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, length, 0);
 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, length,
 170                                                ATTR_MTIME|ATTR_CTIME);
 171         if (!error)
 172                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 173 out_putf:
 174         fput(file);
 175 out:
 176         return error;
 177 }
 178
 179 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 180 {
 181         long ret = do_sys_ftruncate(fd, length, 1);
 182         /* avoid REGPARM breakage on x86: */
 183         asmlinkage_protect(2, ret, fd, length);
 184         return ret;
 185 }
 186
 187 /* LFS versions of truncate are only needed on 32 bit machines */
 188 #if BITS_PER_LONG == 32
 189 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 190 {
 191         return do_sys_truncate(path, length);
 192 }
 193 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 194 asmlinkage long SyS_truncate64(long path, loff_t length)
 195 {
 196         return SYSC_truncate64((const char __user *) path, length);
 197 }
 198 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 199 #endif
 200
 201 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 202 {
 203         long ret = do_sys_ftruncate(fd, length, 0);
 204         /* avoid REGPARM breakage on x86: */
 205         asmlinkage_protect(2, ret, fd, length);
 206         return ret;
 207 }
 208 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 209 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 210 {
 211         return SYSC_ftruncate64((unsigned int) fd, length);
 212 }
 213 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 214 #endif
 215 #endif /* BITS_PER_LONG == 32 */
 216
 217
 218 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 219 {
 220         struct inode *inode = file->f_path.dentry->d_inode;
 221         long ret;
 222
 223         if (offset < 0 || len <= 0)
 224                 return -EINVAL;
 225
 226         /* Return error if mode is not supported */
 227         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 228                 return -EOPNOTSUPP;
 229
 230         if (!(file->f_mode & FMODE_WRITE))
 231                 return -EBADF;
 232         /*
 233          * Revalidate the write permissions, in case security policy has
 234          * changed since the files were opened.
 235          */
 236         ret = security_file_permission(file, MAY_WRITE);
 237         if (ret)
 238                 return ret;
 239
 240         if (S_ISFIFO(inode->i_mode))
 241                 return -ESPIPE;
 242
 243         /*
 244          * Let individual file system decide if it supports preallocation
 245          * for directories or not.
 246          */
 247         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 248                 return -ENODEV;
 249
 250         /* Check for wrap through zero too */
 251         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 252                 return -EFBIG;
 253
 254         if (!inode->i_op->fallocate)
 255                 return -EOPNOTSUPP;
 256
 257         return inode->i_op->fallocate(inode, mode, offset, len);
 258 }
 259
 260 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 261 {
 262         struct file *file;
 263         int error = -EBADF;
 264
 265         file = fget(fd);
 266         if (file) {
 267                 error = do_fallocate(file, mode, offset, len);
 268                 fput(file);
 269         }
 270
 271         return error;
 272 }
 273
 274 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 275 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 276 {
 277         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 278 }
 279 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 280 #endif
 281
 282 /*
 283  * access() needs to use the real uid/gid, not the effective uid/gid.
 284  * We do this by temporarily clearing all FS-related capabilities and
 285  * switching the fsuid/fsgid around to the real ones.
 286  */
 287 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 288 {
 289         const struct cred *old_cred;
 290         struct cred *override_cred;
 291         struct path path;
 292         struct inode *inode;
 293         int res;
 294
 295         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 296                 return -EINVAL;
 297
 298         override_cred = prepare_creds();
 299         if (!override_cred)
 300                 return -ENOMEM;
 301
 302         override_cred->fsuid = override_cred->uid;
 303         override_cred->fsgid = override_cred->gid;
 304
 305         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 306                 /* Clear the capabilities if we switch to a non-root user */
 307                 if (override_cred->uid)
 308                         cap_clear(override_cred->cap_effective);
 309                 else
 310                         override_cred->cap_effective =
 311                                 override_cred->cap_permitted;
 312         }
 313
 314         old_cred = override_creds(override_cred);
 315
 316         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 317         if (res)
 318                 goto out;
 319
 320         inode = path.dentry->d_inode;
 321
 322         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 323                 /*
 324                  * MAY_EXEC on regular files is denied if the fs is mounted
 325                  * with the "noexec" flag.
