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 EXPORT_SYMBOL(do_truncate);
  64
  65 static long do_sys_truncate(const char __user *pathname, loff_t length)
  66 {
  67         struct path path;
  68         struct inode *inode;
  69         int error;
  70
  71         error = -EINVAL;
  72         if (length < 0) /* sorry, but loff_t says... */
  73                 goto out;
  74
  75         error = user_path(pathname, &path);
  76         if (error)
  77                 goto out;
  78         inode = path.dentry->d_inode;
  79
  80         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  81         error = -EISDIR;
  82         if (S_ISDIR(inode->i_mode))
  83                 goto dput_and_out;
  84
  85         error = -EINVAL;
  86         if (!S_ISREG(inode->i_mode))
  87                 goto dput_and_out;
  88
  89         error = mnt_want_write(path.mnt);
  90         if (error)
  91                 goto dput_and_out;
  92
  93         error = inode_permission(inode, MAY_WRITE);
  94         if (error)
  95                 goto mnt_drop_write_and_out;
  96
  97         error = -EPERM;
  98         if (IS_APPEND(inode))
  99                 goto mnt_drop_write_and_out;
 100
 101         error = get_write_access(inode);
 102         if (error)
 103                 goto mnt_drop_write_and_out;
 104
 105         /*
 106          * Make sure that there are no leases.  get_write_access() protects
 107          * against the truncate racing with a lease-granting setlease().
 108          */
 109         error = break_lease(inode, O_WRONLY);
 110         if (error)
 111                 goto put_write_and_out;
 112
 113         error = locks_verify_truncate(inode, NULL, length);
 114         if (!error)
 115                 error = security_path_truncate(&path);
 116         if (!error)
 117                 error = do_truncate(path.dentry, length, 0, NULL);
 118
 119 put_write_and_out:
 120         put_write_access(inode);
 121 mnt_drop_write_and_out:
 122         mnt_drop_write(path.mnt);
 123 dput_and_out:
 124         path_put(&path);
 125 out:
 126         return error;
 127 }
 128
 129 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 130 {
 131         return do_sys_truncate(path, length);
 132 }
 133
 134 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 135 {
 136         struct inode * inode;
 137         struct dentry *dentry;
 138         struct file * file;
 139         int error;
 140
 141         error = -EINVAL;
 142         if (length < 0)
 143                 goto out;
 144         error = -EBADF;
 145         file = fget(fd);
 146         if (!file)
 147                 goto out;
 148
 149         /* explicitly opened as large or we are on 64-bit box */
 150         if (file->f_flags & O_LARGEFILE)
 151                 small = 0;
 152
 153         dentry = file->f_path.dentry;
 154         inode = dentry->d_inode;
 155         error = -EINVAL;
 156         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 157                 goto out_putf;
 158
 159         error = -EINVAL;
 160         /* Cannot ftruncate over 2^31 bytes without large file support */
 161         if (small && length > MAX_NON_LFS)
 162                 goto out_putf;
 163
 164         error = -EPERM;
 165         if (IS_APPEND(inode))
 166                 goto out_putf;
 167
 168         error = locks_verify_truncate(inode, file, length);
 169         if (!error)
 170                 error = security_path_truncate(&file->f_path);
 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 & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 228                 return -EOPNOTSUPP;
 229
 230         /* Punch hole must have keep size set */
 231         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 232             !(mode & FALLOC_FL_KEEP_SIZE))
 233                 return -EOPNOTSUPP;
 234
 235         if (!(file->f_mode & FMODE_WRITE))
 236                 return -EBADF;
 237
 238         /* It's not possible punch hole on append only file */
 239         if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 240                 return -EPERM;
 241
 242         if (IS_IMMUTABLE(inode))
 243                 return -EPERM;
 244
 245         /*
 246          * Revalidate the write permissions, in case security policy has
 247          * changed since the files were opened.
 248          */
 249         ret = security_file_permission(file, MAY_WRITE);
 250         if (ret)
 251                 return ret;
 252
 253         if (S_ISFIFO(inode->i_mode))
 254                 return -ESPIPE;
 255
 256         /*
 257          * Let individual file system decide if it supports preallocation
 258          * for directories or not.
