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 <linux/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 #define CREATE_TRACE_POINTS
  38 #include <trace/events/fs.h>
  39
  40 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  41         struct file *filp)
  42 {
  43         int ret;
  44         struct iattr newattrs;
  45
  46         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  47         if (length < 0)
  48                 return -EINVAL;
  49
  50         newattrs.ia_size = length;
  51         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  52         if (filp) {
  53                 newattrs.ia_file = filp;
  54                 newattrs.ia_valid |= ATTR_FILE;
  55         }
  56
  57         /* Remove suid, sgid, and file capabilities on truncate too */
  58         ret = dentry_needs_remove_privs(dentry);
  59         if (ret < 0)
  60                 return ret;
  61         if (ret)
  62                 newattrs.ia_valid |= ret | ATTR_FORCE;
  63
  64         inode_lock(dentry->d_inode);
  65         /* Note any delegations or leases have already been broken: */
  66         ret = notify_change(dentry, &newattrs, NULL);
  67         inode_unlock(dentry->d_inode);
  68         return ret;
  69 }
  70 EXPORT_SYMBOL_GPL(do_truncate);
  71
  72 long vfs_truncate(const struct path *path, loff_t length)
  73 {
  74         struct inode *inode;
  75         struct dentry *upperdentry;
  76         long error;
  77
  78         inode = path->dentry->d_inode;
  79
  80         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  81         if (S_ISDIR(inode->i_mode))
  82                 return -EISDIR;
  83         if (!S_ISREG(inode->i_mode))
  84                 return -EINVAL;
  85
  86         error = mnt_want_write(path->mnt);
  87         if (error)
  88                 goto out;
  89
  90         error = inode_permission(inode, MAY_WRITE);
  91         if (error)
  92                 goto mnt_drop_write_and_out;
  93
  94         error = -EPERM;
  95         if (IS_APPEND(inode))
  96                 goto mnt_drop_write_and_out;
  97
  98         /*
  99          * If this is an overlayfs then do as if opening the file so we get
 100          * write access on the upper inode, not on the overlay inode.  For
 101          * non-overlay filesystems d_real() is an identity function.
 102          */
 103         upperdentry = d_real(path->dentry, NULL, O_WRONLY, 0);
 104         error = PTR_ERR(upperdentry);
 105         if (IS_ERR(upperdentry))
 106                 goto mnt_drop_write_and_out;
 107
 108         error = get_write_access(upperdentry->d_inode);
 109         if (error)
 110                 goto mnt_drop_write_and_out;
 111
 112         /*
 113          * Make sure that there are no leases.  get_write_access() protects
 114          * against the truncate racing with a lease-granting setlease().
 115          */
 116         error = break_lease(inode, O_WRONLY);
 117         if (error)
 118                 goto put_write_and_out;
 119
 120         error = locks_verify_truncate(inode, NULL, length);
 121         if (!error)
 122                 error = security_path_truncate(path);
 123         if (!error)
 124                 error = do_truncate(path->dentry, length, 0, NULL);
 125
 126 put_write_and_out:
 127         put_write_access(upperdentry->d_inode);
 128 mnt_drop_write_and_out:
 129         mnt_drop_write(path->mnt);
 130 out:
 131         return error;
 132 }
 133 EXPORT_SYMBOL_GPL(vfs_truncate);
 134
 135 static long do_sys_truncate(const char __user *pathname, loff_t length)
 136 {
 137         unsigned int lookup_flags = LOOKUP_FOLLOW;
 138         struct path path;
 139         int error;
 140
 141         if (length < 0) /* sorry, but loff_t says... */
 142                 return -EINVAL;
 143
 144 retry:
 145         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 146         if (!error) {
 147                 error = vfs_truncate(&path, length);
 148                 path_put(&path);
 149         }
 150         if (retry_estale(error, lookup_flags)) {
 151                 lookup_flags |= LOOKUP_REVAL;
 152                 goto retry;
 153         }
 154         return error;
 155 }
 156
 157 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 158 {
 159         return do_sys_truncate(path, length);
 160 }
 161
 162 #ifdef CONFIG_COMPAT
 163 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 164 {
 165         return do_sys_truncate(path, length);
 166 }
 167 #endif
 168
 169 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 170 {
 171         struct inode *inode;
 172         struct dentry *dentry;
 173         struct fd f;
 174         int error;
 175
 176         error = -EINVAL;
 177         if (length < 0)
 178                 goto out;
 179         error = -EBADF;
 180         f = fdget(fd);
 181         if (!f.file)
 182                 goto out;
 183
 184         /* explicitly opened as large or we are on 64-bit box */
 185         if (f.file->f_flags & O_LARGEFILE)
 186                 small = 0;
 187
 188         dentry = f.file->f_path.dentry;
 189         inode = dentry->d_inode;
 190         error = -EINVAL;
 191         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 192                 goto out_putf;
 193
 194         error = -EINVAL;
 195         /* Cannot ftruncate over 2^31 bytes without large file support */