 326                  */
 327                 res = -EACCES;
 328                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 329                         goto out_path_release;
 330         }
 331
 332         res = inode_permission(inode, mode | MAY_ACCESS);
 333         /* SuS v2 requires we report a read only fs too */
 334         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 335                 goto out_path_release;
 336         /*
 337          * This is a rare case where using __mnt_is_readonly()
 338          * is OK without a mnt_want/drop_write() pair.  Since
 339          * no actual write to the fs is performed here, we do
 340          * not need to telegraph to that to anyone.
 341          *
 342          * By doing this, we accept that this access is
 343          * inherently racy and know that the fs may change
 344          * state before we even see this result.
 345          */
 346         if (__mnt_is_readonly(path.mnt))
 347                 res = -EROFS;
 348
 349 out_path_release:
 350         path_put(&path);
 351 out:
 352         revert_creds(old_cred);
 353         put_cred(override_cred);
 354         return res;
 355 }
 356
 357 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 358 {
 359         return sys_faccessat(AT_FDCWD, filename, mode);
 360 }
 361
 362 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 363 {
 364         struct path path;
 365         int error;
 366
 367         error = user_path_dir(filename, &path);
 368         if (error)
 369                 goto out;
 370
 371         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 372         if (error)
 373                 goto dput_and_out;
 374
 375         set_fs_pwd(current->fs, &path);
 376
 377 dput_and_out:
 378         path_put(&path);
 379 out:
 380         return error;
 381 }
 382
 383 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 384 {
 385         struct file *file;
 386         struct inode *inode;
 387         int error;
 388
 389         error = -EBADF;
 390         file = fget(fd);
 391         if (!file)
 392                 goto out;
 393
 394         inode = file->f_path.dentry->d_inode;
 395
 396         error = -ENOTDIR;
 397         if (!S_ISDIR(inode->i_mode))
 398                 goto out_putf;
 399
 400         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 401         if (!error)
 402                 set_fs_pwd(current->fs, &file->f_path);
 403 out_putf:
 404         fput(file);
 405 out:
 406         return error;
 407 }
 408
 409 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 410 {
 411         struct path path;
 412         int error;
 413
 414         error = user_path_dir(filename, &path);
 415         if (error)
 416                 goto out;
 417
 418         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 419         if (error)
 420                 goto dput_and_out;
 421
 422         error = -EPERM;
 423         if (!capable(CAP_SYS_CHROOT))
 424                 goto dput_and_out;
 425         error = security_path_chroot(&path);
 426         if (error)
 427                 goto dput_and_out;
 428
 429         set_fs_root(current->fs, &path);
 430         error = 0;
 431 dput_and_out:
 432         path_put(&path);
 433 out:
 434         return error;
 435 }
 436
 437 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 438 {
 439         struct inode * inode;
 440         struct dentry * dentry;
 441         struct file * file;
 442         int err = -EBADF;
 443         struct iattr newattrs;
 444
 445         file = fget(fd);
 446         if (!file)
 447                 goto out;
 448
 449         dentry = file->f_path.dentry;
 450         inode = dentry->d_inode;
 451
 452         audit_inode(NULL, dentry);
 453
 454         err = mnt_want_write_file(file);
 455         if (err)
 456                 goto out_putf;
 457         mutex_lock(&inode->i_mutex);
 458         err = security_path_chmod(dentry, file->f_vfsmnt, mode);
 459         if (err)
 460                 goto out_unlock;
 461         if (mode == (mode_t) -1)
 462                 mode = inode->i_mode;
 463         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 464         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 465         err = notify_change(dentry, &newattrs);
 466 out_unlock:
 467         mutex_unlock(&inode->i_mutex);
 468         mnt_drop_write(file->f_path.mnt);
 469 out_putf:
 470         fput(file);
 471 out:
 472         return err;
 473 }
 474
 475 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 476 {
 477         struct path path;
 478         struct inode *inode;
 479         int error;
 480         struct iattr newattrs;
 481
 482         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 483         if (error)
 484                 goto out;
 485         inode = path.dentry->d_inode;
 486
 487         error = mnt_want_write(path.mnt);
 488         if (error)
