Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6/linux-2.6-openrd.git] / fs / open.c
blob040cef72bc00e46a0c58790f1c18cef7722756c8
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>
33 #include <linux/ima.h>
35 #include "internal.h"
37 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
39 int retval = -ENODEV;
41 if (dentry) {
42 retval = -ENOSYS;
43 if (dentry->d_sb->s_op->statfs) {
44 memset(buf, 0, sizeof(*buf));
45 retval = security_sb_statfs(dentry);
46 if (retval)
47 return retval;
48 retval = dentry->d_sb->s_op->statfs(dentry, buf);
49 if (retval == 0 && buf->f_frsize == 0)
50 buf->f_frsize = buf->f_bsize;
53 return retval;
56 EXPORT_SYMBOL(vfs_statfs);
58 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
60 struct kstatfs st;
61 int retval;
63 retval = vfs_statfs(dentry, &st);
64 if (retval)
65 return retval;
67 if (sizeof(*buf) == sizeof(st))
68 memcpy(buf, &st, sizeof(st));
69 else {
70 if (sizeof buf->f_blocks == 4) {
71 if ((st.f_blocks | st.f_bfree | st.f_bavail |
72 st.f_bsize | st.f_frsize) &
73 0xffffffff00000000ULL)
74 return -EOVERFLOW;
76 * f_files and f_ffree may be -1; it's okay to stuff
77 * that into 32 bits
79 if (st.f_files != -1 &&
80 (st.f_files & 0xffffffff00000000ULL))
81 return -EOVERFLOW;
82 if (st.f_ffree != -1 &&
83 (st.f_ffree & 0xffffffff00000000ULL))
84 return -EOVERFLOW;
87 buf->f_type = st.f_type;
88 buf->f_bsize = st.f_bsize;
89 buf->f_blocks = st.f_blocks;
90 buf->f_bfree = st.f_bfree;
91 buf->f_bavail = st.f_bavail;
92 buf->f_files = st.f_files;
93 buf->f_ffree = st.f_ffree;
94 buf->f_fsid = st.f_fsid;
95 buf->f_namelen = st.f_namelen;
96 buf->f_frsize = st.f_frsize;
97 memset(buf->f_spare, 0, sizeof(buf->f_spare));
99 return 0;
102 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
104 struct kstatfs st;
105 int retval;
107 retval = vfs_statfs(dentry, &st);
108 if (retval)
109 return retval;
111 if (sizeof(*buf) == sizeof(st))
112 memcpy(buf, &st, sizeof(st));
113 else {
114 buf->f_type = st.f_type;
115 buf->f_bsize = st.f_bsize;
116 buf->f_blocks = st.f_blocks;
117 buf->f_bfree = st.f_bfree;
118 buf->f_bavail = st.f_bavail;
119 buf->f_files = st.f_files;
120 buf->f_ffree = st.f_ffree;
121 buf->f_fsid = st.f_fsid;
122 buf->f_namelen = st.f_namelen;
123 buf->f_frsize = st.f_frsize;
124 memset(buf->f_spare, 0, sizeof(buf->f_spare));
126 return 0;
129 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
131 struct path path;
132 int error;
134 error = user_path(pathname, &path);
135 if (!error) {
136 struct statfs tmp;
137 error = vfs_statfs_native(path.dentry, &tmp);
138 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
139 error = -EFAULT;
140 path_put(&path);
142 return error;
145 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
147 struct path path;
148 long error;
150 if (sz != sizeof(*buf))
151 return -EINVAL;
152 error = user_path(pathname, &path);
153 if (!error) {
154 struct statfs64 tmp;
155 error = vfs_statfs64(path.dentry, &tmp);
156 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
157 error = -EFAULT;
158 path_put(&path);
160 return error;
163 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
165 struct file * file;
166 struct statfs tmp;
167 int error;
169 error = -EBADF;
170 file = fget(fd);
171 if (!file)
172 goto out;
173 error = vfs_statfs_native(file->f_path.dentry, &tmp);
174 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
175 error = -EFAULT;
176 fput(file);
177 out:
178 return error;
181 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
183 struct file * file;
184 struct statfs64 tmp;
185 int error;
187 if (sz != sizeof(*buf))
188 return -EINVAL;
190 error = -EBADF;
191 file = fget(fd);
192 if (!file)
193 goto out;
194 error = vfs_statfs64(file->f_path.dentry, &tmp);
195 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
196 error = -EFAULT;
197 fput(file);
198 out:
199 return error;
202 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
203 struct file *filp)
205 int ret;
206 struct iattr newattrs;
208 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
209 if (length < 0)
210 return -EINVAL;
212 newattrs.ia_size = length;
213 newattrs.ia_valid = ATTR_SIZE | time_attrs;
214 if (filp) {
215 newattrs.ia_file = filp;
216 newattrs.ia_valid |= ATTR_FILE;
219 /* Remove suid/sgid on truncate too */
220 ret = should_remove_suid(dentry);
221 if (ret)
222 newattrs.ia_valid |= ret | ATTR_FORCE;
