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 #include <linux/fs_struct.h>
  33
  34 #include "internal.h"
  35
  36 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  37 {
  38         int retval = -ENODEV;
  39
  40         if (dentry) {
  41                 retval = -ENOSYS;
  42                 if (dentry->d_sb->s_op->statfs) {
  43                         memset(buf, 0, sizeof(*buf));
  44                         retval = security_sb_statfs(dentry);
  45                         if (retval)
  46                                 return retval;
  47                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  48                         if (retval == 0 && buf->f_frsize == 0)
  49                                 buf->f_frsize = buf->f_bsize;
  50                 }
  51         }
  52         return retval;
  53 }
  54
  55 EXPORT_SYMBOL(vfs_statfs);
  56
  57 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  58 {
  59         struct kstatfs st;
  60         int retval;
  61
  62         retval = vfs_statfs(dentry, &st);
  63         if (retval)
  64                 return retval;
  65
  66         if (sizeof(*buf) == sizeof(st))
  67                 memcpy(buf, &st, sizeof(st));
  68         else {
  69                 if (sizeof buf->f_blocks == 4) {
  70                         if ((st.f_blocks | st.f_bfree | st.f_bavail |
  71                              st.f_bsize | st.f_frsize) &
  72                             0xffffffff00000000ULL)
  73                                 return -EOVERFLOW;
  74                         /*
  75                          * f_files and f_ffree may be -1; it's okay to stuff
  76                          * that into 32 bits
  77                          */
  78                         if (st.f_files != -1 &&
  79                             (st.f_files & 0xffffffff00000000ULL))
  80                                 return -EOVERFLOW;
  81                         if (st.f_ffree != -1 &&
  82                             (st.f_ffree & 0xffffffff00000000ULL))
  83                                 return -EOVERFLOW;
  84                 }
  85
  86                 buf->f_type = st.f_type;
  87                 buf->f_bsize = st.f_bsize;
  88                 buf->f_blocks = st.f_blocks;
  89                 buf->f_bfree = st.f_bfree;
  90                 buf->f_bavail = st.f_bavail;
  91                 buf->f_files = st.f_files;
  92                 buf->f_ffree = st.f_ffree;
  93                 buf->f_fsid = st.f_fsid;
  94                 buf->f_namelen = st.f_namelen;
  95                 buf->f_frsize = st.f_frsize;
  96                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  97         }
  98         return 0;
  99 }
 100
 101 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
 102 {
 103         struct kstatfs st;
 104         int retval;
 105
 106         retval = vfs_statfs(dentry, &st);
 107         if (retval)
 108                 return retval;
 109
 110         if (sizeof(*buf) == sizeof(st))
 111                 memcpy(buf, &st, sizeof(st));
 112         else {
 113                 buf->f_type = st.f_type;
 114                 buf->f_bsize = st.f_bsize;
 115                 buf->f_blocks = st.f_blocks;
 116                 buf->f_bfree = st.f_bfree;
 117                 buf->f_bavail = st.f_bavail;
 118                 buf->f_files = st.f_files;
 119                 buf->f_ffree = st.f_ffree;
 120                 buf->f_fsid = st.f_fsid;
 121                 buf->f_namelen = st.f_namelen;
 122                 buf->f_frsize = st.f_frsize;
 123                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 124         }
 125         return 0;
 126 }
 127
 128 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
 129 {
 130         struct path path;
 131         int error;
 132
 133         error = user_path(pathname, &path);
 134         if (!error) {
 135                 struct statfs tmp;
 136                 error = vfs_statfs_native(path.dentry, &tmp);
 137                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 138                         error = -EFAULT;
 139                 path_put(&path);
 140         }
 141         return error;
 142 }
 143
 144 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
 145 {
 146         struct path path;
 147         long error;
 148
 149         if (sz != sizeof(*buf))
 150                 return -EINVAL;
 151         error = user_path(pathname, &path);
 152         if (!error) {
 153                 struct statfs64 tmp;
 154                 error = vfs_statfs64(path.dentry, &tmp);
 155                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 156                         error = -EFAULT;
 157                 path_put(&path);
 158         }
 159         return error;
 160 }
 161
 162 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
 163 {
 164         struct file * file;
 165         struct statfs tmp;
 166         int error;
 167
 168         error = -EBADF;
 169         file = fget(fd);
 170         if (!file)
 171                 goto out;
 172         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 173         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 174                 error = -EFAULT;