 489                 goto dput_and_out;
 490         mutex_lock(&inode->i_mutex);
 491         error = security_path_chmod(path.dentry, path.mnt, mode);
 492         if (error)
 493                 goto out_unlock;
 494         if (mode == (mode_t) -1)
 495                 mode = inode->i_mode;
 496         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 497         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 498         error = notify_change(path.dentry, &newattrs);
 499 out_unlock:
 500         mutex_unlock(&inode->i_mutex);
 501         mnt_drop_write(path.mnt);
 502 dput_and_out:
 503         path_put(&path);
 504 out:
 505         return error;
 506 }
 507
 508 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 509 {
 510         return sys_fchmodat(AT_FDCWD, filename, mode);
 511 }
 512
 513 static int chown_common(struct path *path, uid_t user, gid_t group)
 514 {
 515         struct inode *inode = path->dentry->d_inode;
 516         int error;
 517         struct iattr newattrs;
 518
 519         newattrs.ia_valid =  ATTR_CTIME;
 520         if (user != (uid_t) -1) {
 521                 newattrs.ia_valid |= ATTR_UID;
 522                 newattrs.ia_uid = user;
 523         }
 524         if (group != (gid_t) -1) {
 525                 newattrs.ia_valid |= ATTR_GID;
 526                 newattrs.ia_gid = group;
 527         }
 528         if (!S_ISDIR(inode->i_mode))
 529                 newattrs.ia_valid |=
 530                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 531         mutex_lock(&inode->i_mutex);
 532         error = security_path_chown(path, user, group);
 533         if (!error)
 534                 error = notify_change(path->dentry, &newattrs);
 535         mutex_unlock(&inode->i_mutex);
 536
 537         return error;
 538 }
 539
 540 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 541 {
 542         struct path path;
 543         int error;
 544
 545         error = user_path(filename, &path);
 546         if (error)
 547                 goto out;
 548         error = mnt_want_write(path.mnt);
 549         if (error)
 550                 goto out_release;
 551         error = chown_common(&path, user, group);
 552         mnt_drop_write(path.mnt);
 553 out_release:
 554         path_put(&path);
 555 out:
 556         return error;
 557 }
 558
 559 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 560                 gid_t, group, int, flag)
 561 {
 562         struct path path;
 563         int error = -EINVAL;
 564         int follow;
 565
 566         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 567                 goto out;
 568
 569         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 570         error = user_path_at(dfd, filename, follow, &path);
 571         if (error)
 572                 goto out;
 573         error = mnt_want_write(path.mnt);
 574         if (error)
 575                 goto out_release;
 576         error = chown_common(&path, user, group);
 577         mnt_drop_write(path.mnt);
 578 out_release:
 579         path_put(&path);
 580 out:
 581         return error;
 582 }
 583
 584 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 585 {
 586         struct path path;
 587         int error;
 588
 589         error = user_lpath(filename, &path);
 590         if (error)
 591                 goto out;
 592         error = mnt_want_write(path.mnt);
 593         if (error)
 594                 goto out_release;
 595         error = chown_common(&path, user, group);
 596         mnt_drop_write(path.mnt);
 597 out_release:
 598         path_put(&path);
 599 out:
 600         return error;
 601 }
 602
 603 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 604 {
 605         struct file * file;
 606         int error = -EBADF;
 607         struct dentry * dentry;
 608
 609         file = fget(fd);
 610         if (!file)
 611                 goto out;
 612
 613         error = mnt_want_write_file(file);
 614         if (error)
 615                 goto out_fput;
 616         dentry = file->f_path.dentry;
 617         audit_inode(NULL, dentry);
 618         error = chown_common(&file->f_path, user, group);
 619         mnt_drop_write(file->f_path.mnt);
 620 out_fput:
 621         fput(file);
 622 out:
 623         return error;
 624 }
 625
 626 /*
 627  * You have to be very careful that these write
 628  * counts get cleaned up in error cases and
 629  * upon __fput().  This should probably never
 630  * be called outside of __dentry_open().
 631  */
 632 static inline int __get_file_write_access(struct inode *inode,
 633                                           struct vfsmount *mnt)
 634 {
 635         int error;
 636         error = get_write_access(inode);
 637         if (error)
 638                 return error;
 639         /*
 640          * Do not take mount writer counts on
 641          * special files since no writes to
 642          * the mount itself will occur.
