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
  33 #include "internal.h"
  34
  35 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  36         struct file *filp)
  37 {
  38         int ret;
  39         struct iattr newattrs;
  40
  41         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  42         if (length < 0)
  43                 return -EINVAL;
  44
  45         newattrs.ia_size = length;
  46         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  47         if (filp) {
  48                 newattrs.ia_file = filp;
  49                 newattrs.ia_valid |= ATTR_FILE;
  50         }
  51
  52         /* Remove suid/sgid on truncate too */
  53         ret = should_remove_suid(dentry);
  54         if (ret)
  55                 newattrs.ia_valid |= ret | ATTR_FORCE;
  56
  57         mutex_lock(&dentry->d_inode->i_mutex);
  58         ret = notify_change(dentry, &newattrs);
  59         mutex_unlock(&dentry->d_inode->i_mutex);
  60         return ret;
  61 }
  62
  63 static long do_sys_truncate(const char __user *pathname, loff_t length)
  64 {
  65         struct path path;
  66         struct inode *inode;
  67         int error;
  68
  69         error = -EINVAL;
  70         if (length < 0) /* sorry, but loff_t says... */
  71                 goto out;
  72
  73         error = user_path(pathname, &path);
  74         if (error)
  75                 goto out;
  76         inode = path.dentry->d_inode;
  77
  78         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  79         error = -EISDIR;
  80         if (S_ISDIR(inode->i_mode))
  81                 goto dput_and_out;
  82
  83         error = -EINVAL;
  84         if (!S_ISREG(inode->i_mode))
  85                 goto dput_and_out;
  86
  87         error = mnt_want_write(path.mnt);
  88         if (error)
  89                 goto dput_and_out;
  90
  91         error = inode_permission(inode, MAY_WRITE);
  92         if (error)
  93                 goto mnt_drop_write_and_out;
  94
  95         error = -EPERM;
  96         if (IS_APPEND(inode))
  97                 goto mnt_drop_write_and_out;
  98
  99         error = get_write_access(inode);
 100         if (error)
 101                 goto mnt_drop_write_and_out;
 102
 103         /*
 104          * Make sure that there are no leases.  get_write_access() protects
 105          * against the truncate racing with a lease-granting setlease().
 106          */
 107         error = break_lease(inode, O_WRONLY);
 108         if (error)
 109                 goto put_write_and_out;
 110
 111         error = locks_verify_truncate(inode, NULL, length);
 112         if (!error)
 113                 error = security_path_truncate(&path);
 114         if (!error)
 115                 error = do_truncate(path.dentry, length, 0, NULL);
 116
 117 put_write_and_out:
 118         put_write_access(inode);
 119 mnt_drop_write_and_out:
 120         mnt_drop_write(path.mnt);
 121 dput_and_out:
 122         path_put(&path);
 123 out:
 124         return error;
 125 }
 126
 127 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 128 {
 129         return do_sys_truncate(path, length);
 130 }
 131
 132 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 133 {
 134         struct inode * inode;
 135         struct dentry *dentry;
 136         struct file * file;
 137         int error;
 138
 139         error = -EINVAL;
 140         if (length < 0)
 141                 goto out;
 142         error = -EBADF;
 143         file = fget(fd);
 144         if (!file)
 145                 goto out;
 146
 147         /* explicitly opened as large or we are on 64-bit box */
 148         if (file->f_flags & O_LARGEFILE)
 149                 small = 0;
 150
 151         dentry = file->f_path.dentry;
 152         inode = dentry->d_inode;
 153         error = -EINVAL;
 154         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 155                 goto out_putf;
 156
 157         error = -EINVAL;
 158         /* Cannot ftruncate over 2^31 bytes without large file support */
 159         if (small && length > MAX_NON_LFS)
 160                 goto out_putf;
 161
 162         error = -EPERM;
 163         if (IS_APPEND(inode))
 164                 goto out_putf;
 165
 166         error = locks_verify_truncate(inode, file, length);
 167         if (!error)
 168                 error = security_path_truncate(&file->f_path);
 169         if (!error)
 170                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 171 out_putf:
 172         fput(file);
 173 out:
 174         return error;
 175 }
 176
 177 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 178 {
 179         long ret = do_sys_ftruncate(fd, length, 1);
 180         /* avoid REGPARM breakage on x86: */
 181         asmlinkage_protect(2, ret, fd, length);
 182         return ret;