 259          */
 260         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 261                 return -ENODEV;
 262
 263         /* Check for wrap through zero too */
 264         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 265                 return -EFBIG;
 266
 267         if (!file->f_op->fallocate)
 268                 return -EOPNOTSUPP;
 269
 270         return file->f_op->fallocate(file, mode, offset, len);
 271 }
 272
 273 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 274 {
 275         struct file *file;
 276         int error = -EBADF;
 277
 278         file = fget(fd);
 279         if (file) {
 280                 error = do_fallocate(file, mode, offset, len);
 281                 fput(file);
 282         }
 283
 284         return error;
 285 }
 286
 287 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 288 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 289 {
 290         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 291 }
 292 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 293 #endif
 294
 295 /*
 296  * access() needs to use the real uid/gid, not the effective uid/gid.
 297  * We do this by temporarily clearing all FS-related capabilities and
 298  * switching the fsuid/fsgid around to the real ones.
 299  */
 300 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 301 {
 302         const struct cred *old_cred;
 303         struct cred *override_cred;
 304         struct path path;
 305         struct inode *inode;
 306         int res;
 307
 308         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 309                 return -EINVAL;
 310
 311         override_cred = prepare_creds();
 312         if (!override_cred)
 313                 return -ENOMEM;
 314
 315         override_cred->fsuid = override_cred->uid;
 316         override_cred->fsgid = override_cred->gid;
 317
 318         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 319                 /* Clear the capabilities if we switch to a non-root user */
 320                 if (override_cred->uid)
 321                         cap_clear(override_cred->cap_effective);
 322                 else
 323                         override_cred->cap_effective =
 324                                 override_cred->cap_permitted;
 325         }
 326
 327         old_cred = override_creds(override_cred);
 328
 329         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 330         if (res)
 331                 goto out;
 332
 333         inode = path.dentry->d_inode;
 334
 335         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 336                 /*
 337                  * MAY_EXEC on regular files is denied if the fs is mounted
 338                  * with the "noexec" flag.
 339                  */
 340                 res = -EACCES;
 341                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 342                         goto out_path_release;
 343         }
 344
 345         res = inode_permission(inode, mode | MAY_ACCESS);
 346         /* SuS v2 requires we report a read only fs too */
 347         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 348                 goto out_path_release;
 349         /*
 350          * This is a rare case where using __mnt_is_readonly()
 351          * is OK without a mnt_want/drop_write() pair.  Since
 352          * no actual write to the fs is performed here, we do
 353          * not need to telegraph to that to anyone.
 354          *
 355          * By doing this, we accept that this access is
 356          * inherently racy and know that the fs may change
 357          * state before we even see this result.
 358          */
 359         if (__mnt_is_readonly(path.mnt))
 360                 res = -EROFS;
 361
 362 out_path_release:
 363         path_put(&path);
 364 out:
 365         revert_creds(old_cred);
 366         put_cred(override_cred);
 367         return res;
 368 }
 369
 370 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 371 {
 372         return sys_faccessat(AT_FDCWD, filename, mode);
 373 }
 374
 375 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 376 {
 377         struct path path;
 378         int error;
 379
 380         error = user_path_dir(filename, &path);
 381         if (error)
 382                 goto out;
 383
 384         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 385         if (error)
 386                 goto dput_and_out;
 387
 388         set_fs_pwd(current->fs, &path);
 389
 390 dput_and_out:
 391         path_put(&path);
 392 out:
 393         return error;
 394 }
 395
 396 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 397 {
 398         struct file *file;
 399         struct inode *inode;
 400         int error;
 401
 402         error = -EBADF;
 403         file = fget(fd);
 404         if (!file)
 405                 goto out;
 406
 407         inode = file->f_path.dentry->d_inode;
 408
 409         error = -ENOTDIR;
 410         if (!S_ISDIR(inode->i_mode))
 411                 goto out_putf;
 412
 413         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 414         if (!error)
 415                 set_fs_pwd(current->fs, &file->f_path);
 416 out_putf:
 417         fput(file);
 418 out:
 419         return error;
 420 }
 421
 422 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 423 {
 424         struct path path;
 425         int error;
 426
 427         error = user_path_dir(filename, &path);
 428         if (error)
 429                 goto out;
 430
 431         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 432         if (error)
 433                 goto dput_and_out;
 434
 435         error = -EPERM;
 436         if (!capable(CAP_SYS_CHROOT))
 437                 goto dput_and_out;
