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 #include <linux/ima.h>
  34
  35 #include "internal.h"
  36
  37 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  38 {
  39         int retval = -ENODEV;
  40
  41         if (dentry) {
  42                 retval = -ENOSYS;
  43                 if (dentry->d_sb->s_op->statfs) {
  44                         memset(buf, 0, sizeof(*buf));
  45                         retval = security_sb_statfs(dentry);
  46                         if (retval)
  47                                 return retval;
  48                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  49                         if (retval == 0 && buf->f_frsize == 0)
  50                                 buf->f_frsize = buf->f_bsize;
  51                 }
  52         }
  53         return retval;
  54 }
  55
  56 EXPORT_SYMBOL(vfs_statfs);
  57
  58 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  59 {
  60         struct kstatfs st;
  61         int retval;
  62
  63         retval = vfs_statfs(dentry, &st);
  64         if (retval)
  65                 return retval;
  66
  67         if (sizeof(*buf) == sizeof(st))
  68                 memcpy(buf, &st, sizeof(st));
  69         else {
  70                 if (sizeof buf->f_blocks == 4) {
  71                         if ((st.f_blocks | st.f_bfree | st.f_bavail |
  72                              st.f_bsize | st.f_frsize) &
  73                             0xffffffff00000000ULL)
  74                                 return -EOVERFLOW;
  75                         /*
  76                          * f_files and f_ffree may be -1; it's okay to stuff
  77                          * that into 32 bits
  78                          */
  79                         if (st.f_files != -1 &&
  80                             (st.f_files & 0xffffffff00000000ULL))
  81                                 return -EOVERFLOW;
  82                         if (st.f_ffree != -1 &&
  83                             (st.f_ffree & 0xffffffff00000000ULL))
  84                                 return -EOVERFLOW;
  85                 }
  86
  87                 buf->f_type = st.f_type;
  88                 buf->f_bsize = st.f_bsize;
  89                 buf->f_blocks = st.f_blocks;
  90                 buf->f_bfree = st.f_bfree;
  91                 buf->f_bavail = st.f_bavail;
  92                 buf->f_files = st.f_files;
  93                 buf->f_ffree = st.f_ffree;
  94                 buf->f_fsid = st.f_fsid;
  95                 buf->f_namelen = st.f_namelen;
  96                 buf->f_frsize = st.f_frsize;
  97                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  98         }
  99         return 0;
 100 }
 101
 102 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
 103 {
 104         struct kstatfs st;
 105         int retval;
 106
 107         retval = vfs_statfs(dentry, &st);
 108         if (retval)
 109                 return retval;
 110
 111         if (sizeof(*buf) == sizeof(st))
 112                 memcpy(buf, &st, sizeof(st));
 113         else {
 114                 buf->f_type = st.f_type;
 115                 buf->f_bsize = st.f_bsize;
 116                 buf->f_blocks = st.f_blocks;
 117                 buf->f_bfree = st.f_bfree;
 118                 buf->f_bavail = st.f_bavail;
 119                 buf->f_files = st.f_files;
 120                 buf->f_ffree = st.f_ffree;
 121                 buf->f_fsid = st.f_fsid;
 122                 buf->f_namelen = st.f_namelen;
 123                 buf->f_frsize = st.f_frsize;
 124                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 125         }
 126         return 0;
 127 }
 128
 129 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
 130 {
 131         struct path path;
 132         int error;
 133
 134         error = user_path(pathname, &path);
 135         if (!error) {
 136                 struct statfs tmp;
 137                 error = vfs_statfs_native(path.dentry, &tmp);
 138                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 139                         error = -EFAULT;
 140                 path_put(&path);
 141         }
 142         return error;
 143 }
 144
 145 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
 146 {
 147         struct path path;
 148         long error;
 149
 150         if (sz != sizeof(*buf))
 151                 return -EINVAL;
 152         error = user_path(pathname, &path);
 153         if (!error) {
 154                 struct statfs64 tmp;
 155                 error = vfs_statfs64(path.dentry, &tmp);
 156                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 157                         error = -EFAULT;
 158                 path_put(&path);
 159         }
 160         return error;
 161 }
 162
 163 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
 164 {
 165         struct file * file;
 166         struct statfs tmp;
 167         int error;
 168
 169         error = -EBADF;
 170         file = fget(fd);
 171         if (!file)
 172                 goto out;
 173         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 174         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 175                 error = -EFAULT;
