fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 #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         mutex_lock(&dentry->d_inode->i_mutex);
  65         /* Note any delegations or leases have already been broken: */
  66         ret = notify_change(dentry, &newattrs, NULL);
  67         mutex_unlock(&dentry->d_inode->i_mutex);
  68         return ret;
  69 }
  70 EXPORT_SYMBOL(do_truncate);
  71
  72 long vfs_truncate(struct path *path, loff_t length)
  73 {
  74         struct inode *inode;
  75         long error;
  76
  77         inode = path->dentry->d_inode;
  78
  79         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  80         if (S_ISDIR(inode->i_mode))
  81                 return -EISDIR;
  82         if (!S_ISREG(inode->i_mode))
  83                 return -EINVAL;
  84
  85         error = mnt_want_write(path->mnt);
  86         if (error)
  87                 goto out;
  88
  89         error = inode_permission(inode, MAY_WRITE);
  90         if (error)
  91                 goto mnt_drop_write_and_out;
  92
  93         error = -EPERM;
  94         if (IS_APPEND(inode))
  95                 goto mnt_drop_write_and_out;
  96
  97         error = get_write_access(inode);
  98         if (error)
  99                 goto mnt_drop_write_and_out;
 100
 101         /*
 102          * Make sure that there are no leases.  get_write_access() protects
 103          * against the truncate racing with a lease-granting setlease().
 104          */
 105         error = break_lease(inode, O_WRONLY);
 106         if (error)
 107                 goto put_write_and_out;
 108
 109         error = locks_verify_truncate(inode, NULL, length);
 110         if (!error)
 111                 error = security_path_truncate(path);
 112         if (!error)
 113                 error = do_truncate(path->dentry, length, 0, NULL);
 114
 115 put_write_and_out:
 116         put_write_access(inode);
 117 mnt_drop_write_and_out:
 118         mnt_drop_write(path->mnt);
 119 out:
 120         return error;
 121 }
 122 EXPORT_SYMBOL_GPL(vfs_truncate);
 123
 124 static long do_sys_truncate(const char __user *pathname, loff_t length)
 125 {
 126         unsigned int lookup_flags = LOOKUP_FOLLOW;
 127         struct path path;
 128         int error;
 129
 130         if (length < 0) /* sorry, but loff_t says... */
 131                 return -EINVAL;
 132
 133 retry:
 134         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 135         if (!error) {
 136                 error = vfs_truncate(&path, length);
 137                 path_put(&path);
 138         }
 139         if (retry_estale(error, lookup_flags)) {
 140                 lookup_flags |= LOOKUP_REVAL;
 141                 goto retry;
 142         }
 143         return error;
 144 }
 145
 146 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 147 {
 148         return do_sys_truncate(path, length);
 149 }
 150
 151 #ifdef CONFIG_COMPAT
 152 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 153 {
 154         return do_sys_truncate(path, length);
 155 }
 156 #endif
 157
 158 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 159 {
 160         struct inode *inode;
 161         struct dentry *dentry;
 162         struct fd f;
 163         int error;
 164
 165         error = -EINVAL;
 166         if (length < 0)
 167                 goto out;
 168         error = -EBADF;
 169         f = fdget(fd);
 170         if (!f.file)
 171                 goto out;
 172
 173         /* explicitly opened as large or we are on 64-bit box */
 174         if (f.file->f_flags & O_LARGEFILE)
 175                 small = 0;
 176
 177         dentry = f.file->f_path.dentry;
 178         inode = dentry->d_inode;
 179         error = -EINVAL;
 180         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 181                 goto out_putf;
 182
 183         error = -EINVAL;
 184         /* Cannot ftruncate over 2^31 bytes without large file support */
 185         if (small && length > MAX_NON_LFS)
 186                 goto out_putf;
 187
 188         error = -EPERM;
 189         if (IS_APPEND(inode))
 190                 goto out_putf;
 191
 192         sb_start_write(inode->i_sb);
 193         error = locks_verify_truncate(inode, f.file, length);
 194         if (!error)
 195                 error = security_path_truncate(&f.file->f_path);
 196         if (!error)
 197                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 198         sb_end_write(inode->i_sb);
 199 out_putf:
 200         fdput(f);
 201 out:
 202         return error;
 203 }
 204
