fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid/sgid on truncate too */
  55         ret = should_remove_suid(dentry);
  56         if (ret)
  57                 newattrs.ia_valid |= ret | ATTR_FORCE;
  58
  59         mutex_lock(&dentry->d_inode->i_mutex);
  60         ret = notify_change(dentry, &newattrs);
  61         mutex_unlock(&dentry->d_inode->i_mutex);
  62         return ret;
  63 }
  64
  65 long vfs_truncate(struct path *path, loff_t length)
  66 {
  67         struct inode *inode;
  68         long error;
  69
  70         inode = path->dentry->d_inode;
  71
  72         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  73         if (S_ISDIR(inode->i_mode))
  74                 return -EISDIR;
  75         if (!S_ISREG(inode->i_mode))
  76                 return -EINVAL;
  77
  78         error = mnt_want_write(path->mnt);
  79         if (error)
  80                 goto out;
  81
  82         error = inode_permission(inode, MAY_WRITE);
  83         if (error)
  84                 goto mnt_drop_write_and_out;
  85
  86         error = -EPERM;
  87         if (IS_APPEND(inode))
  88                 goto mnt_drop_write_and_out;
  89
  90         error = get_write_access(inode);
  91         if (error)
  92                 goto mnt_drop_write_and_out;
  93
  94         /*
  95          * Make sure that there are no leases.  get_write_access() protects
  96          * against the truncate racing with a lease-granting setlease().
  97          */
  98         error = break_lease(inode, O_WRONLY);
  99         if (error)
 100                 goto put_write_and_out;
 101
 102         error = locks_verify_truncate(inode, NULL, length);
 103         if (!error)
 104                 error = security_path_truncate(path);
 105         if (!error)
 106                 error = do_truncate(path->dentry, length, 0, NULL);
 107
 108 put_write_and_out:
 109         put_write_access(inode);
 110 mnt_drop_write_and_out:
 111         mnt_drop_write(path->mnt);
 112 out:
 113         return error;
 114 }
 115 EXPORT_SYMBOL_GPL(vfs_truncate);
 116
 117 static long do_sys_truncate(const char __user *pathname, loff_t length)
 118 {
 119         unsigned int lookup_flags = LOOKUP_FOLLOW;
 120         struct path path;
 121         int error;
 122
 123         if (length < 0) /* sorry, but loff_t says... */
 124                 return -EINVAL;
 125
 126 retry:
 127         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 128         if (!error) {
 129                 error = vfs_truncate(&path, length);
 130                 path_put(&path);
 131         }
 132         if (retry_estale(error, lookup_flags)) {
 133                 lookup_flags |= LOOKUP_REVAL;
 134                 goto retry;
 135         }
 136         return error;
 137 }
 138
 139 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 140 {
 141         return do_sys_truncate(path, length);
 142 }
 143
 144 #ifdef CONFIG_COMPAT
 145 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 146 {
 147         return do_sys_truncate(path, length);
 148 }
 149 #endif
 150
 151 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 152 {
 153         struct inode *inode;
 154         struct dentry *dentry;
 155         struct fd f;
 156         int error;
 157
 158         error = -EINVAL;
 159         if (length < 0)
 160                 goto out;
 161         error = -EBADF;
 162         f = fdget(fd);
 163         if (!f.file)
 164                 goto out;
 165
 166         /* explicitly opened as large or we are on 64-bit box */
 167         if (f.file->f_flags & O_LARGEFILE)
 168                 small = 0;
 169
 170         dentry = f.file->f_path.dentry;
 171         inode = dentry->d_inode;
 172         error = -EINVAL;
 173         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 174                 goto out_putf;
 175
 176         error = -EINVAL;
 177         /* Cannot ftruncate over 2^31 bytes without large file support */
 178         if (small && length > MAX_NON_LFS)
 179                 goto out_putf;
 180
 181         error = -EPERM;
 182         if (IS_APPEND(inode))
 183                 goto out_putf;
 184
 185         sb_start_write(inode->i_sb);
 186         error = locks_verify_truncate(inode, f.file, length);
 187         if (!error)
 188                 error = security_path_truncate(&f.file->f_path);
 189         if (!error)
 190                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 191         sb_end_write(inode->i_sb);