224 mutex_lock(&dentry->d_inode->i_mutex);
225 ret = notify_change(dentry, &newattrs);
226 mutex_unlock(&dentry->d_inode->i_mutex);
227 return ret;
230 static long do_sys_truncate(const char __user *pathname, loff_t length)
232 struct path path;
233 struct inode *inode;
234 int error;
236 error = -EINVAL;
237 if (length < 0) /* sorry, but loff_t says... */
238 goto out;
240 error = user_path(pathname, &path);
241 if (error)
242 goto out;
243 inode = path.dentry->d_inode;
245 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
246 error = -EISDIR;
247 if (S_ISDIR(inode->i_mode))
248 goto dput_and_out;
250 error = -EINVAL;
251 if (!S_ISREG(inode->i_mode))
252 goto dput_and_out;
254 error = mnt_want_write(path.mnt);
255 if (error)
256 goto dput_and_out;
258 error = inode_permission(inode, MAY_WRITE);
259 if (error)
260 goto mnt_drop_write_and_out;
262 error = -EPERM;
263 if (IS_APPEND(inode))
264 goto mnt_drop_write_and_out;
266 error = get_write_access(inode);
267 if (error)
268 goto mnt_drop_write_and_out;
271 * Make sure that there are no leases. get_write_access() protects
272 * against the truncate racing with a lease-granting setlease().
274 error = break_lease(inode, FMODE_WRITE);
275 if (error)
276 goto put_write_and_out;
278 error = locks_verify_truncate(inode, NULL, length);
279 if (!error)
280 error = security_path_truncate(&path, length, 0);
281 if (!error) {
282 vfs_dq_init(inode);
283 error = do_truncate(path.dentry, length, 0, NULL);
286 put_write_and_out:
287 put_write_access(inode);
288 mnt_drop_write_and_out:
289 mnt_drop_write(path.mnt);
290 dput_and_out:
291 path_put(&path);
292 out:
293 return error;
296 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
298 return do_sys_truncate(path, length);
301 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
303 struct inode * inode;
304 struct dentry *dentry;
305 struct file * file;
306 int error;
308 error = -EINVAL;
309 if (length < 0)
310 goto out;
311 error = -EBADF;
312 file = fget(fd);
313 if (!file)
314 goto out;
316 /* explicitly opened as large or we are on 64-bit box */
317 if (file->f_flags & O_LARGEFILE)
318 small = 0;
320 dentry = file->f_path.dentry;
321 inode = dentry->d_inode;
322 error = -EINVAL;
323 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
324 goto out_putf;
326 error = -EINVAL;
327 /* Cannot ftruncate over 2^31 bytes without large file support */
328 if (small && length > MAX_NON_LFS)
329 goto out_putf;
331 error = -EPERM;
332 if (IS_APPEND(inode))
333 goto out_putf;
335 error = locks_verify_truncate(inode, file, length);
336 if (!error)
337 error = security_path_truncate(&file->f_path, length,
338 ATTR_MTIME|ATTR_CTIME);
339 if (!error)
340 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
341 out_putf:
342 fput(file);
343 out:
344 return error;
347 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
349 long ret = do_sys_ftruncate(fd, length, 1);
350 /* avoid REGPARM breakage on x86: */
351 asmlinkage_protect(2, ret, fd, length);
352 return ret;
355 /* LFS versions of truncate are only needed on 32 bit machines */
356 #if BITS_PER_LONG == 32
357 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
359 return do_sys_truncate(path, length);
361 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
362 asmlinkage long SyS_truncate64(long path, loff_t length)
364 return SYSC_truncate64((const char __user *) path, length);
366 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
367 #endif
369 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
371 long ret = do_sys_ftruncate(fd, length, 0);
372 /* avoid REGPARM breakage on x86: */
373 asmlinkage_protect(2, ret, fd, length);
374 return ret;
376 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
377 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
379 return SYSC_ftruncate64((unsigned int) fd, length);
381 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
382 #endif
383 #endif /* BITS_PER_LONG == 32 */
386 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
388 struct inode *inode = file->f_path.dentry->d_inode;
389 long ret;
391 if (offset < 0 || len <= 0)
392 return -EINVAL;
394 /* Return error if mode is not supported */
395 if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
396 return -EOPNOTSUPP;
398 if (!(file->f_mode & FMODE_WRITE))
399 return -EBADF;
401 * Revalidate the write permissions, in case security policy has
402 * changed since the files were opened.