 183 }
 184
 185 /* LFS versions of truncate are only needed on 32 bit machines */
 186 #if BITS_PER_LONG == 32
 187 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 188 {
 189         return do_sys_truncate(path, length);
 190 }
 191 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 192 asmlinkage long SyS_truncate64(long path, loff_t length)
 193 {
 194         return SYSC_truncate64((const char __user *) path, length);
 195 }
 196 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 197 #endif
 198
 199 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 200 {
 201         long ret = do_sys_ftruncate(fd, length, 0);
 202         /* avoid REGPARM breakage on x86: */
 203         asmlinkage_protect(2, ret, fd, length);
 204         return ret;
 205 }
 206 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 207 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 208 {
 209         return SYSC_ftruncate64((unsigned int) fd, length);
 210 }
 211 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 212 #endif
 213 #endif /* BITS_PER_LONG == 32 */
 214
 215
 216 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 217 {
 218         struct inode *inode = file->f_path.dentry->d_inode;
 219         long ret;
 220
 221         if (offset < 0 || len <= 0)
 222                 return -EINVAL;
 223
 224         /* Return error if mode is not supported */
 225         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 226                 return -EOPNOTSUPP;
 227
 228         if (!(file->f_mode & FMODE_WRITE))
 229                 return -EBADF;
 230         /*
 231          * Revalidate the write permissions, in case security policy has
 232          * changed since the files were opened.
 233          */
 234         ret = security_file_permission(file, MAY_WRITE);
 235         if (ret)
 236                 return ret;
 237
 238         if (S_ISFIFO(inode->i_mode))
 239                 return -ESPIPE;
 240
 241         /*
 242          * Let individual file system decide if it supports preallocation
 243          * for directories or not.
 244          */
 245         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 246                 return -ENODEV;
 247
 248         /* Check for wrap through zero too */
 249         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 250                 return -EFBIG;
 251
 252         if (!inode->i_op->fallocate)
 253                 return -EOPNOTSUPP;
 254
 255         return inode->i_op->fallocate(inode, mode, offset, len);
 256 }
 257
 258 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 259 {
 260         struct file *file;
 261         int error = -EBADF;
 262
 263         file = fget(fd);
 264         if (file) {
 265                 error = do_fallocate(file, mode, offset, len);
 266                 fput(file);
 267         }
 268
 269         return error;
 270 }
 271
 272 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 273 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 274 {
 275         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 276 }
 277 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 278 #endif
 279
 280 /*
 281  * access() needs to use the real uid/gid, not the effective uid/gid.
 282  * We do this by temporarily clearing all FS-related capabilities and
 283  * switching the fsuid/fsgid around to the real ones.
 284  */
 285 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 286 {
 287         const struct cred *old_cred;
 288         struct cred *override_cred;
 289         struct path path;
 290         struct inode *inode;
 291         int res;
 292
 293         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 294                 return -EINVAL;
 295
 296         override_cred = prepare_creds();
 297         if (!override_cred)
 298                 return -ENOMEM;
 299
 300         override_cred->fsuid = override_cred->uid;
 301         override_cred->fsgid = override_cred->gid;
 302
 303         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 304                 /* Clear the capabilities if we switch to a non-root user */
 305                 if (override_cred->uid)
 306                         cap_clear(override_cred->cap_effective);
 307                 else
 308                         override_cred->cap_effective =
 309                                 override_cred->cap_permitted;
 310         }
 311
 312         old_cred = override_creds(override_cred);
 313
 314         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 315         if (res)
 316                 goto out;
 317
 318         inode = path.dentry->d_inode;
 319
 320         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 321                 /*
 322                  * MAY_EXEC on regular files is denied if the fs is mounted
 323                  * with the "noexec" flag.