 176         fput(file);
 177 out:
 178         return error;
 179 }
 180
 181 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
 182 {
 183         struct file * file;
 184         struct statfs64 tmp;
 185         int error;
 186
 187         if (sz != sizeof(*buf))
 188                 return -EINVAL;
 189
 190         error = -EBADF;
 191         file = fget(fd);
 192         if (!file)
 193                 goto out;
 194         error = vfs_statfs64(file->f_path.dentry, &tmp);
 195         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 196                 error = -EFAULT;
 197         fput(file);
 198 out:
 199         return error;
 200 }
 201
 202 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 203         struct file *filp)
 204 {
 205         int ret;
 206         struct iattr newattrs;
 207
 208         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 209         if (length < 0)
 210                 return -EINVAL;
 211
 212         newattrs.ia_size = length;
 213         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 214         if (filp) {
 215                 newattrs.ia_file = filp;
 216                 newattrs.ia_valid |= ATTR_FILE;
 217         }
 218
 219         /* Remove suid/sgid on truncate too */
 220         ret = should_remove_suid(dentry);
 221         if (ret)
 222                 newattrs.ia_valid |= ret | ATTR_FORCE;
 223
 224         mutex_lock(&dentry->d_inode->i_mutex);
 225         ret = notify_change(dentry, &newattrs);
 226         mutex_unlock(&dentry->d_inode->i_mutex);
 227         return ret;
 228 }
 229
 230 static long do_sys_truncate(const char __user *pathname, loff_t length)
 231 {
 232         struct path path;
 233         struct inode *inode;
 234         int error;
 235
 236         error = -EINVAL;
 237         if (length < 0) /* sorry, but loff_t says... */
 238                 goto out;
 239
 240         error = user_path(pathname, &path);
 241         if (error)
 242                 goto out;
 243         inode = path.dentry->d_inode;
 244
 245         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 246         error = -EISDIR;
 247         if (S_ISDIR(inode->i_mode))
 248                 goto dput_and_out;
 249
 250         error = -EINVAL;
 251         if (!S_ISREG(inode->i_mode))
 252                 goto dput_and_out;
 253
 254         error = mnt_want_write(path.mnt);
 255         if (error)
 256                 goto dput_and_out;
 257
 258         error = inode_permission(inode, MAY_WRITE);
 259         if (error)
 260                 goto mnt_drop_write_and_out;
 261
 262         error = -EPERM;
 263         if (IS_APPEND(inode))
 264                 goto mnt_drop_write_and_out;
 265
 266         error = get_write_access(inode);
 267         if (error)
 268                 goto mnt_drop_write_and_out;
 269
 270         /*
 271          * Make sure that there are no leases.  get_write_access() protects
 272          * against the truncate racing with a lease-granting setlease().
 273          */
 274         error = break_lease(inode, FMODE_WRITE);
 275         if (error)
 276                 goto put_write_and_out;
 277
 278         error = locks_verify_truncate(inode, NULL, length);
 279         if (!error)
 280                 error = security_path_truncate(&path, length, 0);
 281         if (!error) {
 282                 vfs_dq_init(inode);
 283                 error = do_truncate(path.dentry, length, 0, NULL);
 284         }
 285
 286 put_write_and_out:
 287         put_write_access(inode);
 288 mnt_drop_write_and_out:
 289         mnt_drop_write(path.mnt);
 290 dput_and_out:
 291         path_put(&path);
 292 out:
 293         return error;
 294 }
 295
 296 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 297 {
 298         return do_sys_truncate(path, length);
 299 }
 300
 301 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 302 {
 303         struct inode * inode;
 304         struct dentry *dentry;
 305         struct file * file;
 306         int error;
 307
 308         error = -EINVAL;
 309         if (length < 0)
 310                 goto out;
 311         error = -EBADF;
 312         file = fget(fd);
 313         if (!file)
 314                 goto out;
 315
 316         /* explicitly opened as large or we are on 64-bit box */
 317         if (file->f_flags & O_LARGEFILE)
 318                 small = 0;
 319
 320         dentry = file->f_path.dentry;
 321         inode = dentry->d_inode;
 322         error = -EINVAL;
 323         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 324                 goto out_putf;
 325
 326         error = -EINVAL;
 327         /* Cannot ftruncate over 2^31 bytes without large file support */
 328         if (small && length > MAX_NON_LFS)
 329                 goto out_putf;
 330
 331         error = -EPERM;
 332         if (IS_APPEND(inode))
 333                 goto out_putf;
 334
 335         error = locks_verify_truncate(inode, file, length);
 336         if (!error)
 337                 error = security_path_truncate(&file->f_path, length,
 338                                                ATTR_MTIME|ATTR_CTIME);
 339         if (!error)