 438         error = security_path_chroot(&path);
 439         if (error)
 440                 goto dput_and_out;
 441
 442         set_fs_root(current->fs, &path);
 443         error = 0;
 444 dput_and_out:
 445         path_put(&path);
 446 out:
 447         return error;
 448 }
 449
 450 static int chmod_common(struct path *path, umode_t mode)
 451 {
 452         struct inode *inode = path->dentry->d_inode;
 453         struct iattr newattrs;
 454         int error;
 455
 456         error = mnt_want_write(path->mnt);
 457         if (error)
 458                 return error;
 459         mutex_lock(&inode->i_mutex);
 460         error = security_path_chmod(path, mode);
 461         if (error)
 462                 goto out_unlock;
 463         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 464         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 465         error = notify_change(path->dentry, &newattrs);
 466 out_unlock:
 467         mutex_unlock(&inode->i_mutex);
 468         mnt_drop_write(path->mnt);
 469         return error;
 470 }
 471
 472 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 473 {
 474         struct file * file;
 475         int err = -EBADF;
 476
 477         file = fget(fd);
 478         if (file) {
 479                 audit_inode(NULL, file->f_path.dentry);
 480                 err = chmod_common(&file->f_path, mode);
 481                 fput(file);
 482         }
 483         return err;
 484 }
 485
 486 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 487 {
 488         struct path path;
 489         int error;
 490
 491         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 492         if (!error) {
 493                 error = chmod_common(&path, mode);
 494                 path_put(&path);
 495         }
 496         return error;
 497 }
 498
 499 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 500 {
 501         return sys_fchmodat(AT_FDCWD, filename, mode);
 502 }
 503
 504 static int chown_common(struct path *path, uid_t user, gid_t group)
 505 {
 506         struct inode *inode = path->dentry->d_inode;
 507         int error;
 508         struct iattr newattrs;
 509
 510         newattrs.ia_valid =  ATTR_CTIME;
 511         if (user != (uid_t) -1) {
 512                 newattrs.ia_valid |= ATTR_UID;
 513                 newattrs.ia_uid = user;
 514         }
 515         if (group != (gid_t) -1) {
 516                 newattrs.ia_valid |= ATTR_GID;
 517                 newattrs.ia_gid = group;
 518         }
 519         if (!S_ISDIR(inode->i_mode))
 520                 newattrs.ia_valid |=
 521                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 522         mutex_lock(&inode->i_mutex);
 523         error = security_path_chown(path, user, group);
 524         if (!error)
 525                 error = notify_change(path->dentry, &newattrs);
 526         mutex_unlock(&inode->i_mutex);
 527
 528         return error;
 529 }
 530
 531 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 532 {
 533         struct path path;
 534         int error;
 535
 536         error = user_path(filename, &path);
 537         if (error)
 538                 goto out;
 539         error = mnt_want_write(path.mnt);
 540         if (error)
 541                 goto out_release;
 542         error = chown_common(&path, user, group);
 543         mnt_drop_write(path.mnt);
 544 out_release:
 545         path_put(&path);
 546 out:
 547         return error;
 548 }
 549
 550 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 551                 gid_t, group, int, flag)
 552 {
 553         struct path path;
 554         int error = -EINVAL;
 555         int lookup_flags;
 556
 557         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 558                 goto out;
 559
 560         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 561         if (flag & AT_EMPTY_PATH)
 562                 lookup_flags |= LOOKUP_EMPTY;
 563         error = user_path_at(dfd, filename, lookup_flags, &path);
 564         if (error)
 565                 goto out;
 566         error = mnt_want_write(path.mnt);
 567         if (error)
 568                 goto out_release;
 569         error = chown_common(&path, user, group);
 570         mnt_drop_write(path.mnt);
 571 out_release:
 572         path_put(&path);
 573 out:
 574         return error;
 575 }
 576
 577 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 578 {
 579         struct path path;
 580         int error;
 581
 582         error = user_lpath(filename, &path);
 583         if (error)
 584                 goto out;
 585         error = mnt_want_write(path.mnt);
 586         if (error)
 587                 goto out_release;
 588         error = chown_common(&path, user, group);
 589         mnt_drop_write(path.mnt);
 590 out_release:
 591         path_put(&path);
 592 out:
 593         return error;
 594 }
 595
 596 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 597 {
 598         struct file * file;
 599         int error = -EBADF;
 600         struct dentry * dentry;
 601
 602         file = fget(fd);
 603         if (!file)
 604                 goto out;
 605
 606         error = mnt_want_write_file(file);
 607         if (error)
 608                 goto out_fput;
 609         dentry = file->f_path.dentry;
 610         audit_inode(NULL, dentry);
 611         error = chown_common(&file->f_path, user, group);
 612         mnt_drop_write_file(file);
 613 out_fput:
 614         fput(file);
 615 out:
 616         return error;
 617 }
 618
 619 /*
 620  * You have to be very careful that these write
 621  * counts get cleaned up in error cases and
 622  * upon __fput().  This should probably never
 623  * be called outside of __dentry_open().