 196         if (small && length > MAX_NON_LFS)
 197                 goto out_putf;
 198
 199         error = -EPERM;
 200         /* Check IS_APPEND on real upper inode */
 201         if (IS_APPEND(file_inode(f.file)))
 202                 goto out_putf;
 203
 204         sb_start_write(inode->i_sb);
 205         error = locks_verify_truncate(inode, f.file, length);
 206         if (!error)
 207                 error = security_path_truncate(&f.file->f_path);
 208         if (!error)
 209                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 210         sb_end_write(inode->i_sb);
 211 out_putf:
 212         fdput(f);
 213 out:
 214         return error;
 215 }
 216
 217 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 218 {
 219         return do_sys_ftruncate(fd, length, 1);
 220 }
 221
 222 #ifdef CONFIG_COMPAT
 223 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 224 {
 225         return do_sys_ftruncate(fd, length, 1);
 226 }
 227 #endif
 228
 229 /* LFS versions of truncate are only needed on 32 bit machines */
 230 #if BITS_PER_LONG == 32
 231 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 232 {
 233         return do_sys_truncate(path, length);
 234 }
 235
 236 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 237 {
 238         return do_sys_ftruncate(fd, length, 0);
 239 }
 240 #endif /* BITS_PER_LONG == 32 */
 241
 242
 243 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 244 {
 245         struct inode *inode = file_inode(file);
 246         long ret;
 247
 248         if (offset < 0 || len <= 0)
 249                 return -EINVAL;
 250
 251         /* Return error if mode is not supported */
 252         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 253                 return -EOPNOTSUPP;
 254
 255         /* Punch hole and zero range are mutually exclusive */
 256         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 257             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 258                 return -EOPNOTSUPP;
 259
 260         /* Punch hole must have keep size set */
 261         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 262             !(mode & FALLOC_FL_KEEP_SIZE))
 263                 return -EOPNOTSUPP;
 264
 265         /* Collapse range should only be used exclusively. */
 266         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 267             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 268                 return -EINVAL;
 269
 270         /* Insert range should only be used exclusively. */
 271         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 272             (mode & ~FALLOC_FL_INSERT_RANGE))
 273                 return -EINVAL;
 274
 275         /* Unshare range should only be used with allocate mode. */
 276         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 277             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 278                 return -EINVAL;
 279
 280         if (!(file->f_mode & FMODE_WRITE))
 281                 return -EBADF;
 282
 283         /*
 284          * We can only allow pure fallocate on append only files
 285          */
 286         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 287                 return -EPERM;
 288
 289         if (IS_IMMUTABLE(inode))
 290                 return -EPERM;
 291
 292         /*
 293          * We cannot allow any fallocate operation on an active swapfile
 294          */
 295         if (IS_SWAPFILE(inode))
 296                 return -ETXTBSY;
 297
 298         /*
 299          * Revalidate the write permissions, in case security policy has
 300          * changed since the files were opened.
 301          */
 302         ret = security_file_permission(file, MAY_WRITE);
 303         if (ret)
 304                 return ret;
 305
 306         if (S_ISFIFO(inode->i_mode))
 307                 return -ESPIPE;
 308
 309         if (S_ISDIR(inode->i_mode))
 310                 return -EISDIR;
 311
 312         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 313                 return -ENODEV;
 314
 315         /* Check for wrap through zero too */
 316         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 317                 return -EFBIG;
 318
 319         if (!file->f_op->fallocate)
 320                 return -EOPNOTSUPP;
 321
 322         file_start_write(file);
 323         ret = file->f_op->fallocate(file, mode, offset, len);
 324
 325         /*
 326          * Create inotify and fanotify events.
 327          *
 328          * To keep the logic simple always create events if fallocate succeeds.
 329          * This implies that events are even created if the file size remains
 330          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 331          */
 332         if (ret == 0)
 333                 fsnotify_modify(file);
 334
 335         file_end_write(file);
 336         return ret;
 337 }
 338 EXPORT_SYMBOL_GPL(vfs_fallocate);
 339
 340 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 341 {
 342         struct fd f = fdget(fd);
 343         int error = -EBADF;
 344
 345         if (f.file) {
 346                 error = vfs_fallocate(f.file, mode, offset, len);
 347                 fdput(f);
 348         }
 349         return error;
 350 }
 351
 352 /*
 353  * access() needs to use the real uid/gid, not the effective uid/gid.
