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;
 139
 140         error = -EINVAL;
 141         if (length < 0)
 142                 goto out;
 143         error = -EBADF;
 144         file = fget(fd);
 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         error = locks_verify_truncate(inode, file, length);
 168         if (!error)
 169                 error = security_path_truncate(&file->f_path);
 170         if (!error)
 171                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 172 out_putf:
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177
 178 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 179 {
 180         long ret = do_sys_ftruncate(fd, length, 1);
 181         /* avoid REGPARM breakage on x86: */
 182         asmlinkage_protect(2, ret, fd, length);
 183         return ret;
 184 }
 185
 186 /* LFS versions of truncate are only needed on 32 bit machines */
 187 #if BITS_PER_LONG == 32
 188 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 189 {
 190         return do_sys_truncate(path, length);
 191 }
 192 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 193 asmlinkage long SyS_truncate64(long path, loff_t length)
 194 {
 195         return SYSC_truncate64((const char __user *) path, length);
 196 }
 197 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 198 #endif
 199
 200 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 201 {
 202         long ret = do_sys_ftruncate(fd, length, 0);
 203         /* avoid REGPARM breakage on x86: */
 204         asmlinkage_protect(2, ret, fd, length);
 205         return ret;
 206 }
 207 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 208 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 209 {
 210         return SYSC_ftruncate64((unsigned int) fd, length);
 211 }
 212 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 213 #endif
 214 #endif /* BITS_PER_LONG == 32 */
 215
 216
 217 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 218 {
 219         struct inode *inode = file->f_path.dentry->d_inode;
 220         long ret;
 221
 222         if (offset < 0 || len <= 0)
 223                 return -EINVAL;
 224
 225         /* Return error if mode is not supported */
 226         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 227                 return -EOPNOTSUPP;
 228
 229         /* Punch hole must have keep size set */
 230         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 231             !(mode & FALLOC_FL_KEEP_SIZE))
 232                 return -EOPNOTSUPP;
 233
 234         if (!(file->f_mode & FMODE_WRITE))
 235                 return -EBADF;
 236
 237         /* It's not possible punch hole on append only file */
 238         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 239                 return -EPERM;
 240
 241         if (IS_IMMUTABLE(inode))
 242                 return -EPERM;
 243
 244         /*
 245          * Revalidate the write permissions, in case security policy has
 246          * changed since the files were opened.
 247          */
 248         ret = security_file_permission(file, MAY_WRITE);
 249         if (ret)
 250                 return ret;
 251
 252         if (S_ISFIFO(inode->i_mode))
 253                 return -ESPIPE;
 254
 255         /*
 256          * Let individual file system decide if it supports preallocation
 257          * for directories or not.
 258          */
 259         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 260                 return -ENODEV;
 261
 262         /* Check for wrap through zero too */
 263         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 264                 return -EFBIG;
 265
 266         if (!file->f_op->fallocate)
 267                 return -EOPNOTSUPP;
 268
 269         return file->f_op->fallocate(file, mode, offset, len);
 270 }
 271
 272 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 273 {
 274         struct file *file;
 275         int error = -EBADF;
 276
 277         file = fget(fd);
 278         if (file) {
 279                 error = do_fallocate(file, mode, offset, len);
 280                 fput(file);
 281         }
 282
 283         return error;
 284 }
 285
 286 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 287 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 288 {
 289         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 290 }
 291 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 292 #endif
 293
 294 /*
 295  * access() needs to use the real uid/gid, not the effective uid/gid.
 296  * We do this by temporarily clearing all FS-related capabilities and
 297  * switching the fsuid/fsgid around to the real ones.