 624  */
 625 static inline int __get_file_write_access(struct inode *inode,
 626                                           struct vfsmount *mnt)
 627 {
 628         int error;
 629         error = get_write_access(inode);
 630         if (error)
 631                 return error;
 632         /*
 633          * Do not take mount writer counts on
 634          * special files since no writes to
 635          * the mount itself will occur.
 636          */
 637         if (!special_file(inode->i_mode)) {
 638                 /*
 639                  * Balanced in __fput()
 640                  */
 641                 error = mnt_want_write(mnt);
 642                 if (error)
 643                         put_write_access(inode);
 644         }
 645         return error;
 646 }
 647
 648 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 649                                         struct file *f,
 650                                         int (*open)(struct inode *, struct file *),
 651                                         const struct cred *cred)
 652 {
 653         static const struct file_operations empty_fops = {};
 654         struct inode *inode;
 655         int error;
 656
 657         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 658                                 FMODE_PREAD | FMODE_PWRITE;
 659
 660         if (unlikely(f->f_flags & O_PATH))
 661                 f->f_mode = FMODE_PATH;
 662
 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         file_sb_list_add(f, inode->i_sb);
 677
 678         if (unlikely(f->f_mode & FMODE_PATH)) {
 679                 f->f_op = &empty_fops;
 680                 return f;
 681         }
 682
 683         f->f_op = fops_get(inode->i_fop);
 684
 685         error = security_dentry_open(f, cred);
 686         if (error)
 687                 goto cleanup_all;
 688
 689         error = break_lease(inode, f->f_flags);
 690         if (error)
 691                 goto cleanup_all;
 692
 693         if (!open && f->f_op)
 694                 open = f->f_op->open;
 695         if (open) {
 696                 error = open(inode, f);
 697                 if (error)
 698                         goto cleanup_all;
 699         }
 700         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 701                 i_readcount_inc(inode);
 702
 703         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 704
 705         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 706
 707         /* NB: we're sure to have correct a_ops only after f_op->open */
 708         if (f->f_flags & O_DIRECT) {
 709                 if (!f->f_mapping->a_ops ||
 710                     ((!f->f_mapping->a_ops->direct_IO) &&
 711                     (!f->f_mapping->a_ops->get_xip_mem))) {
 712                         fput(f);
 713                         f = ERR_PTR(-EINVAL);
 714                 }
 715         }
 716
 717         return f;
 718
 719 cleanup_all:
 720         fops_put(f->f_op);
 721         if (f->f_mode & FMODE_WRITE) {
 722                 put_write_access(inode);
 723                 if (!special_file(inode->i_mode)) {
 724                         /*
 725                          * We don't consider this a real
 726                          * mnt_want/drop_write() pair
 727                          * because it all happenend right
 728                          * here, so just reset the state.
 729                          */
 730                         file_reset_write(f);
 731                         mnt_drop_write(mnt);
 732                 }
 733         }
 734         file_sb_list_del(f);
 735         f->f_path.dentry = NULL;
 736         f->f_path.mnt = NULL;
 737 cleanup_file:
 738         put_filp(f);
 739         dput(dentry);
 740         mntput(mnt);
 741         return ERR_PTR(error);
 742 }
 743
 744 /**
 745  * lookup_instantiate_filp - instantiates the open intent filp
 746  * @nd: pointer to nameidata
 747  * @dentry: pointer to dentry
 748  * @open: open callback
 749  *
 750  * Helper for filesystems that want to use lookup open intents and pass back
 751  * a fully instantiated struct file to the caller.