404 ret = security_file_permission(file, MAY_WRITE);
405 if (ret)
406 return ret;
408 if (S_ISFIFO(inode->i_mode))
409 return -ESPIPE;
412 * Let individual file system decide if it supports preallocation
413 * for directories or not.
415 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
416 return -ENODEV;
418 /* Check for wrap through zero too */
419 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
420 return -EFBIG;
422 if (!inode->i_op->fallocate)
423 return -EOPNOTSUPP;
425 return inode->i_op->fallocate(inode, mode, offset, len);
428 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
430 struct file *file;
431 int error = -EBADF;
433 file = fget(fd);
434 if (file) {
435 error = do_fallocate(file, mode, offset, len);
436 fput(file);
439 return error;
442 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
443 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
445 return SYSC_fallocate((int)fd, (int)mode, offset, len);
447 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
448 #endif
451 * access() needs to use the real uid/gid, not the effective uid/gid.
452 * We do this by temporarily clearing all FS-related capabilities and
453 * switching the fsuid/fsgid around to the real ones.
455 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
457 const struct cred *old_cred;
458 struct cred *override_cred;
459 struct path path;
460 struct inode *inode;
461 int res;
463 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
464 return -EINVAL;
466 override_cred = prepare_creds();
467 if (!override_cred)
468 return -ENOMEM;
470 override_cred->fsuid = override_cred->uid;
471 override_cred->fsgid = override_cred->gid;
473 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
474 /* Clear the capabilities if we switch to a non-root user */
475 if (override_cred->uid)
476 cap_clear(override_cred->cap_effective);
477 else
478 override_cred->cap_effective =
479 override_cred->cap_permitted;
482 old_cred = override_creds(override_cred);
484 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
485 if (res)
486 goto out;
488 inode = path.dentry->d_inode;
490 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
492 * MAY_EXEC on regular files is denied if the fs is mounted
493 * with the "noexec" flag.
495 res = -EACCES;
496 if (path.mnt->mnt_flags & MNT_NOEXEC)
497 goto out_path_release;
500 res = inode_permission(inode, mode | MAY_ACCESS);
501 /* SuS v2 requires we report a read only fs too */
502 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
503 goto out_path_release;
505 * This is a rare case where using __mnt_is_readonly()
506 * is OK without a mnt_want/drop_write() pair. Since
507 * no actual write to the fs is performed here, we do
508 * not need to telegraph to that to anyone.
510 * By doing this, we accept that this access is
511 * inherently racy and know that the fs may change
512 * state before we even see this result.