 298  */
 299 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 300 {
 301         const struct cred *old_cred;
 302         struct cred *override_cred;
 303         struct path path;
 304         struct inode *inode;
 305         int res;
 306
 307         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 308                 return -EINVAL;
 309
 310         override_cred = prepare_creds();
 311         if (!override_cred)
 312                 return -ENOMEM;
 313
 314         override_cred->fsuid = override_cred->uid;
 315         override_cred->fsgid = override_cred->gid;
 316
 317         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 318                 /* Clear the capabilities if we switch to a non-root user */
 319                 if (override_cred->uid)
 320                         cap_clear(override_cred->cap_effective);
 321                 else
 322                         override_cred->cap_effective =
 323                                 override_cred->cap_permitted;
 324         }
 325
 326         old_cred = override_creds(override_cred);
 327
 328         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 329         if (res)
 330                 goto out;
 331
 332         inode = path.dentry->d_inode;
 333
 334         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 335                 /*
 336                  * MAY_EXEC on regular files is denied if the fs is mounted
 337                  * with the "noexec" flag.
 338                  */
 339                 res = -EACCES;
 340                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 341                         goto out_path_release;
 342         }
 343
 344         res = inode_permission(inode, mode | MAY_ACCESS);
 345         /* SuS v2 requires we report a read only fs too */
 346         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 347                 goto out_path_release;
 348         /*
 349          * This is a rare case where using __mnt_is_readonly()
 350          * is OK without a mnt_want/drop_write() pair.  Since
 351          * no actual write to the fs is performed here, we do
 352          * not need to telegraph to that to anyone.
 353          *
 354          * By doing this, we accept that this access is
 355          * inherently racy and know that the fs may change
 356          * state before we even see this result.
 357          */
 358         if (__mnt_is_readonly(path.mnt))
 359                 res = -EROFS;
 360
 361 out_path_release:
 362         path_put(&path);
 363 out:
 364         revert_creds(old_cred);
 365         put_cred(override_cred);
 366         return res;
 367 }
 368
 369 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 370 {
 371         return sys_faccessat(AT_FDCWD, filename, mode);
 372 }
 373
 374 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 375 {
 376         struct path path;
 377         int error;
 378
 379         error = user_path_dir(filename, &path);
 380         if (error)
 381                 goto out;
 382
 383         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 384         if (error)
 385                 goto dput_and_out;
 386
 387         set_fs_pwd(current->fs, &path);
 388
 389 dput_and_out:
 390         path_put(&path);
 391 out:
 392         return error;
 393 }
 394
 395 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 396 {
 397         struct file *file;
 398         struct inode *inode;
 399         int error;
 400
 401         error = -EBADF;
 402         file = fget(fd);
 403         if (!file)
 404                 goto out;
 405
 406         inode = file->f_path.dentry->d_inode;
 407
 408         error = -ENOTDIR;
 409         if (!S_ISDIR(inode->i_mode))
 410                 goto out_putf;
 411
 412         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 413         if (!error)
 414                 set_fs_pwd(current->fs, &file->f_path);
 415 out_putf:
 416         fput(file);
 417 out:
 418         return error;
 419 }
 420
 421 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 422 {
 423         struct path path;
 424         int error;
 425
 426         error = user_path_dir(filename, &path);
 427         if (error)
 428                 goto out;
 429
 430         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 431         if (error)
 432                 goto dput_and_out;
 433
 434         error = -EPERM;
 435         if (!capable(CAP_SYS_CHROOT))
 436                 goto dput_and_out;
 437         error = security_path_chroot(&path);
 438         if (error)
 439                 goto dput_and_out;
 440
 441         set_fs_root(current->fs, &path);
 442         error = 0;
 443 dput_and_out:
 444         path_put(&path);
 445 out:
 446         return error;
 447 }
 448
 449 static int chmod_common(struct path *path, umode_t mode)
 450 {
 451         struct inode *inode = path->dentry->d_inode;
 452         struct iattr newattrs;
 453         int error;
 454
 455         error = mnt_want_write(path->mnt);
 456         if (error)
 457                 return error;
 458         mutex_lock(&inode->i_mutex);