 752  * This function is meant to be called from within a filesystem's
 753  * lookup method.
 754  * Beware of calling it for non-regular files! Those ->open methods might block
 755  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 756  * leading to a deadlock, as nobody can open that fifo anymore, because
 757  * another process to open fifo will block on locked parent when doing lookup).
 758  * Note that in case of error, nd->intent.open.file is destroyed, but the
 759  * path information remains valid.
 760  * If the open callback is set to NULL, then the standard f_op->open()
 761  * filesystem callback is substituted.
 762  */
 763 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 764                 int (*open)(struct inode *, struct file *))
 765 {
 766         const struct cred *cred = current_cred();
 767
 768         if (IS_ERR(nd->intent.open.file))
 769                 goto out;
 770         if (IS_ERR(dentry))
 771                 goto out_err;
 772         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 773                                              nd->intent.open.file,
 774                                              open, cred);
 775 out:
 776         return nd->intent.open.file;
 777 out_err:
 778         release_open_intent(nd);
 779         nd->intent.open.file = ERR_CAST(dentry);
 780         goto out;
 781 }
 782 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 783
 784 /**
 785  * nameidata_to_filp - convert a nameidata to an open filp.
 786  * @nd: pointer to nameidata
 787  * @flags: open flags
 788  *
 789  * Note that this function destroys the original nameidata
 790  */
 791 struct file *nameidata_to_filp(struct nameidata *nd)
 792 {
 793         const struct cred *cred = current_cred();
 794         struct file *filp;
 795
 796         /* Pick up the filp from the open intent */
 797         filp = nd->intent.open.file;
 798         nd->intent.open.file = NULL;
 799
 800         /* Has the filesystem initialised the file for us? */
 801         if (filp->f_path.dentry == NULL) {
 802                 path_get(&nd->path);
 803                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 804                                      NULL, cred);
 805         }
 806         return filp;
 807 }
 808
 809 /*
 810  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 811  * error.
 812  */
 813 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 814                          const struct cred *cred)
 815 {
 816         int error;
 817         struct file *f;
 818
 819         validate_creds(cred);
 820
 821         /* We must always pass in a valid mount pointer. */
 822         BUG_ON(!mnt);
 823
 824         error = -ENFILE;
 825         f = get_empty_filp();
 826         if (f == NULL) {
 827                 dput(dentry);
 828                 mntput(mnt);
 829                 return ERR_PTR(error);
 830         }
 831
 832         f->f_flags = flags;
 833         return __dentry_open(dentry, mnt, f, NULL, cred);
 834 }
 835 EXPORT_SYMBOL(dentry_open);
 836
 837 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 838 {
 839         struct fdtable *fdt = files_fdtable(files);
 840         __FD_CLR(fd, fdt->open_fds);
 841         if (fd < files->next_fd)
 842                 files->next_fd = fd;
 843 }
 844
 845 void put_unused_fd(unsigned int fd)
 846 {
 847         struct files_struct *files = current->files;
 848         spin_lock(&files->file_lock);
 849         __put_unused_fd(files, fd);
 850         spin_unlock(&files->file_lock);
 851 }
 852
 853 EXPORT_SYMBOL(put_unused_fd);
 854
 855 /*
 856  * Install a file pointer in the fd array.
 857  *
 858  * The VFS is full of places where we drop the files lock between
 859  * setting the open_fds bitmap and installing the file in the file
 860  * array.  At any such point, we are vulnerable to a dup2() race
 861  * installing a file in the array before us.  We need to detect this and
 862  * fput() the struct file we are about to overwrite in this case.
 863  *
 864  * It should never happen - if we allow dup2() do it, _really_ bad things
 865  * will follow.
