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 && !(mode & FALLOC_FL_KEEP_SIZE))
 227                 return -EOPNOTSUPP;
 228
 229         if (!(file->f_mode & FMODE_WRITE))
 230                 return -EBADF;
 231         /*
 232          * Revalidate the write permissions, in case security policy has
 233          * changed since the files were opened.
 234          */
 235         ret = security_file_permission(file, MAY_WRITE);
 236         if (ret)
 237                 return ret;
 238
 239         if (S_ISFIFO(inode->i_mode))
 240                 return -ESPIPE;
 241
 242         /*
 243          * Let individual file system decide if it supports preallocation
 244          * for directories or not.
 245          */
 246         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 247                 return -ENODEV;
 248
 249         /* Check for wrap through zero too */
 250         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 251                 return -EFBIG;
 252
 253         if (!inode->i_op->fallocate)
 254                 return -EOPNOTSUPP;
 255
 256         return inode->i_op->fallocate(inode, mode, offset, len);
 257 }
 258
 259 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 260 {
 261         struct file *file;
 262         int error = -EBADF;
 263
 264         file = fget(fd);
 265         if (file) {
 266                 error = do_fallocate(file, mode, offset, len);
 267                 fput(file);
 268         }
 269
 270         return error;
 271 }
 272
 273 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 274 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 275 {
 276         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 277 }
 278 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 279 #endif
 280
 281 /*
 282  * access() needs to use the real uid/gid, not the effective uid/gid.
 283  * We do this by temporarily clearing all FS-related capabilities and
 284  * switching the fsuid/fsgid around to the real ones.
 285  */
 286 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 287 {
 288         const struct cred *old_cred;
 289         struct cred *override_cred;
 290         struct path path;
 291         struct inode *inode;
 292         int res;
 293
 294         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 295                 return -EINVAL;
 296
 297         override_cred = prepare_creds();
 298         if (!override_cred)
 299                 return -ENOMEM;
 300
 301         override_cred->fsuid = override_cred->uid;
 302         override_cred->fsgid = override_cred->gid;
 303
 304         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 305                 /* Clear the capabilities if we switch to a non-root user */
 306                 if (override_cred->uid)
 307                         cap_clear(override_cred->cap_effective);
 308                 else
 309                         override_cred->cap_effective =
 310                                 override_cred->cap_permitted;
 311         }
 312
 313         old_cred = override_creds(override_cred);
 314
 315         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 316         if (res)
 317                 goto out;
 318
 319         inode = path.dentry->d_inode;
 320
 321         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 322                 /*
 323                  * MAY_EXEC on regular files is denied if the fs is mounted
 324                  * with the "noexec" flag.
 325                  */
 326                 res = -EACCES;
 327                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 328                         goto out_path_release;
 329         }
 330
 331         res = inode_permission(inode, mode | MAY_ACCESS);
 332         /* SuS v2 requires we report a read only fs too */
 333         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 334                 goto out_path_release;
 335         /*
 336          * This is a rare case where using __mnt_is_readonly()
 337          * is OK without a mnt_want/drop_write() pair.  Since
 338          * no actual write to the fs is performed here, we do
 339          * not need to telegraph to that to anyone.
 340          *
 341          * By doing this, we accept that this access is
 342          * inherently racy and know that the fs may change
 343          * state before we even see this result.