 340                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 341 out_putf:
 342         fput(file);
 343 out:
 344         return error;
 345 }
 346
 347 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 348 {
 349         long ret = do_sys_ftruncate(fd, length, 1);
 350         /* avoid REGPARM breakage on x86: */
 351         asmlinkage_protect(2, ret, fd, length);
 352         return ret;
 353 }
 354
 355 /* LFS versions of truncate are only needed on 32 bit machines */
 356 #if BITS_PER_LONG == 32
 357 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 358 {
 359         return do_sys_truncate(path, length);
 360 }
 361 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 362 asmlinkage long SyS_truncate64(long path, loff_t length)
 363 {
 364         return SYSC_truncate64((const char __user *) path, length);
 365 }
 366 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 367 #endif
 368
 369 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 370 {
 371         long ret = do_sys_ftruncate(fd, length, 0);
 372         /* avoid REGPARM breakage on x86: */
 373         asmlinkage_protect(2, ret, fd, length);
 374         return ret;
 375 }
 376 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 377 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 378 {
 379         return SYSC_ftruncate64((unsigned int) fd, length);
 380 }
 381 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 382 #endif
 383 #endif /* BITS_PER_LONG == 32 */
 384
 385
 386 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 387 {
 388         struct inode *inode = file->f_path.dentry->d_inode;
 389         long ret;
 390
 391         if (offset < 0 || len <= 0)
 392                 return -EINVAL;
 393
 394         /* Return error if mode is not supported */
 395         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 396                 return -EOPNOTSUPP;
 397
 398         if (!(file->f_mode & FMODE_WRITE))
 399                 return -EBADF;
 400         /*
 401          * Revalidate the write permissions, in case security policy has
 402          * changed since the files were opened.
 403          */
 404         ret = security_file_permission(file, MAY_WRITE);
 405         if (ret)
 406                 return ret;
 407
 408         if (S_ISFIFO(inode->i_mode))
 409                 return -ESPIPE;
 410
 411         /*
 412          * Let individual file system decide if it supports preallocation
 413          * for directories or not.
 414          */
 415         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 416                 return -ENODEV;
 417
 418         /* Check for wrap through zero too */
 419         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 420                 return -EFBIG;
 421
 422         if (!inode->i_op->fallocate)
 423                 return -EOPNOTSUPP;
 424
 425         return inode->i_op->fallocate(inode, mode, offset, len);
 426 }
 427
 428 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 429 {
 430         struct file *file;
 431         int error = -EBADF;
 432
 433         file = fget(fd);
 434         if (file) {
 435                 error = do_fallocate(file, mode, offset, len);
 436                 fput(file);
 437         }
 438
 439         return error;
 440 }
 441
 442 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 443 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 444 {
 445         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 446 }
 447 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 448 #endif
 449
 450 /*
 451  * access() needs to use the real uid/gid, not the effective uid/gid.
 452  * We do this by temporarily clearing all FS-related capabilities and
 453  * switching the fsuid/fsgid around to the real ones.
 454  */
 455 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 456 {
 457         const struct cred *old_cred;
 458         struct cred *override_cred;
 459         struct path path;
 460         struct inode *inode;
 461         int res;
 462
 463         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 464                 return -EINVAL;
 465
 466         override_cred = prepare_creds();
 467         if (!override_cred)
 468                 return -ENOMEM;
 469
 470         override_cred->fsuid = override_cred->uid;
 471         override_cred->fsgid = override_cred->gid;
 472
 473         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 474                 /* Clear the capabilities if we switch to a non-root user */
 475                 if (override_cred->uid)
 476                         cap_clear(override_cred->cap_effective);
 477                 else
 478                         override_cred->cap_effective =
 479                                 override_cred->cap_permitted;
 480         }
 481
 482         old_cred = override_creds(override_cred);
 483
 484         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 485         if (res)
 486                 goto out;
 487
 488         inode = path.dentry->d_inode;
 489
 490         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 491                 /*
 492                  * MAY_EXEC on regular files is denied if the fs is mounted
 493                  * with the "noexec" flag.
