linux/fs/open.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/open.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/string.h>
   8#include <linux/mm.h>
   9#include <linux/file.h>
  10#include <linux/fdtable.h>
  11#include <linux/fsnotify.h>
  12#include <linux/module.h>
  13#include <linux/tty.h>
  14#include <linux/namei.h>
  15#include <linux/backing-dev.h>
  16#include <linux/capability.h>
  17#include <linux/securebits.h>
  18#include <linux/security.h>
  19#include <linux/mount.h>
  20#include <linux/fcntl.h>
  21#include <linux/slab.h>
  22#include <asm/uaccess.h>
  23#include <linux/fs.h>
  24#include <linux/personality.h>
  25#include <linux/pagemap.h>
  26#include <linux/syscalls.h>
  27#include <linux/rcupdate.h>
  28#include <linux/audit.h>
  29#include <linux/falloc.h>
  30#include <linux/fs_struct.h>
  31#include <linux/ima.h>
  32#include <linux/dnotify.h>
  33#include <linux/compat.h>
  34
  35#include "internal.h"
  36
  37int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38        struct file *filp)
  39{
  40        int ret;
  41        struct iattr newattrs;
  42
  43        /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44        if (length < 0)
  45                return -EINVAL;
  46
  47        newattrs.ia_size = length;
  48        newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49        if (filp) {
  50                newattrs.ia_file = filp;
  51                newattrs.ia_valid |= ATTR_FILE;
  52        }
  53
  54        /* Remove suid/sgid on truncate too */
  55        ret = should_remove_suid(dentry);
  56        if (ret)
  57                newattrs.ia_valid |= ret | ATTR_FORCE;
  58
  59        mutex_lock(&dentry->d_inode->i_mutex);
  60        ret = notify_change(dentry, &newattrs);
  61        mutex_unlock(&dentry->d_inode->i_mutex);
  62        return ret;
  63}
  64
  65long vfs_truncate(struct path *path, loff_t length)
  66{
  67        struct inode *inode;
  68        long error;
  69
  70        inode = path->dentry->d_inode;
  71
  72        /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  73        if (S_ISDIR(inode->i_mode))
  74                return -EISDIR;
  75        if (!S_ISREG(inode->i_mode))
  76                return -EINVAL;
  77
  78        error = mnt_want_write(path->mnt);
  79        if (error)
  80                goto out;
  81
  82        error = inode_permission(inode, MAY_WRITE);
  83        if (error)
  84                goto mnt_drop_write_and_out;
  85
  86        error = -EPERM;
  87        if (IS_APPEND(inode))
  88                goto mnt_drop_write_and_out;
  89
  90        error = get_write_access(inode);
  91        if (error)
  92                goto mnt_drop_write_and_out;
  93
  94        /*
  95         * Make sure that there are no leases.  get_write_access() protects
  96         * against the truncate racing with a lease-granting setlease().