 344          */
 345         if (__mnt_is_readonly(path.mnt))
 346                 res = -EROFS;
 347
 348 out_path_release:
 349         path_put(&path);
 350 out:
 351         revert_creds(old_cred);
 352         put_cred(override_cred);
 353         return res;
 354 }
 355
 356 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 357 {
 358         return sys_faccessat(AT_FDCWD, filename, mode);
 359 }
 360
 361 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 362 {
 363         struct path path;
 364         int error;
 365
 366         error = user_path_dir(filename, &path);
 367         if (error)
 368                 goto out;
 369
 370         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 371         if (error)
 372                 goto dput_and_out;
 373
 374         set_fs_pwd(current->fs, &path);
 375
 376 dput_and_out:
 377         path_put(&path);
 378 out:
 379         return error;
 380 }
 381
 382 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 383 {
 384         struct file *file;
 385         struct inode *inode;
 386         int error;
 387
 388         error = -EBADF;
 389         file = fget(fd);
 390         if (!file)
 391                 goto out;
 392
 393         inode = file->f_path.dentry->d_inode;
 394
 395         error = -ENOTDIR;
 396         if (!S_ISDIR(inode->i_mode))
 397                 goto out_putf;
 398
 399         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 400         if (!error)
 401                 set_fs_pwd(current->fs, &file->f_path);
 402 out_putf:
 403         fput(file);
 404 out:
 405         return error;
 406 }
 407
 408 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 409 {
 410         struct path path;
 411         int error;
 412
 413         error = user_path_dir(filename, &path);
 414         if (error)
 415                 goto out;
 416
 417         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 418         if (error)
 419                 goto dput_and_out;
 420
 421         error = -EPERM;
 422         if (!capable(CAP_SYS_CHROOT))
 423                 goto dput_and_out;
 424         error = security_path_chroot(&path);
 425         if (error)
 426                 goto dput_and_out;
 427
 428         set_fs_root(current->fs, &path);
 429         error = 0;
 430 dput_and_out:
 431         path_put(&path);
 432 out:
 433         return error;
 434 }
 435
 436 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 437 {
 438         struct inode * inode;
 439         struct dentry * dentry;
 440         struct file * file;
 441         int err = -EBADF;
 442         struct iattr newattrs;
 443
 444         file = fget(fd);
 445         if (!file)
 446                 goto out;
 447
 448         dentry = file->f_path.dentry;
 449         inode = dentry->d_inode;
 450
 451         audit_inode(NULL, dentry);
 452
 453         err = mnt_want_write_file(file);
 454         if (err)
 455                 goto out_putf;
 456         mutex_lock(&inode->i_mutex);
 457         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 458         if (err)
 459                 goto out_unlock;
 460         if (mode == (mode_t) -1)
 461                 mode = inode->i_mode;
 462         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 463         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 464         err = notify_change(dentry, &newattrs);
 465 out_unlock:
 466         mutex_unlock(&inode->i_mutex);
 467         mnt_drop_write(file->f_path.mnt);
 468 out_putf:
 469         fput(file);
 470 out:
 471         return err;
 472 }
 473
 474 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 475 {
 476         struct path path;
 477         struct inode *inode;
 478         int error;
 479         struct iattr newattrs;
 480
 481         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 482         if (error)
 483                 goto out;
 484         inode = path.dentry->d_inode;
 485
 486         error = mnt_want_write(path.mnt);
 487         if (error)
 488                 goto dput_and_out;
 489         mutex_lock(&inode->i_mutex);
 490         error = security_path_chmod(path.dentry, path.mnt, mode);
 491         if (error)
 492                 goto out_unlock;
 493         if (mode == (mode_t) -1)
 494                 mode = inode->i_mode;
 495         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 496         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 497         error = notify_change(path.dentry, &newattrs);