 354  * We do this by temporarily clearing all FS-related capabilities and
 355  * switching the fsuid/fsgid around to the real ones.
 356  */
 357 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 358 {
 359         const struct cred *old_cred;
 360         struct cred *override_cred;
 361         struct path path;
 362         struct inode *inode;
 363         int res;
 364         unsigned int lookup_flags = LOOKUP_FOLLOW;
 365
 366         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 367                 return -EINVAL;
 368
 369         override_cred = prepare_creds();
 370         if (!override_cred)
 371                 return -ENOMEM;
 372
 373         override_cred->fsuid = override_cred->uid;
 374         override_cred->fsgid = override_cred->gid;
 375
 376         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 377                 /* Clear the capabilities if we switch to a non-root user */
 378                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 379                 if (!uid_eq(override_cred->uid, root_uid))
 380                         cap_clear(override_cred->cap_effective);
 381                 else
 382                         override_cred->cap_effective =
 383                                 override_cred->cap_permitted;
 384         }
 385
 386         /*
 387          * The new set of credentials can *only* be used in
 388          * task-synchronous circumstances, and does not need
 389          * RCU freeing, unless somebody then takes a separate
 390          * reference to it.
 391          *
 392          * NOTE! This is _only_ true because this credential
 393          * is used purely for override_creds() that installs
 394          * it as the subjective cred. Other threads will be
 395          * accessing ->real_cred, not the subjective cred.
 396          *
 397          * If somebody _does_ make a copy of this (using the
 398          * 'get_current_cred()' function), that will clear the
 399          * non_rcu field, because now that other user may be
 400          * expecting RCU freeing. But normal thread-synchronous
 401          * cred accesses will keep things non-RCY.
 402          */
 403         override_cred->non_rcu = 1;
 404
 405         old_cred = override_creds(override_cred);
 406 retry:
 407         res = user_path_at(dfd, filename, lookup_flags, &path);
 408         if (res)
 409                 goto out;
 410
 411         inode = d_backing_inode(path.dentry);
 412
 413         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 414                 /*
 415                  * MAY_EXEC on regular files is denied if the fs is mounted
 416                  * with the "noexec" flag.
 417                  */
 418                 res = -EACCES;
 419                 if (path_noexec(&path))
 420                         goto out_path_release;
 421         }
 422
 423         res = inode_permission(inode, mode | MAY_ACCESS);
 424         /* SuS v2 requires we report a read only fs too */
 425         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 426                 goto out_path_release;
 427         /*
 428          * This is a rare case where using __mnt_is_readonly()
 429          * is OK without a mnt_want/drop_write() pair.  Since
 430          * no actual write to the fs is performed here, we do
 431          * not need to telegraph to that to anyone.
 432          *
 433          * By doing this, we accept that this access is
 434          * inherently racy and know that the fs may change
 435          * state before we even see this result.
 436          */
 437         if (__mnt_is_readonly(path.mnt))
 438                 res = -EROFS;
 439
 440 out_path_release:
 441         path_put(&path);
 442         if (retry_estale(res, lookup_flags)) {
 443                 lookup_flags |= LOOKUP_REVAL;
 444                 goto retry;
 445         }
 446 out:
 447         revert_creds(old_cred);
 448         put_cred(override_cred);
 449         return res;
 450 }
 451
 452 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 453 {
 454         return sys_faccessat(AT_FDCWD, filename, mode);
 455 }
 456
 457 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 458 {
 459         struct path path;
 460         int error;
 461         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 462 retry:
 463         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 464         if (error)
 465                 goto out;
 466
 467         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 468         if (error)
 469                 goto dput_and_out;
 470
 471         set_fs_pwd(current->fs, &path);
 472
 473 dput_and_out:
 474         path_put(&path);
 475         if (retry_estale(error, lookup_flags)) {
 476                 lookup_flags |= LOOKUP_REVAL;
 477                 goto retry;
 478         }
 479 out:
 480         return error;
 481 }
 482
 483 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 484 {
 485         struct fd f = fdget_raw(fd);
 486         int error;
 487
 488         error = -EBADF;
 489         if (!f.file)
 490                 goto out;
 491
 492         error = -ENOTDIR;
 493         if (!d_can_lookup(f.file->f_path.dentry))
 494                 goto out_putf;
 495
 496         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 497         if (!error)
 498                 set_fs_pwd(current->fs, &f.file->f_path);
 499 out_putf:
 500         fdput(f);
 501 out:
 502         return error;