 866  */
 867
 868 void fd_install(unsigned int fd, struct file *file)
 869 {
 870         struct files_struct *files = current->files;
 871         struct fdtable *fdt;
 872         spin_lock(&files->file_lock);
 873         fdt = files_fdtable(files);
 874         BUG_ON(fdt->fd[fd] != NULL);
 875         rcu_assign_pointer(fdt->fd[fd], file);
 876         spin_unlock(&files->file_lock);
 877 }
 878
 879 EXPORT_SYMBOL(fd_install);
 880
 881 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 882 {
 883         int lookup_flags = 0;
 884         int acc_mode;
 885
 886         if (!(flags & O_CREAT))
 887                 mode = 0;
 888         op->mode = mode;
 889
 890         /* Must never be set by userspace */
 891         flags &= ~FMODE_NONOTIFY;
 892
 893         /*
 894          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 895          * check for O_DSYNC if the need any syncing at all we enforce it's
 896          * always set instead of having to deal with possibly weird behaviour
 897          * for malicious applications setting only __O_SYNC.
 898          */
 899         if (flags & __O_SYNC)
 900                 flags |= O_DSYNC;
 901
 902         /*
 903          * If we have O_PATH in the open flag. Then we
 904          * cannot have anything other than the below set of flags
 905          */
 906         if (flags & O_PATH) {
 907                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 908                 acc_mode = 0;
 909         } else {
 910                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 911         }
 912
 913         op->open_flag = flags;
 914
 915         /* O_TRUNC implies we need access checks for write permissions */
 916         if (flags & O_TRUNC)
 917                 acc_mode |= MAY_WRITE;
 918
 919         /* Allow the LSM permission hook to distinguish append
 920            access from general write access. */
 921         if (flags & O_APPEND)
 922                 acc_mode |= MAY_APPEND;
 923
 924         op->acc_mode = acc_mode;
 925
 926         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 927
 928         if (flags & O_CREAT) {
 929                 op->intent |= LOOKUP_CREATE;
 930                 if (flags & O_EXCL)
 931                         op->intent |= LOOKUP_EXCL;
 932         }
 933
 934         if (flags & O_DIRECTORY)
 935                 lookup_flags |= LOOKUP_DIRECTORY;
 936         if (!(flags & O_NOFOLLOW))
 937                 lookup_flags |= LOOKUP_FOLLOW;
 938         return lookup_flags;
 939 }
 940
 941 /**
 942  * filp_open - open file and return file pointer
 943  *
 944  * @filename:   path to open
 945  * @flags:      open flags as per the open(2) second argument
 946  * @mode:       mode for the new file if O_CREAT is set, else ignored
 947  *
 948  * This is the helper to open a file from kernelspace if you really
 949  * have to.  But in generally you should not do this, so please move
 950  * along, nothing to see here..
 951  */
 952 struct file *filp_open(const char *filename, int flags, umode_t mode)
 953 {
 954         struct open_flags op;
 955         int lookup = build_open_flags(flags, mode, &op);
 956         return do_filp_open(AT_FDCWD, filename, &op, lookup);
 957 }
 958 EXPORT_SYMBOL(filp_open);
 959
 960 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 961                             const char *filename, int flags)
 962 {
 963         struct open_flags op;
 964         int lookup = build_open_flags(flags, 0, &op);
 965         if (flags & O_CREAT)
 966                 return ERR_PTR(-EINVAL);
 967         if (!filename && (flags & O_DIRECTORY))
 968                 if (!dentry->d_inode->i_op->lookup)
 969                         return ERR_PTR(-ENOTDIR);
 970         return do_file_open_root(dentry, mnt, filename, &op, lookup);
 971 }
 972 EXPORT_SYMBOL(file_open_root);
 973
 974 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 975 {
 976         struct open_flags op;
 977         int lookup = build_open_flags(flags, mode, &op);
 978         char *tmp = getname(filename);
 979         int fd = PTR_ERR(tmp);
 980
 981         if (!IS_ERR(tmp)) {
 982                 fd = get_unused_fd_flags(flags);
 983                 if (fd >= 0) {
 984                         struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 985                         if (IS_ERR(f)) {
 986                                 put_unused_fd(fd);
 987                                 fd = PTR_ERR(f);
 988                         } else {
 989                                 fsnotify_open(f);
 990                                 fd_install(fd, f);
 991                         }
 992                 }
 993                 putname(tmp);
 994         }
 995         return fd;
 996 }
 997
 998 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 999 {
1000         long ret;
1001
1002         if (force_o_largefile())
1003                 flags |= O_LARGEFILE;
1004
1005         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1006         /* avoid REGPARM breakage on x86: */
1007         asmlinkage_protect(3, ret, filename, flags, mode);
1008         return ret;
1009 }
1010
1011 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1012                 umode_t, mode)
1013 {
1014         long ret;
1015
1016         if (force_o_largefile())
1017                 flags |= O_LARGEFILE;
1018
1019         ret = do_sys_open(dfd, filename, flags, mode);
1020         /* avoid REGPARM breakage on x86: */
1021         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1022         return ret;
1023 }
1024
1025 #ifndef __alpha__
1026
1027 /*
1028  * For backward compatibility?  Maybe this should be moved
1029  * into arch/i386 instead?