 494                  */
 495                 res = -EACCES;
 496                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 497                         goto out_path_release;
 498         }
 499
 500         res = inode_permission(inode, mode | MAY_ACCESS);
 501         /* SuS v2 requires we report a read only fs too */
 502         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 503                 goto out_path_release;
 504         /*
 505          * This is a rare case where using __mnt_is_readonly()
 506          * is OK without a mnt_want/drop_write() pair.  Since
 507          * no actual write to the fs is performed here, we do
 508          * not need to telegraph to that to anyone.
 509          *
 510          * By doing this, we accept that this access is
 511          * inherently racy and know that the fs may change
 512          * state before we even see this result.
 513          */
 514         if (__mnt_is_readonly(path.mnt))
 515                 res = -EROFS;
 516
 517 out_path_release:
 518         path_put(&path);
 519 out:
 520         revert_creds(old_cred);
 521         put_cred(override_cred);
 522         return res;
 523 }
 524
 525 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 526 {
 527         return sys_faccessat(AT_FDCWD, filename, mode);
 528 }
 529
 530 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 531 {
 532         struct path path;
 533         int error;
 534
 535         error = user_path_dir(filename, &path);
 536         if (error)
 537                 goto out;
 538
 539         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 540         if (error)
 541                 goto dput_and_out;
 542
 543         set_fs_pwd(current->fs, &path);
 544
 545 dput_and_out:
 546         path_put(&path);
 547 out:
 548         return error;
 549 }
 550
 551 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 552 {
 553         struct file *file;
 554         struct inode *inode;
 555         int error;
 556
 557         error = -EBADF;
 558         file = fget(fd);
 559         if (!file)
 560                 goto out;
 561
 562         inode = file->f_path.dentry->d_inode;
 563
 564         error = -ENOTDIR;
 565         if (!S_ISDIR(inode->i_mode))
 566                 goto out_putf;
 567
 568         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 569         if (!error)
 570                 set_fs_pwd(current->fs, &file->f_path);
 571 out_putf:
 572         fput(file);
 573 out:
 574         return error;
 575 }
 576
 577 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 578 {
 579         struct path path;
 580         int error;
 581
 582         error = user_path_dir(filename, &path);
 583         if (error)
 584                 goto out;
 585
 586         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 587         if (error)
 588                 goto dput_and_out;
 589
 590         error = -EPERM;
 591         if (!capable(CAP_SYS_CHROOT))
 592                 goto dput_and_out;
 593         error = security_path_chroot(&path);
 594         if (error)
 595                 goto dput_and_out;
 596
 597         set_fs_root(current->fs, &path);
 598         error = 0;
 599 dput_and_out:
 600         path_put(&path);
 601 out:
 602         return error;
 603 }
 604
 605 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 606 {
 607         struct inode * inode;
 608         struct dentry * dentry;
 609         struct file * file;
 610         int err = -EBADF;
 611         struct iattr newattrs;
 612
 613         file = fget(fd);
 614         if (!file)
 615                 goto out;
 616
 617         dentry = file->f_path.dentry;
 618         inode = dentry->d_inode;
 619
 620         audit_inode(NULL, dentry);
 621
 622         err = mnt_want_write_file(file);
 623         if (err)
 624                 goto out_putf;
 625         mutex_lock(&inode->i_mutex);
 626         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 627         if (err)
 628                 goto out_unlock;
 629         if (mode == (mode_t) -1)
 630                 mode = inode->i_mode;
 631         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 632         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 633         err = notify_change(dentry, &newattrs);
 634 out_unlock:
 635         mutex_unlock(&inode->i_mutex);
 636         mnt_drop_write(file->f_path.mnt);
 637 out_putf:
 638         fput(file);
 639 out:
 640         return err;
 641 }
 642
 643 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 644 {
 645         struct path path;
 646         struct inode *inode;
 647         int error;
 648         struct iattr newattrs;
 649