 503 }
 504
 505 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 506 {
 507         struct path path;
 508         int error;
 509         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 510 retry:
 511         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 512         if (error)
 513                 goto out;
 514
 515         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 516         if (error)
 517                 goto dput_and_out;
 518
 519         error = -EPERM;
 520         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 521                 goto dput_and_out;
 522         error = security_path_chroot(&path);
 523         if (error)
 524                 goto dput_and_out;
 525
 526         set_fs_root(current->fs, &path);
 527         error = 0;
 528 dput_and_out:
 529         path_put(&path);
 530         if (retry_estale(error, lookup_flags)) {
 531                 lookup_flags |= LOOKUP_REVAL;
 532                 goto retry;
 533         }
 534 out:
 535         return error;
 536 }
 537
 538 static int chmod_common(const struct path *path, umode_t mode)
 539 {
 540         struct inode *inode = path->dentry->d_inode;
 541         struct inode *delegated_inode = NULL;
 542         struct iattr newattrs;
 543         int error;
 544
 545         error = mnt_want_write(path->mnt);
 546         if (error)
 547                 return error;
 548 retry_deleg:
 549         inode_lock(inode);
 550         error = security_path_chmod(path, mode);
 551         if (error)
 552                 goto out_unlock;
 553         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 554         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 555         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 556 out_unlock:
 557         inode_unlock(inode);
 558         if (delegated_inode) {
 559                 error = break_deleg_wait(&delegated_inode);
 560                 if (!error)
 561                         goto retry_deleg;
 562         }
 563         mnt_drop_write(path->mnt);
 564         return error;
 565 }
 566
 567 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 568 {
 569         struct fd f = fdget(fd);
 570         int err = -EBADF;
 571
 572         if (f.file) {
 573                 audit_file(f.file);
 574                 err = chmod_common(&f.file->f_path, mode);
 575                 fdput(f);
 576         }
 577         return err;
 578 }
 579
 580 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 581 {
 582         struct path path;
 583         int error;
 584         unsigned int lookup_flags = LOOKUP_FOLLOW;
 585 retry:
 586         error = user_path_at(dfd, filename, lookup_flags, &path);
 587         if (!error) {
 588                 error = chmod_common(&path, mode);
 589                 path_put(&path);
 590                 if (retry_estale(error, lookup_flags)) {
 591                         lookup_flags |= LOOKUP_REVAL;
 592                         goto retry;
 593                 }
 594         }
 595         return error;
 596 }
 597
 598 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 599 {
 600         return sys_fchmodat(AT_FDCWD, filename, mode);
 601 }
 602
 603 static int chown_common(const struct path *path, uid_t user, gid_t group)
 604 {
 605         struct inode *inode = path->dentry->d_inode;
 606         struct inode *delegated_inode = NULL;
 607         int error;
 608         struct iattr newattrs;
 609         kuid_t uid;
 610         kgid_t gid;
 611
 612         uid = make_kuid(current_user_ns(), user);
 613         gid = make_kgid(current_user_ns(), group);
 614
 615 retry_deleg:
 616         newattrs.ia_valid =  ATTR_CTIME;
 617         if (user != (uid_t) -1) {
 618                 if (!uid_valid(uid))
 619                         return -EINVAL;
 620                 newattrs.ia_valid |= ATTR_UID;
 621                 newattrs.ia_uid = uid;
 622         }
 623         if (group != (gid_t) -1) {
 624                 if (!gid_valid(gid))
 625                         return -EINVAL;
 626                 newattrs.ia_valid |= ATTR_GID;
 627                 newattrs.ia_gid = gid;
 628         }
 629         if (!S_ISDIR(inode->i_mode))
 630                 newattrs.ia_valid |=
 631                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 632         inode_lock(inode);
 633         error = security_path_chown(path, uid, gid);
 634         if (!error)
 635                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 636         inode_unlock(inode);
 637         if (delegated_inode) {
 638                 error = break_deleg_wait(&delegated_inode);
 639                 if (!error)
 640                         goto retry_deleg;
 641         }
 642         return error;
 643 }
 644
 645 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 646                 gid_t, group, int, flag)
 647 {
 648         struct path path;
 649         int error = -EINVAL;
 650         int lookup_flags;
 651
 652         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 653                 goto out;
 654
 655         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 656         if (flag & AT_EMPTY_PATH)
 657                 lookup_flags |= LOOKUP_EMPTY;
 658 retry:
 659         error = user_path_at(dfd, filename, lookup_flags, &path);
 660         if (error)
 661                 goto out;
 662         error = mnt_want_write(path.mnt);
 663         if (error)
 664                 goto out_release;