 192 out_putf:
 193         fdput(f);
 194 out:
 195         return error;
 196 }
 197
 198 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 199 {
 200         long ret = do_sys_ftruncate(fd, length, 1);
 201         /* avoid REGPARM breakage on x86: */
 202         asmlinkage_protect(2, ret, fd, length);
 203         return ret;
 204 }
 205
 206 #ifdef CONFIG_COMPAT
 207 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 208 {
 209         return do_sys_ftruncate(fd, length, 1);
 210 }
 211 #endif
 212
 213 /* LFS versions of truncate are only needed on 32 bit machines */
 214 #if BITS_PER_LONG == 32
 215 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 216 {
 217         return do_sys_truncate(path, length);
 218 }
 219 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 220 asmlinkage long SyS_truncate64(long path, loff_t length)
 221 {
 222         return SYSC_truncate64((const char __user *) path, length);
 223 }
 224 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 225 #endif
 226
 227 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 228 {
 229         long ret = do_sys_ftruncate(fd, length, 0);
 230         /* avoid REGPARM breakage on x86: */
 231         asmlinkage_protect(2, ret, fd, length);
 232         return ret;
 233 }
 234 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 235 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 236 {
 237         return SYSC_ftruncate64((unsigned int) fd, length);
 238 }
 239 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 240 #endif
 241 #endif /* BITS_PER_LONG == 32 */
 242
 243
 244 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 245 {
 246         struct inode *inode = file_inode(file);
 247         long ret;
 248
 249         if (offset < 0 || len <= 0)
 250                 return -EINVAL;
 251
 252         /* Return error if mode is not supported */
 253         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 254                 return -EOPNOTSUPP;
 255
 256         /* Punch hole must have keep size set */
 257         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 258             !(mode & FALLOC_FL_KEEP_SIZE))
 259                 return -EOPNOTSUPP;
 260
 261         if (!(file->f_mode & FMODE_WRITE))
 262                 return -EBADF;
 263
 264         /* It's not possible punch hole on append only file */
 265         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 266                 return -EPERM;
 267
 268         if (IS_IMMUTABLE(inode))
 269                 return -EPERM;
 270
 271         /*
 272          * Revalidate the write permissions, in case security policy has
 273          * changed since the files were opened.
 274          */
 275         ret = security_file_permission(file, MAY_WRITE);
 276         if (ret)
 277                 return ret;
 278
 279         if (S_ISFIFO(inode->i_mode))
 280                 return -ESPIPE;
 281
 282         /*
 283          * Let individual file system decide if it supports preallocation
 284          * for directories or not.
 285          */
 286         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 287                 return -ENODEV;
 288
 289         /* Check for wrap through zero too */
 290         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 291                 return -EFBIG;
 292
 293         if (!file->f_op->fallocate)
 294                 return -EOPNOTSUPP;
 295
 296         sb_start_write(inode->i_sb);
 297         ret = file->f_op->fallocate(file, mode, offset, len);
 298         sb_end_write(inode->i_sb);
 299         return ret;
 300 }
 301
 302 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 303 {
 304         struct fd f = fdget(fd);
 305         int error = -EBADF;
 306
 307         if (f.file) {
 308                 error = do_fallocate(f.file, mode, offset, len);
 309                 fdput(f);
 310         }
 311         return error;
 312 }
 313
 314 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 315 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 316 {
 317         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 318 }
 319 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 320 #endif
 321
 322 /*
 323  * access() needs to use the real uid/gid, not the effective uid/gid.
 324  * We do this by temporarily clearing all FS-related capabilities and
 325  * switching the fsuid/fsgid around to the real ones.
