cfq: Optimization for close cooperating queue searching
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / open.c
blobb4b31d277f3aa537efe1ebaa9d847dca86835214
1 /*
2 * linux/fs/open.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 #include <linux/string.h>
8 #include <linux/mm.h>
9 #include <linux/file.h>
10 #include <linux/fdtable.h>
11 #include <linux/quotaops.h>
12 #include <linux/fsnotify.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/tty.h>
16 #include <linux/namei.h>
17 #include <linux/backing-dev.h>
18 #include <linux/capability.h>
19 #include <linux/securebits.h>
20 #include <linux/security.h>
21 #include <linux/mount.h>
22 #include <linux/vfs.h>
23 #include <linux/fcntl.h>
24 #include <asm/uaccess.h>
25 #include <linux/fs.h>
26 #include <linux/personality.h>
27 #include <linux/pagemap.h>
28 #include <linux/syscalls.h>
29 #include <linux/rcupdate.h>
30 #include <linux/audit.h>
31 #include <linux/falloc.h>
32 #include <linux/fs_struct.h>
34 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
36 int retval = -ENODEV;
38 if (dentry) {
39 retval = -ENOSYS;
40 if (dentry->d_sb->s_op->statfs) {
41 memset(buf, 0, sizeof(*buf));
42 retval = security_sb_statfs(dentry);
43 if (retval)
44 return retval;
45 retval = dentry->d_sb->s_op->statfs(dentry, buf);
46 if (retval == 0 && buf->f_frsize == 0)
47 buf->f_frsize = buf->f_bsize;
50 return retval;
53 EXPORT_SYMBOL(vfs_statfs);
55 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
57 struct kstatfs st;
58 int retval;
60 retval = vfs_statfs(dentry, &st);
61 if (retval)
62 return retval;
64 if (sizeof(*buf) == sizeof(st))
65 memcpy(buf, &st, sizeof(st));
66 else {
67 if (sizeof buf->f_blocks == 4) {
68 if ((st.f_blocks | st.f_bfree | st.f_bavail |
69 st.f_bsize | st.f_frsize) &
70 0xffffffff00000000ULL)
71 return -EOVERFLOW;
73 * f_files and f_ffree may be -1; it's okay to stuff
74 * that into 32 bits
76 if (st.f_files != -1 &&
77 (st.f_files & 0xffffffff00000000ULL))
78 return -EOVERFLOW;
79 if (st.f_ffree != -1 &&
80 (st.f_ffree & 0xffffffff00000000ULL))
81 return -EOVERFLOW;
84 buf->f_type = st.f_type;
85 buf->f_bsize = st.f_bsize;
86 buf->f_blocks = st.f_blocks;
87 buf->f_bfree = st.f_bfree;
88 buf->f_bavail = st.f_bavail;
89 buf->f_files = st.f_files;
90 buf->f_ffree = st.f_ffree;
91 buf->f_fsid = st.f_fsid;
92 buf->f_namelen = st.f_namelen;
93 buf->f_frsize = st.f_frsize;
94 memset(buf->f_spare, 0, sizeof(buf->f_spare));
96 return 0;
99 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
101 struct kstatfs st;
102 int retval;
104 retval = vfs_statfs(dentry, &st);
105 if (retval)
106 return retval;
108 if (sizeof(*buf) == sizeof(st))
109 memcpy(buf, &st, sizeof(st));
110 else {
111 buf->f_type = st.f_type;
112 buf->f_bsize = st.f_bsize;
113 buf->f_blocks = st.f_blocks;
114 buf->f_bfree = st.f_bfree;
115 buf->f_bavail = st.f_bavail;
116 buf->f_files = st.f_files;
117 buf->f_ffree = st.f_ffree;
118 buf->f_fsid = st.f_fsid;
119 buf->f_namelen = st.f_namelen;
120 buf->f_frsize = st.f_frsize;
121 memset(buf->f_spare, 0, sizeof(buf->f_spare));
123 return 0;
126 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
128 struct path path;
129 int error;
131 error = user_path(pathname, &path);
132 if (!error) {
133 struct statfs tmp;
134 error = vfs_statfs_native(path.dentry, &tmp);
135 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
136 error = -EFAULT;
137 path_put(&path);
139 return error;
142 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
144 struct path path;
145 long error;
147 if (sz != sizeof(*buf))
148 return -EINVAL;
149 error = user_path(pathname, &path);
150 if (!error) {
151 struct statfs64 tmp;
152 error = vfs_statfs64(path.dentry, &tmp);
153 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
154 error = -EFAULT;
155 path_put(&path);
157 return error;
160 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
162 struct file * file;
163 struct statfs tmp;
164 int error;
166 error = -EBADF;
167 file = fget(fd);
168 if (!file)
169 goto out;
170 error = vfs_statfs_native(file->f_path.dentry, &tmp);
171 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
172 error = -EFAULT;