 665         error = chown_common(&path, user, group);
 666         mnt_drop_write(path.mnt);
 667 out_release:
 668         path_put(&path);
 669         if (retry_estale(error, lookup_flags)) {
 670                 lookup_flags |= LOOKUP_REVAL;
 671                 goto retry;
 672         }
 673 out:
 674         return error;
 675 }
 676
 677 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 678 {
 679         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 680 }
 681
 682 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 683 {
 684         return sys_fchownat(AT_FDCWD, filename, user, group,
 685                             AT_SYMLINK_NOFOLLOW);
 686 }
 687
 688 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 689 {
 690         struct fd f = fdget(fd);
 691         int error = -EBADF;
 692
 693         if (!f.file)
 694                 goto out;
 695
 696         error = mnt_want_write_file_path(f.file);
 697         if (error)
 698                 goto out_fput;
 699         audit_file(f.file);
 700         error = chown_common(&f.file->f_path, user, group);
 701         mnt_drop_write_file_path(f.file);
 702 out_fput:
 703         fdput(f);
 704 out:
 705         return error;
 706 }
 707
 708 int open_check_o_direct(struct file *f)
 709 {
 710         /* NB: we're sure to have correct a_ops only after f_op->open */
 711         if (f->f_flags & O_DIRECT) {
 712                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 713                         return -EINVAL;
 714         }
 715         return 0;
 716 }
 717 EXPORT_SYMBOL_GPL(open_check_o_direct);
 718
 719 static int do_dentry_open(struct file *f,
 720                           struct inode *inode,
 721                           int (*open)(struct inode *, struct file *),
 722                           const struct cred *cred)
 723 {
 724         static const struct file_operations empty_fops = {};
 725         int error;
 726
 727         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 728                                 FMODE_PREAD | FMODE_PWRITE;
 729
 730         path_get(&f->f_path);
 731         f->f_inode = inode;
 732         f->f_mapping = inode->i_mapping;
 733
 734         /* Ensure that we skip any errors that predate opening of the file */
 735         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 736
 737         if (unlikely(f->f_flags & O_PATH)) {
 738                 f->f_mode = FMODE_PATH;
 739                 f->f_op = &empty_fops;
 740                 return 0;
 741         }
 742
 743         /* Any file opened for execve()/uselib() has to be a regular file. */
 744         if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 745                 error = -EACCES;
 746                 goto cleanup_file;
 747         }
 748
 749         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 750                 error = get_write_access(inode);
 751                 if (unlikely(error))
 752                         goto cleanup_file;
 753                 error = __mnt_want_write(f->f_path.mnt);
 754                 if (unlikely(error)) {
 755                         put_write_access(inode);
 756                         goto cleanup_file;
 757                 }
 758                 f->f_mode |= FMODE_WRITER;
 759         }
 760
 761         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 762         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 763                 f->f_mode |= FMODE_ATOMIC_POS;
 764
 765         f->f_op = fops_get(inode->i_fop);
 766         if (unlikely(WARN_ON(!f->f_op))) {
 767                 error = -ENODEV;
 768                 goto cleanup_all;
 769         }
 770
 771         error = security_file_open(f, cred);
 772         if (error)
 773                 goto cleanup_all;
 774
 775         error = break_lease(locks_inode(f), f->f_flags);
 776         if (error)
 777                 goto cleanup_all;
 778
 779         if (!open)
 780                 open = f->f_op->open;
 781         if (open) {
 782                 error = open(inode, f);
 783                 if (error)
 784                         goto cleanup_all;
 785         }
 786         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 787                 i_readcount_inc(inode);
 788         if ((f->f_mode & FMODE_READ) &&
 789              likely(f->f_op->read || f->f_op->read_iter))
 790                 f->f_mode |= FMODE_CAN_READ;
 791         if ((f->f_mode & FMODE_WRITE) &&
 792              likely(f->f_op->write || f->f_op->write_iter))
 793                 f->f_mode |= FMODE_CAN_WRITE;
 794
 795         f->f_write_hint = WRITE_LIFE_NOT_SET;
 796         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 797
 798         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 799
 800         return 0;
 801
 802 cleanup_all:
 803         fops_put(f->f_op);
 804         if (f->f_mode & FMODE_WRITER) {
 805                 put_write_access(inode);
 806                 __mnt_drop_write(f->f_path.mnt);
 807         }
 808 cleanup_file:
 809         path_put(&f->f_path);
 810         f->f_path.mnt = NULL;
 811         f->f_path.dentry = NULL;
 812         f->f_inode = NULL;
 813         return error;
 814 }
 815
 816 /**
 817  * finish_open - finish opening a file
 818  * @file: file pointer
 819  * @dentry: pointer to dentry
 820  * @open: open callback
 821  * @opened: state of open
 822  *
 823  * This can be used to finish opening a file passed to i_op->atomic_open().
