:
 873         release_open_intent(nd);
 874         nd->intent.open.file = (struct file *)dentry;
 875         goto out;
 876 }
 877 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 878
 879 /**
 880  * nameidata_to_filp - convert a nameidata to an open filp.
 881  * @nd: pointer to nameidata
 882  * @flags: open flags
 883  *
 884  * Note that this function destroys the original nameidata
 885  */
 886 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 887 {
 888         struct file *filp;
 889
 890         /* Pick up the filp from the open intent */
 891         filp = nd->intent.open.file;
 892         /* Has the filesystem initialised the file for us? */
 893         if (filp->f_path.dentry == NULL)
 894                 filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
 895         else
 896                 path_release(nd);
 897         return filp;
 898 }
 899
 900 /*
 901  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 902  * error.
 903  */
 904 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 905 {
 906         int error;
 907         struct file *f;
 908
 909         error = -ENFILE;
 910         f = get_empty_filp();
 911         if (f == NULL) {
 912                 dput(dentry);
 913                 mntput(mnt);
 914                 return ERR_PTR(error);
 915         }
 916
 917         return __dentry_open(dentry, mnt, flags, f, NULL);
 918 }
 919 EXPORT_SYMBOL(dentry_open);
 920
 921 /*
 922  * Find an empty file descriptor entry, and mark it busy.
 923  */
 924 int get_unused_fd_flags(int flags)
 925 {
 926         struct files_struct * files = current->files;
 927         int fd, error;
 928         struct fdtable *fdt;
 929
 930         error = -EMFILE;
 931         spin_lock(&files->file_lock);
 932
 933 repeat:
 934         fdt = files_fdtable(files);
 935         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 936                                 files->next_fd);
 937
 938         /*
 939          * N.B. For clone tasks sharing a files structure, this test
 940          * will limit the total number of files that can be opened.
 941          */
 942         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 943                 goto out;
 944
 945         /* Do we need to expand the fd array or fd set?  */
 946         error = expand_files(files, fd);
 947         if (error < 0)
 948                 goto out;
 949
 950         if (error) {
 951                 /*
 952                  * If we needed to expand the fs array we
 953                  * might have blocked - try again.
 954                  */
 955                 error = -EMFILE;
 956                 goto repeat;
 957         }
 958
 959         FD_SET(fd, fdt->open_fds);
 960         if (flags & O_CLOEXEC)
 961                 FD_SET(fd, fdt->close_on_exec);
 962         else
 963                 FD_CLR(fd, fdt->close_on_exec);
 964         files->next_fd = fd + 1;
 965 #if 1
 966         /* Sanity check */
 967         if (fdt->fd[fd] != NULL) {
 968                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 969                 fdt->fd[fd] = NULL;
 970         }
 971 #endif
 972         error = fd;
 973
 974 out:
 975         spin_unlock(&files->file_lock);
 976         return error;
 977 }
 978
 979 int get_unused_fd(void)
 980 {
 981         return get_unused_fd_flags(0);
 982 }
 983
 984 EXPORT_SYMBOL(get_unused_fd);
 985
 986 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 987 {
 988         struct fdtable *fdt = files_fdtable(files);
 989         __FD_CLR(fd, fdt->open_fds);
 990         if (fd < files->next_fd)
 991                 files->next_fd = fd;
 992 }
 993
 994 void fastcall put_unused_fd(unsigned int fd)
 995 {
 996         struct files_struct *files = current->files;
 997         spin_lock(&files->file_lock);
 998         __put_unused_fd(files, fd);
 999         spin_unlock(&files->file_lock);
1000 }
1001
1002 EXPORT_SYMBOL(put_unused_fd);
1003
1004 /*
1005  * Install a file pointer in the fd array.
1006  *
1007  * The VFS is full of places where we drop the files lock between
1008  * setting the open_fds bitmap and installing the file in the file
1009  * array.  At any such point, we are vulnerable to a dup2() race
1010  * installing a file in the array before us.  We need to detect this and
1011  * fput() the struct file we are about to overwrite in this case.
1012  *
1013  * It should never happen - if we allow dup2() do it, _really_ bad things
1014  * will follow.