 175         fput(file);
 176 out:
 177         return error;
 178 }
 179
 180 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
 181 {
 182         struct file * file;
 183         struct statfs64 tmp;
 184         int error;
 185
 186         if (sz != sizeof(*buf))
 187                 return -EINVAL;
 188
 189         error = -EBADF;
 190         file = fget(fd);
 191         if (!file)
 192                 goto out;
 193         error = vfs_statfs64(file->f_path.dentry, &tmp);
 194         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 195                 error = -EFAULT;
 196         fput(file);
 197 out:
 198         return error;
 199 }
 200
 201 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 202         struct file *filp)
 203 {
 204         int ret;
 205         struct iattr newattrs;
 206
 207         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 208         if (length < 0)
 209                 return -EINVAL;
 210
 211         newattrs.ia_size = length;
 212         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 213         if (filp) {
 214                 newattrs.ia_file = filp;
 215                 newattrs.ia_valid |= ATTR_FILE;
 216         }
 217
 218         /* Remove suid/sgid on truncate too */
 219         ret = should_remove_suid(dentry);
 220         if (ret)
 221                 newattrs.ia_valid |= ret | ATTR_FORCE;
 222
 223         mutex_lock(&dentry->d_inode->i_mutex);
 224         ret = notify_change(dentry, &newattrs);
 225         mutex_unlock(&dentry->d_inode->i_mutex);
 226         return ret;
 227 }
 228
 229 static long do_sys_truncate(const char __user *pathname, loff_t length)
 230 {
 231         struct path path;
 232         struct inode *inode;
 233         int error;
 234
 235         error = -EINVAL;
 236         if (length < 0) /* sorry, but loff_t says... */
 237                 goto out;
 238
 239         error = user_path(pathname, &path);
 240         if (error)
 241                 goto out;
 242         inode = path.dentry->d_inode;
 243
 244         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 245         error = -EISDIR;
 246         if (S_ISDIR(inode->i_mode))
 247                 goto dput_and_out;
 248
 249         error = -EINVAL;
 250         if (!S_ISREG(inode->i_mode))
 251                 goto dput_and_out;
 252
 253         error = mnt_want_write(path.mnt);
 254         if (error)
 255                 goto dput_and_out;
 256
 257         error = inode_permission(inode, MAY_WRITE);
 258         if (error)
 259                 goto mnt_drop_write_and_out;
 260
 261         error = -EPERM;
 262         if (IS_APPEND(inode))
 263                 goto mnt_drop_write_and_out;
 264
 265         error = get_write_access(inode);
 266         if (error)
 267                 goto mnt_drop_write_and_out;
 268
 269         /*
 270          * Make sure that there are no leases.  get_write_access() protects
 271          * against the truncate racing with a lease-granting setlease().
 272          */
 273         error = break_lease(inode, FMODE_WRITE);
 274         if (error)
 275                 goto put_write_and_out;
 276
 277         error = locks_verify_truncate(inode, NULL, length);
 278         if (!error)
 279                 error = security_path_truncate(&path, length, 0);
 280         if (!error) {
 281                 vfs_dq_init(inode);
 282                 error = do_truncate(path.dentry, length, 0, NULL);
 283         }
 284
 285 put_write_and_out:
 286         put_write_access(inode);
 287 mnt_drop_write_and_out:
 288         mnt_drop_write(path.mnt);
 289 dput_and_out:
 290         path_put(&path);
 291 out:
 292         return error;
 293 }
 294
 295 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 296 {
 297         return do_sys_truncate(path, length);
 298 }
 299
 300 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 301 {
 302         struct inode * inode;
 303         struct dentry *dentry;
 304         struct file * file;
 305         int error;
 306
 307         error = -EINVAL;
 308         if (length < 0)
 309                 goto out;
 310         error = -EBADF;
 311         file = fget(fd);
 312         if (!file)
 313                 goto out;
 314
 315         /* explicitly opened as large or we are on 64-bit box */
 316         if (file->f_flags & O_LARGEFILE)
 317                 small = 0;
 318
 319         dentry = file->f_path.dentry;
 320         inode = dentry->d_inode;
 321         error = -EINVAL;
 322         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 323                 goto out_putf;
 324
 325         error = -EINVAL;
 326         /* Cannot ftruncate over 2^31 bytes without large file support */
 327         if (small && length > MAX_NON_LFS)
 328                 goto out_putf;
 329
 330         error = -EPERM;
 331         if (IS_APPEND(inode))
 332                 goto out_putf;
 333
 334         error = locks_verify_truncate(inode, file, length);
 335         if (!error)
 336                 error = security_path_truncate(&file->f_path, length,
 337                                                ATTR_MTIME|ATTR_CTIME);
 338         if (!error)