 824  *
 825  * If the open callback is set to NULL, then the standard f_op->open()
 826  * filesystem callback is substituted.
 827  *
 828  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 829  * the return value of d_splice_alias(), then the caller needs to perform dput()
 830  * on it after finish_open().
 831  *
 832  * On successful return @file is a fully instantiated open file.  After this, if
 833  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 834  *
 835  * Returns zero on success or -errno if the open failed.
 836  */
 837 int finish_open(struct file *file, struct dentry *dentry,
 838                 int (*open)(struct inode *, struct file *),
 839                 int *opened)
 840 {
 841         int error;
 842         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 843
 844         file->f_path.dentry = dentry;
 845         error = do_dentry_open(file, d_backing_inode(dentry), open,
 846                                current_cred());
 847         if (!error)
 848                 *opened |= FILE_OPENED;
 849
 850         return error;
 851 }
 852 EXPORT_SYMBOL(finish_open);
 853
 854 /**
 855  * finish_no_open - finish ->atomic_open() without opening the file
 856  *
 857  * @file: file pointer
 858  * @dentry: dentry or NULL (as returned from ->lookup())
 859  *
 860  * This can be used to set the result of a successful lookup in ->atomic_open().
 861  *
 862  * NB: unlike finish_open() this function does consume the dentry reference and
 863  * the caller need not dput() it.
 864  *
 865  * Returns "1" which must be the return value of ->atomic_open() after having
 866  * called this function.
 867  */
 868 int finish_no_open(struct file *file, struct dentry *dentry)
 869 {
 870         file->f_path.dentry = dentry;
 871         return 1;
 872 }
 873 EXPORT_SYMBOL(finish_no_open);
 874
 875 char *file_path(struct file *filp, char *buf, int buflen)
 876 {
 877         return d_path(&filp->f_path, buf, buflen);
 878 }
 879 EXPORT_SYMBOL(file_path);
 880
 881 /**
 882  * vfs_open - open the file at the given path
 883  * @path: path to open
 884  * @file: newly allocated file with f_flag initialized
 885  * @cred: credentials to use
 886  */
 887 int vfs_open(const struct path *path, struct file *file,
 888              const struct cred *cred)
 889 {
 890         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);
 891
 892         if (IS_ERR(dentry))
 893                 return PTR_ERR(dentry);
 894
 895         file->f_path = *path;
 896         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 897 }
 898
 899 struct file *dentry_open(const struct path *path, int flags,
 900                          const struct cred *cred)
 901 {
 902         int error;
 903         struct file *f;
 904
 905         validate_creds(cred);
 906
 907         /* We must always pass in a valid mount pointer. */
 908         BUG_ON(!path->mnt);
 909
 910         f = get_empty_filp();
 911         if (!IS_ERR(f)) {
 912                 f->f_flags = flags;
 913                 error = vfs_open(path, f, cred);
 914                 if (!error) {
 915                         /* from now on we need fput() to dispose of f */
 916                         error = open_check_o_direct(f);
 917                         if (error) {
 918                                 fput(f);
 919                                 f = ERR_PTR(error);
 920                         }
 921                 } else {
 922                         put_filp(f);
 923                         f = ERR_PTR(error);
 924                 }
 925         }
 926         return f;
 927 }
 928 EXPORT_SYMBOL(dentry_open);
 929
 930 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 931 {
 932         int lookup_flags = 0;
 933         int acc_mode = ACC_MODE(flags);
 934
 935         /*
 936          * Clear out all open flags we don't know about so that we don't report
 937          * them in fcntl(F_GETFD) or similar interfaces.
 938          */
 939         flags &= VALID_OPEN_FLAGS;
 940
 941         if (flags & (O_CREAT | __O_TMPFILE))
 942                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 943         else
 944                 op->mode = 0;
 945
 946         /* Must never be set by userspace */
 947         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 948
 949         /*
 950          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 951          * check for O_DSYNC if the need any syncing at all we enforce it's
 952          * always set instead of having to deal with possibly weird behaviour
 953          * for malicious applications setting only __O_SYNC.