 650         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 651         if (error)
 652                 goto out;
 653         inode = path.dentry->d_inode;
 654
 655         error = mnt_want_write(path.mnt);
 656         if (error)
 657                 goto dput_and_out;
 658         mutex_lock(&inode->i_mutex);
 659         error = security_path_chmod(path.dentry, path.mnt, mode);
 660         if (error)
 661                 goto out_unlock;
 662         if (mode == (mode_t) -1)
 663                 mode = inode->i_mode;
 664         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 665         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 666         error = notify_change(path.dentry, &newattrs);
 667 out_unlock:
 668         mutex_unlock(&inode->i_mutex);
 669         mnt_drop_write(path.mnt);
 670 dput_and_out:
 671         path_put(&path);
 672 out:
 673         return error;
 674 }
 675
 676 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 677 {
 678         return sys_fchmodat(AT_FDCWD, filename, mode);
 679 }
 680
 681 static int chown_common(struct path *path, uid_t user, gid_t group)
 682 {
 683         struct inode *inode = path->dentry->d_inode;
 684         int error;
 685         struct iattr newattrs;
 686
 687         newattrs.ia_valid =  ATTR_CTIME;
 688         if (user != (uid_t) -1) {
 689                 newattrs.ia_valid |= ATTR_UID;
 690                 newattrs.ia_uid = user;
 691         }
 692         if (group != (gid_t) -1) {
 693                 newattrs.ia_valid |= ATTR_GID;
 694                 newattrs.ia_gid = group;
 695         }
 696         if (!S_ISDIR(inode->i_mode))
 697                 newattrs.ia_valid |=
 698                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 699         mutex_lock(&inode->i_mutex);
 700         error = security_path_chown(path, user, group);
 701         if (!error)
 702                 error = notify_change(path->dentry, &newattrs);
 703         mutex_unlock(&inode->i_mutex);
 704
 705         return error;
 706 }
 707
 708 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 709 {
 710         struct path path;
 711         int error;
 712
 713         error = user_path(filename, &path);
 714         if (error)
 715                 goto out;
 716         error = mnt_want_write(path.mnt);
 717         if (error)
 718                 goto out_release;
 719         error = chown_common(&path, user, group);
 720         mnt_drop_write(path.mnt);
 721 out_release:
 722         path_put(&path);
 723 out:
 724         return error;
 725 }
 726
 727 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 728                 gid_t, group, int, flag)
 729 {
 730         struct path path;
 731         int error = -EINVAL;
 732         int follow;
 733
 734         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 735                 goto out;
 736
 737         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 738         error = user_path_at(dfd, filename, follow, &path);
 739         if (error)
 740                 goto out;
 741         error = mnt_want_write(path.mnt);
 742         if (error)
 743                 goto out_release;
 744         error = chown_common(&path, user, group);
 745         mnt_drop_write(path.mnt);
 746 out_release:
 747         path_put(&path);
 748 out:
 749         return error;
 750 }
 751
 752 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 753 {
 754         struct path path;
 755         int error;
 756
 757         error = user_lpath(filename, &path);
 758         if (error)
 759                 goto out;
 760         error = mnt_want_write(path.mnt);
 761         if (error)
 762                 goto out_release;
 763         error = chown_common(&path, user, group);
 764         mnt_drop_write(path.mnt);
 765 out_release:
 766         path_put(&path);
 767 out:
 768         return error;
 769 }
 770
 771 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 772 {
 773         struct file * file;
 774         int error = -EBADF;
 775         struct dentry * dentry;
 776
 777         file = fget(fd);
 778         if (!file)
 779                 goto out;
 780
 781         error = mnt_want_write_file(file);
 782         if (error)
 783                 goto out_fput;
 784         dentry = file->f_path.dentry;
 785         audit_inode(NULL, dentry);
 786         error = chown_common(&file->f_path, user, group);
 787         mnt_drop_write(file->f_path.mnt);
 788 out_fput:
 789         fput(file);
 790 out:
 791         return error;
 792 }
 793
 794 /*
 795  * You have to be very careful that these write
 796  * counts get cleaned up in error cases and
 797  * upon __fput().  This should probably never
 798  * be called outside of __dentry_open().