514 if (__mnt_is_readonly(path.mnt))
515 res = -EROFS;
517 out_path_release:
518 path_put(&path);
519 out:
520 revert_creds(old_cred);
521 put_cred(override_cred);
522 return res;
525 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
527 return sys_faccessat(AT_FDCWD, filename, mode);
530 SYSCALL_DEFINE1(chdir, const char __user *, filename)
532 struct path path;
533 int error;
535 error = user_path_dir(filename, &path);
536 if (error)
537 goto out;
539 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
540 if (error)
541 goto dput_and_out;
543 set_fs_pwd(current->fs, &path);
545 dput_and_out:
546 path_put(&path);
547 out:
548 return error;
551 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
553 struct file *file;
554 struct inode *inode;
555 int error;
557 error = -EBADF;
558 file = fget(fd);
559 if (!file)
560 goto out;
562 inode = file->f_path.dentry->d_inode;
564 error = -ENOTDIR;
565 if (!S_ISDIR(inode->i_mode))
566 goto out_putf;
568 error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
569 if (!error)
570 set_fs_pwd(current->fs, &file->f_path);
571 out_putf:
572 fput(file);
573 out:
574 return error;
577 SYSCALL_DEFINE1(chroot, const char __user *, filename)
579 struct path path;
580 int error;
582 error = user_path_dir(filename, &path);
583 if (error)
584 goto out;
586 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
587 if (error)
588 goto dput_and_out;
590 error = -EPERM;
591 if (!capable(CAP_SYS_CHROOT))
592 goto dput_and_out;
593 error = security_path_chroot(&path);
594 if (error)
595 goto dput_and_out;
597 set_fs_root(current->fs, &path);
598 error = 0;
599 dput_and_out:
600 path_put(&path);
601 out:
602 return error;
605 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
607 struct inode * inode;
608 struct dentry * dentry;
609 struct file * file;
610 int err = -EBADF;
611 struct iattr newattrs;
613 file = fget(fd);
614 if (!file)
615 goto out;
617 dentry = file->f_path.dentry;
618 inode = dentry->d_inode;
620 audit_inode(NULL, dentry);
622 err = mnt_want_write_file(file);
623 if (err)
624 goto out_putf;
625 mutex_lock(&inode->i_mutex);
626 err = security_path_chmod(dentry, file->f_vfsmnt, mode);
627 if (err)
628 goto out_unlock;
629 if (mode == (mode_t) -1)
630 mode = inode->i_mode;
631 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
632 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
633 err = notify_change(dentry, &newattrs);
634 out_unlock:
635 mutex_unlock(&inode->i_mutex);
636 mnt_drop_write(file->f_path.mnt);
637 out_putf:
638 fput(file);
639 out:
640 return err;
643 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
645 struct path path;
646 struct inode *inode;
647 int error;
648 struct iattr newattrs;
650 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
651 if (error)
652 goto out;
653 inode = path.dentry->d_inode;
655 error = mnt_want_write(path.mnt);
656 if (error)
657 goto dput_and_out;
658 mutex_lock(&inode->i_mutex);
659 error = security_path_chmod(path.dentry, path.mnt, mode);
660 if (error)
661 goto out_unlock;
662 if (mode == (mode_t) -1)
663 mode = inode->i_mode;
664 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
665 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
666 error = notify_change(path.dentry, &newattrs);
667 out_unlock:
668 mutex_unlock(&inode->i_mutex);
669 mnt_drop_write(path.mnt);
670 dput_and_out:
671 path_put(&path);
672 out:
673 return error;
676 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
678 return sys_fchmodat(AT_FDCWD, filename, mode);
681 static int chown_common(struct path *path, uid_t user, gid_t group)
683 struct inode *inode = path->dentry->d_inode;
684 int error;
685 struct iattr newattrs;
687 newattrs.ia_valid = ATTR_CTIME;
688 if (user != (uid_t) -1) {
689 newattrs.ia_valid |= ATTR_UID;
690 newattrs.ia_uid = user;
692 if (group != (gid_t) -1) {
693 newattrs.ia_valid |= ATTR_GID;
694 newattrs.ia_gid = group;
696 if (!S_ISDIR(inode->i_mode))
697 newattrs.ia_valid |=
698 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
699 mutex_lock(&inode->i_mutex);
700 error = security_path_chown(path, user, group);
701 if (!error)