 326  */
 327 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 328 {
 329         const struct cred *old_cred;
 330         struct cred *override_cred;
 331         struct path path;
 332         struct inode *inode;
 333         int res;
 334         unsigned int lookup_flags = LOOKUP_FOLLOW;
 335
 336         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 337                 return -EINVAL;
 338
 339         override_cred = prepare_creds();
 340         if (!override_cred)
 341                 return -ENOMEM;
 342
 343         override_cred->fsuid = override_cred->uid;
 344         override_cred->fsgid = override_cred->gid;
 345
 346         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 347                 /* Clear the capabilities if we switch to a non-root user */
 348                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 349                 if (!uid_eq(override_cred->uid, root_uid))
 350                         cap_clear(override_cred->cap_effective);
 351                 else
 352                         override_cred->cap_effective =
 353                                 override_cred->cap_permitted;
 354         }
 355
 356         old_cred = override_creds(override_cred);
 357 retry:
 358         res = user_path_at(dfd, filename, lookup_flags, &path);
 359         if (res)
 360                 goto out;
 361
 362         inode = path.dentry->d_inode;
 363
 364         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 365                 /*
 366                  * MAY_EXEC on regular files is denied if the fs is mounted
 367                  * with the "noexec" flag.
 368                  */
 369                 res = -EACCES;
 370                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 371                         goto out_path_release;
 372         }
 373
 374         res = inode_permission(inode, mode | MAY_ACCESS);
 375         /* SuS v2 requires we report a read only fs too */
 376         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 377                 goto out_path_release;
 378         /*
 379          * This is a rare case where using __mnt_is_readonly()
 380          * is OK without a mnt_want/drop_write() pair.  Since
 381          * no actual write to the fs is performed here, we do
 382          * not need to telegraph to that to anyone.
 383          *
 384          * By doing this, we accept that this access is
 385          * inherently racy and know that the fs may change
 386          * state before we even see this result.
 387          */
 388         if (__mnt_is_readonly(path.mnt))
 389                 res = -EROFS;
 390
 391 out_path_release:
 392         path_put(&path);
 393         if (retry_estale(res, lookup_flags)) {
 394                 lookup_flags |= LOOKUP_REVAL;
 395                 goto retry;
 396         }
 397 out:
 398         revert_creds(old_cred);
 399         put_cred(override_cred);
 400         return res;
 401 }
 402
 403 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 404 {
 405         return sys_faccessat(AT_FDCWD, filename, mode);
 406 }
 407
 408 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 409 {
 410         struct path path;
 411         int error;
 412         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 413 retry:
 414         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 415         if (error)
 416                 goto out;
 417
 418         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 419         if (error)
 420                 goto dput_and_out;
 421
 422         set_fs_pwd(current->fs, &path);
 423
 424 dput_and_out:
 425         path_put(&path);
 426         if (retry_estale(error, lookup_flags)) {
 427                 lookup_flags |= LOOKUP_REVAL;
 428                 goto retry;
 429         }
 430 out:
 431         return error;
 432 }
 433
 434 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 435 {
 436         struct fd f = fdget_raw(fd);
 437         struct inode *inode;
 438         int error = -EBADF;
 439
 440         error = -EBADF;
 441         if (!f.file)
 442                 goto out;
 443
 444         inode = file_inode(f.file);
 445
 446         error = -ENOTDIR;
 447         if (!S_ISDIR(inode->i_mode))
 448                 goto out_putf;
 449
 450         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 451         if (!error)
 452                 set_fs_pwd(current->fs, &f.file->f_path);
 453 out_putf:
 454         fdput(f);
 455 out:
 456         return error;
 457 }
 458
 459 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 460 {
 461         struct path path;
 462         int error;
 463         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 464 retry:
 465         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 466         if (error)
 467                 goto out;
 468
 469         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 470         if (error)