 799  */
 800 static inline int __get_file_write_access(struct inode *inode,
 801                                           struct vfsmount *mnt)
 802 {
 803         int error;
 804         error = get_write_access(inode);
 805         if (error)
 806                 return error;
 807         /*
 808          * Do not take mount writer counts on
 809          * special files since no writes to
 810          * the mount itself will occur.
 811          */
 812         if (!special_file(inode->i_mode)) {
 813                 /*
 814                  * Balanced in __fput()
 815                  */
 816                 error = mnt_want_write(mnt);
 817                 if (error)
 818                         put_write_access(inode);
 819         }
 820         return error;
 821 }
 822
 823 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 824                                         int flags, struct file *f,
 825                                         int (*open)(struct inode *, struct file *),
 826                                         const struct cred *cred)
 827 {
 828         struct inode *inode;
 829         int error;
 830
 831         f->f_flags = flags;
 832         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 833                                 FMODE_PREAD | FMODE_PWRITE;
 834         inode = dentry->d_inode;
 835         if (f->f_mode & FMODE_WRITE) {
 836                 error = __get_file_write_access(inode, mnt);
 837                 if (error)
 838                         goto cleanup_file;
 839                 if (!special_file(inode->i_mode))
 840                         file_take_write(f);
 841         }
 842
 843         f->f_mapping = inode->i_mapping;
 844         f->f_path.dentry = dentry;
 845         f->f_path.mnt = mnt;
 846         f->f_pos = 0;
 847         f->f_op = fops_get(inode->i_fop);
 848         file_move(f, &inode->i_sb->s_files);
 849
 850         error = security_dentry_open(f, cred);
 851         if (error)
 852                 goto cleanup_all;
 853
 854         if (!open && f->f_op)
 855                 open = f->f_op->open;
 856         if (open) {
 857                 error = open(inode, f);
 858                 if (error)
 859                         goto cleanup_all;
 860         }
 861         ima_counts_get(f);
 862
 863         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 864
 865         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 866
 867         /* NB: we're sure to have correct a_ops only after f_op->open */
 868         if (f->f_flags & O_DIRECT) {
 869                 if (!f->f_mapping->a_ops ||
 870                     ((!f->f_mapping->a_ops->direct_IO) &&
 871                     (!f->f_mapping->a_ops->get_xip_mem))) {
 872                         fput(f);
 873                         f = ERR_PTR(-EINVAL);
 874                 }
 875         }
 876
 877         return f;
 878
 879 cleanup_all:
 880         fops_put(f->f_op);
 881         if (f->f_mode & FMODE_WRITE) {
 882                 put_write_access(inode);
 883                 if (!special_file(inode->i_mode)) {
 884                         /*
 885                          * We don't consider this a real
 886                          * mnt_want/drop_write() pair
 887                          * because it all happenend right
 888                          * here, so just reset the state.
 889                          */
 890                         file_reset_write(f);
 891                         mnt_drop_write(mnt);
 892                 }
 893         }
 894         file_kill(f);
 895         f->f_path.dentry = NULL;
 896         f->f_path.mnt = NULL;
 897 cleanup_file:
 898         put_filp(f);
 899         dput(dentry);
 900         mntput(mnt);
 901         return ERR_PTR(error);
 902 }
 903
 904 /**
 905  * lookup_instantiate_filp - instantiates the open intent filp
 906  * @nd: pointer to nameidata
 907  * @dentry: pointer to dentry
 908  * @open: open callback
 909  *
 910  * Helper for filesystems that want to use lookup open intents and pass back
 911  * a fully instantiated struct file to the caller.
 912  * This function is meant to be called from within a filesystem's
 913  * lookup method.
 914  * Beware of calling it for non-regular files! Those ->open methods might block
 915  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 916  * leading to a deadlock, as nobody can open that fifo anymore, because
 917  * another process to open fifo will block on locked parent when doing lookup).
 918  * Note that in case of error, nd->intent.open.file is destroyed, but the
 919  * path information remains valid.
 920  * If the open callback is set to NULL, then the standard f_op->open()
 921  * filesystem callback is substituted.