 459         error = security_path_chmod(path, mode);
 460         if (error)
 461                 goto out_unlock;
 462         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 463         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 464         error = notify_change(path->dentry, &newattrs);
 465 out_unlock:
 466         mutex_unlock(&inode->i_mutex);
 467         mnt_drop_write(path->mnt);
 468         return error;
 469 }
 470
 471 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 472 {
 473         struct file * file;
 474         int err = -EBADF;
 475
 476         file = fget(fd);
 477         if (file) {
 478                 audit_inode(NULL, file->f_path.dentry);
 479                 err = chmod_common(&file->f_path, mode);
 480                 fput(file);
 481         }
 482         return err;
 483 }
 484
 485 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 486 {
 487         struct path path;
 488         int error;
 489
 490         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 491         if (!error) {
 492                 error = chmod_common(&path, mode);
 493                 path_put(&path);
 494         }
 495         return error;
 496 }
 497
 498 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 499 {
 500         return sys_fchmodat(AT_FDCWD, filename, mode);
 501 }
 502
 503 static int chown_common(struct path *path, uid_t user, gid_t group)
 504 {
 505         struct inode *inode = path->dentry->d_inode;
 506         int error;
 507         struct iattr newattrs;
 508
 509         newattrs.ia_valid =  ATTR_CTIME;
 510         if (user != (uid_t) -1) {
 511                 newattrs.ia_valid |= ATTR_UID;
 512                 newattrs.ia_uid = user;
 513         }
 514         if (group != (gid_t) -1) {
 515                 newattrs.ia_valid |= ATTR_GID;
 516                 newattrs.ia_gid = group;
 517         }
 518         if (!S_ISDIR(inode->i_mode))
 519                 newattrs.ia_valid |=
 520                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 521         mutex_lock(&inode->i_mutex);
 522         error = security_path_chown(path, user, group);
 523         if (!error)
 524                 error = notify_change(path->dentry, &newattrs);
 525         mutex_unlock(&inode->i_mutex);
 526
 527         return error;
 528 }
 529
 530 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 531 {
 532         struct path path;
 533         int error;
 534
 535         error = user_path(filename, &path);
 536         if (error)
 537                 goto out;
 538         error = mnt_want_write(path.mnt);
 539         if (error)
 540                 goto out_release;
 541         error = chown_common(&path, user, group);
 542         mnt_drop_write(path.mnt);
 543 out_release:
 544         path_put(&path);
 545 out:
 546         return error;
 547 }
 548
 549 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 550                 gid_t, group, int, flag)
 551 {
 552         struct path path;
 553         int error = -EINVAL;
 554         int lookup_flags;
 555
 556         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 557                 goto out;
 558
 559         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 560         if (flag & AT_EMPTY_PATH)
 561                 lookup_flags |= LOOKUP_EMPTY;
 562         error = user_path_at(dfd, filename, lookup_flags, &path);
 563         if (error)
 564                 goto out;
 565         error = mnt_want_write(path.mnt);
 566         if (error)
 567                 goto out_release;
 568         error = chown_common(&path, user, group);
 569         mnt_drop_write(path.mnt);
 570 out_release:
 571         path_put(&path);
 572 out:
 573         return error;
 574 }
 575
 576 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 577 {
 578         struct path path;
 579         int error;
 580
 581         error = user_lpath(filename, &path);
 582         if (error)
 583                 goto out;
 584         error = mnt_want_write(path.mnt);
 585         if (error)
 586                 goto out_release;
 587         error = chown_common(&path, user, group);
 588         mnt_drop_write(path.mnt);
 589 out_release:
 590         path_put(&path);
 591 out:
 592         return error;
 593 }
 594
 595 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 596 {
 597         struct file * file;
 598         int error = -EBADF;
 599         struct dentry * dentry;
 600
 601         file = fget(fd);
 602         if (!file)
 603                 goto out;
 604
 605         error = mnt_want_write_file(file);
 606         if (error)
 607                 goto out_fput;
 608         dentry = file->f_path.dentry;
 609         audit_inode(NULL, dentry);
 610         error = chown_common(&file->f_path, user, group);
 611         mnt_drop_write_file(file);
 612 out_fput:
 613         fput(file);
 614 out:
 615         return error;
 616 }
 617
 618 /*
 619  * You have to be very careful that these write
 620  * counts get cleaned up in error cases and
 621  * upon __fput().  This should probably never
 622  * be called outside of __dentry_open().
