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/quotaops.h>
  12 #include <linux/fsnotify.h>
  13 #include <linux/module.h>
  14 #include <linux/slab.h>
  15 #include <linux/tty.h>
  16 #include <linux/namei.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/capability.h>
  19 #include <linux/securebits.h>
  20 #include <linux/security.h>
  21 #include <linux/mount.h>
  22 #include <linux/vfs.h>
  23 #include <linux/fcntl.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/fs.h>
  26 #include <linux/personality.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/audit.h>
  31 #include <linux/falloc.h>
  32
  33 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  34 {
  35         int retval = -ENODEV;
  36
  37         if (dentry) {
  38                 retval = -ENOSYS;
  39                 if (dentry->d_sb->s_op->statfs) {
  40                         memset(buf, 0, sizeof(*buf));
  41                         retval = security_sb_statfs(dentry);
  42                         if (retval)
  43                                 return retval;
  44                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  45                         if (retval == 0 && buf->f_frsize == 0)
  46                                 buf->f_frsize = buf->f_bsize;
  47                 }
  48         }
  49         return retval;
  50 }
  51
  52 EXPORT_SYMBOL(vfs_statfs);
  53
  54 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  55 {
  56         struct kstatfs st;
  57         int retval;
  58
  59         retval = vfs_statfs(dentry, &st);
  60         if (retval)
  61                 return retval;
  62
  63         if (sizeof(*buf) == sizeof(st))
  64                 memcpy(buf, &st, sizeof(st));
  65         else {
  66                 if (sizeof buf->f_blocks == 4) {
  67                         if ((st.f_blocks | st.f_bfree | st.f_bavail |
  68                              st.f_bsize | st.f_frsize) &
  69                             0xffffffff00000000ULL)
  70                                 return -EOVERFLOW;
  71                         /*
  72                          * f_files and f_ffree may be -1; it's okay to stuff
  73                          * that into 32 bits
  74                          */
  75                         if (st.f_files != -1 &&
  76                             (st.f_files & 0xffffffff00000000ULL))
  77                                 return -EOVERFLOW;
  78                         if (st.f_ffree != -1 &&
  79                             (st.f_ffree & 0xffffffff00000000ULL))
  80                                 return -EOVERFLOW;
  81                 }
  82
  83                 buf->f_type = st.f_type;
  84                 buf->f_bsize = st.f_bsize;
  85                 buf->f_blocks = st.f_blocks;
  86                 buf->f_bfree = st.f_bfree;
  87                 buf->f_bavail = st.f_bavail;
  88                 buf->f_files = st.f_files;
  89                 buf->f_ffree = st.f_ffree;
  90                 buf->f_fsid = st.f_fsid;
  91                 buf->f_namelen = st.f_namelen;
  92                 buf->f_frsize = st.f_frsize;
  93                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  94         }
  95         return 0;
  96 }
  97
  98 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
  99 {
 100         struct kstatfs st;
 101         int retval;
 102
 103         retval = vfs_statfs(dentry, &st);
 104         if (retval)
 105                 return retval;
 106
 107         if (sizeof(*buf) == sizeof(st))
 108                 memcpy(buf, &st, sizeof(st));
 109         else {
 110                 buf->f_type = st.f_type;
 111                 buf->f_bsize = st.f_bsize;
 112                 buf->f_blocks = st.f_blocks;
 113                 buf->f_bfree = st.f_bfree;
 114                 buf->f_bavail = st.f_bavail;
 115                 buf->f_files = st.f_files;
 116                 buf->f_ffree = st.f_ffree;
 117                 buf->f_fsid = st.f_fsid;
 118                 buf->f_namelen = st.f_namelen;
 119                 buf->f_frsize = st.f_frsize;
 120                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 121         }
 122         return 0;
 123 }
 124
 125 asmlinkage long sys_statfs(const char __user *pathname, struct statfs __user * buf)
 126 {
 127         struct path path;
 128         int error;
 129
 130         error = user_path(pathname, &path);