 643          */
 644         if (!special_file(inode->i_mode)) {
 645                 /*
 646                  * Balanced in __fput()
 647                  */
 648                 error = mnt_want_write(mnt);
 649                 if (error)
 650                         put_write_access(inode);
 651         }
 652         return error;
 653 }
 654
 655 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 656                                         struct file *f,
 657                                         int (*open)(struct inode *, struct file *),
 658                                         const struct cred *cred)
 659 {
 660         struct inode *inode;
 661         int error;
 662
 663         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 664                                 FMODE_PREAD | FMODE_PWRITE;
 665         inode = dentry->d_inode;
 666         if (f->f_mode & FMODE_WRITE) {
 667                 error = __get_file_write_access(inode, mnt);
 668                 if (error)
 669                         goto cleanup_file;
 670                 if (!special_file(inode->i_mode))
 671                         file_take_write(f);
 672         }
 673
 674         f->f_mapping = inode->i_mapping;
 675         f->f_path.dentry = dentry;
 676         f->f_path.mnt = mnt;
 677         f->f_pos = 0;
 678         f->f_op = fops_get(inode->i_fop);
 679         file_move(f, &inode->i_sb->s_files);
 680
 681         error = security_dentry_open(f, cred);
 682         if (error)
 683                 goto cleanup_all;
 684
 685         if (!open && f->f_op)
 686                 open = f->f_op->open;
 687         if (open) {
 688                 error = open(inode, f);
 689                 if (error)
 690                         goto cleanup_all;
 691         }
 692         ima_counts_get(f);
 693
 694         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 695
 696         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 697
 698         /* NB: we're sure to have correct a_ops only after f_op->open */
 699         if (f->f_flags & O_DIRECT) {
 700                 if (!f->f_mapping->a_ops ||
 701                     ((!f->f_mapping->a_ops->direct_IO) &&
 702                     (!f->f_mapping->a_ops->get_xip_mem))) {
 703                         fput(f);
 704                         f = ERR_PTR(-EINVAL);
 705                 }
 706         }
 707
 708         return f;
 709
 710 cleanup_all:
 711         fops_put(f->f_op);
 712         if (f->f_mode & FMODE_WRITE) {
 713                 put_write_access(inode);
 714                 if (!special_file(inode->i_mode)) {
 715                         /*
 716                          * We don't consider this a real
 717                          * mnt_want/drop_write() pair
 718                          * because it all happenend right
 719                          * here, so just reset the state.
 720                          */
 721                         file_reset_write(f);
 722                         mnt_drop_write(mnt);
 723                 }
 724         }
 725         file_kill(f);
 726         f->f_path.dentry = NULL;
 727         f->f_path.mnt = NULL;
 728 cleanup_file:
 729         put_filp(f);
 730         dput(dentry);
 731         mntput(mnt);
 732         return ERR_PTR(error);
 733 }
 734
 735 /**
 736  * lookup_instantiate_filp - instantiates the open intent filp
 737  * @nd: pointer to nameidata
 738  * @dentry: pointer to dentry
 739  * @open: open callback
 740  *
 741  * Helper for filesystems that want to use lookup open intents and pass back
 742  * a fully instantiated struct file to the caller.
 743  * This function is meant to be called from within a filesystem's
 744  * lookup method.
 745  * Beware of calling it for non-regular files! Those ->open methods might block
 746  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 747  * leading to a deadlock, as nobody can open that fifo anymore, because
 748  * another process to open fifo will block on locked parent when doing lookup).
 749  * Note that in case of error, nd->intent.open.file is destroyed, but the
 750  * path information remains valid.
 751  * If the open callback is set to NULL, then the standard f_op->open()
 752  * filesystem callback is substituted.
 753  */
 754 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 755                 int (*open)(struct inode *, struct file *))
 756 {
 757         const struct cred *cred = current_cred();
 758
 759         if (IS_ERR(nd->intent.open.file))
 760                 goto out;
 761         if (IS_ERR(dentry))
 762                 goto out_err;
 763         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 764                                              nd->intent.open.file,
 765                                              open, cred);
 766 out:
 767         return nd->intent.open.file;
 768 out_err:
 769         release_open_intent(nd);
 770         nd->intent.open.file = (struct file *)dentry;
 771         goto out;
 772 }
 773 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 774
 775 /**
 776  * nameidata_to_filp - convert a nameidata to an open filp.