 339                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 340 out_putf:
 341         fput(file);
 342 out:
 343         return error;
 344 }
 345
 346 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 347 {
 348         long ret = do_sys_ftruncate(fd, length, 1);
 349         /* avoid REGPARM breakage on x86: */
 350         asmlinkage_protect(2, ret, fd, length);
 351         return ret;
 352 }
 353
 354 /* LFS versions of truncate are only needed on 32 bit machines */
 355 #if BITS_PER_LONG == 32
 356 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 357 {
 358         return do_sys_truncate(path, length);
 359 }
 360 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 361 asmlinkage long SyS_truncate64(long path, loff_t length)
 362 {
 363         return SYSC_truncate64((const char __user *) path, length);
 364 }
 365 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 366 #endif
 367
 368 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 369 {
 370         long ret = do_sys_ftruncate(fd, length, 0);
 371         /* avoid REGPARM breakage on x86: */
 372         asmlinkage_protect(2, ret, fd, length);
 373         return ret;
 374 }
 375 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 376 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 377 {
 378         return SYSC_ftruncate64((unsigned int) fd, length);
 379 }
 380 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 381 #endif
 382 #endif /* BITS_PER_LONG == 32 */
 383
 384
 385 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 386 {
 387         struct inode *inode = file->f_path.dentry->d_inode;
 388         long ret;
 389
 390         if (offset < 0 || len <= 0)
 391                 return -EINVAL;
 392
 393         /* Return error if mode is not supported */
 394         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 395                 return -EOPNOTSUPP;
 396
 397         if (!(file->f_mode & FMODE_WRITE))
 398                 return -EBADF;
 399         /*
 400          * Revalidate the write permissions, in case security policy has
 401          * changed since the files were opened.
 402          */
 403         ret = security_file_permission(file, MAY_WRITE);
 404         if (ret)
 405                 return ret;
 406
 407         if (S_ISFIFO(inode->i_mode))
 408                 return -ESPIPE;
 409
 410         /*
 411          * Let individual file system decide if it supports preallocation
 412          * for directories or not.
 413          */
 414         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 415                 return -ENODEV;
 416
 417         /* Check for wrap through zero too */
 418         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 419                 return -EFBIG;
 420
 421         if (!inode->i_op->fallocate)
 422                 return -EOPNOTSUPP;
 423
 424         return inode->i_op->fallocate(inode, mode, offset, len);
 425 }
 426
 427 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 428 {
 429         struct file *file;
 430         int error = -EBADF;
 431
 432         file = fget(fd);
 433         if (file) {
 434                 error = do_fallocate(file, mode, offset, len);
 435                 fput(file);
 436         }
 437
 438         return error;
 439 }
 440
 441 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 442 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 443 {
 444         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 445 }
 446 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 447 #endif
 448
 449 /*
 450  * access() needs to use the real uid/gid, not the effective uid/gid.
 451  * We do this by temporarily clearing all FS-related capabilities and
 452  * switching the fsuid/fsgid around to the real ones.
 453  */
 454 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 455 {
 456         const struct cred *old_cred;
 457         struct cred *override_cred;
 458         struct path path;
 459         struct inode *inode;
 460         int res;
 461
 462         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 463                 return -EINVAL;
 464
 465         override_cred = prepare_creds();
 466         if (!override_cred)
 467                 return -ENOMEM;
 468
 469         override_cred->fsuid = override_cred->uid;
 470         override_cred->fsgid = override_cred->gid;
 471
 472         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 473                 /* Clear the capabilities if we switch to a non-root user */
 474                 if (override_cred->uid)
 475                         cap_clear(override_cred->cap_effective);
 476                 else
 477                         override_cred->cap_effective =
 478                                 override_cred->cap_permitted;
 479         }
 480
 481         old_cred = override_creds(override_cred);
 482
 483         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 484         if (res)
 485                 goto out;
 486
 487         inode = path.dentry->d_inode;
 488
 489         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 490                 /*
 491                  * MAY_EXEC on regular files is denied if the fs is mounted
 492                  * with the "noexec" flag.