 205 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 206 {
 207         return do_sys_ftruncate(fd, length, 1);
 208 }
 209
 210 #ifdef CONFIG_COMPAT
 211 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 212 {
 213         return do_sys_ftruncate(fd, length, 1);
 214 }
 215 #endif
 216
 217 /* LFS versions of truncate are only needed on 32 bit machines */
 218 #if BITS_PER_LONG == 32
 219 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 220 {
 221         return do_sys_truncate(path, length);
 222 }
 223
 224 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 225 {
 226         return do_sys_ftruncate(fd, length, 0);
 227 }
 228 #endif /* BITS_PER_LONG == 32 */
 229
 230
 231 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 232 {
 233         struct inode *inode = file_inode(file);
 234         long ret;
 235
 236         if (offset < 0 || len <= 0)
 237                 return -EINVAL;
 238
 239         /* Return error if mode is not supported */
 240         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 241                 return -EOPNOTSUPP;
 242
 243         /* Punch hole and zero range are mutually exclusive */
 244         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 245             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 246                 return -EOPNOTSUPP;
 247
 248         /* Punch hole must have keep size set */
 249         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 250             !(mode & FALLOC_FL_KEEP_SIZE))
 251                 return -EOPNOTSUPP;
 252
 253         /* Collapse range should only be used exclusively. */
 254         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 255             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 256                 return -EINVAL;
 257
 258         /* Insert range should only be used exclusively. */
 259         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 260             (mode & ~FALLOC_FL_INSERT_RANGE))
 261                 return -EINVAL;
 262
 263         if (!(file->f_mode & FMODE_WRITE))
 264                 return -EBADF;
 265
 266         /*
 267          * We can only allow pure fallocate on append only files
 268          */
 269         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 270                 return -EPERM;
 271
 272         if (IS_IMMUTABLE(inode))
 273                 return -EPERM;
 274
 275         /*
 276          * We cannot allow any fallocate operation on an active swapfile
 277          */
 278         if (IS_SWAPFILE(inode))
 279                 return -ETXTBSY;
 280
 281         /*
 282          * Revalidate the write permissions, in case security policy has
 283          * changed since the files were opened.
 284          */
 285         ret = security_file_permission(file, MAY_WRITE);
 286         if (ret)
 287                 return ret;
 288
 289         if (S_ISFIFO(inode->i_mode))
 290                 return -ESPIPE;
 291
 292         /*
 293          * Let individual file system decide if it supports preallocation
 294          * for directories or not.
 295          */
 296         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 297                 return -ENODEV;
 298
 299         /* Check for wrap through zero too */
 300         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 301                 return -EFBIG;
 302
 303         if (!file->f_op->fallocate)
 304                 return -EOPNOTSUPP;
 305
 306         sb_start_write(inode->i_sb);
 307         ret = file->f_op->fallocate(file, mode, offset, len);
 308
 309         /*
 310          * Create inotify and fanotify events.
 311          *
 312          * To keep the logic simple always create events if fallocate succeeds.
 313          * This implies that events are even created if the file size remains
 314          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 315          */
 316         if (ret == 0)
 317                 fsnotify_modify(file);
 318
 319         sb_end_write(inode->i_sb);
 320         return ret;
 321 }
 322 EXPORT_SYMBOL_GPL(vfs_fallocate);
 323
 324 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 325 {
 326         struct fd f = fdget(fd);
 327         int error = -EBADF;
 328
 329         if (f.file) {
 330                 error = vfs_fallocate(f.file, mode, offset, len);
 331                 fdput(f);
 332         }
 333         return error;
 334 }
 335
 336 /*
 337  * access() needs to use the real uid/gid, not the effective uid/gid.
 338  * We do this by temporarily clearing all FS-related capabilities and
 339  * switching the fsuid/fsgid around to the real ones.