 324                  */
 325                 res = -EACCES;
 326                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 327                         goto out_path_release;
 328         }
 329
 330         res = inode_permission(inode, mode | MAY_ACCESS);
 331         /* SuS v2 requires we report a read only fs too */
 332         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 333                 goto out_path_release;
 334         /*
 335          * This is a rare case where using __mnt_is_readonly()
 336          * is OK without a mnt_want/drop_write() pair.  Since
 337          * no actual write to the fs is performed here, we do
 338          * not need to telegraph to that to anyone.
 339          *
 340          * By doing this, we accept that this access is
 341          * inherently racy and know that the fs may change
 342          * state before we even see this result.
 343          */
 344         if (__mnt_is_readonly(path.mnt))
 345                 res = -EROFS;
 346
 347 out_path_release:
 348         path_put(&path);
 349 out:
 350         revert_creds(old_cred);
 351         put_cred(override_cred);
 352         return res;
 353 }
 354
 355 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 356 {
 357         return sys_faccessat(AT_FDCWD, filename, mode);
 358 }
 359
 360 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 361 {
 362         struct path path;
 363         int error;
 364
 365         error = user_path_dir(filename, &path);
 366         if (error)
 367                 goto out;
 368
 369         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 370         if (error)
 371                 goto dput_and_out;
 372
 373         set_fs_pwd(current->fs, &path);
 374
 375 dput_and_out:
 376         path_put(&path);
 377 out:
 378         return error;
 379 }
 380
 381 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 382 {
 383         struct file *file;
 384         struct inode *inode;
 385         int error;
 386
 387         error = -EBADF;
 388         file = fget(fd);
 389         if (!file)
 390                 goto out;
 391
 392         inode = file->f_path.dentry->d_inode;
 393
 394         error = -ENOTDIR;
 395         if (!S_ISDIR(inode->i_mode))
 396                 goto out_putf;
 397
 398         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 399         if (!error)
 400                 set_fs_pwd(current->fs, &file->f_path);
 401 out_putf:
 402         fput(file);
 403 out:
 404         return error;
 405 }
 406
 407 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 408 {
 409         struct path path;
 410         int error;
 411
 412         error = user_path_dir(filename, &path);
 413         if (error)
 414                 goto out;
 415
 416         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 417         if (error)
 418                 goto dput_and_out;
 419
 420         error = -EPERM;
 421         if (!capable(CAP_SYS_CHROOT))
 422                 goto dput_and_out;
 423         error = security_path_chroot(&path);
 424         if (error)
 425                 goto dput_and_out;
 426
 427         set_fs_root(current->fs, &path);
 428         error = 0;
 429 dput_and_out:
 430         path_put(&path);
 431 out:
 432         return error;
 433 }
 434
 435 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 436 {
 437         struct inode * inode;
 438         struct dentry * dentry;
 439         struct file * file;
 440         int err = -EBADF;
 441         struct iattr newattrs;
 442
 443         file = fget(fd);
 444         if (!file)
 445                 goto out;
 446
 447         dentry = file->f_path.dentry;
 448         inode = dentry->d_inode;
 449
 450         audit_inode(NULL, dentry);
 451
 452         err = mnt_want_write_file(file);
 453         if (err)
 454                 goto out_putf;
 455         mutex_lock(&inode->i_mutex);
 456         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 457         if (err)
 458                 goto out_unlock;
 459         if (mode == (mode_t) -1)
 460                 mode = inode->i_mode;
 461         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 462         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 463         err = notify_change(dentry, &newattrs);
 464 out_unlock:
 465         mutex_unlock(&inode->i_mutex);
 466         mnt_drop_write(file->f_path.mnt);
 467 out_putf:
 468         fput(file);
 469 out:
 470         return err;
 471 }
 472
 473 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 474 {
 475         struct path path;
 476         struct inode *inode;
 477         int error;
 478         struct iattr newattrs;
 479
 480         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 481         if (error)
 482                 goto out;
 483         inode = path.dentry->d_inode;
 484
 485         error = mnt_want_write(path.mnt);
 486         if (error)