 493                  */
 494                 res = -EACCES;
 495                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 496                         goto out_path_release;
 497         }
 498
 499         res = inode_permission(inode, mode | MAY_ACCESS);
 500         /* SuS v2 requires we report a read only fs too */
 501         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 502                 goto out_path_release;
 503         /*
 504          * This is a rare case where using __mnt_is_readonly()
 505          * is OK without a mnt_want/drop_write() pair.  Since
 506          * no actual write to the fs is performed here, we do
 507          * not need to telegraph to that to anyone.
 508          *
 509          * By doing this, we accept that this access is
 510          * inherently racy and know that the fs may change
 511          * state before we even see this result.
 512          */
 513         if (__mnt_is_readonly(path.mnt))
 514                 res = -EROFS;
 515
 516 out_path_release:
 517         path_put(&path);
 518 out:
 519         revert_creds(old_cred);
 520         put_cred(override_cred);
 521         return res;
 522 }
 523
 524 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 525 {
 526         return sys_faccessat(AT_FDCWD, filename, mode);
 527 }
 528
 529 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 530 {
 531         struct path path;
 532         int error;
 533
 534         error = user_path_dir(filename, &path);
 535         if (error)
 536                 goto out;
 537
 538         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 539         if (error)
 540                 goto dput_and_out;
 541
 542         set_fs_pwd(current->fs, &path);
 543
 544 dput_and_out:
 545         path_put(&path);
 546 out:
 547         return error;
 548 }
 549
 550 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 551 {
 552         struct file *file;
 553         struct inode *inode;
 554         int error;
 555
 556         error = -EBADF;
 557         file = fget(fd);
 558         if (!file)
 559                 goto out;
 560
 561         inode = file->f_path.dentry->d_inode;
 562
 563         error = -ENOTDIR;
 564         if (!S_ISDIR(inode->i_mode))
 565                 goto out_putf;
 566
 567         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 568         if (!error)
 569                 set_fs_pwd(current->fs, &file->f_path);
 570 out_putf:
 571         fput(file);
 572 out:
 573         return error;
 574 }
 575
 576 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 577 {
 578         struct path path;
 579         int error;
 580
 581         error = user_path_dir(filename, &path);
 582         if (error)
 583                 goto out;
 584
 585         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 586         if (error)
 587                 goto dput_and_out;
 588
 589         error = -EPERM;
 590         if (!capable(CAP_SYS_CHROOT))
 591                 goto dput_and_out;
 592         error = security_path_chroot(&path);
 593         if (error)
 594                 goto dput_and_out;
 595
 596         set_fs_root(current->fs, &path);
 597         error = 0;
 598 dput_and_out:
 599         path_put(&path);
 600 out:
 601         return error;
 602 }
 603
 604 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 605 {
 606         struct inode * inode;
 607         struct dentry * dentry;
 608         struct file * file;
 609         int err = -EBADF;
 610         struct iattr newattrs;
 611
 612         file = fget(fd);
 613         if (!file)
 614                 goto out;
 615
 616         dentry = file->f_path.dentry;
 617         inode = dentry->d_inode;
 618
 619         audit_inode(NULL, dentry);
 620
 621         err = mnt_want_write_file(file);
 622         if (err)
 623                 goto out_putf;
 624         mutex_lock(&inode->i_mutex);
 625         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 626         if (err)
 627                 goto out_unlock;
 628         if (mode == (mode_t) -1)
 629                 mode = inode->i_mode;
 630         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 631         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 632         err = notify_change(dentry, &newattrs);
 633 out_unlock:
 634         mutex_unlock(&inode->i_mutex);
 635         mnt_drop_write(file->f_path.mnt);
 636 out_putf:
 637         fput(file);
 638 out:
 639         return err;
 640 }
 641
 642 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 643 {
 644         struct path path;
 645         struct inode *inode;
 646         int error;
 647         struct iattr newattrs;
 648