 340  */
 341 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 342 {
 343         const struct cred *old_cred;
 344         struct cred *override_cred;
 345         struct path path;
 346         struct inode *inode;
 347         int res;
 348         unsigned int lookup_flags = LOOKUP_FOLLOW;
 349
 350         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 351                 return -EINVAL;
 352
 353         override_cred = prepare_creds();
 354         if (!override_cred)
 355                 return -ENOMEM;
 356
 357         override_cred->fsuid = override_cred->uid;
 358         override_cred->fsgid = override_cred->gid;
 359
 360         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 361                 /* Clear the capabilities if we switch to a non-root user */
 362                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 363                 if (!uid_eq(override_cred->uid, root_uid))
 364                         cap_clear(override_cred->cap_effective);
 365                 else
 366                         override_cred->cap_effective =
 367                                 override_cred->cap_permitted;
 368         }
 369
 370         old_cred = override_creds(override_cred);
 371 retry:
 372         res = user_path_at(dfd, filename, lookup_flags, &path);
 373         if (res)
 374                 goto out;
 375
 376         inode = d_backing_inode(path.dentry);
 377
 378         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 379                 /*
 380                  * MAY_EXEC on regular files is denied if the fs is mounted
 381                  * with the "noexec" flag.
 382                  */
 383                 res = -EACCES;
 384                 if (path_noexec(&path))
 385                         goto out_path_release;
 386         }
 387
 388         res = inode_permission(inode, mode | MAY_ACCESS);
 389         /* SuS v2 requires we report a read only fs too */
 390         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 391                 goto out_path_release;
 392         /*
 393          * This is a rare case where using __mnt_is_readonly()
 394          * is OK without a mnt_want/drop_write() pair.  Since
 395          * no actual write to the fs is performed here, we do
 396          * not need to telegraph to that to anyone.
 397          *
 398          * By doing this, we accept that this access is
 399          * inherently racy and know that the fs may change
 400          * state before we even see this result.
 401          */
 402         if (__mnt_is_readonly(path.mnt))
 403                 res = -EROFS;
 404
 405 out_path_release:
 406         path_put(&path);
 407         if (retry_estale(res, lookup_flags)) {
 408                 lookup_flags |= LOOKUP_REVAL;
 409                 goto retry;
 410         }
 411 out:
 412         revert_creds(old_cred);
 413         put_cred(override_cred);
 414         return res;
 415 }
 416
 417 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 418 {
 419         return sys_faccessat(AT_FDCWD, filename, mode);
 420 }
 421
 422 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 423 {
 424         struct path path;
 425         int error;
 426         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 427 retry:
 428         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 429         if (error)
 430                 goto out;
 431
 432         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 433         if (error)
 434                 goto dput_and_out;
 435
 436         set_fs_pwd(current->fs, &path);
 437
 438 dput_and_out:
 439         path_put(&path);
 440         if (retry_estale(error, lookup_flags)) {
 441                 lookup_flags |= LOOKUP_REVAL;
 442                 goto retry;
 443         }
 444 out:
 445         return error;
 446 }
 447
 448 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 449 {
 450         struct fd f = fdget_raw(fd);
 451         struct inode *inode;
 452         int error = -EBADF;
 453
 454         error = -EBADF;
 455         if (!f.file)
 456                 goto out;
 457
 458         inode = file_inode(f.file);
 459
 460         error = -ENOTDIR;
 461         if (!S_ISDIR(inode->i_mode))
 462                 goto out_putf;
 463
 464         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 465         if (!error)
 466                 set_fs_pwd(current->fs, &f.file->f_path);
 467 out_putf:
 468         fdput(f);
 469 out:
 470         return error;
 471 }
 472
 473 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 474 {
 475         struct path path;
 476         int error;
 477         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 478 retry:
 479         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 480         if (error)
 481                 goto out;
 482
 483         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 484         if (error)
 485                 goto dput_and_out;
 486
 487         error = -EPERM;