702 error = notify_change(path->dentry, &newattrs);
703 mutex_unlock(&inode->i_mutex);
705 return error;
708 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
710 struct path path;
711 int error;
713 error = user_path(filename, &path);
714 if (error)
715 goto out;
716 error = mnt_want_write(path.mnt);
717 if (error)
718 goto out_release;
719 error = chown_common(&path, user, group);
720 mnt_drop_write(path.mnt);
721 out_release:
722 path_put(&path);
723 out:
724 return error;
727 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
728 gid_t, group, int, flag)
730 struct path path;
731 int error = -EINVAL;
732 int follow;
734 if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
735 goto out;
737 follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
738 error = user_path_at(dfd, filename, follow, &path);
739 if (error)
740 goto out;
741 error = mnt_want_write(path.mnt);
742 if (error)
743 goto out_release;
744 error = chown_common(&path, user, group);
745 mnt_drop_write(path.mnt);
746 out_release:
747 path_put(&path);
748 out:
749 return error;
752 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
754 struct path path;
755 int error;
757 error = user_lpath(filename, &path);
758 if (error)
759 goto out;
760 error = mnt_want_write(path.mnt);
761 if (error)
762 goto out_release;
763 error = chown_common(&path, user, group);
764 mnt_drop_write(path.mnt);
765 out_release:
766 path_put(&path);
767 out:
768 return error;
771 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
773 struct file * file;
774 int error = -EBADF;
775 struct dentry * dentry;
777 file = fget(fd);
778 if (!file)
779 goto out;
781 error = mnt_want_write_file(file);
782 if (error)
783 goto out_fput;
784 dentry = file->f_path.dentry;
785 audit_inode(NULL, dentry);
786 error = chown_common(&file->f_path, user, group);
787 mnt_drop_write(file->f_path.mnt);
788 out_fput:
789 fput(file);
790 out:
791 return error;
795 * You have to be very careful that these write
796 * counts get cleaned up in error cases and
797 * upon __fput(). This should probably never
798 * be called outside of __dentry_open().
800 static inline int __get_file_write_access(struct inode *inode,
801 struct vfsmount *mnt)
803 int error;
804 error = get_write_access(inode);
805 if (error)
806 return error;
808 * Do not take mount writer counts on
809 * special files since no writes to
810 * the mount itself will occur.
812 if (!special_file(inode->i_mode)) {
814 * Balanced in __fput()
816 error = mnt_want_write(mnt);
817 if (error)
818 put_write_access(inode);
820 return error;
823 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
824 struct file *f,
825 int (*open)(struct inode *, struct file *),
826 const struct cred *cred)
828 struct inode *inode;
829 int error;
831 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
832 FMODE_PREAD | FMODE_PWRITE;
833 inode = dentry->d_inode;
834 if (f->f_mode & FMODE_WRITE) {
835 error = __get_file_write_access(inode, mnt);
836 if (error)
837 goto cleanup_file;
838 if (!special_file(inode->i_mode))
839 file_take_write(f);
842 f->f_mapping = inode->i_mapping;
843 f->f_path.dentry = dentry;
844 f->f_path.mnt = mnt;
845 f->f_pos = 0;
846 f->f_op = fops_get(inode->i_fop);
847 file_move(f, &inode->i_sb->s_files);
849 error = security_dentry_open(f, cred);
850 if (error)
851 goto cleanup_all;
853 if (!open && f->f_op)
854 open = f->f_op->open;
855 if (open) {
856 error = open(inode, f);
857 if (error)
858 goto cleanup_all;
860 ima_counts_get(f);
862 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
864 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
866 /* NB: we're sure to have correct a_ops only after f_op->open */
867 if (f->f_flags & O_DIRECT) {
868 if (!f->f_mapping->a_ops ||
869 ((!f->f_mapping->a_ops->direct_IO) &&
870 (!f->f_mapping->a_ops->get_xip_mem))) {
871 fput(f);
872 f = ERR_PTR(-EINVAL);
876 return f;
878 cleanup_all:
879 fops_put(f->f_op);
880 if (f->f_mode & FMODE_WRITE) {
881 put_write_access(inode);
882 if (!special_file(inode->i_mode)) {
884 * We don't consider this a real
885 * mnt_want/drop_write() pair
886 * because it all happenend right
887 * here, so just reset the state.