 623  */
 624 static inline int __get_file_write_access(struct inode *inode,
 625                                           struct vfsmount *mnt)
 626 {
 627         int error;
 628         error = get_write_access(inode);
 629         if (error)
 630                 return error;
 631         /*
 632          * Do not take mount writer counts on
 633          * special files since no writes to
 634          * the mount itself will occur.
 635          */
 636         if (!special_file(inode->i_mode)) {
 637                 /*
 638                  * Balanced in __fput()
 639                  */
 640                 error = mnt_want_write(mnt);
 641                 if (error)
 642                         put_write_access(inode);
 643         }
 644         return error;
 645 }
 646
 647 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 648                                         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         inode = dentry->d_inode;
 663         if (f->f_mode & FMODE_WRITE) {
 664                 error = __get_file_write_access(inode, mnt);
 665                 if (error)
 666                         goto cleanup_file;
 667                 if (!special_file(inode->i_mode))
 668                         file_take_write(f);
 669         }
 670
 671         f->f_mapping = inode->i_mapping;
 672         f->f_path.dentry = dentry;
 673         f->f_path.mnt = mnt;
 674         f->f_pos = 0;
 675         file_sb_list_add(f, inode->i_sb);
 676
 677         if (unlikely(f->f_mode & FMODE_PATH)) {
 678                 f->f_op = &empty_fops;
 679                 return f;
 680         }
 681
 682         f->f_op = fops_get(inode->i_fop);
 683
 684         error = security_dentry_open(f, cred);
 685         if (error)
 686                 goto cleanup_all;
 687
 688         error = break_lease(inode, f->f_flags);
 689         if (error)
 690                 goto cleanup_all;
 691
 692         if (!open && f->f_op)
 693                 open = f->f_op->open;
 694         if (open) {
 695                 error = open(inode, f);
 696                 if (error)
 697                         goto cleanup_all;
 698         }
 699         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 700                 i_readcount_inc(inode);
 701
 702         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 703
 704         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 705
 706         /* NB: we're sure to have correct a_ops only after f_op->open */
 707         if (f->f_flags & O_DIRECT) {
 708                 if (!f->f_mapping->a_ops ||
 709                     ((!f->f_mapping->a_ops->direct_IO) &&
 710                     (!f->f_mapping->a_ops->get_xip_mem))) {
 711                         fput(f);
 712                         f = ERR_PTR(-EINVAL);
 713                 }
 714         }
 715
 716         return f;
 717
 718 cleanup_all:
 719         fops_put(f->f_op);
 720         if (f->f_mode & FMODE_WRITE) {
 721                 put_write_access(inode);
 722                 if (!special_file(inode->i_mode)) {
 723                         /*
 724                          * We don't consider this a real
 725                          * mnt_want/drop_write() pair
 726                          * because it all happenend right
 727                          * here, so just reset the state.
 728                          */
 729                         file_reset_write(f);
 730                         mnt_drop_write(mnt);
 731                 }
 732         }
 733         file_sb_list_del(f);
 734         f->f_path.dentry = NULL;
 735         f->f_path.mnt = NULL;
 736 cleanup_file:
 737         put_filp(f);
 738         dput(dentry);
 739         mntput(mnt);
 740         return ERR_PTR(error);
 741 }
 742
 743 /**
 744  * lookup_instantiate_filp - instantiates the open intent filp
 745  * @nd: pointer to nameidata
 746  * @dentry: pointer to dentry
 747  * @open: open callback
 748  *
 749  * Helper for filesystems that want to use lookup open intents and pass back
 750  * a fully instantiated struct file to the caller.
 751  * This function is meant to be called from within a filesystem's
 752  * lookup method.
 753  * Beware of calling it for non-regular files! Those ->open methods might block
 754  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 755  * leading to a deadlock, as nobody can open that fifo anymore, because
 756  * another process to open fifo will block on locked parent when doing lookup).