 488         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 489                 goto dput_and_out;
 490         error = security_path_chroot(&path);
 491         if (error)
 492                 goto dput_and_out;
 493
 494         set_fs_root(current->fs, &path);
 495         error = 0;
 496 dput_and_out:
 497         path_put(&path);
 498         if (retry_estale(error, lookup_flags)) {
 499                 lookup_flags |= LOOKUP_REVAL;
 500                 goto retry;
 501         }
 502 out:
 503         return error;
 504 }
 505
 506 static int chmod_common(struct path *path, umode_t mode)
 507 {
 508         struct inode *inode = path->dentry->d_inode;
 509         struct inode *delegated_inode = NULL;
 510         struct iattr newattrs;
 511         int error;
 512
 513         error = mnt_want_write(path->mnt);
 514         if (error)
 515                 return error;
 516 retry_deleg:
 517         mutex_lock(&inode->i_mutex);
 518         error = security_path_chmod(path, mode);
 519         if (error)
 520                 goto out_unlock;
 521         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 522         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 523         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 524 out_unlock:
 525         mutex_unlock(&inode->i_mutex);
 526         if (delegated_inode) {
 527                 error = break_deleg_wait(&delegated_inode);
 528                 if (!error)
 529                         goto retry_deleg;
 530         }
 531         mnt_drop_write(path->mnt);
 532         return error;
 533 }
 534
 535 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 536 {
 537         struct fd f = fdget(fd);
 538         int err = -EBADF;
 539
 540         if (f.file) {
 541                 audit_file(f.file);
 542                 err = chmod_common(&f.file->f_path, mode);
 543                 fdput(f);
 544         }
 545         return err;
 546 }
 547
 548 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 549 {
 550         struct path path;
 551         int error;
 552         unsigned int lookup_flags = LOOKUP_FOLLOW;
 553 retry:
 554         error = user_path_at(dfd, filename, lookup_flags, &path);
 555         if (!error) {
 556                 error = chmod_common(&path, mode);
 557                 path_put(&path);
 558                 if (retry_estale(error, lookup_flags)) {
 559                         lookup_flags |= LOOKUP_REVAL;
 560                         goto retry;
 561                 }
 562         }
 563         return error;
 564 }
 565
 566 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 567 {
 568         return sys_fchmodat(AT_FDCWD, filename, mode);
 569 }
 570
 571 static int chown_common(struct path *path, uid_t user, gid_t group)
 572 {
 573         struct inode *inode = path->dentry->d_inode;
 574         struct inode *delegated_inode = NULL;
 575         int error;
 576         struct iattr newattrs;
 577         kuid_t uid;
 578         kgid_t gid;
 579
 580         uid = make_kuid(current_user_ns(), user);
 581         gid = make_kgid(current_user_ns(), group);
 582
 583 retry_deleg:
 584         newattrs.ia_valid =  ATTR_CTIME;
 585         if (user != (uid_t) -1) {
 586                 if (!uid_valid(uid))
 587                         return -EINVAL;
 588                 newattrs.ia_valid |= ATTR_UID;
 589                 newattrs.ia_uid = uid;
 590         }
 591         if (group != (gid_t) -1) {
 592                 if (!gid_valid(gid))
 593                         return -EINVAL;
 594                 newattrs.ia_valid |= ATTR_GID;
 595                 newattrs.ia_gid = gid;
 596         }
 597         if (!S_ISDIR(inode->i_mode))
 598                 newattrs.ia_valid |=
 599                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 600         mutex_lock(&inode->i_mutex);
 601         error = security_path_chown(path, uid, gid);
 602         if (!error)
 603                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 604         mutex_unlock(&inode->i_mutex);
 605         if (delegated_inode) {
 606                 error = break_deleg_wait(&delegated_inode);
 607                 if (!error)
 608                         goto retry_deleg;
 609         }
 610         return error;
 611 }
 612
 613 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 614                 gid_t, group, int, flag)
 615 {
 616         struct path path;
 617         int error = -EINVAL;
 618         int lookup_flags;
 619
 620         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 621                 goto out;
 622
 623         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 624         if (flag & AT_EMPTY_PATH)
 625                 lookup_flags |= LOOKUP_EMPTY;
 626 retry:
 627         error = user_path_at(dfd, filename, lookup_flags, &path);
 628         if (error)
 629                 goto out;
 630         error = mnt_want_write(path.mnt);
 631         if (error)
 632                 goto out_release;