 498 out_unlock:
 499         mutex_unlock(&inode->i_mutex);
 500         mnt_drop_write(path.mnt);
 501 dput_and_out:
 502         path_put(&path);
 503 out:
 504         return error;
 505 }
 506
 507 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 508 {
 509         return sys_fchmodat(AT_FDCWD, filename, mode);
 510 }
 511
 512 static int chown_common(struct path *path, uid_t user, gid_t group)
 513 {
 514         struct inode *inode = path->dentry->d_inode;
 515         int error;
 516         struct iattr newattrs;
 517
 518         newattrs.ia_valid =  ATTR_CTIME;
 519         if (user != (uid_t) -1) {
 520                 newattrs.ia_valid |= ATTR_UID;
 521                 newattrs.ia_uid = user;
 522         }
 523         if (group != (gid_t) -1) {
 524                 newattrs.ia_valid |= ATTR_GID;
 525                 newattrs.ia_gid = group;
 526         }
 527         if (!S_ISDIR(inode->i_mode))
 528                 newattrs.ia_valid |=
 529                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 530         mutex_lock(&inode->i_mutex);
 531         error = security_path_chown(path, user, group);
 532         if (!error)
 533                 error = notify_change(path->dentry, &newattrs);
 534         mutex_unlock(&inode->i_mutex);
 535
 536         return error;
 537 }
 538
 539 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 540 {
 541         struct path path;
 542         int error;
 543
 544         error = user_path(filename, &path);
 545         if (error)
 546                 goto out;
 547         error = mnt_want_write(path.mnt);
 548         if (error)
 549                 goto out_release;
 550         error = chown_common(&path, user, group);
 551         mnt_drop_write(path.mnt);
 552 out_release:
 553         path_put(&path);
 554 out:
 555         return error;
 556 }
 557
 558 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 559                 gid_t, group, int, flag)
 560 {
 561         struct path path;
 562         int error = -EINVAL;
 563         int follow;
 564
 565         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 566                 goto out;
 567
 568         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 569         error = user_path_at(dfd, filename, follow, &path);
 570         if (error)
 571                 goto out;
 572         error = mnt_want_write(path.mnt);
 573         if (error)
 574                 goto out_release;
 575         error = chown_common(&path, user, group);
 576         mnt_drop_write(path.mnt);
 577 out_release:
 578         path_put(&path);
 579 out:
 580         return error;
 581 }
 582
 583 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 584 {
 585         struct path path;
 586         int error;
 587
 588         error = user_lpath(filename, &path);
 589         if (error)
 590                 goto out;
 591         error = mnt_want_write(path.mnt);
 592         if (error)
 593                 goto out_release;
 594         error = chown_common(&path, user, group);
 595         mnt_drop_write(path.mnt);
 596 out_release:
 597         path_put(&path);
 598 out:
 599         return error;
 600 }
 601
 602 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 603 {
 604         struct file * file;
 605         int error = -EBADF;
 606         struct dentry * dentry;
 607
 608         file = fget(fd);
 609         if (!file)
 610                 goto out;
 611
 612         error = mnt_want_write_file(file);
 613         if (error)
 614                 goto out_fput;
 615         dentry = file->f_path.dentry;
 616         audit_inode(NULL, dentry);
 617         error = chown_common(&file->f_path, user, group);
 618         mnt_drop_write(file->f_path.mnt);
 619 out_fput:
 620         fput(file);
 621 out:
 622         return error;
 623 }
 624
 625 /*
 626  * You have to be very careful that these write
 627  * counts get cleaned up in error cases and
 628  * upon __fput().  This should probably never
 629  * be called outside of __dentry_open().
 630  */
 631 static inline int __get_file_write_access(struct inode *inode,
 632                                           struct vfsmount *mnt)
 633 {
 634         int error;
 635         error = get_write_access(inode);
 636         if (error)
 637                 return error;
 638         /*
 639          * Do not take mount writer counts on
 640          * special files since no writes to
 641          * the mount itself will occur.