 131         if (!error) {
 132                 struct statfs tmp;
 133                 error = vfs_statfs_native(path.dentry, &tmp);
 134                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 135                         error = -EFAULT;
 136                 path_put(&path);
 137         }
 138         return error;
 139 }
 140
 141
 142 asmlinkage long sys_statfs64(const char __user *pathname, size_t sz, struct statfs64 __user *buf)
 143 {
 144         struct path path;
 145         long error;
 146
 147         if (sz != sizeof(*buf))
 148                 return -EINVAL;
 149         error = user_path(pathname, &path);
 150         if (!error) {
 151                 struct statfs64 tmp;
 152                 error = vfs_statfs64(path.dentry, &tmp);
 153                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 154                         error = -EFAULT;
 155                 path_put(&path);
 156         }
 157         return error;
 158 }
 159
 160
 161 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 162 {
 163         struct file * file;
 164         struct statfs tmp;
 165         int error;
 166
 167         error = -EBADF;
 168         file = fget(fd);
 169         if (!file)
 170                 goto out;
 171         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 172         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 173                 error = -EFAULT;
 174         fput(file);
 175 out:
 176         return error;
 177 }
 178
 179 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 180 {
 181         struct file * file;
 182         struct statfs64 tmp;
 183         int error;
 184
 185         if (sz != sizeof(*buf))
 186                 return -EINVAL;
 187
 188         error = -EBADF;
 189         file = fget(fd);
 190         if (!file)
 191                 goto out;
 192         error = vfs_statfs64(file->f_path.dentry, &tmp);
 193         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 194                 error = -EFAULT;
 195         fput(file);
 196 out:
 197         return error;
 198 }
 199
 200 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 201         struct file *filp)
 202 {
 203         int err;
 204         struct iattr newattrs;
 205
 206         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 207         if (length < 0)
 208                 return -EINVAL;
 209
 210         newattrs.ia_size = length;
 211         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 212         if (filp) {
 213                 newattrs.ia_file = filp;
 214                 newattrs.ia_valid |= ATTR_FILE;
 215         }
 216
 217         /* Remove suid/sgid on truncate too */
 218         newattrs.ia_valid |= should_remove_suid(dentry);
 219
 220         mutex_lock(&dentry->d_inode->i_mutex);
 221         err = notify_change(dentry, &newattrs);
 222         mutex_unlock(&dentry->d_inode->i_mutex);
 223         return err;
 224 }
 225
 226 static long do_sys_truncate(const char __user *pathname, loff_t length)
 227 {
 228         struct path path;
 229         struct inode *inode;
 230         int error;
 231
 232         error = -EINVAL;
 233         if (length < 0) /* sorry, but loff_t says... */
 234                 goto out;
 235
 236         error = user_path(pathname, &path);
 237         if (error)
 238                 goto out;
 239         inode = path.dentry->d_inode;
 240
 241         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 242         error = -EISDIR;
 243         if (S_ISDIR(inode->i_mode))
 244                 goto dput_and_out;
 245
 246         error = -EINVAL;
 247         if (!S_ISREG(inode->i_mode))
 248                 goto dput_and_out;
 249
 250         error = mnt_want_write(path.mnt);
 251         if (error)
 252                 goto dput_and_out;
 253
 254         error = inode_permission(inode, MAY_WRITE);
 255         if (error)
 256                 goto mnt_drop_write_and_out;
 257
 258         error = -EPERM;
 259         if (IS_APPEND(inode))
 260                 goto mnt_drop_write_and_out;
 261
 262         error = get_write_access(inode);
 263         if (error)
 264                 goto mnt_drop_write_and_out;
 265
 266         /*
 267          * Make sure that there are no leases.  get_write_access() protects
 268          * against the truncate racing with a lease-granting setlease().