173 fput(file);
174 out:
175 return error;
178 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
180 struct file * file;
181 struct statfs64 tmp;
182 int error;
184 if (sz != sizeof(*buf))
185 return -EINVAL;
187 error = -EBADF;
188 file = fget(fd);
189 if (!file)
190 goto out;
191 error = vfs_statfs64(file->f_path.dentry, &tmp);
192 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
193 error = -EFAULT;
194 fput(file);
195 out:
196 return error;
199 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
200 struct file *filp)
202 int ret;
203 struct iattr newattrs;
205 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
206 if (length < 0)
207 return -EINVAL;
209 newattrs.ia_size = length;
210 newattrs.ia_valid = ATTR_SIZE | time_attrs;
211 if (filp) {
212 newattrs.ia_file = filp;
213 newattrs.ia_valid |= ATTR_FILE;
216 /* Remove suid/sgid on truncate too */
217 ret = should_remove_suid(dentry);
218 if (ret)
219 newattrs.ia_valid |= ret | ATTR_FORCE;
221 mutex_lock(&dentry->d_inode->i_mutex);
222 ret = notify_change(dentry, &newattrs);
223 mutex_unlock(&dentry->d_inode->i_mutex);
224 return ret;
227 static long do_sys_truncate(const char __user *pathname, loff_t length)
229 struct path path;
230 struct inode *inode;
231 int error;
233 error = -EINVAL;
234 if (length < 0) /* sorry, but loff_t says... */
235 goto out;
237 error = user_path(pathname, &path);
238 if (error)
239 goto out;
240 inode = path.dentry->d_inode;
242 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
243 error = -EISDIR;
244 if (S_ISDIR(inode->i_mode))
245 goto dput_and_out;
247 error = -EINVAL;
248 if (!S_ISREG(inode->i_mode))
249 goto dput_and_out;
251 error = mnt_want_write(path.mnt);
252 if (error)
253 goto dput_and_out;
255 error = inode_permission(inode, MAY_WRITE);
256 if (error)
257 goto mnt_drop_write_and_out;
259 error = -EPERM;
260 if (IS_APPEND(inode))
261 goto mnt_drop_write_and_out;
263 error = get_write_access(inode);
264 if (error)
265 goto mnt_drop_write_and_out;
268 * Make sure that there are no leases. get_write_access() protects
269 * against the truncate racing with a lease-granting setlease().
271 error = break_lease(inode, FMODE_WRITE);
272 if (error)
273 goto put_write_and_out;
275 error = locks_verify_truncate(inode, NULL, length);
276 if (!error)
277 error = security_path_truncate(&path, length, 0);
278 if (!error) {
279 vfs_dq_init(inode);
280 error = do_truncate(path.dentry, length, 0, NULL);
283 put_write_and_out:
284 put_write_access(inode);
285 mnt_drop_write_and_out:
286 mnt_drop_write(path.mnt);
287 dput_and_out:
288 path_put(&path);
289 out:
290 return error;
293 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
295 return do_sys_truncate(path, length);
298 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
300 struct inode * inode;
301 struct dentry *dentry;
302 struct file * file;
303 int error;
305 error = -EINVAL;
306 if (length < 0)
307 goto out;
308 error = -EBADF;
309 file = fget(fd);
310 if (!file)
311 goto out;
313 /* explicitly opened as large or we are on 64-bit box */
314 if (file->f_flags & O_LARGEFILE)
315 small = 0;
317 dentry = file->f_path.dentry;
318 inode = dentry->d_inode;
319 error = -EINVAL;
320 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
321 goto out_putf;
323 error = -EINVAL;
324 /* Cannot ftruncate over 2^31 bytes without large file support */
325 if (small && length > MAX_NON_LFS)
326 goto out_putf;
328 error = -EPERM;
329 if (IS_APPEND(inode))
330 goto out_putf;
332 error = locks_verify_truncate(inode, file, length);
333 if (!error)
334 error = security_path_truncate(&file->f_path, length,
335 ATTR_MTIME|ATTR_CTIME);
336 if (!error)
337 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
338 out_putf:
339 fput(file);
340 out:
341 return error;
344 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
346 long ret = do_sys_ftruncate(fd, length, 1);
347 /* avoid REGPARM breakage on x86: */
348 asmlinkage_protect(2, ret, fd, length);
349 return ret;
352 /* LFS versions of truncate are only needed on 32 bit machines */
353 #if BITS_PER_LONG == 32
354 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