 642          */
 643         if (!special_file(inode->i_mode)) {
 644                 /*
 645                  * Balanced in __fput()
 646                  */
 647                 error = mnt_want_write(mnt);
 648                 if (error)
 649                         put_write_access(inode);
 650         }
 651         return error;
 652 }
 653
 654 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 655                                         struct file *f,
 656                                         int (*open)(struct inode *, struct file *),
 657                                         const struct cred *cred)
 658 {
 659         struct inode *inode;
 660         int error;
 661
 662         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 663                                 FMODE_PREAD | FMODE_PWRITE;
 664         inode = dentry->d_inode;
 665         if (f->f_mode & FMODE_WRITE) {
 666                 error = __get_file_write_access(inode, mnt);
 667                 if (error)
 668                         goto cleanup_file;
 669                 if (!special_file(inode->i_mode))
 670                         file_take_write(f);
 671         }
 672
 673         f->f_mapping = inode->i_mapping;
 674         f->f_path.dentry = dentry;
 675         f->f_path.mnt = mnt;
 676         f->f_pos = 0;
 677         f->f_op = fops_get(inode->i_fop);
 678         file_move(f, &inode->i_sb->s_files);
 679
 680         error = security_dentry_open(f, cred);
 681         if (error)
 682                 goto cleanup_all;
 683
 684         if (!open && f->f_op)
 685                 open = f->f_op->open;
 686         if (open) {
 687                 error = open(inode, f);
 688                 if (error)
 689                         goto cleanup_all;
 690         }
 691         ima_counts_get(f);
 692
 693         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 694
 695         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 696
 697         /* NB: we're sure to have correct a_ops only after f_op->open */
 698         if (f->f_flags & O_DIRECT) {
 699                 if (!f->f_mapping->a_ops ||
 700                     ((!f->f_mapping->a_ops->direct_IO) &&
 701                     (!f->f_mapping->a_ops->get_xip_mem))) {
 702                         fput(f);
 703                         f = ERR_PTR(-EINVAL);
 704                 }
 705         }
 706
 707         return f;
 708
 709 cleanup_all:
 710         fops_put(f->f_op);
 711         if (f->f_mode & FMODE_WRITE) {
 712                 put_write_access(inode);
 713                 if (!special_file(inode->i_mode)) {
 714                         /*
 715                          * We don't consider this a real
 716                          * mnt_want/drop_write() pair
 717                          * because it all happenend right
 718                          * here, so just reset the state.
 719                          */
 720                         file_reset_write(f);
 721                         mnt_drop_write(mnt);
 722                 }
 723         }
 724         file_kill(f);
 725         f->f_path.dentry = NULL;
 726         f->f_path.mnt = NULL;
 727 cleanup_file:
 728         put_filp(f);
 729         dput(dentry);
 730         mntput(mnt);
 731         return ERR_PTR(error);
 732 }
 733
 734 /**
 735  * lookup_instantiate_filp - instantiates the open intent filp
 736  * @nd: pointer to nameidata
 737  * @dentry: pointer to dentry
 738  * @open: open callback
 739  *
 740  * Helper for filesystems that want to use lookup open intents and pass back
 741  * a fully instantiated struct file to the caller.
 742  * This function is meant to be called from within a filesystem's
 743  * lookup method.
 744  * Beware of calling it for non-regular files! Those ->open methods might block
 745  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 746  * leading to a deadlock, as nobody can open that fifo anymore, because
 747  * another process to open fifo will block on locked parent when doing lookup).
 748  * Note that in case of error, nd->intent.open.file is destroyed, but the
 749  * path information remains valid.
 750  * If the open callback is set to NULL, then the standard f_op->open()
 751  * filesystem callback is substituted.
 752  */
 753 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 754                 int (*open)(struct inode *, struct file *))
 755 {
 756         const struct cred *cred = current_cred();
 757
 758         if (IS_ERR(nd->intent.open.file))
 759                 goto out;
 760         if (IS_ERR(dentry))
 761                 goto out_err;
 762         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 763                                              nd->intent.open.file,
 764                                              open, cred);
 765 out:
 766         return nd->intent.open.file;
 767 out_err:
 768         release_open_intent(nd);
 769         nd->intent.open.file = (struct file *)dentry;
 770         goto out;
 771 }
 772 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 773
 774 /**
 775  * nameidata_to_filp - convert a nameidata to an open filp.