 269          */
 270         error = break_lease(inode, FMODE_WRITE);
 271         if (error)
 272                 goto put_write_and_out;
 273
 274         error = locks_verify_truncate(inode, NULL, length);
 275         if (!error) {
 276                 DQUOT_INIT(inode);
 277                 error = do_truncate(path.dentry, length, 0, NULL);
 278         }
 279
 280 put_write_and_out:
 281         put_write_access(inode);
 282 mnt_drop_write_and_out:
 283         mnt_drop_write(path.mnt);
 284 dput_and_out:
 285         path_put(&path);
 286 out:
 287         return error;
 288 }
 289
 290 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 291 {
 292         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 293         return do_sys_truncate(path, (long)length);
 294 }
 295
 296 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 297 {
 298         struct inode * inode;
 299         struct dentry *dentry;
 300         struct file * file;
 301         int error;
 302
 303         error = -EINVAL;
 304         if (length < 0)
 305                 goto out;
 306         error = -EBADF;
 307         file = fget(fd);
 308         if (!file)
 309                 goto out;
 310
 311         /* explicitly opened as large or we are on 64-bit box */
 312         if (file->f_flags & O_LARGEFILE)
 313                 small = 0;
 314
 315         dentry = file->f_path.dentry;
 316         inode = dentry->d_inode;
 317         error = -EINVAL;
 318         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 319                 goto out_putf;
 320
 321         error = -EINVAL;
 322         /* Cannot ftruncate over 2^31 bytes without large file support */
 323         if (small && length > MAX_NON_LFS)
 324                 goto out_putf;
 325
 326         error = -EPERM;
 327         if (IS_APPEND(inode))
 328                 goto out_putf;
 329
 330         error = locks_verify_truncate(inode, file, length);
 331         if (!error)
 332                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 333 out_putf:
 334         fput(file);
 335 out:
 336         return error;
 337 }
 338
 339 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 340 {
 341         long ret = do_sys_ftruncate(fd, length, 1);
 342         /* avoid REGPARM breakage on x86: */
 343         asmlinkage_protect(2, ret, fd, length);
 344         return ret;
 345 }
 346
 347 /* LFS versions of truncate are only needed on 32 bit machines */
 348 #if BITS_PER_LONG == 32
 349 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 350 {
 351         return do_sys_truncate(path, length);
 352 }
 353
 354 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 355 {
 356         long ret = do_sys_ftruncate(fd, length, 0);
 357         /* avoid REGPARM breakage on x86: */
 358         asmlinkage_protect(2, ret, fd, length);
 359         return ret;
 360 }
 361 #endif
 362
 363 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 364 {
 365         struct file *file;
 366         struct inode *inode;
 367         long ret = -EINVAL;
 368
 369         if (offset < 0 || len <= 0)
 370                 goto out;
 371
 372         /* Return error if mode is not supported */
 373         ret = -EOPNOTSUPP;
 374         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 375                 goto out;
 376
 377         ret = -EBADF;
 378         file = fget(fd);
 379         if (!file)
 380                 goto out;
 381         if (!(file->f_mode & FMODE_WRITE))
 382                 goto out_fput;
 383         /*
 384          * Revalidate the write permissions, in case security policy has
 385          * changed since the files were opened.
 386          */
 387         ret = security_file_permission(file, MAY_WRITE);
 388         if (ret)
 389                 goto out_fput;
 390
 391         inode = file->f_path.dentry->d_inode;
 392
 393         ret = -ESPIPE;
 394         if (S_ISFIFO(inode->i_mode))
 395                 goto out_fput;
 396
 397         ret = -ENODEV;
 398         /*
 399          * Let individual file system decide if it supports preallocation
 400          * for directories or not.
 401          */
 402         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 403                 goto out_fput;
 404
 405         ret = -EFBIG;
 406         /* Check for wrap through zero too */
 407         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 408                 goto out_fput;
 409
 410         if (inode->i_op && inode->i_op->fallocate)
 411                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 412         else
 413                 ret = -EOPNOTSUPP;
 414
 415 out_fput:
 416         fput(file);
 417 out:
 418         return ret;
 419 }
 420
 421 /*
 422  * access() needs to use the real uid/gid, not the effective uid/gid.
 423  * We do this by temporarily clearing all FS-related capabilities and
 424  * switching the fsuid/fsgid around to the real ones.
 425  */
 426 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 427 {
 428         struct cred *cred = current->cred;
 429         struct path path;
 430         struct inode *inode;
 431         int old_fsuid, old_fsgid;
 432         kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
 433         int res;
 434
 435         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 436                 return -EINVAL;
 437
 438         old_fsuid = cred->fsuid;
 439         old_fsgid = cred->fsgid;
 440
 441         cred->fsuid = cred->uid;
 442         cred->fsgid = cred->gid;
 443
 444         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 445                 /* Clear the capabilities if we switch to a non-root user */
 446                 if (current->cred->uid)
 447                         old_cap = cap_set_effective(__cap_empty_set);
 448                 else
 449                         old_cap = cap_set_effective(cred->cap_permitted);
 450         }
 451
 452         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 453         if (res)
 454                 goto out;
 455
 456         inode = path.dentry->d_inode;
 457
 458         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 459                 /*
 460                  * MAY_EXEC on regular files is denied if the fs is mounted
 461                  * with the "noexec" flag.