 633         error = chown_common(&path, user, group);
 634         mnt_drop_write(path.mnt);
 635 out_release:
 636         path_put(&path);
 637         if (retry_estale(error, lookup_flags)) {
 638                 lookup_flags |= LOOKUP_REVAL;
 639                 goto retry;
 640         }
 641 out:
 642         return error;
 643 }
 644
 645 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 646 {
 647         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 648 }
 649
 650 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 651 {
 652         return sys_fchownat(AT_FDCWD, filename, user, group,
 653                             AT_SYMLINK_NOFOLLOW);
 654 }
 655
 656 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 657 {
 658         struct fd f = fdget(fd);
 659         int error = -EBADF;
 660
 661         if (!f.file)
 662                 goto out;
 663
 664         error = mnt_want_write_file(f.file);
 665         if (error)
 666                 goto out_fput;
 667         audit_file(f.file);
 668         error = chown_common(&f.file->f_path, user, group);
 669         mnt_drop_write_file(f.file);
 670 out_fput:
 671         fdput(f);
 672 out:
 673         return error;
 674 }
 675
 676 int open_check_o_direct(struct file *f)
 677 {
 678         /* NB: we're sure to have correct a_ops only after f_op->open */
 679         if (f->f_flags & O_DIRECT) {
 680                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 681                         return -EINVAL;
 682         }
 683         return 0;
 684 }
 685 EXPORT_SYMBOL(open_check_o_direct);
 686
 687 static int do_dentry_open(struct file *f,
 688                           struct inode *inode,
 689                           int (*open)(struct inode *, struct file *),
 690                           const struct cred *cred)
 691 {
 692         static const struct file_operations empty_fops = {};
 693         int error;
 694
 695         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 696                                 FMODE_PREAD | FMODE_PWRITE;
 697
 698         path_get(&f->f_path);
 699         f->f_inode = inode;
 700         f->f_mapping = inode->i_mapping;
 701
 702         if (unlikely(f->f_flags & O_PATH)) {
 703                 f->f_mode = FMODE_PATH;
 704                 f->f_op = &empty_fops;
 705                 return 0;
 706         }
 707
 708         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 709                 error = get_write_access(inode);
 710                 if (unlikely(error))
 711                         goto cleanup_file;
 712                 error = __mnt_want_write(f->f_path.mnt);
 713                 if (unlikely(error)) {
 714                         put_write_access(inode);
 715                         goto cleanup_file;
 716                 }
 717                 f->f_mode |= FMODE_WRITER;
 718         }
 719
 720         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 721         if (S_ISREG(inode->i_mode))
 722                 f->f_mode |= FMODE_ATOMIC_POS;
 723
 724         f->f_op = fops_get(inode->i_fop);
 725         if (unlikely(WARN_ON(!f->f_op))) {
 726                 error = -ENODEV;
 727                 goto cleanup_all;
 728         }
 729
 730         error = security_file_open(f, cred);
 731         if (error)
 732                 goto cleanup_all;
 733
 734         error = break_lease(inode, f->f_flags);
 735         if (error)
 736                 goto cleanup_all;
 737
 738         if (!open)
 739                 open = f->f_op->open;
 740         if (open) {
 741                 error = open(inode, f);
 742                 if (error)
 743                         goto cleanup_all;
 744         }
 745         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 746                 i_readcount_inc(inode);
 747         if ((f->f_mode & FMODE_READ) &&
 748              likely(f->f_op->read || f->f_op->read_iter))
 749                 f->f_mode |= FMODE_CAN_READ;
 750         if ((f->f_mode & FMODE_WRITE) &&
 751              likely(f->f_op->write || f->f_op->write_iter))
 752                 f->f_mode |= FMODE_CAN_WRITE;
 753
 754         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 755
 756         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 757
 758         return 0;
 759
 760 cleanup_all:
 761         fops_put(f->f_op);
 762         if (f->f_mode & FMODE_WRITER) {
 763                 put_write_access(inode);
 764                 __mnt_drop_write(f->f_path.mnt);
 765         }
 766 cleanup_file:
 767         path_put(&f->f_path);
 768         f->f_path.mnt = NULL;
 769         f->f_path.dentry = NULL;
 770         f->f_inode = NULL;
 771         return error;
 772 }
 773
 774 /**
 775  * finish_open - finish opening a file
 776  * @file: file pointer
 777  * @dentry: pointer to dentry
 778  * @open: open callback
 779  * @opened: state of open
 780  *
 781  * This can be used to finish opening a file passed to i_op->atomic_open().