  97         */
  98        error = break_lease(inode, O_WRONLY);
  99        if (error)
 100                goto put_write_and_out;
 101
 102        error = locks_verify_truncate(inode, NULL, length);
 103        if (!error)
 104                error = security_path_truncate(path);
 105        if (!error)
 106                error = do_truncate(path->dentry, length, 0, NULL);
 107
 108put_write_and_out:
 109        put_write_access(inode);
 110mnt_drop_write_and_out:
 111        mnt_drop_write(path->mnt);
 112out:
 113        return error;
 114}
 115EXPORT_SYMBOL_GPL(vfs_truncate);
 116
 117static long do_sys_truncate(const char __user *pathname, loff_t length)
 118{
 119        unsigned int lookup_flags = LOOKUP_FOLLOW;
 120        struct path path;
 121        int error;
 122
 123        if (length < 0) /* sorry, but loff_t says... */
 124                return -EINVAL;
 125
 126retry:
 127        error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 128        if (!error) {
 129                error = vfs_truncate(&path, length);
 130                path_put(&path);
 131        }
 132        if (retry_estale(error, lookup_flags)) {
 133                lookup_flags |= LOOKUP_REVAL;
 134                goto retry;
 135        }
 136        return error;
 137}
 138
 139SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 140{
 141        return do_sys_truncate(path, length);
 142}
 143
 144#ifdef CONFIG_COMPAT
 145COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 146{
 147        return do_sys_truncate(path, length);
 148}
 149#endif
 150
 151static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 152{
 153        struct inode *inode;
 154        struct dentry *dentry;
 155        struct fd f;
 156        int error;
 157
 158        error = -EINVAL;
 159        if (length < 0)
 160                goto out;
 161        error = -EBADF;
 162        f = fdget(fd);
 163        if (!f.file)
 164                goto out;
 165
 166        /* explicitly opened as large or we are on 64-bit box */
 167        if (f.file->f_flags & O_LARGEFILE)
 168                small = 0;
 169
 170        dentry = f.file->f_path.dentry;
 171        inode = dentry->d_inode;
 172        error = -EINVAL;
 173        if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 174                goto out_putf;
 175
 176        error = -EINVAL;
 177        /* Cannot ftruncate over 2^31 bytes without large file support */
 178        if (small && length > MAX_NON_LFS)
 179                goto out_putf;
 180
 181        error = -EPERM;
 182        if (IS_APPEND(inode))
 183                goto out_putf;
 184
 185        sb_start_write(inode->i_sb);
 186        error = locks_verify_truncate(inode, f.file, length);
 187        if (!error)
 188                error = security_path_truncate(&f.file->f_path);
 189        if (!error)
 190                error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 191        sb_end_write(inode->i_sb);
 192out_putf:
 193        fdput(f);
 194out:
 195        return error;
 196}
 197
 198SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 199{
 200        return do_sys_ftruncate(fd, length, 1);
 201}
 202
 203#ifdef CONFIG_COMPAT
 204COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 205{
 206        return do_sys_ftruncate(fd, length, 1);
 207}
 208#endif
 209
 210/* LFS versions of truncate are only needed on 32 bit machines */
 211#if BITS_PER_LONG == 32
 212SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 213{
 214        return do_sys_truncate(path, length);
 215}
 216
 217SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 218{
 219        return do_sys_ftruncate(fd, length, 0);
 220}
 221#endif /* BITS_PER_LONG == 32 */
 222
 223
 224int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 225{
 226        struct inode *inode = file_inode(file);
 227        long ret;
 228
 229        if (offset < 0 || len <= 0)
 230                return -EINVAL;
 231
 232        /* Return error if mode is not supported */
 233        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 234                return -EOPNOTSUPP;
 235
 236        /* Punch hole must have keep size set */
 237        if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 238            !(mode & FALLOC_FL_KEEP_SIZE))
 239                return -EOPNOTSUPP;
 240
 241        if (!(file->f_mode & FMODE_WRITE))
 242                return -EBADF;
 243
 244        /* It's not possible punch hole on append only file */
 245        if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 246                return -EPERM;
 247
 248        if (IS_IMMUTABLE(inode))
 249                return -EPERM;
 250
 251        /*
 252         * Revalidate the write permissions, in case security policy has
 253         * changed since the files were opened.
 254         */
 255        ret = security_file_permission(file, MAY_WRITE);
 256        if (ret)
 257                return ret;
 258
 259        if (S_ISFIFO(inode->i_mode))
 260                return -ESPIPE;
 261
 262        /*
 263         * Let individual file system decide if it supports preallocation
 264         * for directories or not.
 265         */
 266        if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 267                return -ENODEV;
 268
 269        /* Check for wrap through zero too */
 270        if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 271                return -EFBIG;
 272
 273        if (!file->f_op->fallocate)
 274                return -EOPNOTSUPP;
 275
 276        sb_start_write(inode->i_sb);
 277        ret = file->f_op->fallocate(file, mode, offset, len);
 278        sb_end_write(inode->i_sb);
 279        return ret;
 280}
 281
 282SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 283{
 284        struct fd f = fdget(fd);
 285        int error = -EBADF;
 286
 287        if (f.file) {
 288                error = do_fallocate(f.file, mode, offset, len);
 289                fdput(f);
 290        }
 291        return error;
 292}
 293
 294/*
 295 * access() needs to use the real uid/gid, not the effective uid/gid.