 462                  */
 463                 res = -EACCES;
 464                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 465                         goto out_path_release;
 466         }
 467
 468         res = inode_permission(inode, mode | MAY_ACCESS);
 469         /* SuS v2 requires we report a read only fs too */
 470         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 471                 goto out_path_release;
 472         /*
 473          * This is a rare case where using __mnt_is_readonly()
 474          * is OK without a mnt_want/drop_write() pair.  Since
 475          * no actual write to the fs is performed here, we do
 476          * not need to telegraph to that to anyone.
 477          *
 478          * By doing this, we accept that this access is
 479          * inherently racy and know that the fs may change
 480          * state before we even see this result.
 481          */
 482         if (__mnt_is_readonly(path.mnt))
 483                 res = -EROFS;
 484
 485 out_path_release:
 486         path_put(&path);
 487 out:
 488         cred->fsuid = old_fsuid;
 489         cred->fsgid = old_fsgid;
 490
 491         if (!issecure(SECURE_NO_SETUID_FIXUP))
 492                 cap_set_effective(old_cap);
 493
 494         return res;
 495 }
 496
 497 asmlinkage long sys_access(const char __user *filename, int mode)
 498 {
 499         return sys_faccessat(AT_FDCWD, filename, mode);
 500 }
 501
 502 asmlinkage long sys_chdir(const char __user * filename)
 503 {
 504         struct path path;
 505         int error;
 506
 507         error = user_path_dir(filename, &path);
 508         if (error)
 509                 goto out;
 510
 511         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 512         if (error)
 513                 goto dput_and_out;
 514
 515         set_fs_pwd(current->fs, &path);
 516
 517 dput_and_out:
 518         path_put(&path);
 519 out:
 520         return error;
 521 }
 522
 523 asmlinkage long sys_fchdir(unsigned int fd)
 524 {
 525         struct file *file;
 526         struct inode *inode;
 527         int error;
 528
 529         error = -EBADF;
 530         file = fget(fd);
 531         if (!file)
 532                 goto out;
 533
 534         inode = file->f_path.dentry->d_inode;
 535
 536         error = -ENOTDIR;
 537         if (!S_ISDIR(inode->i_mode))
 538                 goto out_putf;
 539
 540         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 541         if (!error)
 542                 set_fs_pwd(current->fs, &file->f_path);
 543 out_putf:
 544         fput(file);
 545 out:
 546         return error;
 547 }
 548
 549 asmlinkage long sys_chroot(const char __user * filename)
 550 {
 551         struct path path;
 552         int error;
 553
 554         error = user_path_dir(filename, &path);
 555         if (error)
 556                 goto out;
 557
 558         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 559         if (error)
 560                 goto dput_and_out;
 561
 562         error = -EPERM;
 563         if (!capable(CAP_SYS_CHROOT))
 564                 goto dput_and_out;
 565
 566         set_fs_root(current->fs, &path);
 567         error = 0;
 568 dput_and_out:
 569         path_put(&path);
 570 out:
 571         return error;
 572 }
 573
 574 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 575 {
 576         struct inode * inode;
 577         struct dentry * dentry;
 578         struct file * file;
 579         int err = -EBADF;
 580         struct iattr newattrs;
 581
 582         file = fget(fd);
 583         if (!file)
 584                 goto out;
 585
 586         dentry = file->f_path.dentry;
 587         inode = dentry->d_inode;
 588
 589         audit_inode(NULL, dentry);
 590
 591         err = mnt_want_write(file->f_path.mnt);
 592         if (err)
 593                 goto out_putf;
 594         mutex_lock(&inode->i_mutex);
 595         if (mode == (mode_t) -1)
 596                 mode = inode->i_mode;
 597         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 598         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 599         err = notify_change(dentry, &newattrs);
 600         mutex_unlock(&inode->i_mutex);
 601         mnt_drop_write(file->f_path.mnt);
 602 out_putf:
 603         fput(file);
 604 out:
 605         return err;
 606 }
 607
 608 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 609                              mode_t mode)
 610 {
 611         struct path path;
 612         struct inode *inode;
 613         int error;