 777  * @nd: pointer to nameidata
 778  * @flags: open flags
 779  *
 780  * Note that this function destroys the original nameidata
 781  */
 782 struct file *nameidata_to_filp(struct nameidata *nd)
 783 {
 784         const struct cred *cred = current_cred();
 785         struct file *filp;
 786
 787         /* Pick up the filp from the open intent */
 788         filp = nd->intent.open.file;
 789         /* Has the filesystem initialised the file for us? */
 790         if (filp->f_path.dentry == NULL)
 791                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 792                                      NULL, cred);
 793         else
 794                 path_put(&nd->path);
 795         return filp;
 796 }
 797
 798 /*
 799  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 800  * error.
 801  */
 802 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 803                          const struct cred *cred)
 804 {
 805         int error;
 806         struct file *f;
 807
 808         validate_creds(cred);
 809
 810         /*
 811          * We must always pass in a valid mount pointer.   Historically
 812          * callers got away with not passing it, but we must enforce this at
 813          * the earliest possible point now to avoid strange problems deep in the
 814          * filesystem stack.
 815          */
 816         if (!mnt) {
 817                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 818                 dump_stack();
 819                 return ERR_PTR(-EINVAL);
 820         }
 821
 822         error = -ENFILE;
 823         f = get_empty_filp();
 824         if (f == NULL) {
 825                 dput(dentry);
 826                 mntput(mnt);
 827                 return ERR_PTR(error);
 828         }
 829
 830         f->f_flags = flags;
 831         return __dentry_open(dentry, mnt, f, NULL, cred);
 832 }
 833 EXPORT_SYMBOL(dentry_open);
 834
 835 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 836 {
 837         struct fdtable *fdt = files_fdtable(files);
 838         __FD_CLR(fd, fdt->open_fds);
 839         if (fd < files->next_fd)
 840                 files->next_fd = fd;
 841 }
 842
 843 void put_unused_fd(unsigned int fd)
 844 {
 845         struct files_struct *files = current->files;
 846         spin_lock(&files->file_lock);
 847         __put_unused_fd(files, fd);
 848         spin_unlock(&files->file_lock);
 849 }
 850
 851 EXPORT_SYMBOL(put_unused_fd);
 852
 853 /*
 854  * Install a file pointer in the fd array.
 855  *
 856  * The VFS is full of places where we drop the files lock between
 857  * setting the open_fds bitmap and installing the file in the file
 858  * array.  At any such point, we are vulnerable to a dup2() race
 859  * installing a file in the array before us.  We need to detect this and
 860  * fput() the struct file we are about to overwrite in this case.
 861  *
 862  * It should never happen - if we allow dup2() do it, _really_ bad things
 863  * will follow.
 864  */
 865
 866 void fd_install(unsigned int fd, struct file *file)
 867 {
 868         struct files_struct *files = current->files;
 869         struct fdtable *fdt;
 870         spin_lock(&files->file_lock);
 871         fdt = files_fdtable(files);
 872         BUG_ON(fdt->fd[fd] != NULL);
 873         rcu_assign_pointer(fdt->fd[fd], file);
 874         spin_unlock(&files->file_lock);
 875 }
 876
 877 EXPORT_SYMBOL(fd_install);
 878
 879 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 880 {
 881         char *tmp = getname(filename);
 882         int fd = PTR_ERR(tmp);
 883
 884         if (!IS_ERR(tmp)) {
 885                 fd = get_unused_fd_flags(flags);
 886                 if (fd >= 0) {
 887                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
 888                         if (IS_ERR(f)) {
 889                                 put_unused_fd(fd);
 890                                 fd = PTR_ERR(f);
 891                         } else {
 892                                 fsnotify_open(f);
 893                                 fd_install(fd, f);
 894                         }
 895                 }
 896                 putname(tmp);
 897         }
 898         return fd;
 899 }
 900
 901 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 902 {
 903         long ret;
 904
 905         if (force_o_largefile())
 906                 flags |= O_LARGEFILE;
 907
 908         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 909         /* avoid REGPARM breakage on x86: */
 910         asmlinkage_protect(3, ret, filename, flags, mode);
 911         return ret;
 912 }
 913
 914 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 915                 int, mode)
 916 {
 917         long ret;
 918
 919         if (force_o_largefile())
 920                 flags |= O_LARGEFILE;
 921
 922         ret = do_sys_open(dfd, filename, flags, mode);
 923         /* avoid REGPARM breakage on x86: */
 924         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
 925         return ret;
 926 }
 927
 928 #ifndef __alpha__
 929
 930 /*
 931  * For backward compatibility?  Maybe this should be moved
 932  * into arch/i386 instead?