 776  * @nd: pointer to nameidata
 777  * @flags: open flags
 778  *
 779  * Note that this function destroys the original nameidata
 780  */
 781 struct file *nameidata_to_filp(struct nameidata *nd)
 782 {
 783         const struct cred *cred = current_cred();
 784         struct file *filp;
 785
 786         /* Pick up the filp from the open intent */
 787         filp = nd->intent.open.file;
 788         /* Has the filesystem initialised the file for us? */
 789         if (filp->f_path.dentry == NULL)
 790                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 791                                      NULL, cred);
 792         else
 793                 path_put(&nd->path);
 794         return filp;
 795 }
 796
 797 /*
 798  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 799  * error.
 800  */
 801 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 802                          const struct cred *cred)
 803 {
 804         int error;
 805         struct file *f;
 806
 807         validate_creds(cred);
 808
 809         /*
 810          * We must always pass in a valid mount pointer.   Historically
 811          * callers got away with not passing it, but we must enforce this at
 812          * the earliest possible point now to avoid strange problems deep in the
 813          * filesystem stack.
 814          */
 815         if (!mnt) {
 816                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 817                 dump_stack();
 818                 return ERR_PTR(-EINVAL);
 819         }
 820
 821         error = -ENFILE;
 822         f = get_empty_filp();
 823         if (f == NULL) {
 824                 dput(dentry);
 825                 mntput(mnt);
 826                 return ERR_PTR(error);
 827         }
 828
 829         f->f_flags = flags;
 830         return __dentry_open(dentry, mnt, f, NULL, cred);
 831 }
 832 EXPORT_SYMBOL(dentry_open);
 833
 834 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 835 {
 836         struct fdtable *fdt = files_fdtable(files);
 837         __FD_CLR(fd, fdt->open_fds);
 838         if (fd < files->next_fd)
 839                 files->next_fd = fd;
 840 }
 841
 842 void put_unused_fd(unsigned int fd)
 843 {
 844         struct files_struct *files = current->files;
 845         spin_lock(&files->file_lock);
 846         __put_unused_fd(files, fd);
 847         spin_unlock(&files->file_lock);
 848 }
 849
 850 EXPORT_SYMBOL(put_unused_fd);
 851
 852 /*
 853  * Install a file pointer in the fd array.
 854  *
 855  * The VFS is full of places where we drop the files lock between
 856  * setting the open_fds bitmap and installing the file in the file
 857  * array.  At any such point, we are vulnerable to a dup2() race
 858  * installing a file in the array before us.  We need to detect this and
 859  * fput() the struct file we are about to overwrite in this case.
 860  *
 861  * It should never happen - if we allow dup2() do it, _really_ bad things
 862  * will follow.
 863  */
 864
 865 void fd_install(unsigned int fd, struct file *file)
 866 {
 867         struct files_struct *files = current->files;
 868         struct fdtable *fdt;
 869         spin_lock(&files->file_lock);
 870         fdt = files_fdtable(files);
 871         BUG_ON(fdt->fd[fd] != NULL);
 872         rcu_assign_pointer(fdt->fd[fd], file);
 873         spin_unlock(&files->file_lock);
 874 }
 875
 876 EXPORT_SYMBOL(fd_install);
 877
 878 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 879 {
 880         char *tmp = getname(filename);
 881         int fd = PTR_ERR(tmp);
 882
 883         if (!IS_ERR(tmp)) {
 884                 fd = get_unused_fd_flags(flags);
 885                 if (fd >= 0) {
 886                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
 887                         if (IS_ERR(f)) {
 888                                 put_unused_fd(fd);
 889                                 fd = PTR_ERR(f);
 890                         } else {
 891                                 fsnotify_open(f);
 892                                 fd_install(fd, f);
 893                         }
 894                 }
 895                 putname(tmp);
 896         }
 897         return fd;
 898 }
 899
 900 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 901 {
 902         long ret;
 903
 904         if (force_o_largefile())
 905                 flags |= O_LARGEFILE;
 906
 907         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 908         /* avoid REGPARM breakage on x86: */
 909         asmlinkage_protect(3, ret, filename, flags, mode);
 910         return ret;
 911 }
 912
 913 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 914                 int, mode)
 915 {
 916         long ret;
 917
 918         if (force_o_largefile())
 919                 flags |= O_LARGEFILE;
 920
 921         ret = do_sys_open(dfd, filename, flags, mode);
 922         /* avoid REGPARM breakage on x86: */
 923         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 924         return ret;
 925 }
 926
 927 #ifndef __alpha__
 928
 929 /*
 930  * For backward compatibility?  Maybe this should be moved
 931  * into arch/i386 instead?