 614         struct iattr newattrs;
 615
 616         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 617         if (error)
 618                 goto out;
 619         inode = path.dentry->d_inode;
 620
 621         error = mnt_want_write(path.mnt);
 622         if (error)
 623                 goto dput_and_out;
 624         mutex_lock(&inode->i_mutex);
 625         if (mode == (mode_t) -1)
 626                 mode = inode->i_mode;
 627         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 628         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 629         error = notify_change(path.dentry, &newattrs);
 630         mutex_unlock(&inode->i_mutex);
 631         mnt_drop_write(path.mnt);
 632 dput_and_out:
 633         path_put(&path);
 634 out:
 635         return error;
 636 }
 637
 638 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 639 {
 640         return sys_fchmodat(AT_FDCWD, filename, mode);
 641 }
 642
 643 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 644 {
 645         struct inode *inode = dentry->d_inode;
 646         int error;
 647         struct iattr newattrs;
 648
 649         newattrs.ia_valid =  ATTR_CTIME;
 650         if (user != (uid_t) -1) {
 651                 newattrs.ia_valid |= ATTR_UID;
 652                 newattrs.ia_uid = user;
 653         }
 654         if (group != (gid_t) -1) {
 655                 newattrs.ia_valid |= ATTR_GID;
 656                 newattrs.ia_gid = group;
 657         }
 658         if (!S_ISDIR(inode->i_mode))
 659                 newattrs.ia_valid |=
 660                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 661         mutex_lock(&inode->i_mutex);
 662         error = notify_change(dentry, &newattrs);
 663         mutex_unlock(&inode->i_mutex);
 664
 665         return error;
 666 }
 667
 668 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 669 {
 670         struct path path;
 671         int error;
 672
 673         error = user_path(filename, &path);
 674         if (error)
 675                 goto out;
 676         error = mnt_want_write(path.mnt);
 677         if (error)
 678                 goto out_release;
 679         error = chown_common(path.dentry, user, group);
 680         mnt_drop_write(path.mnt);
 681 out_release:
 682         path_put(&path);
 683 out:
 684         return error;
 685 }
 686
 687 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 688                              gid_t group, int flag)
 689 {
 690         struct path path;
 691         int error = -EINVAL;
 692         int follow;
 693
 694         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 695                 goto out;
 696
 697         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 698         error = user_path_at(dfd, filename, follow, &path);
 699         if (error)
 700                 goto out;
 701         error = mnt_want_write(path.mnt);
 702         if (error)
 703                 goto out_release;
 704         error = chown_common(path.dentry, user, group);
 705         mnt_drop_write(path.mnt);
 706 out_release:
 707         path_put(&path);
 708 out:
 709         return error;
 710 }
 711
 712 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 713 {
 714         struct path path;
 715         int error;
 716
 717         error = user_lpath(filename, &path);
 718         if (error)
 719                 goto out;
 720         error = mnt_want_write(path.mnt);
 721         if (error)
 722                 goto out_release;
 723         error = chown_common(path.dentry, user, group);
 724         mnt_drop_write(path.mnt);
 725 out_release:
 726         path_put(&path);
 727 out:
 728         return error;
 729 }
 730
 731
 732 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 733 {
 734         struct file * file;
 735         int error = -EBADF;
 736         struct dentry * dentry;
 737
 738         file = fget(fd);
 739         if (!file)
 740                 goto out;
 741
 742         error = mnt_want_write(file->f_path.mnt);
 743         if (error)
 744                 goto out_fput;
 745         dentry = file->f_path.dentry;
 746         audit_inode(NULL, dentry);
 747         error = chown_common(dentry, user, group);
 748         mnt_drop_write(file->f_path.mnt);
 749 out_fput:
 750         fput(file);
 751 out:
 752         return error;
 753 }
 754
 755 /*
 756  * You have to be very careful that these write
 757  * counts get cleaned up in error cases and
 758  * upon __fput().  This should probably never
 759  * be called outside of __dentry_open().