1015  */
1016
1017 void fastcall fd_install(unsigned int fd, struct file * file)
1018 {
1019         struct files_struct *files = current->files;
1020         struct fdtable *fdt;
1021         spin_lock(&files->file_lock);
1022         fdt = files_fdtable(files);
1023         BUG_ON(fdt->fd[fd] != NULL);
1024         rcu_assign_pointer(fdt->fd[fd], file);
1025         spin_unlock(&files->file_lock);
1026 }
1027
1028 EXPORT_SYMBOL(fd_install);
1029
1030 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1031 {
1032         char *tmp = getname(filename);
1033         int fd = PTR_ERR(tmp);
1034
1035         if (!IS_ERR(tmp)) {
1036                 fd = get_unused_fd_flags(flags);
1037                 if (fd >= 0) {
1038                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1039                         if (IS_ERR(f)) {
1040                                 put_unused_fd(fd);
1041                                 fd = PTR_ERR(f);
1042                         } else {
1043                                 fsnotify_open(f->f_path.dentry);
1044                                 fd_install(fd, f);
1045                         }
1046                 }
1047                 putname(tmp);
1048         }
1049         return fd;
1050 }
1051
1052 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1053 {
1054         long ret;
1055
1056         if (force_o_largefile())
1057                 flags |= O_LARGEFILE;
1058
1059         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1060         /* avoid REGPARM breakage on x86: */
1061         prevent_tail_call(ret);
1062         return ret;
1063 }
1064 EXPORT_SYMBOL_GPL(sys_open);
1065
1066 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1067                            int mode)
1068 {
1069         long ret;
1070
1071         if (force_o_largefile())
1072                 flags |= O_LARGEFILE;
1073
1074         ret = do_sys_open(dfd, filename, flags, mode);
1075         /* avoid REGPARM breakage on x86: */
1076         prevent_tail_call(ret);
1077         return ret;
1078 }
1079
1080 #ifndef __alpha__
1081
1082 /*
1083  * For backward compatibility?  Maybe this should be moved
1084  * into arch/i386 instead?
1085  */
1086 asmlinkage long sys_creat(const char __user * pathname, int mode)
1087 {
1088         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1089 }
1090
1091 #endif
1092
1093 /*
1094  * "id" is the POSIX thread ID. We use the
1095  * files pointer for this..
1096  */
1097 int filp_close(struct file *filp, fl_owner_t id)
1098 {
1099         int retval = 0;
1100
1101         if (!file_count(filp)) {
1102                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1103                 return 0;
1104         }
1105
1106         if (filp->f_op && filp->f_op->flush)
1107                 retval = filp->f_op->flush(filp, id);
1108
1109         dnotify_flush(filp, id);
1110         locks_remove_posix(filp, id);
1111         fput(filp);
1112         return retval;
1113 }
1114
1115 EXPORT_SYMBOL(filp_close);
1116
1117 /*
1118  * Careful here! We test whether the file pointer is NULL before
1119  * releasing the fd. This ensures that one clone task can't release
1120  * an fd while another clone is opening it.
1121  */
1122 asmlinkage long sys_close(unsigned int fd)
1123 {
1124         struct file * filp;
1125         struct files_struct *files = current->files;
1126         struct fdtable *fdt;
1127         int retval;
1128
1129         spin_lock(&files->file_lock);
1130         fdt = files_fdtable(files);
1131         if (fd >= fdt->max_fds)
1132                 goto out_unlock;
1133         filp = fdt->fd[fd];
1134         if (!filp)
1135                 goto out_unlock;
1136         rcu_assign_pointer(fdt->fd[fd], NULL);
1137         FD_CLR(fd, fdt->close_on_exec);
1138         __put_unused_fd(files, fd);
1139         spin_unlock(&files->file_lock);
1140         retval = filp_close(filp, files);
1141
1142         /* can't restart close syscall because file table entry was cleared */
1143         if (unlikely(retval == -ERESTARTSYS ||
1144                      retval == -ERESTARTNOINTR ||
1145                      retval == -ERESTARTNOHAND ||
1146                      retval == -ERESTART_RESTARTBLOCK))
1147                 retval = -EINTR;
1148
1149         return retval;
1150
1151 out_unlock:
1152         spin_unlock(&files->file_lock);
1153         return -EBADF;
1154 }
1155
1156 EXPORT_SYMBOL(sys_close);
1157
1158 /*
1159  * This routine simulates a hangup on the tty, to arrange that users
1160  * are given clean terminals at login time.
1161  */
1162 asmlinkage long sys_vhangup(void)
1163 {
1164         if (capable(CAP_SYS_TTY_CONFIG)) {
1165                 /* XXX: this needs locking */
1166                 tty_vhangup(current->signal->tty);
1167                 return 0;
1168         }
1169         return -EPERM;
1170 }
1171
1172 /*
1173  * Called when an inode is about to be open.
1174  * We use this to disallow opening large files on 32bit systems if
1175  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1176  * on this flag in sys_open.
1177  */
1178 int generic_file_open(struct inode * inode, struct file * filp)
1179 {
1180         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1181                 return -EOVERFLOW;
1182         return 0;
1183 }
1184
1185 EXPORT_SYMBOL(generic_file_open);
1186
1187 /*
1188  * This is used by subsystems that don't want seekable
1189  * file descriptors
1190  */
1191 int nonseekable_open(struct inode *inode, struct file *filp)
1192 {
1193         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1194         return 0;
1195 }
1196
1197 EXPORT_SYMBOL(nonseekable_open);