 296 * We do this by temporarily clearing all FS-related capabilities and
 297 * switching the fsuid/fsgid around to the real ones.
 298 */
 299SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 300{
 301        const struct cred *old_cred;
 302        struct cred *override_cred;
 303        struct path path;
 304        struct inode *inode;
 305        int res;
 306        unsigned int lookup_flags = LOOKUP_FOLLOW;
 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                kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 321                if (!uid_eq(override_cred->uid, root_uid))
 322                        cap_clear(override_cred->cap_effective);
 323                else
 324                        override_cred->cap_effective =
 325                                override_cred->cap_permitted;
 326        }
 327
 328        old_cred = override_creds(override_cred);
 329retry:
 330        res = user_path_at(dfd, filename, lookup_flags, &path);
 331        if (res)
 332                goto out;
 333
 334        inode = path.dentry->d_inode;
 335
 336        if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 337                /*
 338                 * MAY_EXEC on regular files is denied if the fs is mounted
 339                 * with the "noexec" flag.
 340                 */
 341                res = -EACCES;
 342                if (path.mnt->mnt_flags & MNT_NOEXEC)
 343                        goto out_path_release;
 344        }
 345
 346        res = inode_permission(inode, mode | MAY_ACCESS);
 347        /* SuS v2 requires we report a read only fs too */
 348        if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 349                goto out_path_release;
 350        /*
 351         * This is a rare case where using __mnt_is_readonly()
 352         * is OK without a mnt_want/drop_write() pair.  Since
 353         * no actual write to the fs is performed here, we do
 354         * not need to telegraph to that to anyone.
 355         *
 356         * By doing this, we accept that this access is
 357         * inherently racy and know that the fs may change
 358         * state before we even see this result.
 359         */
 360        if (__mnt_is_readonly(path.mnt))
 361                res = -EROFS;
 362
 363out_path_release:
 364        path_put(&path);
 365        if (retry_estale(res, lookup_flags)) {
 366                lookup_flags |= LOOKUP_REVAL;
 367                goto retry;
 368        }
 369out:
 370        revert_creds(old_cred);
 371        put_cred(override_cred);
 372        return res;
 373}
 374
 375SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 376{
 377        return sys_faccessat(AT_FDCWD, filename, mode);
 378}
 379
 380SYSCALL_DEFINE1(chdir, const char __user *, filename)
 381{
 382        struct path path;
 383        int error;
 384        unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 385retry:
 386        error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 387        if (error)
 388                goto out;
 389
 390        error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 391        if (error)
 392                goto dput_and_out;
 393
 394        set_fs_pwd(current->fs, &path);
 395
 396dput_and_out:
 397        path_put(&path);
 398        if (retry_estale(error, lookup_flags)) {
 399                lookup_flags |= LOOKUP_REVAL;
 400                goto retry;
 401        }
 402out:
 403        return error;
 404}
 405
 406SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 407{
 408        struct fd f = fdget_raw(fd);
 409        struct inode *inode;
 410        int error = -EBADF;
 411
 412        error = -EBADF;
 413        if (!f.file)
 414                goto out;
 415
 416        inode = file_inode(f.file);
 417
 418        error = -ENOTDIR;
 419        if (!S_ISDIR(inode->i_mode))
 420                goto out_putf;
 421
 422        error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 423        if (!error)
 424                set_fs_pwd(current->fs, &f.file->f_path);
 425out_putf:
 426        fdput(f);
 427out:
 428        return error;
 429}
 430
 431SYSCALL_DEFINE1(chroot, const char __user *, filename)
 432{
 433        struct path path;
 434        int error;
 435        unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 436retry:
 437        error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 438        if (error)
 439                goto out;
 440
 441        error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 442        if (error)
 443                goto dput_and_out;
 444
 445        error = -EPERM;
 446        if (!nsown_capable(CAP_SYS_CHROOT))