 760  */
 761 static inline int __get_file_write_access(struct inode *inode,
 762                                           struct vfsmount *mnt)
 763 {
 764         int error;
 765         error = get_write_access(inode);
 766         if (error)
 767                 return error;
 768         /*
 769          * Do not take mount writer counts on
 770          * special files since no writes to
 771          * the mount itself will occur.
 772          */
 773         if (!special_file(inode->i_mode)) {
 774                 /*
 775                  * Balanced in __fput()
 776                  */
 777                 error = mnt_want_write(mnt);
 778                 if (error)
 779                         put_write_access(inode);
 780         }
 781         return error;
 782 }
 783
 784 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 785                                         int flags, struct file *f,
 786                                         int (*open)(struct inode *, struct file *),
 787                                         const struct cred *cred)
 788 {
 789         struct inode *inode;
 790         int error;
 791
 792         f->f_flags = flags;
 793         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 794                                 FMODE_PREAD | FMODE_PWRITE;
 795         inode = dentry->d_inode;
 796         if (f->f_mode & FMODE_WRITE) {
 797                 error = __get_file_write_access(inode, mnt);
 798                 if (error)
 799                         goto cleanup_file;
 800                 if (!special_file(inode->i_mode))
 801                         file_take_write(f);
 802         }
 803
 804         f->f_mapping = inode->i_mapping;
 805         f->f_path.dentry = dentry;
 806         f->f_path.mnt = mnt;
 807         f->f_pos = 0;
 808         f->f_op = fops_get(inode->i_fop);
 809         file_move(f, &inode->i_sb->s_files);
 810
 811         error = security_dentry_open(f, cred);
 812         if (error)
 813                 goto cleanup_all;
 814
 815         if (!open && f->f_op)
 816                 open = f->f_op->open;
 817         if (open) {
 818                 error = open(inode, f);
 819                 if (error)
 820                         goto cleanup_all;
 821         }
 822
 823         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 824
 825         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 826
 827         /* NB: we're sure to have correct a_ops only after f_op->open */
 828         if (f->f_flags & O_DIRECT) {
 829                 if (!f->f_mapping->a_ops ||
 830                     ((!f->f_mapping->a_ops->direct_IO) &&
 831                     (!f->f_mapping->a_ops->get_xip_mem))) {
 832                         fput(f);
 833                         f = ERR_PTR(-EINVAL);
 834                 }
 835         }
 836
 837         return f;
 838
 839 cleanup_all:
 840         fops_put(f->f_op);
 841         if (f->f_mode & FMODE_WRITE) {
 842                 put_write_access(inode);
 843                 if (!special_file(inode->i_mode)) {
 844                         /*
 845                          * We don't consider this a real
 846                          * mnt_want/drop_write() pair
 847                          * because it all happenend right
 848                          * here, so just reset the state.
 849                          */
 850                         file_reset_write(f);
 851                         mnt_drop_write(mnt);
 852                 }
 853         }
 854         file_kill(f);
 855         f->f_path.dentry = NULL;
 856         f->f_path.mnt = NULL;
 857 cleanup_file:
 858         put_filp(f);
 859         dput(dentry);
 860         mntput(mnt);
 861         return ERR_PTR(error);
 862 }
 863
 864 /**
 865  * lookup_instantiate_filp - instantiates the open intent filp
 866  * @nd: pointer to nameidata
 867  * @dentry: pointer to dentry
 868  * @open: open callback
 869  *
 870  * Helper for filesystems that want to use lookup open intents and pass back
 871  * a fully instantiated struct file to the caller.
 872  * This function is meant to be called from within a filesystem's
 873  * lookup method.
 874  * Beware of calling it for non-regular files! Those ->open methods might block
 875  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 876  * leading to a deadlock, as nobody can open that fifo anymore, because
 877  * another process to open fifo will block on locked parent when doing lookup).
 878  * Note that in case of error, nd->intent.open.file is destroyed, but the
 879  * path information remains valid.
 880  * If the open callback is set to NULL, then the standard f_op->open()
 881  * filesystem callback is substituted.