 471                 goto dput_and_out;
 472
 473         error = -EPERM;
 474         if (!nsown_capable(CAP_SYS_CHROOT))
 475                 goto dput_and_out;
 476         error = security_path_chroot(&path);
 477         if (error)
 478                 goto dput_and_out;
 479
 480         set_fs_root(current->fs, &path);
 481         error = 0;
 482 dput_and_out:
 483         path_put(&path);
 484         if (retry_estale(error, lookup_flags)) {
 485                 lookup_flags |= LOOKUP_REVAL;
 486                 goto retry;
 487         }
 488 out:
 489         return error;
 490 }
 491
 492 static int chmod_common(struct path *path, umode_t mode)
 493 {
 494         struct inode *inode = path->dentry->d_inode;
 495         struct iattr newattrs;
 496         int error;
 497
 498         error = mnt_want_write(path->mnt);
 499         if (error)
 500                 return error;
 501         mutex_lock(&inode->i_mutex);
 502         error = security_path_chmod(path, mode);
 503         if (error)
 504                 goto out_unlock;
 505         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 506         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 507         error = notify_change(path->dentry, &newattrs);
 508 out_unlock:
 509         mutex_unlock(&inode->i_mutex);
 510         mnt_drop_write(path->mnt);
 511         return error;
 512 }
 513
 514 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 515 {
 516         struct file * file;
 517         int err = -EBADF;
 518
 519         file = fget(fd);
 520         if (file) {
 521                 audit_inode(NULL, file->f_path.dentry, 0);
 522                 err = chmod_common(&file->f_path, mode);
 523                 fput(file);
 524         }
 525         return err;
 526 }
 527
 528 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 529 {
 530         struct path path;
 531         int error;
 532         unsigned int lookup_flags = LOOKUP_FOLLOW;
 533 retry:
 534         error = user_path_at(dfd, filename, lookup_flags, &path);
 535         if (!error) {
 536                 error = chmod_common(&path, mode);
 537                 path_put(&path);
 538                 if (retry_estale(error, lookup_flags)) {
 539                         lookup_flags |= LOOKUP_REVAL;
 540                         goto retry;
 541                 }
 542         }
 543         return error;
 544 }
 545
 546 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 547 {
 548         return sys_fchmodat(AT_FDCWD, filename, mode);
 549 }
 550
 551 static int chown_common(struct path *path, uid_t user, gid_t group)
 552 {
 553         struct inode *inode = path->dentry->d_inode;
 554         int error;
 555         struct iattr newattrs;
 556         kuid_t uid;
 557         kgid_t gid;
 558
 559         uid = make_kuid(current_user_ns(), user);
 560         gid = make_kgid(current_user_ns(), group);
 561
 562         newattrs.ia_valid =  ATTR_CTIME;
 563         if (user != (uid_t) -1) {
 564                 if (!uid_valid(uid))
 565                         return -EINVAL;
 566                 newattrs.ia_valid |= ATTR_UID;
 567                 newattrs.ia_uid = uid;
 568         }
 569         if (group != (gid_t) -1) {
 570                 if (!gid_valid(gid))
 571                         return -EINVAL;
 572                 newattrs.ia_valid |= ATTR_GID;
 573                 newattrs.ia_gid = gid;
 574         }
 575         if (!S_ISDIR(inode->i_mode))
 576                 newattrs.ia_valid |=
 577                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 578         mutex_lock(&inode->i_mutex);
 579         error = security_path_chown(path, uid, gid);
 580         if (!error)
 581                 error = notify_change(path->dentry, &newattrs);
 582         mutex_unlock(&inode->i_mutex);
 583
 584         return error;
 585 }
 586
 587 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 588                 gid_t, group, int, flag)
 589 {
 590         struct path path;
 591         int error = -EINVAL;
 592         int lookup_flags;
 593
 594         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 595                 goto out;
 596
 597         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 598         if (flag & AT_EMPTY_PATH)
 599                 lookup_flags |= LOOKUP_EMPTY;
 600 retry:
 601         error = user_path_at(dfd, filename, lookup_flags, &path);
 602         if (error)
 603                 goto out;
 604         error = mnt_want_write(path.mnt);
 605         if (error)
 606                 goto out_release;
 607         error = chown_common(&path, user, group);
 608         mnt_drop_write(path.mnt);
 609 out_release:
 610         path_put(&path);
 611         if (retry_estale(error, lookup_flags)) {
 612                 lookup_flags |= LOOKUP_REVAL;
 613                 goto retry;
 614         }
 615 out:
 616         return error;
 617 }
 618
 619 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 620 {
 621         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 622 }
 623
 624 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 625 {
 626         return sys_fchownat(AT_FDCWD, filename, user, group,
 627                             AT_SYMLINK_NOFOLLOW);
 628 }
 629
 630 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 631 {
 632         struct fd f = fdget(fd);
 633         int error = -EBADF;
 634
 635         if (!f.file)
 636                 goto out;
 637
 638         error = mnt_want_write_file(f.file);
 639         if (error)
 640                 goto out_fput;
 641         audit_inode(NULL, f.file->f_path.dentry, 0);
 642         error = chown_common(&f.file->f_path, user, group);
 643         mnt_drop_write_file(f.file);
 644 out_fput:
 645         fdput(f);
 646 out:
 647         return error;
 648 }
 649
 650 /*
 651  * You have to be very careful that these write
 652  * counts get cleaned up in error cases and
 653  * upon __fput().  This should probably never
 654  * be called outside of __dentry_open().
 655  */
 656 static inline int __get_file_write_access(struct inode *inode,
 657                                           struct vfsmount *mnt)
 658 {
 659         int error;
 660         error = get_write_access(inode);
 661         if (error)
 662                 return error;
 663         /*
 664          * Do not take mount writer counts on
 665          * special files since no writes to
 666          * the mount itself will occur.
 667          */
 668         if (!special_file(inode->i_mode)) {
 669                 /*
 670                  * Balanced in __fput()
 671                  */
 672                 error = __mnt_want_write(mnt);
 673                 if (error)
 674                         put_write_access(inode);
 675         }
 676         return error;
 677 }
 678
 679 int open_check_o_direct(struct file *f)
 680 {
 681         /* NB: we're sure to have correct a_ops only after f_op->open */
 682         if (f->f_flags & O_DIRECT) {
 683                 if (!f->f_mapping->a_ops ||
 684                     ((!f->f_mapping->a_ops->direct_IO) &&
 685                     (!f->f_mapping->a_ops->get_xip_mem))) {
 686                         return -EINVAL;
 687                 }
 688         }
 689         return 0;
 690 }
 691
 692 static int do_dentry_open(struct file *f,
 693                           int (*open)(struct inode *, struct file *),
 694                           const struct cred *cred)
 695 {
 696         static const struct file_operations empty_fops = {};
 697         struct inode *inode;
 698         int error;
 699
 700         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 701                                 FMODE_PREAD | FMODE_PWRITE;
 702
 703         if (unlikely(f->f_flags & O_PATH))
 704                 f->f_mode = FMODE_PATH;
 705
 706         path_get(&f->f_path);
 707         inode = file_inode(f);
 708         if (f->f_mode & FMODE_WRITE) {
 709                 error = __get_file_write_access(inode, f->f_path.mnt);
 710                 if (error)
 711                         goto cleanup_file;
 712                 if (!special_file(inode->i_mode))
 713                         file_take_write(f);
 714         }
 715
 716         f->f_mapping = inode->i_mapping;
 717         file_sb_list_add(f, inode->i_sb);
 718
 719         if (unlikely(f->f_mode & FMODE_PATH)) {
 720                 f->f_op = &empty_fops;
 721                 return 0;
 722         }
 723
 724         f->f_op = fops_get(inode->i_fop);
 725
 726         error = security_file_open(f, cred);
 727         if (error)
 728                 goto cleanup_all;
 729
 730         error = break_lease(inode, f->f_flags);
 731         if (error)
 732                 goto cleanup_all;
 733
 734         if (!open && f->f_op)
 735                 open = f->f_op->open;
 736         if (open) {
 737                 error = open(inode, f);
 738                 if (error)
 739                         goto cleanup_all;
 740         }
 741         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 742                 i_readcount_inc(inode);
 743
 744         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 745
 746         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 747
 748         return 0;
 749
 750 cleanup_all:
 751         fops_put(f->f_op);
 752         file_sb_list_del(f);
 753         if (f->f_mode & FMODE_WRITE) {
 754                 put_write_access(inode);
 755                 if (!special_file(inode->i_mode)) {
 756                         /*
 757                          * We don't consider this a real
 758                          * mnt_want/drop_write() pair
 759                          * because it all happenend right
 760                          * here, so just reset the state.