356 return do_sys_truncate(path, length);
358 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
359 asmlinkage long SyS_truncate64(long path, loff_t length)
361 return SYSC_truncate64((const char __user *) path, length);
363 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
364 #endif
366 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
368 long ret = do_sys_ftruncate(fd, length, 0);
369 /* avoid REGPARM breakage on x86: */
370 asmlinkage_protect(2, ret, fd, length);
371 return ret;
373 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
374 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
376 return SYSC_ftruncate64((unsigned int) fd, length);
378 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
379 #endif
380 #endif /* BITS_PER_LONG == 32 */
383 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
385 struct inode *inode = file->f_path.dentry->d_inode;
386 long ret;
388 if (offset < 0 || len <= 0)
389 return -EINVAL;
391 /* Return error if mode is not supported */
392 if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
393 return -EOPNOTSUPP;
395 if (!(file->f_mode & FMODE_WRITE))
396 return -EBADF;
398 * Revalidate the write permissions, in case security policy has
399 * changed since the files were opened.
401 ret = security_file_permission(file, MAY_WRITE);
402 if (ret)
403 return ret;
405 if (S_ISFIFO(inode->i_mode))
406 return -ESPIPE;
409 * Let individual file system decide if it supports preallocation
410 * for directories or not.
412 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
413 return -ENODEV;
415 /* Check for wrap through zero too */
416 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
417 return -EFBIG;
419 if (!inode->i_op->fallocate)
420 return -EOPNOTSUPP;
422 return inode->i_op->fallocate(inode, mode, offset, len);
425 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
427 struct file *file;
428 int error = -EBADF;
430 file = fget(fd);
431 if (file) {
432 error = do_fallocate(file, mode, offset, len);
433 fput(file);
436 return error;
439 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
440 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
442 return SYSC_fallocate((int)fd, (int)mode, offset, len);
444 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
445 #endif
448 * access() needs to use the real uid/gid, not the effective uid/gid.
449 * We do this by temporarily clearing all FS-related capabilities and
450 * switching the fsuid/fsgid around to the real ones.
452 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
454 const struct cred *old_cred;
455 struct cred *override_cred;
456 struct path path;
457 struct inode *inode;
458 int res;
460 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
461 return -EINVAL;
463 override_cred = prepare_creds();
464 if (!override_cred)
465 return -ENOMEM;
467 override_cred->fsuid = override_cred->uid;
468 override_cred->fsgid = override_cred->gid;
470 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
471 /* Clear the capabilities if we switch to a non-root user */
472 if (override_cred->uid)
473 cap_clear(override_cred->cap_effective);
474 else
475 override_cred->cap_effective =
476 override_cred->cap_permitted;
479 old_cred = override_creds(override_cred);
481 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
482 if (res)
483 goto out;
485 inode = path.dentry->d_inode;
487 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
489 * MAY_EXEC on regular files is denied if the fs is mounted
490 * with the "noexec" flag.
492 res = -EACCES;
493 if (path.mnt->mnt_flags & MNT_NOEXEC)
494 goto out_path_release;
497 res = inode_permission(inode, mode | MAY_ACCESS);
498 /* SuS v2 requires we report a read only fs too */
499 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
500 goto out_path_release;
502 * This is a rare case where using __mnt_is_readonly()
503 * is OK without a mnt_want/drop_write() pair. Since
504 * no actual write to the fs is performed here, we do
505 * not need to telegraph to that to anyone.
507 * By doing this, we accept that this access is
508 * inherently racy and know that the fs may change
509 * state before we even see this result.