 487                 goto dput_and_out;
 488         mutex_lock(&inode->i_mutex);
 489         error = security_path_chmod(path.dentry, path.mnt, mode);
 490         if (error)
 491                 goto out_unlock;
 492         if (mode == (mode_t) -1)
 493                 mode = inode->i_mode;
 494         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 495         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 496         error = notify_change(path.dentry, &newattrs);
 497 out_unlock:
 498         mutex_unlock(&inode->i_mutex);
 499         mnt_drop_write(path.mnt);
 500 dput_and_out:
 501         path_put(&path);
 502 out:
 503         return error;
 504 }
 505
 506 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 507 {
 508         return sys_fchmodat(AT_FDCWD, filename, mode);
 509 }
 510
 511 static int chown_common(struct path *path, uid_t user, gid_t group)
 512 {
 513         struct inode *inode = path->dentry->d_inode;
 514         int error;
 515         struct iattr newattrs;
 516
 517         newattrs.ia_valid =  ATTR_CTIME;
 518         if (user != (uid_t) -1) {
 519                 newattrs.ia_valid |= ATTR_UID;
 520                 newattrs.ia_uid = user;
 521         }
 522         if (group != (gid_t) -1) {
 523                 newattrs.ia_valid |= ATTR_GID;
 524                 newattrs.ia_gid = group;
 525         }
 526         if (!S_ISDIR(inode->i_mode))
 527                 newattrs.ia_valid |=
 528                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 529         mutex_lock(&inode->i_mutex);
 530         error = security_path_chown(path, user, group);
 531         if (!error)
 532                 error = notify_change(path->dentry, &newattrs);
 533         mutex_unlock(&inode->i_mutex);
 534
 535         return error;
 536 }
 537
 538 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 539 {
 540         struct path path;
 541         int error;
 542
 543         error = user_path(filename, &path);
 544         if (error)
 545                 goto out;
 546         error = mnt_want_write(path.mnt);
 547         if (error)
 548                 goto out_release;
 549         error = chown_common(&path, user, group);
 550         mnt_drop_write(path.mnt);
 551 out_release:
 552         path_put(&path);
 553 out:
 554         return error;
 555 }
 556
 557 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 558                 gid_t, group, int, flag)
 559 {
 560         struct path path;
 561         int error = -EINVAL;
 562         int follow;
 563
 564         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 565                 goto out;
 566
 567         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 568         error = user_path_at(dfd, filename, follow, &path);
 569         if (error)
 570                 goto out;
 571         error = mnt_want_write(path.mnt);
 572         if (error)
 573                 goto out_release;
 574         error = chown_common(&path, user, group);
 575         mnt_drop_write(path.mnt);
 576 out_release:
 577         path_put(&path);
 578 out:
 579         return error;
 580 }
 581
 582 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 583 {
 584         struct path path;
 585         int error;
 586
 587         error = user_lpath(filename, &path);
 588         if (error)
 589                 goto out;
 590         error = mnt_want_write(path.mnt);
 591         if (error)
 592                 goto out_release;
 593         error = chown_common(&path, user, group);
 594         mnt_drop_write(path.mnt);
 595 out_release:
 596         path_put(&path);
 597 out:
 598         return error;
 599 }
 600
 601 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 602 {
 603         struct file * file;
 604         int error = -EBADF;
 605         struct dentry * dentry;
 606
 607         file = fget(fd);
 608         if (!file)
 609                 goto out;
 610
 611         error = mnt_want_write_file(file);
 612         if (error)
 613                 goto out_fput;
 614         dentry = file->f_path.dentry;
 615         audit_inode(NULL, dentry);
 616         error = chown_common(&file->f_path, user, group);
 617         mnt_drop_write(file->f_path.mnt);
 618 out_fput:
 619         fput(file);
 620 out:
 621         return error;
 622 }
 623
 624 /*
 625  * You have to be very careful that these write
 626  * counts get cleaned up in error cases and
 627  * upon __fput().  This should probably never
 628  * be called outside of __dentry_open().
 629  */
 630 static inline int __get_file_write_access(struct inode *inode,
 631                                           struct vfsmount *mnt)
 632 {
 633         int error;
 634         error = get_write_access(inode);
 635         if (error)
 636                 return error;
 637         /*
 638          * Do not take mount writer counts on
 639          * special files since no writes to
 640          * the mount itself will occur.