 761                          */
 762                         file_reset_write(f);
 763                         __mnt_drop_write(f->f_path.mnt);
 764                 }
 765         }
 766 cleanup_file:
 767         path_put(&f->f_path);
 768         f->f_path.mnt = NULL;
 769         f->f_path.dentry = NULL;
 770         return error;
 771 }
 772
 773 /**
 774  * finish_open - finish opening a file
 775  * @od: opaque open data
 776  * @dentry: pointer to dentry
 777  * @open: open callback
 778  *
 779  * This can be used to finish opening a file passed to i_op->atomic_open().
 780  *
 781  * If the open callback is set to NULL, then the standard f_op->open()
 782  * filesystem callback is substituted.
 783  */
 784 int finish_open(struct file *file, struct dentry *dentry,
 785                 int (*open)(struct inode *, struct file *),
 786                 int *opened)
 787 {
 788         int error;
 789         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 790
 791         file->f_path.dentry = dentry;
 792         error = do_dentry_open(file, open, current_cred());
 793         if (!error)
 794                 *opened |= FILE_OPENED;
 795
 796         return error;
 797 }
 798 EXPORT_SYMBOL(finish_open);
 799
 800 /**
 801  * finish_no_open - finish ->atomic_open() without opening the file
 802  *
 803  * @od: opaque open data
 804  * @dentry: dentry or NULL (as returned from ->lookup())
 805  *
 806  * This can be used to set the result of a successful lookup in ->atomic_open().
 807  * The filesystem's atomic_open() method shall return NULL after calling this.
 808  */
 809 int finish_no_open(struct file *file, struct dentry *dentry)
 810 {
 811         file->f_path.dentry = dentry;
 812         return 1;
 813 }
 814 EXPORT_SYMBOL(finish_no_open);
 815
 816 struct file *dentry_open(const struct path *path, int flags,
 817                          const struct cred *cred)
 818 {
 819         int error;
 820         struct file *f;
 821
 822         validate_creds(cred);
 823
 824         /* We must always pass in a valid mount pointer. */
 825         BUG_ON(!path->mnt);
 826
 827         f = get_empty_filp();
 828         if (!IS_ERR(f)) {
 829                 f->f_flags = flags;
 830                 f->f_path = *path;
 831                 error = do_dentry_open(f, NULL, cred);
 832                 if (!error) {
 833                         /* from now on we need fput() to dispose of f */
 834                         error = open_check_o_direct(f);
 835                         if (error) {
 836                                 fput(f);
 837                                 f = ERR_PTR(error);
 838                         }
 839                 } else {
 840                         put_filp(f);
 841                         f = ERR_PTR(error);
 842                 }
 843         }
 844         return f;
 845 }
 846 EXPORT_SYMBOL(dentry_open);
 847
 848 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 849 {
 850         int lookup_flags = 0;
 851         int acc_mode;
 852
 853         if (flags & O_CREAT)
 854                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 855         else
 856                 op->mode = 0;
 857
 858         /* Must never be set by userspace */
 859         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 860
 861         /*
 862          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 863          * check for O_DSYNC if the need any syncing at all we enforce it's
 864          * always set instead of having to deal with possibly weird behaviour
 865          * for malicious applications setting only __O_SYNC.