 782  *
 783  * If the open callback is set to NULL, then the standard f_op->open()
 784  * filesystem callback is substituted.
 785  *
 786  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 787  * the return value of d_splice_alias(), then the caller needs to perform dput()
 788  * on it after finish_open().
 789  *
 790  * On successful return @file is a fully instantiated open file.  After this, if
 791  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 792  *
 793  * Returns zero on success or -errno if the open failed.
 794  */
 795 int finish_open(struct file *file, struct dentry *dentry,
 796                 int (*open)(struct inode *, struct file *),
 797                 int *opened)
 798 {
 799         int error;
 800         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 801
 802         file->f_path.dentry = dentry;
 803         error = do_dentry_open(file, d_backing_inode(dentry), open,
 804                                current_cred());
 805         if (!error)
 806                 *opened |= FILE_OPENED;
 807
 808         return error;
 809 }
 810 EXPORT_SYMBOL(finish_open);
 811
 812 /**
 813  * finish_no_open - finish ->atomic_open() without opening the file
 814  *
 815  * @file: file pointer
 816  * @dentry: dentry or NULL (as returned from ->lookup())
 817  *
 818  * This can be used to set the result of a successful lookup in ->atomic_open().
 819  *
 820  * NB: unlike finish_open() this function does consume the dentry reference and
 821  * the caller need not dput() it.
 822  *
 823  * Returns "1" which must be the return value of ->atomic_open() after having
 824  * called this function.
 825  */
 826 int finish_no_open(struct file *file, struct dentry *dentry)
 827 {
 828         file->f_path.dentry = dentry;
 829         return 1;
 830 }
 831 EXPORT_SYMBOL(finish_no_open);
 832
 833 char *file_path(struct file *filp, char *buf, int buflen)
 834 {
 835         return d_path(&filp->f_path, buf, buflen);
 836 }
 837 EXPORT_SYMBOL(file_path);
 838
 839 /**
 840  * vfs_open - open the file at the given path
 841  * @path: path to open
 842  * @file: newly allocated file with f_flag initialized
 843  * @cred: credentials to use
 844  */
 845 int vfs_open(const struct path *path, struct file *file,
 846              const struct cred *cred)
 847 {
 848         struct inode *inode = vfs_select_inode(path->dentry, file->f_flags);
 849
 850         if (IS_ERR(inode))
 851                 return PTR_ERR(inode);
 852
 853         file->f_path = *path;
 854         return do_dentry_open(file, inode, NULL, cred);
 855 }
 856
 857 struct file *dentry_open(const struct path *path, int flags,
 858                          const struct cred *cred)
 859 {
 860         int error;
 861         struct file *f;
 862
 863         validate_creds(cred);
 864
 865         /* We must always pass in a valid mount pointer. */
 866         BUG_ON(!path->mnt);
 867
 868         f = get_empty_filp();
 869         if (!IS_ERR(f)) {
 870                 f->f_flags = flags;
 871                 error = vfs_open(path, f, cred);
 872                 if (!error) {
 873                         /* from now on we need fput() to dispose of f */
 874                         error = open_check_o_direct(f);
 875                         if (error) {
 876                                 fput(f);
 877                                 f = ERR_PTR(error);
 878                         }
 879                 } else {
 880                         put_filp(f);
 881                         f = ERR_PTR(error);
 882                 }
 883         }
 884         return f;
 885 }
 886 EXPORT_SYMBOL(dentry_open);
 887
 888 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 889 {
 890         int lookup_flags = 0;
 891         int acc_mode;
 892
 893         /*
 894          * Clear out all open flags we don't know about so that we don't report
 895          * them in fcntl(F_GETFD) or similar interfaces.