 641          */
 642         if (!special_file(inode->i_mode)) {
 643                 /*
 644                  * Balanced in __fput()
 645                  */
 646                 error = mnt_want_write(mnt);
 647                 if (error)
 648                         put_write_access(inode);
 649         }
 650         return error;
 651 }
 652
 653 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 654                                         struct file *f,
 655                                         int (*open)(struct inode *, struct file *),
 656                                         const struct cred *cred)
 657 {
 658         struct inode *inode;
 659         int error;
 660
 661         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 662                                 FMODE_PREAD | FMODE_PWRITE;
 663         inode = dentry->d_inode;
 664         if (f->f_mode & FMODE_WRITE) {
 665                 error = __get_file_write_access(inode, mnt);
 666                 if (error)
 667                         goto cleanup_file;
 668                 if (!special_file(inode->i_mode))
 669                         file_take_write(f);
 670         }
 671
 672         f->f_mapping = inode->i_mapping;
 673         f->f_path.dentry = dentry;
 674         f->f_path.mnt = mnt;
 675         f->f_pos = 0;
 676         f->f_op = fops_get(inode->i_fop);
 677         file_move(f, &inode->i_sb->s_files);
 678
 679         error = security_dentry_open(f, cred);
 680         if (error)
 681                 goto cleanup_all;
 682
 683         if (!open && f->f_op)
 684                 open = f->f_op->open;
 685         if (open) {
 686                 error = open(inode, f);
 687                 if (error)
 688                         goto cleanup_all;
 689         }
 690         ima_counts_get(f);
 691
 692         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 693
 694         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 695
 696         /* NB: we're sure to have correct a_ops only after f_op->open */
 697         if (f->f_flags & O_DIRECT) {
 698                 if (!f->f_mapping->a_ops ||
 699                     ((!f->f_mapping->a_ops->direct_IO) &&
 700                     (!f->f_mapping->a_ops->get_xip_mem))) {
 701                         fput(f);
 702                         f = ERR_PTR(-EINVAL);
 703                 }
 704         }
 705
 706         return f;
 707
 708 cleanup_all:
 709         fops_put(f->f_op);
 710         if (f->f_mode & FMODE_WRITE) {
 711                 put_write_access(inode);
 712                 if (!special_file(inode->i_mode)) {
 713                         /*
 714                          * We don't consider this a real
 715                          * mnt_want/drop_write() pair
 716                          * because it all happenend right
 717                          * here, so just reset the state.
 718                          */
 719                         file_reset_write(f);
 720                         mnt_drop_write(mnt);
 721                 }
 722         }
 723         file_kill(f);
 724         f->f_path.dentry = NULL;
 725         f->f_path.mnt = NULL;
 726 cleanup_file:
 727         put_filp(f);
 728         dput(dentry);
 729         mntput(mnt);
 730         return ERR_PTR(error);
 731 }
 732
 733 /**
 734  * lookup_instantiate_filp - instantiates the open intent filp
 735  * @nd: pointer to nameidata
 736  * @dentry: pointer to dentry
 737  * @open: open callback
 738  *
 739  * Helper for filesystems that want to use lookup open intents and pass back
 740  * a fully instantiated struct file to the caller.
 741  * This function is meant to be called from within a filesystem's
 742  * lookup method.
 743  * Beware of calling it for non-regular files! Those ->open methods might block
 744  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 745  * leading to a deadlock, as nobody can open that fifo anymore, because
 746  * another process to open fifo will block on locked parent when doing lookup).
 747  * Note that in case of error, nd->intent.open.file is destroyed, but the
 748  * path information remains valid.
 749  * If the open callback is set to NULL, then the standard f_op->open()
 750  * filesystem callback is substituted.
 751  */
 752 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 753                 int (*open)(struct inode *, struct file *))
 754 {
 755         const struct cred *cred = current_cred();
 756
 757         if (IS_ERR(nd->intent.open.file))
 758                 goto out;
 759         if (IS_ERR(dentry))
 760                 goto out_err;
 761         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 762                                              nd->intent.open.file,
 763                                              open, cred);
 764 out:
 765         return nd->intent.open.file;
 766 out_err:
 767         release_open_intent(nd);
 768         nd->intent.open.file = (struct file *)dentry;
 769         goto out;
 770 }
 771 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 772
 773 /**
 774  * nameidata_to_filp - convert a nameidata to an open filp.