 954          */
 955         if (flags & __O_SYNC)
 956                 flags |= O_DSYNC;
 957
 958         if (flags & __O_TMPFILE) {
 959                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 960                         return -EINVAL;
 961                 if (!(acc_mode & MAY_WRITE))
 962                         return -EINVAL;
 963         } else if (flags & O_PATH) {
 964                 /*
 965                  * If we have O_PATH in the open flag. Then we
 966                  * cannot have anything other than the below set of flags
 967                  */
 968                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 969                 acc_mode = 0;
 970         }
 971
 972         op->open_flag = flags;
 973
 974         /* O_TRUNC implies we need access checks for write permissions */
 975         if (flags & O_TRUNC)
 976                 acc_mode |= MAY_WRITE;
 977
 978         /* Allow the LSM permission hook to distinguish append
 979            access from general write access. */
 980         if (flags & O_APPEND)
 981                 acc_mode |= MAY_APPEND;
 982
 983         op->acc_mode = acc_mode;
 984
 985         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 986
 987         if (flags & O_CREAT) {
 988                 op->intent |= LOOKUP_CREATE;
 989                 if (flags & O_EXCL)
 990                         op->intent |= LOOKUP_EXCL;
 991         }
 992
 993         if (flags & O_DIRECTORY)
 994                 lookup_flags |= LOOKUP_DIRECTORY;
 995         if (!(flags & O_NOFOLLOW))
 996                 lookup_flags |= LOOKUP_FOLLOW;
 997         op->lookup_flags = lookup_flags;
 998         return 0;
 999 }
1000
1001 /**
1002  * file_open_name - open file and return file pointer
1003  *
1004  * @name:       struct filename containing path to open
1005  * @flags:      open flags as per the open(2) second argument
1006  * @mode:       mode for the new file if O_CREAT is set, else ignored
1007  *
1008  * This is the helper to open a file from kernelspace if you really
1009  * have to.  But in generally you should not do this, so please move
1010  * along, nothing to see here..
1011  */
1012 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1013 {
1014         struct open_flags op;
1015         int err = build_open_flags(flags, mode, &op);
1016         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1017 }
1018
1019 /**
1020  * filp_open - open file and return file pointer
1021  *
1022  * @filename:   path to open
1023  * @flags:      open flags as per the open(2) second argument
1024  * @mode:       mode for the new file if O_CREAT is set, else ignored
1025  *
1026  * This is the helper to open a file from kernelspace if you really
1027  * have to.  But in generally you should not do this, so please move
1028  * along, nothing to see here..
1029  */
1030 struct file *filp_open(const char *filename, int flags, umode_t mode)
1031 {
1032         struct filename *name = getname_kernel(filename);
1033         struct file *file = ERR_CAST(name);
1034
1035         if (!IS_ERR(name)) {
1036                 file = file_open_name(name, flags, mode);
1037                 putname(name);
1038         }
1039         return file;
1040 }
1041 EXPORT_SYMBOL(filp_open);
1042
1043 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1044                             const char *filename, int flags, umode_t mode)
1045 {
1046         struct open_flags op;
1047         int err = build_open_flags(flags, mode, &op);
1048         if (err)
1049                 return ERR_PTR(err);
1050         return do_file_open_root(dentry, mnt, filename, &op);
1051 }
1052 EXPORT_SYMBOL(file_open_root);
1053
1054 struct file *filp_clone_open(struct file *oldfile)
1055 {
1056         struct file *file;
1057         int retval;
1058
1059         file = get_empty_filp();
1060         if (IS_ERR(file))
1061                 return file;
1062
1063         file->f_flags = oldfile->f_flags;
1064         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1065         if (retval) {
1066                 put_filp(file);
1067                 return ERR_PTR(retval);
1068         }
1069
1070         return file;
1071 }
1072 EXPORT_SYMBOL(filp_clone_open);
1073
1074 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1075 {
1076         struct open_flags op;
1077         int fd = build_open_flags(flags, mode, &op);
1078         struct filename *tmp;
1079
1080         if (fd)
1081                 return fd;
1082
1083         tmp = getname(filename);
1084         if (IS_ERR(tmp))
1085                 return PTR_ERR(tmp);
1086
1087         fd = get_unused_fd_flags(flags);
1088         if (fd >= 0) {
1089                 struct file *f = do_filp_open(dfd, tmp, &op);
1090                 if (IS_ERR(f)) {
1091                         put_unused_fd(fd);
1092                         fd = PTR_ERR(f);
1093                 } else {
1094                         fsnotify_open(f);
1095                         fd_install(fd, f);
1096                         trace_do_sys_open(tmp->name, flags, mode);
1097                 }
1098         }
1099         putname(tmp);
1100         return fd;
1101 }
1102
1103 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1104 {
1105         if (force_o_largefile())
1106                 flags |= O_LARGEFILE;
1107
1108         return do_sys_open(AT_FDCWD, filename, flags, mode);
1109 }
1110
1111 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1112                 umode_t, mode)
1113 {
1114         if (force_o_largefile())
1115                 flags |= O_LARGEFILE;
1116
1117         return do_sys_open(dfd, filename, flags, mode);
1118 }
1119
1120 #ifdef CONFIG_COMPAT
1121 /*
1122  * Exactly like sys_open(), except that it doesn't set the
1123  * O_LARGEFILE flag.