511 if (__mnt_is_readonly(path.mnt))
512 res = -EROFS;
514 out_path_release:
515 path_put(&path);
516 out:
517 revert_creds(old_cred);
518 put_cred(override_cred);
519 return res;
522 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
524 return sys_faccessat(AT_FDCWD, filename, mode);
527 SYSCALL_DEFINE1(chdir, const char __user *, filename)
529 struct path path;
530 int error;
532 error = user_path_dir(filename, &path);
533 if (error)
534 goto out;
536 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
537 if (error)
538 goto dput_and_out;
540 set_fs_pwd(current->fs, &path);
542 dput_and_out:
543 path_put(&path);
544 out:
545 return error;
548 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
550 struct file *file;
551 struct inode *inode;
552 int error;
554 error = -EBADF;
555 file = fget(fd);
556 if (!file)
557 goto out;
559 inode = file->f_path.dentry->d_inode;
561 error = -ENOTDIR;
562 if (!S_ISDIR(inode->i_mode))
563 goto out_putf;
565 error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
566 if (!error)
567 set_fs_pwd(current->fs, &file->f_path);
568 out_putf:
569 fput(file);
570 out:
571 return error;
574 SYSCALL_DEFINE1(chroot, const char __user *, filename)
576 struct path path;
577 int error;
579 error = user_path_dir(filename, &path);
580 if (error)
581 goto out;
583 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
584 if (error)
585 goto dput_and_out;
587 error = -EPERM;
588 if (!capable(CAP_SYS_CHROOT))
589 goto dput_and_out;
590 error = security_path_chroot(&path);
591 if (error)
592 goto dput_and_out;
594 set_fs_root(current->fs, &path);
595 error = 0;
596 dput_and_out:
597 path_put(&path);
598 out:
599 return error;
602 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
604 struct inode * inode;
605 struct dentry * dentry;
606 struct file * file;
607 int err = -EBADF;
608 struct iattr newattrs;
610 file = fget(fd);
611 if (!file)
612 goto out;
614 dentry = file->f_path.dentry;
615 inode = dentry->d_inode;
617 audit_inode(NULL, dentry);
619 err = mnt_want_write_file(file);
620 if (err)
621 goto out_putf;
622 mutex_lock(&inode->i_mutex);
623 err = security_path_chmod(dentry, file->f_vfsmnt, mode);
624 if (err)
625 goto out_unlock;
626 if (mode == (mode_t) -1)
627 mode = inode->i_mode;
628 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
629 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
630 err = notify_change(dentry, &newattrs);
631 out_unlock:
632 mutex_unlock(&inode->i_mutex);
633 mnt_drop_write(file->f_path.mnt);
634 out_putf:
635 fput(file);
636 out:
637 return err;
640 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
642 struct path path;
643 struct inode *inode;
644 int error;
645 struct iattr newattrs;
647 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
648 if (error)
649 goto out;
650 inode = path.dentry->d_inode;
652 error = mnt_want_write(path.mnt);
653 if (error)
654 goto dput_and_out;
655 mutex_lock(&inode->i_mutex);
656 error = security_path_chmod(path.dentry, path.mnt, mode);
657 if (error)
658 goto out_unlock;
659 if (mode == (mode_t) -1)
660 mode = inode->i_mode;
661 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
662 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
663 error = notify_change(path.dentry, &newattrs);
664 out_unlock:
665 mutex_unlock(&inode->i_mutex);
666 mnt_drop_write(path.mnt);
667 dput_and_out:
668 path_put(&path);
669 out:
670 return error;
673 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
675 return sys_fchmodat(AT_FDCWD, filename, mode);
678 static int chown_common(struct path *path, uid_t user, gid_t group)
680 struct inode *inode = path->dentry->d_inode;
681 int error;
682 struct iattr newattrs;
684 newattrs.ia_valid = ATTR_CTIME;
685 if (user != (uid_t) -1) {
686 newattrs.ia_valid |= ATTR_UID;
687 newattrs.ia_uid = user;
689 if (group != (gid_t) -1) {
690 newattrs.ia_valid |= ATTR_GID;
691 newattrs.ia_gid = group;
693 if (!S_ISDIR(inode->i_mode))
694 newattrs.ia_valid |=
695 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
696 mutex_lock(&inode->i_mutex);
697 error = security_path_chown(path, user, group);
698 if (!error)
699 error = notify_change(path->dentry, &newattrs);