 649         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 650         if (error)
 651                 goto out;
 652         inode = path.dentry->d_inode;
 653
 654         error = mnt_want_write(path.mnt);
 655         if (error)
 656                 goto dput_and_out;
 657         mutex_lock(&inode->i_mutex);
 658         error = security_path_chmod(path.dentry, path.mnt, mode);
 659         if (error)
 660                 goto out_unlock;
 661         if (mode == (mode_t) -1)
 662                 mode = inode->i_mode;
 663         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 664         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 665         error = notify_change(path.dentry, &newattrs);
 666 out_unlock:
 667         mutex_unlock(&inode->i_mutex);
 668         mnt_drop_write(path.mnt);
 669 dput_and_out:
 670         path_put(&path);
 671 out:
 672         return error;
 673 }
 674
 675 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 676 {
 677         return sys_fchmodat(AT_FDCWD, filename, mode);
 678 }
 679
 680 static int chown_common(struct path *path, uid_t user, gid_t group)
 681 {
 682         struct inode *inode = path->dentry->d_inode;
 683         int error;
 684         struct iattr newattrs;
 685
 686         newattrs.ia_valid =  ATTR_CTIME;
 687         if (user != (uid_t) -1) {
 688                 newattrs.ia_valid |= ATTR_UID;
 689                 newattrs.ia_uid = user;
 690         }
 691         if (group != (gid_t) -1) {
 692                 newattrs.ia_valid |= ATTR_GID;
 693                 newattrs.ia_gid = group;
 694         }
 695         if (!S_ISDIR(inode->i_mode))
 696                 newattrs.ia_valid |=
 697                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 698         mutex_lock(&inode->i_mutex);
 699         error = security_path_chown(path, user, group);
 700         if (!error)
 701                 error = notify_change(path->dentry, &newattrs);
 702         mutex_unlock(&inode->i_mutex);
 703
 704         return error;
 705 }
 706
 707 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 708 {
 709         struct path path;
 710         int error;
 711
 712         error = user_path(filename, &path);
 713         if (error)
 714                 goto out;
 715         error = mnt_want_write(path.mnt);
 716         if (error)
 717                 goto out_release;
 718         error = chown_common(&path, user, group);
 719         mnt_drop_write(path.mnt);
 720 out_release:
 721         path_put(&path);
 722 out:
 723         return error;
 724 }
 725
 726 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 727                 gid_t, group, int, flag)
 728 {
 729         struct path path;
 730         int error = -EINVAL;
 731         int follow;
 732
 733         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 734                 goto out;
 735
 736         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 737         error = user_path_at(dfd, filename, follow, &path);
 738         if (error)
 739                 goto out;
 740         error = mnt_want_write(path.mnt);
 741         if (error)
 742                 goto out_release;
 743         error = chown_common(&path, user, group);
 744         mnt_drop_write(path.mnt);
 745 out_release:
 746         path_put(&path);
 747 out:
 748         return error;
 749 }
 750
 751 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 752 {
 753         struct path path;
 754         int error;
 755
 756         error = user_lpath(filename, &path);
 757         if (error)
 758                 goto out;
 759         error = mnt_want_write(path.mnt);
 760         if (error)
 761                 goto out_release;
 762         error = chown_common(&path, user, group);
 763         mnt_drop_write(path.mnt);
 764 out_release:
 765         path_put(&path);
 766 out:
 767         return error;
 768 }
 769
 770 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 771 {
 772         struct file * file;
 773         int error = -EBADF;
 774         struct dentry * dentry;
 775
 776         file = fget(fd);
 777         if (!file)
 778                 goto out;
 779
 780         error = mnt_want_write_file(file);
 781         if (error)
 782                 goto out_fput;
 783         dentry = file->f_path.dentry;
 784         audit_inode(NULL, dentry);
 785         error = chown_common(&file->f_path, user, group);
 786         mnt_drop_write(file->f_path.mnt);
 787 out_fput:
 788         fput(file);
 789 out:
 790         return error;
 791 }
 792
 793 /*
 794  * You have to be very careful that these write
 795  * counts get cleaned up in error cases and
 796  * upon __fput().  This should probably never
 797  * be called outside of __dentry_open().