 775  * @nd: pointer to nameidata
 776  * @flags: open flags
 777  *
 778  * Note that this function destroys the original nameidata
 779  */
 780 struct file *nameidata_to_filp(struct nameidata *nd)
 781 {
 782         const struct cred *cred = current_cred();
 783         struct file *filp;
 784
 785         /* Pick up the filp from the open intent */
 786         filp = nd->intent.open.file;
 787         /* Has the filesystem initialised the file for us? */
 788         if (filp->f_path.dentry == NULL)
 789                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 790                                      NULL, cred);
 791         else
 792                 path_put(&nd->path);
 793         return filp;
 794 }
 795
 796 /*
 797  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 798  * error.
 799  */
 800 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 801                          const struct cred *cred)
 802 {
 803         int error;
 804         struct file *f;
 805
 806         validate_creds(cred);
 807
 808         /*
 809          * We must always pass in a valid mount pointer.   Historically
 810          * callers got away with not passing it, but we must enforce this at
 811          * the earliest possible point now to avoid strange problems deep in the
 812          * filesystem stack.
 813          */
 814         if (!mnt) {
 815                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 816                 dump_stack();
 817                 return ERR_PTR(-EINVAL);
 818         }
 819
 820         error = -ENFILE;
 821         f = get_empty_filp();
 822         if (f == NULL) {
 823                 dput(dentry);
 824                 mntput(mnt);
 825                 return ERR_PTR(error);
 826         }
 827
 828         f->f_flags = flags;
 829         return __dentry_open(dentry, mnt, f, NULL, cred);
 830 }
 831 EXPORT_SYMBOL(dentry_open);
 832
 833 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 834 {
 835         struct fdtable *fdt = files_fdtable(files);
 836         __FD_CLR(fd, fdt->open_fds);
 837         if (fd < files->next_fd)
 838                 files->next_fd = fd;
 839 }
 840
 841 void put_unused_fd(unsigned int fd)
 842 {
 843         struct files_struct *files = current->files;
 844         spin_lock(&files->file_lock);
 845         __put_unused_fd(files, fd);
 846         spin_unlock(&files->file_lock);
 847 }
 848
 849 EXPORT_SYMBOL(put_unused_fd);
 850
 851 /*
 852  * Install a file pointer in the fd array.
 853  *
 854  * The VFS is full of places where we drop the files lock between
 855  * setting the open_fds bitmap and installing the file in the file
 856  * array.  At any such point, we are vulnerable to a dup2() race
 857  * installing a file in the array before us.  We need to detect this and
 858  * fput() the struct file we are about to overwrite in this case.
 859  *
 860  * It should never happen - if we allow dup2() do it, _really_ bad things
 861  * will follow.
 862  */
 863
 864 void fd_install(unsigned int fd, struct file *file)
 865 {
 866         struct files_struct *files = current->files;
 867         struct fdtable *fdt;
 868         spin_lock(&files->file_lock);
 869         fdt = files_fdtable(files);
 870         BUG_ON(fdt->fd[fd] != NULL);
 871         rcu_assign_pointer(fdt->fd[fd], file);
 872         spin_unlock(&files->file_lock);
 873 }
 874
 875 EXPORT_SYMBOL(fd_install);
 876
 877 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 878 {
 879         char *tmp = getname(filename);
 880         int fd = PTR_ERR(tmp);
 881
 882         if (!IS_ERR(tmp)) {
 883                 fd = get_unused_fd_flags(flags);
 884                 if (fd >= 0) {
 885                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
 886                         if (IS_ERR(f)) {
 887                                 put_unused_fd(fd);
 888                                 fd = PTR_ERR(f);
 889                         } else {
 890                                 fsnotify_open(f->f_path.dentry);
 891                                 fd_install(fd, f);
 892                         }
 893                 }
 894                 putname(tmp);
 895         }
 896         return fd;
 897 }
 898
 899 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 900 {
 901         long ret;
 902
 903         if (force_o_largefile())
 904                 flags |= O_LARGEFILE;
 905
 906         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 907         /* avoid REGPARM breakage on x86: */
 908         asmlinkage_protect(3, ret, filename, flags, mode);
 909         return ret;
 910 }
 911
 912 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 913                 int, mode)
 914 {
 915         long ret;
 916
 917         if (force_o_largefile())
 918                 flags |= O_LARGEFILE;
 919
 920         ret = do_sys_open(dfd, filename, flags, mode);
 921         /* avoid REGPARM breakage on x86: */
 922         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 923         return ret;
 924 }
 925
 926 #ifndef __alpha__
 927
 928 /*
 929  * For backward compatibility?  Maybe this should be moved
 930  * into arch/i386 instead?
