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