700 mutex_unlock(&inode->i_mutex);
702 return error;
705 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
707 struct path path;
708 int error;
710 error = user_path(filename, &path);
711 if (error)
712 goto out;
713 error = mnt_want_write(path.mnt);
714 if (error)
715 goto out_release;
716 error = chown_common(&path, user, group);
717 mnt_drop_write(path.mnt);
718 out_release:
719 path_put(&path);
720 out:
721 return error;
724 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
725 gid_t, group, int, flag)
727 struct path path;
728 int error = -EINVAL;
729 int follow;
731 if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
732 goto out;
734 follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
735 error = user_path_at(dfd, filename, follow, &path);
736 if (error)
737 goto out;
738 error = mnt_want_write(path.mnt);
739 if (error)
740 goto out_release;
741 error = chown_common(&path, user, group);
742 mnt_drop_write(path.mnt);
743 out_release:
744 path_put(&path);
745 out:
746 return error;
749 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
751 struct path path;
752 int error;
754 error = user_lpath(filename, &path);
755 if (error)
756 goto out;
757 error = mnt_want_write(path.mnt);
758 if (error)
759 goto out_release;
760 error = chown_common(&path, user, group);
761 mnt_drop_write(path.mnt);
762 out_release:
763 path_put(&path);
764 out:
765 return error;
768 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
770 struct file * file;
771 int error = -EBADF;
772 struct dentry * dentry;
774 file = fget(fd);
775 if (!file)
776 goto out;
778 error = mnt_want_write_file(file);
779 if (error)
780 goto out_fput;
781 dentry = file->f_path.dentry;
782 audit_inode(NULL, dentry);
783 error = chown_common(&file->f_path, user, group);
784 mnt_drop_write(file->f_path.mnt);
785 out_fput:
786 fput(file);
787 out:
788 return error;
792 * You have to be very careful that these write
793 * counts get cleaned up in error cases and
794 * upon __fput(). This should probably never
795 * be called outside of __dentry_open().
797 static inline int __get_file_write_access(struct inode *inode,
798 struct vfsmount *mnt)
800 int error;
801 error = get_write_access(inode);
802 if (error)
803 return error;
805 * Do not take mount writer counts on
806 * special files since no writes to
807 * the mount itself will occur.
809 if (!special_file(inode->i_mode)) {
811 * Balanced in __fput()
813 error = mnt_want_write(mnt);
814 if (error)
815 put_write_access(inode);
817 return error;
820 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
821 int flags, struct file *f,
822 int (*open)(struct inode *, struct file *),
823 const struct cred *cred)
825 struct inode *inode;
826 int error;
828 f->f_flags = flags;
829 f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
830 FMODE_PREAD | FMODE_PWRITE;
831 inode = dentry->d_inode;
832 if (f->f_mode & FMODE_WRITE) {
833 error = __get_file_write_access(inode, mnt);
834 if (error)
835 goto cleanup_file;
836 if (!special_file(inode->i_mode))
837 file_take_write(f);
840 f->f_mapping = inode->i_mapping;
841 f->f_path.dentry = dentry;
842 f->f_path.mnt = mnt;
843 f->f_pos = 0;
844 f->f_op = fops_get(inode->i_fop);
845 file_move(f, &inode->i_sb->s_files);
847 error = security_dentry_open(f, cred);
848 if (error)
849 goto cleanup_all;
851 if (!open && f->f_op)
852 open = f->f_op->open;
853 if (open) {
854 error = open(inode, f);
855 if (error)
856 goto cleanup_all;
859 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
861 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
863 /* NB: we're sure to have correct a_ops only after f_op->open */
864 if (f->f_flags & O_DIRECT) {
865 if (!f->f_mapping->a_ops ||
866 ((!f->f_mapping->a_ops->direct_IO) &&
867 (!f->f_mapping->a_ops->get_xip_mem))) {
868 fput(f);
869 f = ERR_PTR(-EINVAL);
873 return f;
875 cleanup_all:
876 fops_put(f->f_op);
877 if (f->f_mode & FMODE_WRITE) {
878 put_write_access(inode);
879 if (!special_file(inode->i_mode)) {
881 * We don't consider this a real
882 * mnt_want/drop_write() pair
883 * because it all happenend right
884 * here, so just reset the state.