 931  */
 932 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
 933 {
 934         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 935 }
 936
 937 #endif
 938
 939 /*
 940  * "id" is the POSIX thread ID. We use the
 941  * files pointer for this..
 942  */
 943 int filp_close(struct file *filp, fl_owner_t id)
 944 {
 945         int retval = 0;
 946
 947         if (!file_count(filp)) {
 948                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 949                 return 0;
 950         }
 951
 952         if (filp->f_op && filp->f_op->flush)
 953                 retval = filp->f_op->flush(filp, id);
 954
 955         dnotify_flush(filp, id);
 956         locks_remove_posix(filp, id);
 957         fput(filp);
 958         return retval;
 959 }
 960
 961 EXPORT_SYMBOL(filp_close);
 962
 963 /*
 964  * Careful here! We test whether the file pointer is NULL before
 965  * releasing the fd. This ensures that one clone task can't release
 966  * an fd while another clone is opening it.
 967  */
 968 SYSCALL_DEFINE1(close, unsigned int, fd)
 969 {
 970         struct file * filp;
 971         struct files_struct *files = current->files;
 972         struct fdtable *fdt;
 973         int retval;
 974
 975         spin_lock(&files->file_lock);
 976         fdt = files_fdtable(files);
 977         if (fd >= fdt->max_fds)
 978                 goto out_unlock;
 979         filp = fdt->fd[fd];
 980         if (!filp)
 981                 goto out_unlock;
 982         rcu_assign_pointer(fdt->fd[fd], NULL);
 983         FD_CLR(fd, fdt->close_on_exec);
 984         __put_unused_fd(files, fd);
 985         spin_unlock(&files->file_lock);
 986         retval = filp_close(filp, files);
 987
 988         /* can't restart close syscall because file table entry was cleared */
 989         if (unlikely(retval == -ERESTARTSYS ||
 990                      retval == -ERESTARTNOINTR ||
 991                      retval == -ERESTARTNOHAND ||
 992                      retval == -ERESTART_RESTARTBLOCK))
 993                 retval = -EINTR;
 994
 995         return retval;
 996
 997 out_unlock:
 998         spin_unlock(&files->file_lock);
 999         return -EBADF;
1000 }
1001 EXPORT_SYMBOL(sys_close);
1002
1003 /*
1004  * This routine simulates a hangup on the tty, to arrange that users
1005  * are given clean terminals at login time.
1006  */
1007 SYSCALL_DEFINE0(vhangup)
1008 {
1009         if (capable(CAP_SYS_TTY_CONFIG)) {
1010                 tty_vhangup_self();
1011                 return 0;
1012         }
1013         return -EPERM;
1014 }
1015
1016 /*
1017  * Called when an inode is about to be open.
1018  * We use this to disallow opening large files on 32bit systems if
1019  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1020  * on this flag in sys_open.
1021  */
1022 int generic_file_open(struct inode * inode, struct file * filp)
1023 {
1024         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1025                 return -EOVERFLOW;
1026         return 0;
1027 }
1028
1029 EXPORT_SYMBOL(generic_file_open);
1030
1031 /*
1032  * This is used by subsystems that don't want seekable
1033  * file descriptors
1034  */
1035 int nonseekable_open(struct inode *inode, struct file *filp)
1036 {
1037         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1038         return 0;
1039 }
1040
1041 EXPORT_SYMBOL(nonseekable_open);