 866          */
 867         if (flags & __O_SYNC)
 868                 flags |= O_DSYNC;
 869
 870         /*
 871          * If we have O_PATH in the open flag. Then we
 872          * cannot have anything other than the below set of flags
 873          */
 874         if (flags & O_PATH) {
 875                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 876                 acc_mode = 0;
 877         } else {
 878                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 879         }
 880
 881         op->open_flag = flags;
 882
 883         /* O_TRUNC implies we need access checks for write permissions */
 884         if (flags & O_TRUNC)
 885                 acc_mode |= MAY_WRITE;
 886
 887         /* Allow the LSM permission hook to distinguish append
 888            access from general write access. */
 889         if (flags & O_APPEND)
 890                 acc_mode |= MAY_APPEND;
 891
 892         op->acc_mode = acc_mode;
 893
 894         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 895
 896         if (flags & O_CREAT) {
 897                 op->intent |= LOOKUP_CREATE;
 898                 if (flags & O_EXCL)
 899                         op->intent |= LOOKUP_EXCL;
 900         }
 901
 902         if (flags & O_DIRECTORY)
 903                 lookup_flags |= LOOKUP_DIRECTORY;
 904         if (!(flags & O_NOFOLLOW))
 905                 lookup_flags |= LOOKUP_FOLLOW;
 906         return lookup_flags;
 907 }
 908
 909 /**
 910  * file_open_name - open file and return file pointer
 911  *
 912  * @name:       struct filename containing path to open
 913  * @flags:      open flags as per the open(2) second argument
 914  * @mode:       mode for the new file if O_CREAT is set, else ignored
 915  *
 916  * This is the helper to open a file from kernelspace if you really
 917  * have to.  But in generally you should not do this, so please move
 918  * along, nothing to see here..
 919  */
 920 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 921 {
 922         struct open_flags op;
 923         int lookup = build_open_flags(flags, mode, &op);
 924         return do_filp_open(AT_FDCWD, name, &op, lookup);
 925 }
 926
 927 /**
 928  * filp_open - open file and return file pointer
 929  *
 930  * @filename:   path to open
 931  * @flags:      open flags as per the open(2) second argument
 932  * @mode:       mode for the new file if O_CREAT is set, else ignored
 933  *
 934  * This is the helper to open a file from kernelspace if you really
 935  * have to.  But in generally you should not do this, so please move
 936  * along, nothing to see here..
 937  */
 938 struct file *filp_open(const char *filename, int flags, umode_t mode)
 939 {
 940         struct filename name = {.name = filename};
 941         return file_open_name(&name, flags, mode);
 942 }
 943 EXPORT_SYMBOL(filp_open);
 944
 945 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 946                             const char *filename, int flags)
 947 {
 948         struct open_flags op;
 949         int lookup = build_open_flags(flags, 0, &op);
 950         if (flags & O_CREAT)
 951                 return ERR_PTR(-EINVAL);
 952         if (!filename && (flags & O_DIRECTORY))
 953                 if (!dentry->d_inode->i_op->lookup)
 954                         return ERR_PTR(-ENOTDIR);
 955         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 956 }
 957 EXPORT_SYMBOL(file_open_root);
 958
 959 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 960 {
 961         struct open_flags op;
 962         int lookup = build_open_flags(flags, mode, &op);
 963         struct filename *tmp = getname(filename);
 964         int fd = PTR_ERR(tmp);
 965
 966         if (!IS_ERR(tmp)) {
 967                 fd = get_unused_fd_flags(flags);
 968                 if (fd >= 0) {
 969                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 970                         if (IS_ERR(f)) {
 971                                 put_unused_fd(fd);
 972                                 fd = PTR_ERR(f);
 973                         } else {
 974                                 fsnotify_open(f);
 975                                 fd_install(fd, f);
 976                         }
 977                 }
 978                 putname(tmp);
 979         }
 980         return fd;
 981 }
 982
 983 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 984 {
 985         long ret;
 986
 987         if (force_o_largefile())
 988                 flags |= O_LARGEFILE;
 989
 990         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 991         /* avoid REGPARM breakage on x86: */
 992         asmlinkage_protect(3, ret, filename, flags, mode);
 993         return ret;
 994 }
 995
 996 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 997                 umode_t, mode)
 998 {
 999         long ret;
1000
1001         if (force_o_largefile())
1002                 flags |= O_LARGEFILE;
1003
1004         ret = do_sys_open(dfd, filename, flags, mode);
1005         /* avoid REGPARM breakage on x86: */
1006         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1007         return ret;
1008 }
1009
1010 #ifndef __alpha__
1011
1012 /*
1013  * For backward compatibility?  Maybe this should be moved
1014  * into arch/i386 instead?