 447                goto dput_and_out;
 448        error = security_path_chroot(&path);
 449        if (error)
 450                goto dput_and_out;
 451
 452        set_fs_root(current->fs, &path);
 453        error = 0;
 454dput_and_out:
 455        path_put(&path);
 456        if (retry_estale(error, lookup_flags)) {
 457                lookup_flags |= LOOKUP_REVAL;
 458                goto retry;
 459        }
 460out:
 461        return error;
 462}
 463
 464static int chmod_common(struct path *path, umode_t mode)
 465{
 466        struct inode *inode = path->dentry->d_inode;
 467        struct iattr newattrs;
 468        int error;
 469
 470        error = mnt_want_write(path->mnt);
 471        if (error)
 472                return error;
 473        mutex_lock(&inode->i_mutex);
 474        error = security_path_chmod(path, mode);
 475        if (error)
 476                goto out_unlock;
 477        newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 478        newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 479        error = notify_change(path->dentry, &newattrs);
 480out_unlock:
 481        mutex_unlock(&inode->i_mutex);
 482        mnt_drop_write(path->mnt);
 483        return error;
 484}
 485
 486SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 487{
 488        struct file * file;
 489        int err = -EBADF;
 490
 491        file = fget(fd);
 492        if (file) {
 493                audit_inode(NULL, file->f_path.dentry, 0);
 494                err = chmod_common(&file->f_path, mode);
 495                fput(file);
 496        }
 497        return err;
 498}
 499
 500SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 501{
 502        struct path path;
 503        int error;
 504        unsigned int lookup_flags = LOOKUP_FOLLOW;
 505retry:
 506        error = user_path_at(dfd, filename, lookup_flags, &path);
 507        if (!error) {
 508                error = chmod_common(&path, mode);
 509                path_put(&path);
 510                if (retry_estale(error, lookup_flags)) {
 511                        lookup_flags |= LOOKUP_REVAL;
 512                        goto retry;
 513                }
 514        }
 515        return error;
 516}
 517
 518SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 519{
 520        return sys_fchmodat(AT_FDCWD, filename, mode);
 521}
 522
 523static int chown_common(struct path *path, uid_t user, gid_t group)
 524{
 525        struct inode *inode = path->dentry->d_inode;
 526        int error;
 527        struct iattr newattrs;
 528        kuid_t uid;
 529        kgid_t gid;
 530
 531        uid = make_kuid(current_user_ns(), user);
 532        gid = make_kgid(current_user_ns(), group);
 533
 534        newattrs.ia_valid =  ATTR_CTIME;
 535        if (user != (uid_t) -1) {
 536                if (!uid_valid(uid))
 537                        return -EINVAL;
 538                newattrs.ia_valid |= ATTR_UID;
 539                newattrs.ia_uid = uid;
 540        }
 541        if (group != (gid_t) -1) {
 542                if (!gid_valid(gid))
 543                        return -EINVAL;
 544                newattrs.ia_valid |= ATTR_GID;
 545                newattrs.ia_gid = gid;
 546        }
 547        if (!S_ISDIR(inode->i_mode))
 548                newattrs.ia_valid |=
 549                        ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 550        mutex_lock(&inode->i_mutex);
 551        error = security_path_chown(path, uid, gid);
 552        if (!error)
 553                error = notify_change(path->dentry, &newattrs);
 554        mutex_unlock(&inode->i_mutex);
 555
 556        return error;
 557}
 558
 559SYSCALL_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 lookup_flags;
 565
 566        if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 567                goto out;
 568
 569        lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 570        if (flag & AT_EMPTY_PATH)
 571                lookup_flags |= LOOKUP_EMPTY;
 572retry:
 573        error = user_path_at(dfd, filename, lookup_flags, &path);
 574        if (error)
 575                goto out;
 576        error = mnt_want_write(path.mnt);
 577        if (error)
 578                goto out_release;
 579        error = chown_common(&path, user, group);
 580        mnt_drop_write(path.mnt);
 581out_release:
 582        path_put(&path);
 583        if (retry_estale(error, lookup_flags)) {
 584                lookup_flags |= LOOKUP_REVAL;
 585                goto retry;
 586        }
 587out:
 588        return error;
 589}
 590
 591SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 592{
 593        return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 594}