 798  */
 799 static inline int __get_file_write_access(struct inode *inode,
 800                                           struct vfsmount *mnt)
 801 {
 802         int error;
 803         error = get_write_access(inode);
 804         if (error)
 805                 return error;
 806         /*
 807          * Do not take mount writer counts on
 808          * special files since no writes to
 809          * the mount itself will occur.
 810          */
 811         if (!special_file(inode->i_mode)) {
 812                 /*
 813                  * Balanced in __fput()
 814                  */
 815                 error = mnt_want_write(mnt);
 816                 if (error)
 817                         put_write_access(inode);
 818         }
 819         return error;
 820 }
 821
 822 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 823                                         int flags, struct file *f,
 824                                         int (*open)(struct inode *, struct file *),
 825                                         const struct cred *cred)
 826 {
 827         struct inode *inode;
 828         int error;
 829
 830         f->f_flags = flags;
 831         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 832                                 FMODE_PREAD | FMODE_PWRITE;
 833         inode = dentry->d_inode;
 834         if (f->f_mode & FMODE_WRITE) {
 835                 error = __get_file_write_access(inode, mnt);
 836                 if (error)
 837                         goto cleanup_file;
 838                 if (!special_file(inode->i_mode))
 839                         file_take_write(f);
 840         }
 841
 842         f->f_mapping = inode->i_mapping;
 843         f->f_path.dentry = dentry;
 844         f->f_path.mnt = mnt;
 845         f->f_pos = 0;
 846         f->f_op = fops_get(inode->i_fop);
 847         file_move(f, &inode->i_sb->s_files);
 848
 849         error = security_dentry_open(f, cred);
 850         if (error)
 851                 goto cleanup_all;
 852
 853         if (!open && f->f_op)
 854                 open = f->f_op->open;
 855         if (open) {
 856                 error = open(inode, f);
 857                 if (error)
 858                         goto cleanup_all;
 859         }
 860
 861         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 862
 863         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 864
 865         /* NB: we're sure to have correct a_ops only after f_op->open */
 866         if (f->f_flags & O_DIRECT) {
 867                 if (!f->f_mapping->a_ops ||
 868                     ((!f->f_mapping->a_ops->direct_IO) &&
 869                     (!f->f_mapping->a_ops->get_xip_mem))) {
 870                         fput(f);
 871                         f = ERR_PTR(-EINVAL);
 872                 }
 873         }
 874
 875         return f;
 876
 877 cleanup_all:
 878         fops_put(f->f_op);
 879         if (f->f_mode & FMODE_WRITE) {
 880                 put_write_access(inode);
 881                 if (!special_file(inode->i_mode)) {
 882                         /*
 883                          * We don't consider this a real
 884                          * mnt_want/drop_write() pair
 885                          * because it all happenend right
 886                          * here, so just reset the state.
 887                          */
 888                         file_reset_write(f);
 889                         mnt_drop_write(mnt);
 890                 }
 891         }
 892         file_kill(f);
 893         f->f_path.dentry = NULL;
 894         f->f_path.mnt = NULL;
 895 cleanup_file:
 896         put_filp(f);
 897         dput(dentry);
 898         mntput(mnt);
 899         return ERR_PTR(error);
 900 }
 901
 902 /**
 903  * lookup_instantiate_filp - instantiates the open intent filp
 904  * @nd: pointer to nameidata
 905  * @dentry: pointer to dentry
 906  * @open: open callback
 907  *
 908  * Helper for filesystems that want to use lookup open intents and pass back
 909  * a fully instantiated struct file to the caller.
 910  * This function is meant to be called from within a filesystem's
 911  * lookup method.
 912  * Beware of calling it for non-regular files! Those ->open methods might block
 913  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 914  * leading to a deadlock, as nobody can open that fifo anymore, because
 915  * another process to open fifo will block on locked parent when doing lookup).
 916  * Note that in case of error, nd->intent.open.file is destroyed, but the
 917  * path information remains valid.
 918  * If the open callback is set to NULL, then the standard f_op->open()
 919  * filesystem callback is substituted.