1015  */
1016 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1017 {
1018         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1019 }
1020
1021 #endif
1022
1023 /*
1024  * "id" is the POSIX thread ID. We use the
1025  * files pointer for this..
1026  */
1027 int filp_close(struct file *filp, fl_owner_t id)
1028 {
1029         int retval = 0;
1030
1031         if (!file_count(filp)) {
1032                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1033                 return 0;
1034         }
1035
1036         if (filp->f_op && filp->f_op->flush)
1037                 retval = filp->f_op->flush(filp, id);
1038
1039         if (likely(!(filp->f_mode & FMODE_PATH))) {
1040                 dnotify_flush(filp, id);
1041                 locks_remove_posix(filp, id);
1042         }
1043         fput(filp);
1044         return retval;
1045 }
1046
1047 EXPORT_SYMBOL(filp_close);
1048
1049 /*
1050  * Careful here! We test whether the file pointer is NULL before
1051  * releasing the fd. This ensures that one clone task can't release
1052  * an fd while another clone is opening it.
1053  */
1054 SYSCALL_DEFINE1(close, unsigned int, fd)
1055 {
1056         int retval = __close_fd(current->files, fd);
1057
1058         /* can't restart close syscall because file table entry was cleared */
1059         if (unlikely(retval == -ERESTARTSYS ||
1060                      retval == -ERESTARTNOINTR ||
1061                      retval == -ERESTARTNOHAND ||
1062                      retval == -ERESTART_RESTARTBLOCK))
1063                 retval = -EINTR;
1064
1065         return retval;
1066 }
1067 EXPORT_SYMBOL(sys_close);
1068
1069 /*
1070  * This routine simulates a hangup on the tty, to arrange that users
1071  * are given clean terminals at login time.
1072  */
1073 SYSCALL_DEFINE0(vhangup)
1074 {
1075         if (capable(CAP_SYS_TTY_CONFIG)) {
1076                 tty_vhangup_self();
1077                 return 0;
1078         }
1079         return -EPERM;
1080 }
1081
1082 /*
1083  * Called when an inode is about to be open.
1084  * We use this to disallow opening large files on 32bit systems if
1085  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1086  * on this flag in sys_open.
1087  */
1088 int generic_file_open(struct inode * inode, struct file * filp)
1089 {
1090         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1091                 return -EOVERFLOW;
1092         return 0;
1093 }
1094
1095 EXPORT_SYMBOL(generic_file_open);
1096
1097 /*
1098  * This is used by subsystems that don't want seekable
1099  * file descriptors. The function is not supposed to ever fail, the only
1100  * reason it returns an 'int' and not 'void' is so that it can be plugged
1101  * directly into file_operations structure.
1102  */
1103 int nonseekable_open(struct inode *inode, struct file *filp)
1104 {
1105         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1106         return 0;
1107 }
1108
1109 EXPORT_SYMBOL(nonseekable_open);