 595
 596SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 597{
 598        return sys_fchownat(AT_FDCWD, filename, user, group,
 599                            AT_SYMLINK_NOFOLLOW);
 600}
 601
 602SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 603{
 604        struct fd f = fdget(fd);
 605        int error = -EBADF;
 606
 607        if (!f.file)
 608                goto out;
 609
 610        error = mnt_want_write_file(f.file);
 611        if (error)
 612                goto out_fput;
 613        audit_inode(NULL, f.file->f_path.dentry, 0);
 614        error = chown_common(&f.file->f_path, user, group);
 615        mnt_drop_write_file(f.file);
 616out_fput:
 617        fdput(f);
 618out:
 619        return error;
 620}
 621
 622/*
 623 * You have to be very careful that these write
 624 * counts get cleaned up in error cases and
 625 * upon __fput().  This should probably never
 626 * be called outside of __dentry_open().
 627 */
 628static inline int __get_file_write_access(struct inode *inode,
 629                                          struct vfsmount *mnt)
 630{
 631        int error;
 632        error = get_write_access(inode);
 633        if (error)
 634                return error;
 635        /*
 636         * Do not take mount writer counts on
 637         * special files since no writes to
 638         * the mount itself will occur.
 639         */
 640        if (!special_file(inode->i_mode)) {
 641                /*
 642                 * Balanced in __fput()
 643                 */
 644                error = __mnt_want_write(mnt);
 645                if (error)
 646                        put_write_access(inode);
 647        }
 648        return error;
 649}
 650
 651int open_check_o_direct(struct file *f)
 652{
 653        /* NB: we're sure to have correct a_ops only after f_op->open */
 654        if (f->f_flags & O_DIRECT) {
 655                if (!f->f_mapping->a_ops ||
 656                    ((!f->f_mapping->a_ops->direct_IO) &&
 657                    (!f->f_mapping->a_ops->get_xip_mem))) {
 658                        return -EINVAL;
 659                }
 660        }
 661        return 0;
 662}
 663
 664static int do_dentry_open(struct file *f,
 665                          int (*open)(struct inode *, struct file *),
 666                          const struct cred *cred)
 667{
 668        static const struct file_operations empty_fops = {};
 669        struct inode *inode;
 670        int error;
 671
 672        f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 673                                FMODE_PREAD | FMODE_PWRITE;
 674
 675        if (unlikely(f->f_flags & O_PATH))
 676                f->f_mode = FMODE_PATH;
 677
 678        path_get(&f->f_path);
 679        inode = f->f_inode = f->f_path.dentry->d_inode;
 680        if (f->f_mode & FMODE_WRITE) {
 681                error = __get_file_write_access(inode, f->f_path.mnt);
 682                if (error)
 683                        goto cleanup_file;
 684                if (!special_file(inode->i_mode))
 685                        file_take_write(f);
 686        }
 687
 688        f->f_mapping = inode->i_mapping;
 689        file_sb_list_add(f, inode->i_sb);
 690
 691        if (unlikely(f->f_mode & FMODE_PATH)) {
 692                f->f_op = &empty_fops;
 693                return 0;
 694        }
 695
 696        f->f_op = fops_get(inode->i_fop);
 697
 698        error = security_file_open(f, cred);
 699        if (error)
 700                goto cleanup_all;
 701
 702        error = break_lease(inode, f->f_flags);
 703        if (error)
 704                goto cleanup_all;
 705
 706        if (!open && f->f_op)
 707                open = f->f_op->open;
 708        if (open) {
 709                error = open(inode, f);
 710                if (error)
 711                        goto cleanup_all;
 712        }
 713        if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 714                i_readcount_inc(inode);
 715
 716        f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 717
 718        file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 719
 720        return 0;
 721
 722cleanup_all:
 723        fops_put(f->f_op);
 724        file_sb_list_del(f);
 725        if (f->f_mode & FMODE_WRITE) {
 726                put_write_access(inode);
 727                if (!special_file(inode->i_mode)) {
 728                        /*
 729                         * We don't consider this a real
 730                         * mnt_want/drop_write() pair
 731                         * because it all happenend right
 732                         * here, so just reset the state.