889 file_reset_write(f);
890 mnt_drop_write(mnt);
893 file_kill(f);
894 f->f_path.dentry = NULL;
895 f->f_path.mnt = NULL;
896 cleanup_file:
897 put_filp(f);
898 dput(dentry);
899 mntput(mnt);
900 return ERR_PTR(error);
904 * lookup_instantiate_filp - instantiates the open intent filp
905 * @nd: pointer to nameidata
906 * @dentry: pointer to dentry
907 * @open: open callback
909 * Helper for filesystems that want to use lookup open intents and pass back
910 * a fully instantiated struct file to the caller.
911 * This function is meant to be called from within a filesystem's
912 * lookup method.
913 * Beware of calling it for non-regular files! Those ->open methods might block
914 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
915 * leading to a deadlock, as nobody can open that fifo anymore, because
916 * another process to open fifo will block on locked parent when doing lookup).
917 * Note that in case of error, nd->intent.open.file is destroyed, but the
918 * path information remains valid.
919 * If the open callback is set to NULL, then the standard f_op->open()
920 * filesystem callback is substituted.
922 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
923 int (*open)(struct inode *, struct file *))
925 const struct cred *cred = current_cred();
927 if (IS_ERR(nd->intent.open.file))
928 goto out;
929 if (IS_ERR(dentry))
930 goto out_err;
931 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
932 nd->intent.open.file,
933 open, cred);
934 out:
935 return nd->intent.open.file;
936 out_err:
937 release_open_intent(nd);
938 nd->intent.open.file = (struct file *)dentry;
939 goto out;
941 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
944 * nameidata_to_filp - convert a nameidata to an open filp.
945 * @nd: pointer to nameidata
946 * @flags: open flags
948 * Note that this function destroys the original nameidata
950 struct file *nameidata_to_filp(struct nameidata *nd)
952 const struct cred *cred = current_cred();
953 struct file *filp;
955 /* Pick up the filp from the open intent */
956 filp = nd->intent.open.file;
957 /* Has the filesystem initialised the file for us? */
958 if (filp->f_path.dentry == NULL)
959 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
960 NULL, cred);
961 else
962 path_put(&nd->path);
963 return filp;
967 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
968 * error.
970 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
971 const struct cred *cred)
973 int error;
974 struct file *f;
976 validate_creds(cred);
979 * We must always pass in a valid mount pointer. Historically
980 * callers got away with not passing it, but we must enforce this at
981 * the earliest possible point now to avoid strange problems deep in the
982 * filesystem stack.
984 if (!mnt) {
985 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
986 dump_stack();
987 return ERR_PTR(-EINVAL);
990 error = -ENFILE;
991 f = get_empty_filp();
992 if (f == NULL) {
993 dput(dentry);
994 mntput(mnt);
995 return ERR_PTR(error);
998 f->f_flags = flags;
999 return __dentry_open(dentry, mnt, f, NULL, cred);
1001 EXPORT_SYMBOL(dentry_open);
1003 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1005 struct fdtable *fdt = files_fdtable(files);
1006 __FD_CLR(fd, fdt->open_fds);
1007 if (fd < files->next_fd)
1008 files->next_fd = fd;
1011 void put_unused_fd(unsigned int fd)
1013 struct files_struct *files = current->files;
1014 spin_lock(&files->file_lock);
1015 __put_unused_fd(files, fd);
1016 spin_unlock(&files->file_lock);
1019 EXPORT_SYMBOL(put_unused_fd);
1022 * Install a file pointer in the fd array.
1024 * The VFS is full of places where we drop the files lock between
1025 * setting the open_fds bitmap and installing the file in the file
1026 * array. At any such point, we are vulnerable to a dup2() race
1027 * installing a file in the array before us. We need to detect this and
1028 * fput() the struct file we are about to overwrite in this case.
1030 * It should never happen - if we allow dup2() do it, _really_ bad things
1031 * will follow.