 757  * Note that in case of error, nd->intent.open.file is destroyed, but the
 758  * path information remains valid.
 759  * If the open callback is set to NULL, then the standard f_op->open()
 760  * filesystem callback is substituted.
 761  */
 762 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 763                 int (*open)(struct inode *, struct file *))
 764 {
 765         const struct cred *cred = current_cred();
 766
 767         if (IS_ERR(nd->intent.open.file))
 768                 goto out;
 769         if (IS_ERR(dentry))
 770                 goto out_err;
 771         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 772                                              nd->intent.open.file,
 773                                              open, cred);
 774 out:
 775         return nd->intent.open.file;
 776 out_err:
 777         release_open_intent(nd);
 778         nd->intent.open.file = ERR_CAST(dentry);
 779         goto out;
 780 }
 781 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 782
 783 /**
 784  * nameidata_to_filp - convert a nameidata to an open filp.
 785  * @nd: pointer to nameidata
 786  * @flags: open flags
 787  *
 788  * Note that this function destroys the original nameidata
 789  */
 790 struct file *nameidata_to_filp(struct nameidata *nd)
 791 {
 792         const struct cred *cred = current_cred();
 793         struct file *filp;
 794
 795         /* Pick up the filp from the open intent */
 796         filp = nd->intent.open.file;
 797         nd->intent.open.file = NULL;
 798
 799         /* Has the filesystem initialised the file for us? */
 800         if (filp->f_path.dentry == NULL) {
 801                 path_get(&nd->path);
 802                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 803                                      NULL, cred);
 804         }
 805         return filp;
 806 }
 807
 808 /*
 809  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 810  * error.
 811  */
 812 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 813                          const struct cred *cred)
 814 {
 815         int error;
 816         struct file *f;
 817
 818         validate_creds(cred);
 819
 820         /* We must always pass in a valid mount pointer. */
 821         BUG_ON(!mnt);
 822
 823         error = -ENFILE;
 824         f = get_empty_filp();
 825         if (f == NULL) {
 826                 dput(dentry);
 827                 mntput(mnt);
 828                 return ERR_PTR(error);
 829         }
 830
 831         f->f_flags = flags;
 832         return __dentry_open(dentry, mnt, f, NULL, cred);
 833 }
 834 EXPORT_SYMBOL(dentry_open);
 835
 836 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 837 {
 838         struct fdtable *fdt = files_fdtable(files);
 839         __clear_open_fd(fd, fdt);
 840         if (fd < files->next_fd)
 841                 files->next_fd = fd;
 842 }
 843
 844 void put_unused_fd(unsigned int fd)
 845 {
 846         struct files_struct *files = current->files;
 847         spin_lock(&files->file_lock);
 848         __put_unused_fd(files, fd);
 849         spin_unlock(&files->file_lock);
 850 }
 851
 852 EXPORT_SYMBOL(put_unused_fd);
 853
 854 /*
 855  * Install a file pointer in the fd array.
 856  *
 857  * The VFS is full of places where we drop the files lock between
 858  * setting the open_fds bitmap and installing the file in the file
 859  * array.  At any such point, we are vulnerable to a dup2() race
 860  * installing a file in the array before us.  We need to detect this and
 861  * fput() the struct file we are about to overwrite in this case.
 862  *
 863  * It should never happen - if we allow dup2() do it, _really_ bad things
 864  * will follow.
 865  */
 866
 867 void fd_install(unsigned int fd, struct file *file)
 868 {
 869         struct files_struct *files = current->files;
 870         struct fdtable *fdt;
 871         spin_lock(&files->file_lock);
 872         fdt = files_fdtable(files);
 873         BUG_ON(fdt->fd[fd] != NULL);
 874         rcu_assign_pointer(fdt->fd[fd], file);
 875         spin_unlock(&files->file_lock);
 876 }
 877
 878 EXPORT_SYMBOL(fd_install);
 879
 880 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 881 {
 882         int lookup_flags = 0;
 883         int acc_mode;
 884
 885         if (!(flags & O_CREAT))
 886                 mode = 0;
 887         op->mode = mode;
 888
 889         /* Must never be set by userspace */
 890         flags &= ~FMODE_NONOTIFY;
 891
 892         /*
 893          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 894          * check for O_DSYNC if the need any syncing at all we enforce it's
 895          * always set instead of having to deal with possibly weird behaviour
 896          * for malicious applications setting only __O_SYNC.