 733                         */
 734                        file_reset_write(f);
 735                        __mnt_drop_write(f->f_path.mnt);
 736                }
 737        }
 738cleanup_file:
 739        path_put(&f->f_path);
 740        f->f_path.mnt = NULL;
 741        f->f_path.dentry = NULL;
 742        f->f_inode = NULL;
 743        return error;
 744}
 745
 746/**
 747 * finish_open - finish opening a file
 748 * @od: opaque open data
 749 * @dentry: pointer to dentry
 750 * @open: open callback
 751 *
 752 * This can be used to finish opening a file passed to i_op->atomic_open().
 753 *
 754 * If the open callback is set to NULL, then the standard f_op->open()
 755 * filesystem callback is substituted.
 756 */
 757int finish_open(struct file *file, struct dentry *dentry,
 758                int (*open)(struct inode *, struct file *),
 759                int *opened)
 760{
 761        int error;
 762        BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 763
 764        file->f_path.dentry = dentry;
 765        error = do_dentry_open(file, open, current_cred());
 766        if (!error)
 767                *opened |= FILE_OPENED;
 768
 769        return error;
 770}
 771EXPORT_SYMBOL(finish_open);
 772
 773/**
 774 * finish_no_open - finish ->atomic_open() without opening the file
 775 *
 776 * @od: opaque open data
 777 * @dentry: dentry or NULL (as returned from ->lookup())
 778 *
 779 * This can be used to set the result of a successful lookup in ->atomic_open().
 780 * The filesystem's atomic_open() method shall return NULL after calling this.
 781 */
 782int finish_no_open(struct file *file, struct dentry *dentry)
 783{
 784        file->f_path.dentry = dentry;
 785        return 1;
 786}
 787EXPORT_SYMBOL(finish_no_open);
 788
 789struct file *dentry_open(const struct path *path, int flags,
 790                         const struct cred *cred)
 791{
 792        int error;
 793        struct file *f;
 794
 795        validate_creds(cred);
 796
 797        /* We must always pass in a valid mount pointer. */
 798        BUG_ON(!path->mnt);
 799
 800        f = get_empty_filp();
 801        if (!IS_ERR(f)) {
 802                f->f_flags = flags;
 803                f->f_path = *path;
 804                error = do_dentry_open(f, NULL, cred);
 805                if (!error) {
 806                        /* from now on we need fput() to dispose of f */
 807                        error = open_check_o_direct(f);
 808                        if (error) {
 809                                fput(f);
 810                                f = ERR_PTR(error);
 811                        }
 812                } else { 
 813                        put_filp(f);
 814                        f = ERR_PTR(error);
 815                }
 816        }
 817        return f;
 818}
 819EXPORT_SYMBOL(dentry_open);
 820
 821static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 822{
 823        int lookup_flags = 0;
 824        int acc_mode;
 825
 826        if (flags & (O_CREAT | __O_TMPFILE))
 827                op->mode = (mode & S_IALLUGO) | S_IFREG;
 828        else
 829                op->mode = 0;
 830
 831        /* Must never be set by userspace */
 832        flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 833
 834        /*
 835         * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 836         * check for O_DSYNC if the need any syncing at all we enforce it's
 837         * always set instead of having to deal with possibly weird behaviour
 838         * for malicious applications setting only __O_SYNC.