 922  */
 923 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 924                 int (*open)(struct inode *, struct file *))
 925 {
 926         const struct cred *cred = current_cred();
 927
 928         if (IS_ERR(nd->intent.open.file))
 929                 goto out;
 930         if (IS_ERR(dentry))
 931                 goto out_err;
 932         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 933                                              nd->intent.open.flags - 1,
 934                                              nd->intent.open.file,
 935                                              open, cred);
 936 out:
 937         return nd->intent.open.file;
 938 out_err:
 939         release_open_intent(nd);
 940         nd->intent.open.file = (struct file *)dentry;
 941         goto out;
 942 }
 943 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 944
 945 /**
 946  * nameidata_to_filp - convert a nameidata to an open filp.
 947  * @nd: pointer to nameidata
 948  * @flags: open flags
 949  *
 950  * Note that this function destroys the original nameidata
 951  */
 952 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 953 {
 954         const struct cred *cred = current_cred();
 955         struct file *filp;
 956
 957         /* Pick up the filp from the open intent */
 958         filp = nd->intent.open.file;
 959         /* Has the filesystem initialised the file for us? */
 960         if (filp->f_path.dentry == NULL)
 961                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 962                                      NULL, cred);
 963         else
 964                 path_put(&nd->path);
 965         return filp;
 966 }
 967
 968 /*
 969  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 970  * error.
 971  */
 972 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 973                          const struct cred *cred)
 974 {
 975         int error;
 976         struct file *f;
 977
 978         validate_creds(cred);
 979
 980         /*
 981          * We must always pass in a valid mount pointer.   Historically
 982          * callers got away with not passing it, but we must enforce this at
 983          * the earliest possible point now to avoid strange problems deep in the
 984          * filesystem stack.
 985          */
 986         if (!mnt) {
 987                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 988                 dump_stack();
 989                 return ERR_PTR(-EINVAL);
 990         }
 991
 992         error = -ENFILE;
 993         f = get_empty_filp();
 994         if (f == NULL) {
 995                 dput(dentry);
 996                 mntput(mnt);
 997                 return ERR_PTR(error);
 998         }
 999
1000         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
1001 }
1002 EXPORT_SYMBOL(dentry_open);
1003
1004 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1005 {
1006         struct fdtable *fdt = files_fdtable(files);
1007         __FD_CLR(fd, fdt->open_fds);
1008         if (fd < files->next_fd)
1009                 files->next_fd = fd;
1010 }
1011
1012 void put_unused_fd(unsigned int fd)
1013 {
1014         struct files_struct *files = current->files;
1015         spin_lock(&files->file_lock);
1016         __put_unused_fd(files, fd);
1017         spin_unlock(&files->file_lock);
1018 }
1019
1020 EXPORT_SYMBOL(put_unused_fd);
1021
1022 /*
1023  * Install a file pointer in the fd array.
1024  *
1025  * The VFS is full of places where we drop the files lock between
1026  * setting the open_fds bitmap and installing the file in the file
1027  * array.  At any such point, we are vulnerable to a dup2() race
1028  * installing a file in the array before us.  We need to detect this and
1029  * fput() the struct file we are about to overwrite in this case.
1030  *
1031  * It should never happen - if we allow dup2() do it, _really_ bad things
1032  * will follow.
1033  */
1034
1035 void fd_install(unsigned int fd, struct file *file)
1036 {
1037         struct files_struct *files = current->files;
1038         struct fdtable *fdt;
1039         spin_lock(&files->file_lock);
1040         fdt = files_fdtable(files);
1041         BUG_ON(fdt->fd[fd] != NULL);
1042         rcu_assign_pointer(fdt->fd[fd], file);
1043         spin_unlock(&files->file_lock);
1044 }
1045
1046 EXPORT_SYMBOL(fd_install);
1047
1048 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1049 {
1050         char *tmp = getname(filename);
1051         int fd = PTR_ERR(tmp);
1052
1053         if (!IS_ERR(tmp)) {
1054                 fd = get_unused_fd_flags(flags);
1055                 if (fd >= 0) {
1056                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1057                         if (IS_ERR(f)) {
1058                                 put_unused_fd(fd);
1059                                 fd = PTR_ERR(f);
1060                         } else {
1061                                 fsnotify_open(f->f_path.dentry);
1062                                 fd_install(fd, f);
1063                         }
1064                 }
1065                 putname(tmp);
1066         }
1067         return fd;
1068 }
1069
1070 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1071 {
1072         long ret;
1073
1074         if (force_o_largefile())
1075                 flags |= O_LARGEFILE;
1076
1077         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1078         /* avoid REGPARM breakage on x86: */
1079         asmlinkage_protect(3, ret, filename, flags, mode);
1080         return ret;
1081 }
1082
1083 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1084                 int, mode)
1085 {
1086         long ret;
1087
1088         if (force_o_largefile())
1089                 flags |= O_LARGEFILE;
1090
1091         ret = do_sys_open(dfd, filename, flags, mode);
1092         /* avoid REGPARM breakage on x86: */
1093         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1094         return ret;
1095 }
1096
1097 #ifndef __alpha__
1098
1099 /*
1100  * For backward compatibility?  Maybe this should be moved
1101  * into arch/i386 instead?