 882  */
 883 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 884                 int (*open)(struct inode *, struct file *))
 885 {
 886         const struct cred *cred = current_cred();
 887
 888         if (IS_ERR(nd->intent.open.file))
 889                 goto out;
 890         if (IS_ERR(dentry))
 891                 goto out_err;
 892         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 893                                              nd->intent.open.flags - 1,
 894                                              nd->intent.open.file,
 895                                              open, cred);
 896 out:
 897         return nd->intent.open.file;
 898 out_err:
 899         release_open_intent(nd);
 900         nd->intent.open.file = (struct file *)dentry;
 901         goto out;
 902 }
 903 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 904
 905 /**
 906  * nameidata_to_filp - convert a nameidata to an open filp.
 907  * @nd: pointer to nameidata
 908  * @flags: open flags
 909  *
 910  * Note that this function destroys the original nameidata
 911  */
 912 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 913 {
 914         const struct cred *cred = current_cred();
 915         struct file *filp;
 916
 917         /* Pick up the filp from the open intent */
 918         filp = nd->intent.open.file;
 919         /* Has the filesystem initialised the file for us? */
 920         if (filp->f_path.dentry == NULL)
 921                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 922                                      NULL, cred);
 923         else
 924                 path_put(&nd->path);
 925         return filp;
 926 }
 927
 928 /*
 929  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 930  * error.
 931  */
 932 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 933                          const struct cred *cred)
 934 {
 935         int error;
 936         struct file *f;
 937
 938         /*
 939          * We must always pass in a valid mount pointer.   Historically
 940          * callers got away with not passing it, but we must enforce this at
 941          * the earliest possible point now to avoid strange problems deep in the
 942          * filesystem stack.
 943          */
 944         if (!mnt) {
 945                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 946                 dump_stack();
 947                 return ERR_PTR(-EINVAL);
 948         }
 949
 950         error = -ENFILE;
 951         f = get_empty_filp();
 952         if (f == NULL) {
 953                 dput(dentry);
 954                 mntput(mnt);
 955                 return ERR_PTR(error);
 956         }
 957
 958         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 959 }
 960 EXPORT_SYMBOL(dentry_open);
 961
 962 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 963 {
 964         struct fdtable *fdt = files_fdtable(files);
 965         __FD_CLR(fd, fdt->open_fds);
 966         if (fd < files->next_fd)
 967                 files->next_fd = fd;
 968 }
 969
 970 void put_unused_fd(unsigned int fd)
 971 {
 972         struct files_struct *files = current->files;
 973         spin_lock(&files->file_lock);
 974         __put_unused_fd(files, fd);
 975         spin_unlock(&files->file_lock);
 976 }
 977
 978 EXPORT_SYMBOL(put_unused_fd);
 979
 980 /*
 981  * Install a file pointer in the fd array.
 982  *
 983  * The VFS is full of places where we drop the files lock between
 984  * setting the open_fds bitmap and installing the file in the file
 985  * array.  At any such point, we are vulnerable to a dup2() race
 986  * installing a file in the array before us.  We need to detect this and
 987  * fput() the struct file we are about to overwrite in this case.
 988  *
 989  * It should never happen - if we allow dup2() do it, _really_ bad things
 990  * will follow.
 991  */
 992
 993 void fd_install(unsigned int fd, struct file *file)
 994 {
 995         struct files_struct *files = current->files;
 996         struct fdtable *fdt;
 997         spin_lock(&files->file_lock);
 998         fdt = files_fdtable(files);
 999         BUG_ON(fdt->fd[fd] != NULL);
1000         rcu_assign_pointer(fdt->fd[fd], file);
1001         spin_unlock(&files->file_lock);
1002 }
1003
1004 EXPORT_SYMBOL(fd_install);
1005
1006 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1007 {
1008         char *tmp = getname(filename);
1009         int fd = PTR_ERR(tmp);
1010
1011         if (!IS_ERR(tmp)) {
1012                 fd = get_unused_fd_flags(flags);
1013                 if (fd >= 0) {
1014                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1015                         if (IS_ERR(f)) {
1016                                 put_unused_fd(fd);
1017                                 fd = PTR_ERR(f);
1018                         } else {
1019                                 fsnotify_open(f->f_path.dentry);
1020                                 fd_install(fd, f);
1021                         }
1022                 }
1023                 putname(tmp);
1024         }
1025         return fd;
1026 }
1027
1028 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1029 {
1030         long ret;
1031
1032         if (force_o_largefile())
1033                 flags |= O_LARGEFILE;
1034
1035         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1036         /* avoid REGPARM breakage on x86: */
1037         asmlinkage_protect(3, ret, filename, flags, mode);
1038         return ret;
1039 }
1040
1041 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1042                            int mode)
1043 {
1044         long ret;
1045
1046         if (force_o_largefile())
1047                 flags |= O_LARGEFILE;
1048
1049         ret = do_sys_open(dfd, filename, flags, mode);
1050         /* avoid REGPARM breakage on x86: */
1051         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1052         return ret;
1053 }
1054
1055 #ifndef __alpha__
1056
1057 /*
1058  * For backward compatibility?  Maybe this should be moved
1059  * into arch/i386 instead?