 839         */
 840        if (flags & __O_SYNC)
 841                flags |= O_DSYNC;
 842
 843        if (flags & __O_TMPFILE) {
 844                if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 845                        return -EINVAL;
 846                acc_mode = MAY_OPEN | ACC_MODE(flags);
 847                if (!(acc_mode & MAY_WRITE))
 848                        return -EINVAL;
 849        } else if (flags & O_PATH) {
 850                /*
 851                 * If we have O_PATH in the open flag. Then we
 852                 * cannot have anything other than the below set of flags
 853                 */
 854                flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 855                acc_mode = 0;
 856        } else {
 857                acc_mode = MAY_OPEN | ACC_MODE(flags);
 858        }
 859
 860        op->open_flag = flags;
 861
 862        /* O_TRUNC implies we need access checks for write permissions */
 863        if (flags & O_TRUNC)
 864                acc_mode |= MAY_WRITE;
 865
 866        /* Allow the LSM permission hook to distinguish append
 867           access from general write access. */
 868        if (flags & O_APPEND)
 869                acc_mode |= MAY_APPEND;
 870
 871        op->acc_mode = acc_mode;
 872
 873        op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 874
 875        if (flags & O_CREAT) {
 876                op->intent |= LOOKUP_CREATE;
 877                if (flags & O_EXCL)
 878                        op->intent |= LOOKUP_EXCL;
 879        }
 880
 881        if (flags & O_DIRECTORY)
 882                lookup_flags |= LOOKUP_DIRECTORY;
 883        if (!(flags & O_NOFOLLOW))
 884                lookup_flags |= LOOKUP_FOLLOW;
 885        op->lookup_flags = lookup_flags;
 886        return 0;
 887}
 888
 889/**
 890 * file_open_name - open file and return file pointer
 891 *
 892 * @name:       struct filename containing path to open
 893 * @flags:      open flags as per the open(2) second argument
 894 * @mode:       mode for the new file if O_CREAT is set, else ignored
 895 *
 896 * This is the helper to open a file from kernelspace if you really
 897 * have to.  But in generally you should not do this, so please move
 898 * along, nothing to see here..
 899 */
 900struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 901{
 902        struct open_flags op;
 903        int err = build_open_flags(flags, mode, &op);
 904        return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 905}
 906
 907/**
 908 * filp_open - open file and return file pointer
 909 *
 910 * @filename:   path to open
 911 * @flags:      open flags as per the open(2) second argument
 912 * @mode:       mode for the new file if O_CREAT is set, else ignored
 913 *
 914 * This is the helper to open a file from kernelspace if you really
 915 * have to.  But in generally you should not do this, so please move
 916 * along, nothing to see here..
 917 */
 918struct file *filp_open(const char *filename, int flags, umode_t mode)
 919{
 920        struct filename name = {.name = filename};
 921        return file_open_name(&name, flags, mode);
 922}
 923EXPORT_SYMBOL(filp_open);
 924
 925struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 926                            const char *filename, int flags)
 927{
 928        struct open_flags op;
 929        int err = build_open_flags(flags, 0, &op);
 930        if (err)
 931                return ERR_PTR(err);
 932        if (flags & O_CREAT)
 933                return ERR_PTR(-EINVAL);
 934        if (!filename && (flags & O_DIRECTORY))
 935                if (!dentry->d_inode->i_op->lookup)
 936                        return ERR_PTR(-ENOTDIR);
 937        return do_file_open_root(dentry, mnt, filename, &op);
 938}
 939EXPORT_SYMBOL(file_open_root);
 940
 941long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 942{
 943        struct open_flags op;
 944        int fd = build_open_flags(flags, mode, &op);
 945        struct filename *tmp;
 946
 947        if (fd)
 948                return fd;
 949
 950        tmp = getname(filename);
 951        if (IS_ERR(tmp))
 952                return PTR_ERR(tmp);
 953
 954        fd = get_unused_fd_flags(flags);
 955        if (fd >= 0) {
 956                struct file *f = do_filp_open(dfd, tmp, &op);
 957                if (IS_ERR(f)) {
 958                        put_unused_fd(fd);
 959                        fd = PTR_ERR(f);
 960                } else {
 961                        fsnotify_open(f);
 962                        fd_install(fd, f);
 963                }
 964        }
 965        putname(tmp);
 966        return fd;
 967}
 968
 969SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
 970{
 971        if (force_o_largefile())
 972                flags |= O_LARGEFILE;
 973
 974        return do_sys_open(AT_FDCWD, filename, flags, mode);
 975}
 976
 977SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
 978                umode_t, mode)
 979{
 980        if (force_o_largefile())
 981                flags |= O_LARGEFILE;
 982
 983        return do_sys_open(dfd, filename, flags, mode);
 984}
 985
 986#ifndef __alpha__
 987
 988/*
 989 * For backward compatibility?  Maybe this should be moved
 990 * into arch/i386 instead?