 920  */
 921 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 922                 int (*open)(struct inode *, struct file *))
 923 {
 924         const struct cred *cred = current_cred();
 925
 926         if (IS_ERR(nd->intent.open.file))
 927                 goto out;
 928         if (IS_ERR(dentry))
 929                 goto out_err;
 930         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 931                                              nd->intent.open.flags - 1,
 932                                              nd->intent.open.file,
 933                                              open, cred);
 934 out:
 935         return nd->intent.open.file;
 936 out_err:
 937         release_open_intent(nd);
 938         nd->intent.open.file = (struct file *)dentry;
 939         goto out;
 940 }
 941 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 942
 943 /**
 944  * nameidata_to_filp - convert a nameidata to an open filp.
 945  * @nd: pointer to nameidata
 946  * @flags: open flags
 947  *
 948  * Note that this function destroys the original nameidata
 949  */
 950 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 951 {
 952         const struct cred *cred = current_cred();
 953         struct file *filp;
 954
 955         /* Pick up the filp from the open intent */
 956         filp = nd->intent.open.file;
 957         /* Has the filesystem initialised the file for us? */
 958         if (filp->f_path.dentry == NULL)
 959                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 960                                      NULL, cred);
 961         else
 962                 path_put(&nd->path);
 963         return filp;
 964 }
 965
 966 /*
 967  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 968  * error.
 969  */
 970 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 971                          const struct cred *cred)
 972 {
 973         int error;
 974         struct file *f;
 975
 976         validate_creds(cred);
 977
 978         /*
 979          * We must always pass in a valid mount pointer.   Historically
 980          * callers got away with not passing it, but we must enforce this at
 981          * the earliest possible point now to avoid strange problems deep in the
 982          * filesystem stack.
 983          */
 984         if (!mnt) {
 985                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 986                 dump_stack();
 987                 return ERR_PTR(-EINVAL);
 988         }
 989
 990         error = -ENFILE;
 991         f = get_empty_filp();
 992         if (f == NULL) {
 993                 dput(dentry);
 994                 mntput(mnt);
 995                 return ERR_PTR(error);
 996         }
 997
 998         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 999 }
1000 EXPORT_SYMBOL(dentry_open);
1001
1002 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1003 {
1004         struct fdtable *fdt = files_fdtable(files);
1005         __FD_CLR(fd, fdt->open_fds);
1006         if (fd < files->next_fd)
1007                 files->next_fd = fd;
1008 }
1009
1010 void put_unused_fd(unsigned int fd)
1011 {
1012         struct files_struct *files = current->files;
1013         spin_lock(&files->file_lock);
1014         __put_unused_fd(files, fd);
1015         spin_unlock(&files->file_lock);
1016 }
1017
1018 EXPORT_SYMBOL(put_unused_fd);
1019
1020 /*
1021  * Install a file pointer in the fd array.
1022  *
1023  * The VFS is full of places where we drop the files lock between
1024  * setting the open_fds bitmap and installing the file in the file
1025  * array.  At any such point, we are vulnerable to a dup2() race
1026  * installing a file in the array before us.  We need to detect this and
1027  * fput() the struct file we are about to overwrite in this case.
1028  *
1029  * It should never happen - if we allow dup2() do it, _really_ bad things
1030  * will follow.
1031  */
1032
1033 void fd_install(unsigned int fd, struct file *file)
1034 {
1035         struct files_struct *files = current->files;
1036         struct fdtable *fdt;
1037         spin_lock(&files->file_lock);
1038         fdt = files_fdtable(files);
1039         BUG_ON(fdt->fd[fd] != NULL);
1040         rcu_assign_pointer(fdt->fd[fd], file);
1041         spin_unlock(&files->file_lock);
1042 }
1043
1044 EXPORT_SYMBOL(fd_install);
1045
1046 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1047 {
1048         char *tmp = getname(filename);
1049         int fd = PTR_ERR(tmp);
1050
1051         if (!IS_ERR(tmp)) {
1052                 fd = get_unused_fd_flags(flags);
1053                 if (fd >= 0) {
1054                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1055                         if (IS_ERR(f)) {
1056                                 put_unused_fd(fd);
1057                                 fd = PTR_ERR(f);
1058                         } else {
1059                                 fsnotify_open(f->f_path.dentry);
1060                                 fd_install(fd, f);
1061                         }
1062                 }
1063                 putname(tmp);
1064         }
1065         return fd;
1066 }
1067
1068 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1069 {
1070         long ret;
1071
1072         if (force_o_largefile())
1073                 flags |= O_LARGEFILE;
1074
1075         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1076         /* avoid REGPARM breakage on x86: */
1077         asmlinkage_protect(3, ret, filename, flags, mode);
1078         return ret;
1079 }
1080
1081 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1082                 int, mode)
1083 {
1084         long ret;
1085
1086         if (force_o_largefile())
1087                 flags |= O_LARGEFILE;
1088
1089         ret = do_sys_open(dfd, filename, flags, mode);
1090         /* avoid REGPARM breakage on x86: */
1091         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1092         return ret;
1093 }
1094
1095 #ifndef __alpha__
1096
1097 /*
1098  * For backward compatibility?  Maybe this should be moved
1099  * into arch/i386 instead?