1124  */
1125 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1126 {
1127         return do_sys_open(AT_FDCWD, filename, flags, mode);
1128 }
1129
1130 /*
1131  * Exactly like sys_openat(), except that it doesn't set the
1132  * O_LARGEFILE flag.
1133  */
1134 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1135 {
1136         return do_sys_open(dfd, filename, flags, mode);
1137 }
1138 #endif
1139
1140 #ifndef __alpha__
1141
1142 /*
1143  * For backward compatibility?  Maybe this should be moved
1144  * into arch/i386 instead?
1145  */
1146 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1147 {
1148         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1149 }
1150
1151 #endif
1152
1153 /*
1154  * "id" is the POSIX thread ID. We use the
1155  * files pointer for this..
1156  */
1157 int filp_close(struct file *filp, fl_owner_t id)
1158 {
1159         int retval = 0;
1160
1161         if (!file_count(filp)) {
1162                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1163                 return 0;
1164         }
1165
1166         if (filp->f_op->flush)
1167                 retval = filp->f_op->flush(filp, id);
1168
1169         if (likely(!(filp->f_mode & FMODE_PATH))) {
1170                 dnotify_flush(filp, id);
1171                 locks_remove_posix(filp, id);
1172         }
1173         fput(filp);
1174         return retval;
1175 }
1176
1177 EXPORT_SYMBOL(filp_close);
1178
1179 /*
1180  * Careful here! We test whether the file pointer is NULL before
1181  * releasing the fd. This ensures that one clone task can't release
1182  * an fd while another clone is opening it.
1183  */
1184 SYSCALL_DEFINE1(close, unsigned int, fd)
1185 {
1186         int retval = __close_fd(current->files, fd);
1187
1188         /* can't restart close syscall because file table entry was cleared */
1189         if (unlikely(retval == -ERESTARTSYS ||
1190                      retval == -ERESTARTNOINTR ||
1191                      retval == -ERESTARTNOHAND ||
1192                      retval == -ERESTART_RESTARTBLOCK))
1193                 retval = -EINTR;
1194
1195         return retval;
1196 }
1197 EXPORT_SYMBOL(sys_close);
1198
1199 /*
1200  * This routine simulates a hangup on the tty, to arrange that users
1201  * are given clean terminals at login time.
1202  */
1203 SYSCALL_DEFINE0(vhangup)
1204 {
1205         if (capable(CAP_SYS_TTY_CONFIG)) {
1206                 tty_vhangup_self();
1207                 return 0;
1208         }
1209         return -EPERM;
1210 }
1211
1212 /*
1213  * Called when an inode is about to be open.
1214  * We use this to disallow opening large files on 32bit systems if
1215  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1216  * on this flag in sys_open.
1217  */
1218 int generic_file_open(struct inode * inode, struct file * filp)
1219 {
1220         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1221                 return -EOVERFLOW;
1222         return 0;
1223 }
1224
1225 EXPORT_SYMBOL(generic_file_open);
1226
1227 /*
1228  * This is used by subsystems that don't want seekable
1229  * file descriptors. The function is not supposed to ever fail, the only
1230  * reason it returns an 'int' and not 'void' is so that it can be plugged
1231  * directly into file_operations structure.
1232  */
1233 int nonseekable_open(struct inode *inode, struct file *filp)
1234 {
1235         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1236         return 0;
1237 }
1238
1239 EXPORT_SYMBOL(nonseekable_open);
1240
1241 /*
1242  * stream_open is used by subsystems that want stream-like file descriptors.
1243  * Such file descriptors are not seekable and don't have notion of position
1244  * (file.f_pos is always 0). Contrary to file descriptors of other regular
1245  * files, .read() and .write() can run simultaneously.
1246  *
1247  * stream_open never fails and is marked to return int so that it could be
1248  * directly used as file_operations.open .
1249  */
1250 int stream_open(struct inode *inode, struct file *filp)
1251 {
1252         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1253         filp->f_mode |= FMODE_STREAM;
1254         return 0;
1255 }
1256
1257 EXPORT_SYMBOL(stream_open);