1030  */
1031 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1032 {
1033         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1034 }
1035
1036 #endif
1037
1038 /*
1039  * "id" is the POSIX thread ID. We use the
1040  * files pointer for this..
1041  */
1042 int filp_close(struct file *filp, fl_owner_t id)
1043 {
1044         int retval = 0;
1045
1046         if (!file_count(filp)) {
1047                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1048                 return 0;
1049         }
1050
1051         if (filp->f_op && filp->f_op->flush)
1052                 retval = filp->f_op->flush(filp, id);
1053
1054         if (likely(!(filp->f_mode & FMODE_PATH))) {
1055                 dnotify_flush(filp, id);
1056                 locks_remove_posix(filp, id);
1057         }
1058         fput(filp);
1059         return retval;
1060 }
1061
1062 EXPORT_SYMBOL(filp_close);
1063
1064 /*
1065  * Careful here! We test whether the file pointer is NULL before
1066  * releasing the fd. This ensures that one clone task can't release
1067  * an fd while another clone is opening it.
1068  */
1069 SYSCALL_DEFINE1(close, unsigned int, fd)
1070 {
1071         struct file * filp;
1072         struct files_struct *files = current->files;
1073         struct fdtable *fdt;
1074         int retval;
1075
1076         spin_lock(&files->file_lock);
1077         fdt = files_fdtable(files);
1078         if (fd >= fdt->max_fds)
1079                 goto out_unlock;
1080         filp = fdt->fd[fd];
1081         if (!filp)
1082                 goto out_unlock;
1083         rcu_assign_pointer(fdt->fd[fd], NULL);
1084         FD_CLR(fd, fdt->close_on_exec);
1085         __put_unused_fd(files, fd);
1086         spin_unlock(&files->file_lock);
1087         retval = filp_close(filp, files);
1088
1089         /* can't restart close syscall because file table entry was cleared */
1090         if (unlikely(retval == -ERESTARTSYS ||
1091                      retval == -ERESTARTNOINTR ||
1092                      retval == -ERESTARTNOHAND ||
1093                      retval == -ERESTART_RESTARTBLOCK))
1094                 retval = -EINTR;
1095
1096         return retval;
1097
1098 out_unlock:
1099         spin_unlock(&files->file_lock);
1100         return -EBADF;
1101 }
1102 EXPORT_SYMBOL(sys_close);
1103
1104 /*
1105  * This routine simulates a hangup on the tty, to arrange that users
1106  * are given clean terminals at login time.
1107  */
1108 SYSCALL_DEFINE0(vhangup)
1109 {
1110         if (capable(CAP_SYS_TTY_CONFIG)) {
1111                 tty_vhangup_self();
1112                 return 0;
1113         }
1114         return -EPERM;
1115 }
1116
1117 /*
1118  * Called when an inode is about to be open.
1119  * We use this to disallow opening large files on 32bit systems if
1120  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1121  * on this flag in sys_open.
1122  */
1123 int generic_file_open(struct inode * inode, struct file * filp)
1124 {
1125         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1126                 return -EOVERFLOW;
1127         return 0;
1128 }
1129
1130 EXPORT_SYMBOL(generic_file_open);
1131
1132 /*
1133  * This is used by subsystems that don't want seekable
1134  * file descriptors. The function is not supposed to ever fail, the only
1135  * reason it returns an 'int' and not 'void' is so that it can be plugged
1136  * directly into file_operations structure.
1137  */
1138 int nonseekable_open(struct inode *inode, struct file *filp)
1139 {
1140         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1141         return 0;
1142 }
1143
1144 EXPORT_SYMBOL(nonseekable_open);