 897          */
 898         if (flags & __O_SYNC)
 899                 flags |= O_DSYNC;
 900
 901         /*
 902          * If we have O_PATH in the open flag. Then we
 903          * cannot have anything other than the below set of flags
 904          */
 905         if (flags & O_PATH) {
 906                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 907                 acc_mode = 0;
 908         } else {
 909                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 910         }
 911
 912         op->open_flag = flags;
 913
 914         /* O_TRUNC implies we need access checks for write permissions */
 915         if (flags & O_TRUNC)
 916                 acc_mode |= MAY_WRITE;
 917
 918         /* Allow the LSM permission hook to distinguish append
 919            access from general write access. */
 920         if (flags & O_APPEND)
 921                 acc_mode |= MAY_APPEND;
 922
 923         op->acc_mode = acc_mode;
 924
 925         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 926
 927         if (flags & O_CREAT) {
 928                 op->intent |= LOOKUP_CREATE;
 929                 if (flags & O_EXCL)
 930                         op->intent |= LOOKUP_EXCL;
 931         }
 932
 933         if (flags & O_DIRECTORY)
 934                 lookup_flags |= LOOKUP_DIRECTORY;
 935         if (!(flags & O_NOFOLLOW))
 936                 lookup_flags |= LOOKUP_FOLLOW;
 937         return lookup_flags;
 938 }
 939
 940 /**
 941  * filp_open - open file and return file pointer
 942  *
 943  * @filename:   path to open
 944  * @flags:      open flags as per the open(2) second argument
 945  * @mode:       mode for the new file if O_CREAT is set, else ignored
 946  *
 947  * This is the helper to open a file from kernelspace if you really
 948  * have to.  But in generally you should not do this, so please move
 949  * along, nothing to see here..
 950  */
 951 struct file *filp_open(const char *filename, int flags, umode_t mode)
 952 {
 953         struct open_flags op;
 954         int lookup = build_open_flags(flags, mode, &op);
 955         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 956 }
 957 EXPORT_SYMBOL(filp_open);
 958
 959 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 960                             const char *filename, int flags)
 961 {
 962         struct open_flags op;
 963         int lookup = build_open_flags(flags, 0, &op);
 964         if (flags & O_CREAT)
 965                 return ERR_PTR(-EINVAL);
 966         if (!filename && (flags & O_DIRECTORY))
 967                 if (!dentry->d_inode->i_op->lookup)
 968                         return ERR_PTR(-ENOTDIR);
 969         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 970 }
 971 EXPORT_SYMBOL(file_open_root);
 972
 973 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 974 {
 975         struct open_flags op;
 976         int lookup = build_open_flags(flags, mode, &op);
 977         char *tmp = getname(filename);
 978         int fd = PTR_ERR(tmp);
 979
 980         if (!IS_ERR(tmp)) {
 981                 fd = get_unused_fd_flags(flags);
 982                 if (fd >= 0) {
 983                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 984                         if (IS_ERR(f)) {
 985                                 put_unused_fd(fd);
 986                                 fd = PTR_ERR(f);
 987                         } else {
 988                                 fsnotify_open(f);
 989                                 fd_install(fd, f);
 990                         }
 991                 }
 992                 putname(tmp);
 993         }
 994         return fd;
 995 }
 996
 997 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 998 {
 999         long ret;
1000
1001         if (force_o_largefile())
1002                 flags |= O_LARGEFILE;
1003
1004         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1005         /* avoid REGPARM breakage on x86: */
1006         asmlinkage_protect(3, ret, filename, flags, mode);
1007         return ret;
1008 }
1009
1010 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1011                 umode_t, mode)
1012 {
1013         long ret;
1014
1015         if (force_o_largefile())
1016                 flags |= O_LARGEFILE;
1017
1018         ret = do_sys_open(dfd, filename, flags, mode);
1019         /* avoid REGPARM breakage on x86: */
1020         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1021         return ret;
1022 }
1023
1024 #ifndef __alpha__
1025
1026 /*
1027  * For backward compatibility?  Maybe this should be moved
1028  * into arch/i386 instead?