886 file_reset_write(f);
887 mnt_drop_write(mnt);
890 file_kill(f);
891 f->f_path.dentry = NULL;
892 f->f_path.mnt = NULL;
893 cleanup_file:
894 put_filp(f);
895 dput(dentry);
896 mntput(mnt);
897 return ERR_PTR(error);
901 * lookup_instantiate_filp - instantiates the open intent filp
902 * @nd: pointer to nameidata
903 * @dentry: pointer to dentry
904 * @open: open callback
906 * Helper for filesystems that want to use lookup open intents and pass back
907 * a fully instantiated struct file to the caller.
908 * This function is meant to be called from within a filesystem's
909 * lookup method.
910 * Beware of calling it for non-regular files! Those ->open methods might block
911 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
912 * leading to a deadlock, as nobody can open that fifo anymore, because
913 * another process to open fifo will block on locked parent when doing lookup).
914 * Note that in case of error, nd->intent.open.file is destroyed, but the
915 * path information remains valid.
916 * If the open callback is set to NULL, then the standard f_op->open()
917 * filesystem callback is substituted.
919 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
920 int (*open)(struct inode *, struct file *))
922 const struct cred *cred = current_cred();
924 if (IS_ERR(nd->intent.open.file))
925 goto out;
926 if (IS_ERR(dentry))
927 goto out_err;
928 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
929 nd->intent.open.flags - 1,
930 nd->intent.open.file,
931 open, cred);
932 out:
933 return nd->intent.open.file;
934 out_err:
935 release_open_intent(nd);
936 nd->intent.open.file = (struct file *)dentry;
937 goto out;
939 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
942 * nameidata_to_filp - convert a nameidata to an open filp.
943 * @nd: pointer to nameidata
944 * @flags: open flags
946 * Note that this function destroys the original nameidata
948 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
950 const struct cred *cred = current_cred();
951 struct file *filp;
953 /* Pick up the filp from the open intent */
954 filp = nd->intent.open.file;
955 /* Has the filesystem initialised the file for us? */
956 if (filp->f_path.dentry == NULL)
957 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
958 NULL, cred);
959 else
960 path_put(&nd->path);
961 return filp;
965 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
966 * error.
968 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
969 const struct cred *cred)
971 int error;
972 struct file *f;
974 validate_creds(cred);
977 * We must always pass in a valid mount pointer. Historically
978 * callers got away with not passing it, but we must enforce this at
979 * the earliest possible point now to avoid strange problems deep in the
980 * filesystem stack.
982 if (!mnt) {
983 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
984 dump_stack();
985 return ERR_PTR(-EINVAL);
988 error = -ENFILE;
989 f = get_empty_filp();
990 if (f == NULL) {
991 dput(dentry);
992 mntput(mnt);
993 return ERR_PTR(error);
996 return __dentry_open(dentry, mnt, flags, f, NULL, cred);
998 EXPORT_SYMBOL(dentry_open);
1000 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1002 struct fdtable *fdt = files_fdtable(files);
1003 __FD_CLR(fd, fdt->open_fds);
1004 if (fd < files->next_fd)
1005 files->next_fd = fd;
1008 void put_unused_fd(unsigned int fd)
1010 struct files_struct *files = current->files;
1011 spin_lock(&files->file_lock);
1012 __put_unused_fd(files, fd);
1013 spin_unlock(&files->file_lock);
1016 EXPORT_SYMBOL(put_unused_fd);
1019 * Install a file pointer in the fd array.
1021 * The VFS is full of places where we drop the files lock between
1022 * setting the open_fds bitmap and installing the file in the file
1023 * array. At any such point, we are vulnerable to a dup2() race
1024 * installing a file in the array before us. We need to detect this and
1025 * fput() the struct file we are about to overwrite in this case.
1027 * It should never happen - if we allow dup2() do it, _really_ bad things
1028 * will follow.