1060  */
1061 asmlinkage long sys_creat(const char __user * pathname, int mode)
1062 {
1063         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1064 }
1065
1066 #endif
1067
1068 /*
1069  * "id" is the POSIX thread ID. We use the
1070  * files pointer for this..
1071  */
1072 int filp_close(struct file *filp, fl_owner_t id)
1073 {
1074         int retval = 0;
1075
1076         if (!file_count(filp)) {
1077                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1078                 return 0;
1079         }
1080
1081         if (filp->f_op && filp->f_op->flush)
1082                 retval = filp->f_op->flush(filp, id);
1083
1084         dnotify_flush(filp, id);
1085         locks_remove_posix(filp, id);
1086         fput(filp);
1087         return retval;
1088 }
1089
1090 EXPORT_SYMBOL(filp_close);
1091
1092 /*
1093  * Careful here! We test whether the file pointer is NULL before
1094  * releasing the fd. This ensures that one clone task can't release
1095  * an fd while another clone is opening it.
1096  */
1097 asmlinkage long sys_close(unsigned int fd)
1098 {
1099         struct file * filp;
1100         struct files_struct *files = current->files;
1101         struct fdtable *fdt;
1102         int retval;
1103
1104         spin_lock(&files->file_lock);
1105         fdt = files_fdtable(files);
1106         if (fd >= fdt->max_fds)
1107                 goto out_unlock;
1108         filp = fdt->fd[fd];
1109         if (!filp)
1110                 goto out_unlock;
1111         rcu_assign_pointer(fdt->fd[fd], NULL);
1112         FD_CLR(fd, fdt->close_on_exec);
1113         __put_unused_fd(files, fd);
1114         spin_unlock(&files->file_lock);
1115         retval = filp_close(filp, files);
1116
1117         /* can't restart close syscall because file table entry was cleared */
1118         if (unlikely(retval == -ERESTARTSYS ||
1119                      retval == -ERESTARTNOINTR ||
1120                      retval == -ERESTARTNOHAND ||
1121                      retval == -ERESTART_RESTARTBLOCK))
1122                 retval = -EINTR;
1123
1124         return retval;
1125
1126 out_unlock:
1127         spin_unlock(&files->file_lock);
1128         return -EBADF;
1129 }
1130
1131 EXPORT_SYMBOL(sys_close);
1132
1133 /*
1134  * This routine simulates a hangup on the tty, to arrange that users
1135  * are given clean terminals at login time.
1136  */
1137 asmlinkage long sys_vhangup(void)
1138 {
1139         if (capable(CAP_SYS_TTY_CONFIG)) {
1140                 tty_vhangup_self();
1141                 return 0;
1142         }
1143         return -EPERM;
1144 }
1145
1146 /*
1147  * Called when an inode is about to be open.
1148  * We use this to disallow opening large files on 32bit systems if
1149  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1150  * on this flag in sys_open.
1151  */
1152 int generic_file_open(struct inode * inode, struct file * filp)
1153 {
1154         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1155                 return -EOVERFLOW;
1156         return 0;
1157 }
1158
1159 EXPORT_SYMBOL(generic_file_open);
1160
1161 /*
1162  * This is used by subsystems that don't want seekable
1163  * file descriptors
1164  */
1165 int nonseekable_open(struct inode *inode, struct file *filp)
1166 {
1167         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1168         return 0;
1169 }
1170
1171 EXPORT_SYMBOL(nonseekable_open);