 896          */
 897         flags &= VALID_OPEN_FLAGS;
 898
 899         if (flags & (O_CREAT | __O_TMPFILE))
 900                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 901         else
 902                 op->mode = 0;
 903
 904         /* Must never be set by userspace */
 905         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 906
 907         /*
 908          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 909          * check for O_DSYNC if the need any syncing at all we enforce it's
 910          * always set instead of having to deal with possibly weird behaviour
 911          * for malicious applications setting only __O_SYNC.
 912          */
 913         if (flags & __O_SYNC)
 914                 flags |= O_DSYNC;
 915
 916         if (flags & __O_TMPFILE) {
 917                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 918                         return -EINVAL;
 919                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 920                 if (!(acc_mode & MAY_WRITE))
 921                         return -EINVAL;
 922         } else if (flags & O_PATH) {
 923                 /*
 924                  * If we have O_PATH in the open flag. Then we
 925                  * cannot have anything other than the below set of flags
 926                  */
 927                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 928                 acc_mode = 0;
 929         } else {
 930                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 931         }
 932
 933         op->open_flag = flags;
 934
 935         /* O_TRUNC implies we need access checks for write permissions */
 936         if (flags & O_TRUNC)
 937                 acc_mode |= MAY_WRITE;
 938
 939         /* Allow the LSM permission hook to distinguish append
 940            access from general write access. */
 941         if (flags & O_APPEND)
 942                 acc_mode |= MAY_APPEND;
 943
 944         op->acc_mode = acc_mode;
 945
 946         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 947
 948         if (flags & O_CREAT) {
 949                 op->intent |= LOOKUP_CREATE;
 950                 if (flags & O_EXCL)
 951                         op->intent |= LOOKUP_EXCL;
 952         }
 953
 954         if (flags & O_DIRECTORY)
 955                 lookup_flags |= LOOKUP_DIRECTORY;
 956         if (!(flags & O_NOFOLLOW))
 957                 lookup_flags |= LOOKUP_FOLLOW;
 958         op->lookup_flags = lookup_flags;
 959         return 0;
 960 }
 961
 962 /**
 963  * file_open_name - open file and return file pointer
 964  *
 965  * @name:       struct filename containing path to open
 966  * @flags:      open flags as per the open(2) second argument
 967  * @mode:       mode for the new file if O_CREAT is set, else ignored
 968  *
 969  * This is the helper to open a file from kernelspace if you really
 970  * have to.  But in generally you should not do this, so please move
 971  * along, nothing to see here..
 972  */
 973 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 974 {
 975         struct open_flags op;
 976         int err = build_open_flags(flags, mode, &op);
 977         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 978 }
 979
 980 /**
 981  * filp_open - open file and return file pointer
 982  *
 983  * @filename:   path to open
 984  * @flags:      open flags as per the open(2) second argument
 985  * @mode:       mode for the new file if O_CREAT is set, else ignored
 986  *
 987  * This is the helper to open a file from kernelspace if you really
 988  * have to.  But in generally you should not do this, so please move
 989  * along, nothing to see here..