1031 void fd_install(unsigned int fd, struct file *file)
1033 struct files_struct *files = current->files;
1034 struct fdtable *fdt;
1035 spin_lock(&files->file_lock);
1036 fdt = files_fdtable(files);
1037 BUG_ON(fdt->fd[fd] != NULL);
1038 rcu_assign_pointer(fdt->fd[fd], file);
1039 spin_unlock(&files->file_lock);
1042 EXPORT_SYMBOL(fd_install);
1044 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1046 char *tmp = getname(filename);
1047 int fd = PTR_ERR(tmp);
1049 if (!IS_ERR(tmp)) {
1050 fd = get_unused_fd_flags(flags);
1051 if (fd >= 0) {
1052 struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1053 if (IS_ERR(f)) {
1054 put_unused_fd(fd);
1055 fd = PTR_ERR(f);
1056 } else {
1057 fsnotify_open(f->f_path.dentry);
1058 fd_install(fd, f);
1061 putname(tmp);
1063 return fd;
1066 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1068 long ret;
1070 if (force_o_largefile())
1071 flags |= O_LARGEFILE;
1073 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1074 /* avoid REGPARM breakage on x86: */
1075 asmlinkage_protect(3, ret, filename, flags, mode);
1076 return ret;
1079 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1080 int, mode)
1082 long ret;
1084 if (force_o_largefile())
1085 flags |= O_LARGEFILE;
1087 ret = do_sys_open(dfd, filename, flags, mode);
1088 /* avoid REGPARM breakage on x86: */
1089 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1090 return ret;
1093 #ifndef __alpha__
1096 * For backward compatibility? Maybe this should be moved
1097 * into arch/i386 instead?
1099 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1101 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1104 #endif
1107 * "id" is the POSIX thread ID. We use the
1108 * files pointer for this..
1110 int filp_close(struct file *filp, fl_owner_t id)
1112 int retval = 0;
1114 if (!file_count(filp)) {
1115 printk(KERN_ERR "VFS: Close: file count is 0\n");
1116 return 0;
1119 if (filp->f_op && filp->f_op->flush)
1120 retval = filp->f_op->flush(filp, id);
1122 dnotify_flush(filp, id);
1123 locks_remove_posix(filp, id);
1124 fput(filp);
1125 return retval;
1128 EXPORT_SYMBOL(filp_close);
1131 * Careful here! We test whether the file pointer is NULL before
1132 * releasing the fd. This ensures that one clone task can't release
1133 * an fd while another clone is opening it.
1135 SYSCALL_DEFINE1(close, unsigned int, fd)
1137 struct file * filp;
1138 struct files_struct *files = current->files;
1139 struct fdtable *fdt;
1140 int retval;
1142 spin_lock(&files->file_lock);
1143 fdt = files_fdtable(files);
1144 if (fd >= fdt->max_fds)
1145 goto out_unlock;
1146 filp = fdt->fd[fd];
1147 if (!filp)
1148 goto out_unlock;
1149 rcu_assign_pointer(fdt->fd[fd], NULL);
1150 FD_CLR(fd, fdt->close_on_exec);
1151 __put_unused_fd(files, fd);
1152 spin_unlock(&files->file_lock);
1153 retval = filp_close(filp, files);
1155 /* can't restart close syscall because file table entry was cleared */
1156 if (unlikely(retval == -ERESTARTSYS ||
1157 retval == -ERESTARTNOINTR ||
1158 retval == -ERESTARTNOHAND ||
1159 retval == -ERESTART_RESTARTBLOCK))
1160 retval = -EINTR;
1162 return retval;
1164 out_unlock:
1165 spin_unlock(&files->file_lock);
1166 return -EBADF;
1168 EXPORT_SYMBOL(sys_close);
1171 * This routine simulates a hangup on the tty, to arrange that users
1172 * are given clean terminals at login time.
1174 SYSCALL_DEFINE0(vhangup)
1176 if (capable(CAP_SYS_TTY_CONFIG)) {
1177 tty_vhangup_self();
1178 return 0;
1180 return -EPERM;
1184 * Called when an inode is about to be open.
1185 * We use this to disallow opening large files on 32bit systems if
1186 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1187 * on this flag in sys_open.
1189 int generic_file_open(struct inode * inode, struct file * filp)
1191 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1192 return -EOVERFLOW;
1193 return 0;
1196 EXPORT_SYMBOL(generic_file_open);
1199 * This is used by subsystems that don't want seekable
1200 * file descriptors
1202 int nonseekable_open(struct inode *inode, struct file *filp)
1204 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1205 return 0;
1208 EXPORT_SYMBOL(nonseekable_open);