1034 void fd_install(unsigned int fd, struct file *file)
1036 struct files_struct *files = current->files;
1037 struct fdtable *fdt;
1038 spin_lock(&files->file_lock);
1039 fdt = files_fdtable(files);
1040 BUG_ON(fdt->fd[fd] != NULL);
1041 rcu_assign_pointer(fdt->fd[fd], file);
1042 spin_unlock(&files->file_lock);
1045 EXPORT_SYMBOL(fd_install);
1047 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1049 char *tmp = getname(filename);
1050 int fd = PTR_ERR(tmp);
1052 if (!IS_ERR(tmp)) {
1053 fd = get_unused_fd_flags(flags);
1054 if (fd >= 0) {
1055 struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1056 if (IS_ERR(f)) {
1057 put_unused_fd(fd);
1058 fd = PTR_ERR(f);
1059 } else {
1060 fsnotify_open(f->f_path.dentry);
1061 fd_install(fd, f);
1064 putname(tmp);
1066 return fd;
1069 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1071 long ret;
1073 if (force_o_largefile())
1074 flags |= O_LARGEFILE;
1076 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1077 /* avoid REGPARM breakage on x86: */
1078 asmlinkage_protect(3, ret, filename, flags, mode);
1079 return ret;
1082 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1083 int, mode)
1085 long ret;
1087 if (force_o_largefile())
1088 flags |= O_LARGEFILE;
1090 ret = do_sys_open(dfd, filename, flags, mode);
1091 /* avoid REGPARM breakage on x86: */
1092 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1093 return ret;
1096 #ifndef __alpha__
1099 * For backward compatibility? Maybe this should be moved
1100 * into arch/i386 instead?
1102 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1104 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1107 #endif
1110 * "id" is the POSIX thread ID. We use the
1111 * files pointer for this..
1113 int filp_close(struct file *filp, fl_owner_t id)
1115 int retval = 0;
1117 if (!file_count(filp)) {
1118 printk(KERN_ERR "VFS: Close: file count is 0\n");
1119 return 0;
1122 if (filp->f_op && filp->f_op->flush)
1123 retval = filp->f_op->flush(filp, id);
1125 dnotify_flush(filp, id);
1126 locks_remove_posix(filp, id);
1127 fput(filp);
1128 return retval;
1131 EXPORT_SYMBOL(filp_close);
1134 * Careful here! We test whether the file pointer is NULL before
1135 * releasing the fd. This ensures that one clone task can't release
1136 * an fd while another clone is opening it.
1138 SYSCALL_DEFINE1(close, unsigned int, fd)
1140 struct file * filp;
1141 struct files_struct *files = current->files;
1142 struct fdtable *fdt;
1143 int retval;
1145 spin_lock(&files->file_lock);
1146 fdt = files_fdtable(files);
1147 if (fd >= fdt->max_fds)
1148 goto out_unlock;
1149 filp = fdt->fd[fd];
1150 if (!filp)
1151 goto out_unlock;
1152 rcu_assign_pointer(fdt->fd[fd], NULL);
1153 FD_CLR(fd, fdt->close_on_exec);
1154 __put_unused_fd(files, fd);
1155 spin_unlock(&files->file_lock);
1156 retval = filp_close(filp, files);
1158 /* can't restart close syscall because file table entry was cleared */
1159 if (unlikely(retval == -ERESTARTSYS ||
1160 retval == -ERESTARTNOINTR ||
1161 retval == -ERESTARTNOHAND ||
1162 retval == -ERESTART_RESTARTBLOCK))
1163 retval = -EINTR;
1165 return retval;
1167 out_unlock:
1168 spin_unlock(&files->file_lock);
1169 return -EBADF;
1171 EXPORT_SYMBOL(sys_close);
1174 * This routine simulates a hangup on the tty, to arrange that users
1175 * are given clean terminals at login time.
1177 SYSCALL_DEFINE0(vhangup)
1179 if (capable(CAP_SYS_TTY_CONFIG)) {
1180 tty_vhangup_self();
1181 return 0;
1183 return -EPERM;
1187 * Called when an inode is about to be open.
1188 * We use this to disallow opening large files on 32bit systems if
1189 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1190 * on this flag in sys_open.
1192 int generic_file_open(struct inode * inode, struct file * filp)
1194 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1195 return -EOVERFLOW;
1196 return 0;
1199 EXPORT_SYMBOL(generic_file_open);
1202 * This is used by subsystems that don't want seekable
1203 * file descriptors
1205 int nonseekable_open(struct inode *inode, struct file *filp)
1207 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1208 return 0;
1211 EXPORT_SYMBOL(nonseekable_open);