 932  */
 933 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
 934 {
 935         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 936 }
 937
 938 #endif
 939
 940 /*
 941  * "id" is the POSIX thread ID. We use the
 942  * files pointer for this..
 943  */
 944 int filp_close(struct file *filp, fl_owner_t id)
 945 {
 946         int retval = 0;
 947
 948         if (!file_count(filp)) {
 949                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 950                 return 0;
 951         }
 952
 953         if (filp->f_op && filp->f_op->flush)
 954                 retval = filp->f_op->flush(filp, id);
 955
 956         dnotify_flush(filp, id);
 957         locks_remove_posix(filp, id);
 958         fput(filp);
 959         return retval;
 960 }
 961
 962 EXPORT_SYMBOL(filp_close);
 963
 964 /*
 965  * Careful here! We test whether the file pointer is NULL before
 966  * releasing the fd. This ensures that one clone task can't release
 967  * an fd while another clone is opening it.
 968  */
 969 SYSCALL_DEFINE1(close, unsigned int, fd)
 970 {
 971         struct file * filp;
 972         struct files_struct *files = current->files;
 973         struct fdtable *fdt;
 974         int retval;
 975
 976         spin_lock(&files->file_lock);
 977         fdt = files_fdtable(files);
 978         if (fd >= fdt->max_fds)
 979                 goto out_unlock;
 980         filp = fdt->fd[fd];
 981         if (!filp)
 982                 goto out_unlock;
 983         rcu_assign_pointer(fdt->fd[fd], NULL);
 984         FD_CLR(fd, fdt->close_on_exec);
 985         __put_unused_fd(files, fd);
 986         spin_unlock(&files->file_lock);
 987         retval = filp_close(filp, files);
 988
 989         /* can't restart close syscall because file table entry was cleared */
 990         if (unlikely(retval == -ERESTARTSYS ||
 991                      retval == -ERESTARTNOINTR ||
 992                      retval == -ERESTARTNOHAND ||
 993                      retval == -ERESTART_RESTARTBLOCK))
 994                 retval = -EINTR;
 995
 996         return retval;
 997
 998 out_unlock:
 999         spin_unlock(&files->file_lock);
1000         return -EBADF;
1001 }
1002 EXPORT_SYMBOL(sys_close);
1003
1004 /*
1005  * This routine simulates a hangup on the tty, to arrange that users
1006  * are given clean terminals at login time.
1007  */
1008 SYSCALL_DEFINE0(vhangup)
1009 {
1010         if (capable(CAP_SYS_TTY_CONFIG)) {
1011                 tty_vhangup_self();
1012                 return 0;
1013         }
1014         return -EPERM;
1015 }
1016
1017 /*
1018  * Called when an inode is about to be open.
1019  * We use this to disallow opening large files on 32bit systems if
1020  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1021  * on this flag in sys_open.
1022  */
1023 int generic_file_open(struct inode * inode, struct file * filp)
1024 {
1025         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1026                 return -EOVERFLOW;
1027         return 0;
1028 }
1029
1030 EXPORT_SYMBOL(generic_file_open);
1031
1032 /*
1033  * This is used by subsystems that don't want seekable
1034  * file descriptors. The function is not supposed to ever fail, the only
1035  * reason it returns an 'int' and not 'void' is so that it can be plugged
1036  * directly into file_operations structure.
1037  */
1038 int nonseekable_open(struct inode *inode, struct file *filp)
1039 {
1040         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1041         return 0;
1042 }
1043
1044 EXPORT_SYMBOL(nonseekable_open);