 991 */
 992SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
 993{
 994        return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 995}
 996
 997#endif
 998
 999/*
1000 * "id" is the POSIX thread ID. We use the
1001 * files pointer for this..
1002 */
1003int filp_close(struct file *filp, fl_owner_t id)
1004{
1005        int retval = 0;
1006
1007        if (!file_count(filp)) {
1008                printk(KERN_ERR "VFS: Close: file count is 0\n");
1009                return 0;
1010        }
1011
1012        if (filp->f_op && filp->f_op->flush)
1013                retval = filp->f_op->flush(filp, id);
1014
1015        if (likely(!(filp->f_mode & FMODE_PATH))) {
1016                dnotify_flush(filp, id);
1017                locks_remove_posix(filp, id);
1018        }
1019        fput(filp);
1020        return retval;
1021}
1022
1023EXPORT_SYMBOL(filp_close);
1024
1025/*
1026 * Careful here! We test whether the file pointer is NULL before
1027 * releasing the fd. This ensures that one clone task can't release
1028 * an fd while another clone is opening it.
1029 */
1030SYSCALL_DEFINE1(close, unsigned int, fd)
1031{
1032        int retval = __close_fd(current->files, fd);
1033
1034        /* can't restart close syscall because file table entry was cleared */
1035        if (unlikely(retval == -ERESTARTSYS ||
1036                     retval == -ERESTARTNOINTR ||
1037                     retval == -ERESTARTNOHAND ||
1038                     retval == -ERESTART_RESTARTBLOCK))
1039                retval = -EINTR;
1040
1041        return retval;
1042}
1043EXPORT_SYMBOL(sys_close);
1044
1045/*
1046 * This routine simulates a hangup on the tty, to arrange that users
1047 * are given clean terminals at login time.
1048 */
1049SYSCALL_DEFINE0(vhangup)
1050{
1051        if (capable(CAP_SYS_TTY_CONFIG)) {
1052                tty_vhangup_self();
1053                return 0;
1054        }
1055        return -EPERM;
1056}
1057
1058/*
1059 * Called when an inode is about to be open.
1060 * We use this to disallow opening large files on 32bit systems if
1061 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1062 * on this flag in sys_open.
1063 */
1064int generic_file_open(struct inode * inode, struct file * filp)
1065{
1066        if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1067                return -EOVERFLOW;
1068        return 0;
1069}
1070
1071EXPORT_SYMBOL(generic_file_open);
1072
1073/*
1074 * This is used by subsystems that don't want seekable
1075 * file descriptors. The function is not supposed to ever fail, the only
1076 * reason it returns an 'int' and not 'void' is so that it can be plugged
1077 * directly into file_operations structure.
1078 */
1079int nonseekable_open(struct inode *inode, struct file *filp)
1080{
1081        filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1082        return 0;
1083}
1084
1085EXPORT_SYMBOL(nonseekable_open);
1086
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.