1100  */
1101 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1102 {
1103         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1104 }
1105
1106 #endif
1107
1108 /*
1109  * "id" is the POSIX thread ID. We use the
1110  * files pointer for this..
1111  */
1112 int filp_close(struct file *filp, fl_owner_t id)
1113 {
1114         int retval = 0;
1115
1116         if (!file_count(filp)) {
1117                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1118                 return 0;
1119         }
1120
1121         if (filp->f_op && filp->f_op->flush)
1122                 retval = filp->f_op->flush(filp, id);
1123
1124         dnotify_flush(filp, id);
1125         locks_remove_posix(filp, id);
1126         fput(filp);
1127         return retval;
1128 }
1129
1130 EXPORT_SYMBOL(filp_close);
1131
1132 /*
1133  * Careful here! We test whether the file pointer is NULL before
1134  * releasing the fd. This ensures that one clone task can't release
1135  * an fd while another clone is opening it.
1136  */
1137 SYSCALL_DEFINE1(close, unsigned int, fd)
1138 {
1139         struct file * filp;
1140         struct files_struct *files = current->files;
1141         struct fdtable *fdt;
1142         int retval;
1143
1144         spin_lock(&files->file_lock);
1145         fdt = files_fdtable(files);
1146         if (fd >= fdt->max_fds)
1147                 goto out_unlock;
1148         filp = fdt->fd[fd];
1149         if (!filp)
1150                 goto out_unlock;
1151         rcu_assign_pointer(fdt->fd[fd], NULL);
1152         FD_CLR(fd, fdt->close_on_exec);
1153         __put_unused_fd(files, fd);
1154         spin_unlock(&files->file_lock);
1155         retval = filp_close(filp, files);
1156
1157         /* can't restart close syscall because file table entry was cleared */
1158         if (unlikely(retval == -ERESTARTSYS ||
1159                      retval == -ERESTARTNOINTR ||
1160                      retval == -ERESTARTNOHAND ||
1161                      retval == -ERESTART_RESTARTBLOCK))
1162                 retval = -EINTR;
1163
1164         return retval;
1165
1166 out_unlock:
1167         spin_unlock(&files->file_lock);
1168         return -EBADF;
1169 }
1170 EXPORT_SYMBOL(sys_close);
1171
1172 /*
1173  * This routine simulates a hangup on the tty, to arrange that users
1174  * are given clean terminals at login time.
1175  */
1176 SYSCALL_DEFINE0(vhangup)
1177 {
1178         if (capable(CAP_SYS_TTY_CONFIG)) {
1179                 tty_vhangup_self();
1180                 return 0;
1181         }
1182         return -EPERM;
1183 }
1184
1185 /*
1186  * Called when an inode is about to be open.
1187  * We use this to disallow opening large files on 32bit systems if
1188  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1189  * on this flag in sys_open.
1190  */
1191 int generic_file_open(struct inode * inode, struct file * filp)
1192 {
1193         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1194                 return -EOVERFLOW;
1195         return 0;
1196 }
1197
1198 EXPORT_SYMBOL(generic_file_open);
1199
1200 /*
1201  * This is used by subsystems that don't want seekable
1202  * file descriptors
1203  */
1204 int nonseekable_open(struct inode *inode, struct file *filp)
1205 {
1206         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1207         return 0;
1208 }
1209
1210 EXPORT_SYMBOL(nonseekable_open);