1102  */
1103 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1104 {
1105         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1106 }
1107
1108 #endif
1109
1110 /*
1111  * "id" is the POSIX thread ID. We use the
1112  * files pointer for this..
1113  */
1114 int filp_close(struct file *filp, fl_owner_t id)
1115 {
1116         int retval = 0;
1117
1118         if (!file_count(filp)) {
1119                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1120                 return 0;
1121         }
1122
1123         if (filp->f_op && filp->f_op->flush)
1124                 retval = filp->f_op->flush(filp, id);
1125
1126         dnotify_flush(filp, id);
1127         locks_remove_posix(filp, id);
1128         fput(filp);
1129         return retval;
1130 }
1131
1132 EXPORT_SYMBOL(filp_close);
1133
1134 /*
1135  * Careful here! We test whether the file pointer is NULL before
1136  * releasing the fd. This ensures that one clone task can't release
1137  * an fd while another clone is opening it.
1138  */
1139 SYSCALL_DEFINE1(close, unsigned int, fd)
1140 {
1141         struct file * filp;
1142         struct files_struct *files = current->files;
1143         struct fdtable *fdt;
1144         int retval;
1145
1146         spin_lock(&files->file_lock);
1147         fdt = files_fdtable(files);
1148         if (fd >= fdt->max_fds)
1149                 goto out_unlock;
1150         filp = fdt->fd[fd];
1151         if (!filp)
1152                 goto out_unlock;
1153         rcu_assign_pointer(fdt->fd[fd], NULL);
1154         FD_CLR(fd, fdt->close_on_exec);
1155         __put_unused_fd(files, fd);
1156         spin_unlock(&files->file_lock);
1157         retval = filp_close(filp, files);
1158
1159         /* can't restart close syscall because file table entry was cleared */
1160         if (unlikely(retval == -ERESTARTSYS ||
1161                      retval == -ERESTARTNOINTR ||
1162                      retval == -ERESTARTNOHAND ||
1163                      retval == -ERESTART_RESTARTBLOCK))
1164                 retval = -EINTR;
1165
1166         return retval;
1167
1168 out_unlock:
1169         spin_unlock(&files->file_lock);
1170         return -EBADF;
1171 }
1172 EXPORT_SYMBOL(sys_close);
1173
1174 /*
1175  * This routine simulates a hangup on the tty, to arrange that users
1176  * are given clean terminals at login time.
1177  */
1178 SYSCALL_DEFINE0(vhangup)
1179 {
1180         if (capable(CAP_SYS_TTY_CONFIG)) {
1181                 tty_vhangup_self();
1182                 return 0;
1183         }
1184         return -EPERM;
1185 }
1186
1187 /*
1188  * Called when an inode is about to be open.
1189  * We use this to disallow opening large files on 32bit systems if
1190  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1191  * on this flag in sys_open.
1192  */
1193 int generic_file_open(struct inode * inode, struct file * filp)
1194 {
1195         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1196                 return -EOVERFLOW;
1197         return 0;
1198 }
1199
1200 EXPORT_SYMBOL(generic_file_open);
1201
1202 /*
1203  * This is used by subsystems that don't want seekable
1204  * file descriptors
1205  */
1206 int nonseekable_open(struct inode *inode, struct file *filp)
1207 {
1208         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1209         return 0;
1210 }
1211
1212 EXPORT_SYMBOL(nonseekable_open);