 990  */
 991 struct file *filp_open(const char *filename, int flags, umode_t mode)
 992 {
 993         struct filename *name = getname_kernel(filename);
 994         struct file *file = ERR_CAST(name);
 995
 996         if (!IS_ERR(name)) {
 997                 file = file_open_name(name, flags, mode);
 998                 putname(name);
 999         }
1000         return file;
1001 }
1002 EXPORT_SYMBOL(filp_open);
1003
1004 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1005                             const char *filename, int flags, umode_t mode)
1006 {
1007         struct open_flags op;
1008         int err = build_open_flags(flags, mode, &op);
1009         if (err)
1010                 return ERR_PTR(err);
1011         return do_file_open_root(dentry, mnt, filename, &op);
1012 }
1013 EXPORT_SYMBOL(file_open_root);
1014
1015 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1016 {
1017         struct open_flags op;
1018         int fd = build_open_flags(flags, mode, &op);
1019         struct filename *tmp;
1020
1021         if (fd)
1022                 return fd;
1023
1024         tmp = getname(filename);
1025         if (IS_ERR(tmp))
1026                 return PTR_ERR(tmp);
1027
1028         fd = get_unused_fd_flags(flags);
1029         if (fd >= 0) {
1030                 struct file *f = do_filp_open(dfd, tmp, &op);
1031                 if (IS_ERR(f)) {
1032                         put_unused_fd(fd);
1033                         fd = PTR_ERR(f);
1034                 } else {
1035                         fsnotify_open(f);
1036                         fd_install(fd, f);
1037                         trace_do_sys_open(tmp->name, flags, mode);
1038                 }
1039         }
1040         putname(tmp);
1041         return fd;
1042 }
1043
1044 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1045 {
1046         if (force_o_largefile())
1047                 flags |= O_LARGEFILE;
1048
1049         return do_sys_open(AT_FDCWD, filename, flags, mode);
1050 }
1051
1052 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1053                 umode_t, mode)
1054 {
1055         if (force_o_largefile())
1056                 flags |= O_LARGEFILE;
1057
1058         return do_sys_open(dfd, filename, flags, mode);
1059 }
1060
1061 #ifndef __alpha__
1062
1063 /*
1064  * For backward compatibility?  Maybe this should be moved
1065  * into arch/i386 instead?
1066  */
1067 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1068 {
1069         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1070 }
1071
1072 #endif
1073
1074 /*
1075  * "id" is the POSIX thread ID. We use the
1076  * files pointer for this..
1077  */
1078 int filp_close(struct file *filp, fl_owner_t id)
1079 {
1080         int retval = 0;
1081
1082         if (!file_count(filp)) {
1083                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1084                 return 0;
1085         }
1086
1087         if (filp->f_op->flush)
1088                 retval = filp->f_op->flush(filp, id);
1089
1090         if (likely(!(filp->f_mode & FMODE_PATH))) {
1091                 dnotify_flush(filp, id);
1092                 locks_remove_posix(filp, id);
1093         }
1094         fput(filp);
1095         return retval;
1096 }
1097
1098 EXPORT_SYMBOL(filp_close);
1099
1100 /*
1101  * Careful here! We test whether the file pointer is NULL before
1102  * releasing the fd. This ensures that one clone task can't release
1103  * an fd while another clone is opening it.
1104  */
1105 SYSCALL_DEFINE1(close, unsigned int, fd)
1106 {
1107         int retval = __close_fd(current->files, fd);
1108
1109         /* can't restart close syscall because file table entry was cleared */
1110         if (unlikely(retval == -ERESTARTSYS ||
1111                      retval == -ERESTARTNOINTR ||
1112                      retval == -ERESTARTNOHAND ||
1113                      retval == -ERESTART_RESTARTBLOCK))
1114                 retval = -EINTR;
1115
1116         return retval;
1117 }
1118 EXPORT_SYMBOL(sys_close);
1119
1120 /*
1121  * This routine simulates a hangup on the tty, to arrange that users
1122  * are given clean terminals at login time.
1123  */
1124 SYSCALL_DEFINE0(vhangup)
1125 {
1126         if (capable(CAP_SYS_TTY_CONFIG)) {
1127                 tty_vhangup_self();
1128                 return 0;
1129         }
1130         return -EPERM;
1131 }
1132
1133 /*
1134  * Called when an inode is about to be open.
1135  * We use this to disallow opening large files on 32bit systems if
1136  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1137  * on this flag in sys_open.
1138  */
1139 int generic_file_open(struct inode * inode, struct file * filp)
1140 {
1141         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1142                 return -EOVERFLOW;
1143         return 0;
1144 }
1145
1146 EXPORT_SYMBOL(generic_file_open);
1147
1148 /*
1149  * This is used by subsystems that don't want seekable
1150  * file descriptors. The function is not supposed to ever fail, the only
1151  * reason it returns an 'int' and not 'void' is so that it can be plugged
1152  * directly into file_operations structure.
1153  */
1154 int nonseekable_open(struct inode *inode, struct file *filp)
1155 {
1156         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1157         return 0;
1158 }
1159
1160 EXPORT_SYMBOL(nonseekable_open);