 933  */
 934 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
 935 {
 936         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 937 }
 938
 939 #endif
 940
 941 /*
 942  * "id" is the POSIX thread ID. We use the
 943  * files pointer for this..
 944  */
 945 int filp_close(struct file *filp, fl_owner_t id)
 946 {
 947         int retval = 0;
 948
 949         if (!file_count(filp)) {
 950                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 951                 return 0;
 952         }
 953
 954         if (filp->f_op && filp->f_op->flush)
 955                 retval = filp->f_op->flush(filp, id);
 956
 957         dnotify_flush(filp, id);
 958         locks_remove_posix(filp, id);
 959         fput(filp);
 960         return retval;
 961 }
 962
 963 EXPORT_SYMBOL(filp_close);
 964
 965 /*
 966  * Careful here! We test whether the file pointer is NULL before
 967  * releasing the fd. This ensures that one clone task can't release
 968  * an fd while another clone is opening it.
 969  */
 970 SYSCALL_DEFINE1(close, unsigned int, fd)
 971 {
 972         struct file * filp;
 973         struct files_struct *files = current->files;
 974         struct fdtable *fdt;
 975         int retval;
 976
 977         spin_lock(&files->file_lock);
 978         fdt = files_fdtable(files);
 979         if (fd >= fdt->max_fds)
 980                 goto out_unlock;
 981         filp = fdt->fd[fd];
 982         if (!filp)
 983                 goto out_unlock;
 984         rcu_assign_pointer(fdt->fd[fd], NULL);
 985         FD_CLR(fd, fdt->close_on_exec);
 986         __put_unused_fd(files, fd);
 987         spin_unlock(&files->file_lock);
 988         retval = filp_close(filp, files);
 989
 990         /* can't restart close syscall because file table entry was cleared */
 991         if (unlikely(retval == -ERESTARTSYS ||
 992                      retval == -ERESTARTNOINTR ||
 993                      retval == -ERESTARTNOHAND ||
 994                      retval == -ERESTART_RESTARTBLOCK))
 995                 retval = -EINTR;
 996
 997         return retval;
 998
 999 out_unlock:
1000         spin_unlock(&files->file_lock);
1001         return -EBADF;
1002 }
1003 EXPORT_SYMBOL(sys_close);
1004
1005 /*
1006  * This routine simulates a hangup on the tty, to arrange that users
1007  * are given clean terminals at login time.
1008  */
1009 SYSCALL_DEFINE0(vhangup)
1010 {
1011         if (capable(CAP_SYS_TTY_CONFIG)) {
1012                 tty_vhangup_self();
1013                 return 0;
1014         }
1015         return -EPERM;
1016 }
1017
1018 /*
1019  * Called when an inode is about to be open.
1020  * We use this to disallow opening large files on 32bit systems if
1021  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1022  * on this flag in sys_open.
1023  */
1024 int generic_file_open(struct inode * inode, struct file * filp)
1025 {
1026         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1027                 return -EOVERFLOW;
1028         return 0;
1029 }
1030
1031 EXPORT_SYMBOL(generic_file_open);
1032
1033 /*
1034  * This is used by subsystems that don't want seekable
1035  * file descriptors
1036  */
1037 int nonseekable_open(struct inode *inode, struct file *filp)
1038 {
1039         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1040         return 0;
1041 }
1042
1043 EXPORT_SYMBOL(nonseekable_open);