1029  */
1030 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1031 {
1032         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1033 }
1034
1035 #endif
1036
1037 /*
1038  * "id" is the POSIX thread ID. We use the
1039  * files pointer for this..
1040  */
1041 int filp_close(struct file *filp, fl_owner_t id)
1042 {
1043         int retval = 0;
1044
1045         if (!file_count(filp)) {
1046                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1047                 return 0;
1048         }
1049
1050         if (filp->f_op && filp->f_op->flush)
1051                 retval = filp->f_op->flush(filp, id);
1052
1053         if (likely(!(filp->f_mode & FMODE_PATH))) {
1054                 dnotify_flush(filp, id);
1055                 locks_remove_posix(filp, id);
1056         }
1057         fput(filp);
1058         return retval;
1059 }
1060
1061 EXPORT_SYMBOL(filp_close);
1062
1063 /*
1064  * Careful here! We test whether the file pointer is NULL before
1065  * releasing the fd. This ensures that one clone task can't release
1066  * an fd while another clone is opening it.
1067  */
1068 SYSCALL_DEFINE1(close, unsigned int, fd)
1069 {
1070         struct file * filp;
1071         struct files_struct *files = current->files;
1072         struct fdtable *fdt;
1073         int retval;
1074
1075         spin_lock(&files->file_lock);
1076         fdt = files_fdtable(files);
1077         if (fd >= fdt->max_fds)
1078                 goto out_unlock;
1079         filp = fdt->fd[fd];
1080         if (!filp)
1081                 goto out_unlock;
1082         rcu_assign_pointer(fdt->fd[fd], NULL);
1083         __clear_close_on_exec(fd, fdt);
1084         __put_unused_fd(files, fd);
1085         spin_unlock(&files->file_lock);
1086         retval = filp_close(filp, files);
1087
1088         /* can't restart close syscall because file table entry was cleared */
1089         if (unlikely(retval == -ERESTARTSYS ||
1090                      retval == -ERESTARTNOINTR ||
1091                      retval == -ERESTARTNOHAND ||
1092                      retval == -ERESTART_RESTARTBLOCK))
1093                 retval = -EINTR;
1094
1095         return retval;
1096
1097 out_unlock:
1098         spin_unlock(&files->file_lock);
1099         return -EBADF;
1100 }
1101 EXPORT_SYMBOL(sys_close);
1102
1103 /*
1104  * This routine simulates a hangup on the tty, to arrange that users
1105  * are given clean terminals at login time.
1106  */
1107 SYSCALL_DEFINE0(vhangup)
1108 {
1109         if (capable(CAP_SYS_TTY_CONFIG)) {
1110                 tty_vhangup_self();
1111                 return 0;
1112         }
1113         return -EPERM;
1114 }
1115
1116 /*
1117  * Called when an inode is about to be open.
1118  * We use this to disallow opening large files on 32bit systems if
1119  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1120  * on this flag in sys_open.
1121  */
1122 int generic_file_open(struct inode * inode, struct file * filp)
1123 {
1124         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1125                 return -EOVERFLOW;
1126         return 0;
1127 }
1128
1129 EXPORT_SYMBOL(generic_file_open);
1130
1131 /*
1132  * This is used by subsystems that don't want seekable
1133  * file descriptors. The function is not supposed to ever fail, the only
1134  * reason it returns an 'int' and not 'void' is so that it can be plugged
1135  * directly into file_operations structure.
1136  */
1137 int nonseekable_open(struct inode *inode, struct file *filp)
1138 {
1139         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1140         return 0;
1141 }
1142
1143 EXPORT_SYMBOL(nonseekable_open);