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
do_truncate2(struct vfsmount * mnt,struct dentry * dentry,loff_t length,unsigned int time_attrs,struct file * filp)37 int do_truncate2(struct vfsmount *mnt, struct dentry *dentry, loff_t length,
38 unsigned int time_attrs, 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_change2(mnt, dentry, &newattrs);
61 mutex_unlock(&dentry->d_inode->i_mutex);
62 return ret;
63 }
do_truncate(struct dentry * dentry,loff_t length,unsigned int time_attrs,struct file * filp)64 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
65 struct file *filp)
66 {
67 return do_truncate2(NULL, dentry, length, time_attrs, filp);
68 }
69
vfs_truncate(struct path * path,loff_t length)70 long vfs_truncate(struct path *path, loff_t length)
71 {
72 struct inode *inode;
73 struct vfsmount *mnt;
74 long error;
75
76 inode = path->dentry->d_inode;
77 mnt = path->mnt;
78
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 if (S_ISDIR(inode->i_mode))
81 return -EISDIR;
82 if (!S_ISREG(inode->i_mode))
83 return -EINVAL;
84
85 error = mnt_want_write(path->mnt);
86 if (error)
87 goto out;
88
89 error = inode_permission2(mnt, inode, MAY_WRITE);
90 if (error)
91 goto mnt_drop_write_and_out;
92
93 error = -EPERM;
94 if (IS_APPEND(inode))
95 goto mnt_drop_write_and_out;
96
97 error = get_write_access(inode);
98 if (error)
99 goto mnt_drop_write_and_out;
100
101 /*
102 * Make sure that there are no leases. get_write_access() protects
103 * against the truncate racing with a lease-granting setlease().
104 */
105 error = break_lease(inode, O_WRONLY);
106 if (error)
107 goto put_write_and_out;
108
109 error = locks_verify_truncate(inode, NULL, length);
110 if (!error)
111 error = security_path_truncate(path);
112 if (!error)
113 error = do_truncate2(mnt, path->dentry, length, 0, NULL);
114
115 put_write_and_out:
116 put_write_access(inode);
117 mnt_drop_write_and_out:
118 mnt_drop_write(path->mnt);
119 out:
120 return error;
121 }
122 EXPORT_SYMBOL_GPL(vfs_truncate);
123
do_sys_truncate(const char __user * pathname,loff_t length)124 static long do_sys_truncate(const char __user *pathname, loff_t length)
125 {
126 unsigned int lookup_flags = LOOKUP_FOLLOW;
127 struct path path;
128 int error;
129
130 if (length < 0) /* sorry, but loff_t says... */
131 return -EINVAL;
132
133 retry:
134 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
135 if (!error) {
136 error = vfs_truncate(&path, length);
137 path_put(&path);
138 }
139 if (retry_estale(error, lookup_flags)) {
140 lookup_flags |= LOOKUP_REVAL;
141 goto retry;
142 }
143 return error;
144 }
145
SYSCALL_DEFINE2(truncate,const char __user *,path,long,length)146 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
147 {
148 return do_sys_truncate(path, length);
149 }
150
151 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(truncate,const char __user *,path,compat_off_t,length)152 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
153 {
154 return do_sys_truncate(path, length);
155 }
156 #endif
157
do_sys_ftruncate(unsigned int fd,loff_t length,int small)158 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
159 {
160 struct inode *inode;
161 struct dentry *dentry;
162 struct vfsmount *mnt;
163 struct fd f;
164 int error;
165
166 error = -EINVAL;
167 if (length < 0)
168 goto out;
169 error = -EBADF;
170 f = fdget(fd);
171 if (!f.file)
172 goto out;
173
174 /* explicitly opened as large or we are on 64-bit box */
175 if (f.file->f_flags & O_LARGEFILE)
176 small = 0;
177
178 dentry = f.file->f_path.dentry;
179 mnt = f.file->f_path.mnt;
180 inode = dentry->d_inode;
181 error = -EINVAL;
182 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
183 goto out_putf;
184
185 error = -EINVAL;
186 /* Cannot ftruncate over 2^31 bytes without large file support */
187 if (small && length > MAX_NON_LFS)
188 goto out_putf;
189
190 error = -EPERM;
191 if (IS_APPEND(inode))
192 goto out_putf;
193
194 sb_start_write(inode->i_sb);
195 error = locks_verify_truncate(inode, f.file, length);
196 if (!error)
197 error = security_path_truncate(&f.file->f_path);
198 if (!error)
199 error = do_truncate2(mnt, dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
200 sb_end_write(inode->i_sb);
201 out_putf:
202 fdput(f);
203 out:
204 return error;
205 }
206
SYSCALL_DEFINE2(ftruncate,unsigned int,fd,unsigned long,length)207 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
208 {
209 return do_sys_ftruncate(fd, length, 1);
210 }
211
212 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(ftruncate,unsigned int,fd,compat_ulong_t,length)213 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
214 {
215 return do_sys_ftruncate(fd, length, 1);
216 }
217 #endif
218
219 /* LFS versions of truncate are only needed on 32 bit machines */
220 #if BITS_PER_LONG == 32
SYSCALL_DEFINE2(truncate64,const char __user *,path,loff_t,length)221 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
222 {
223 return do_sys_truncate(path, length);
224 }
225
SYSCALL_DEFINE2(ftruncate64,unsigned int,fd,loff_t,length)226 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
227 {
228 return do_sys_ftruncate(fd, length, 0);
229 }
230 #endif /* BITS_PER_LONG == 32 */
231
232
do_fallocate(struct file * file,int mode,loff_t offset,loff_t len)233 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
234 {
235 struct inode *inode = file_inode(file);
236 long ret;
237
238 if (offset < 0 || len <= 0)
239 return -EINVAL;
240
241 /* Return error if mode is not supported */
242 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
243 return -EOPNOTSUPP;
244
245 /* Punch hole must have keep size set */
246 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
247 !(mode & FALLOC_FL_KEEP_SIZE))
248 return -EOPNOTSUPP;
249
250 if (!(file->f_mode & FMODE_WRITE))
251 return -EBADF;
252
253 /* It's not possible punch hole on append only file */
254 if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
255 return -EPERM;
256
257 if (IS_IMMUTABLE(inode))
258 return -EPERM;
259
260 /*
261 * Revalidate the write permissions, in case security policy has
262 * changed since the files were opened.
263 */
264 ret = security_file_permission(file, MAY_WRITE);
265 if (ret)
266 return ret;
267
268 if (S_ISFIFO(inode->i_mode))
269 return -ESPIPE;
270
271 /*
272 * Let individual file system decide if it supports preallocation
273 * for directories or not.
274 */
275 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
276 return -ENODEV;
277
278 /* Check for wrap through zero too */
279 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
280 return -EFBIG;
281
282 if (!file->f_op->fallocate)
283 return -EOPNOTSUPP;
284
285 sb_start_write(inode->i_sb);
286 ret = file->f_op->fallocate(file, mode, offset, len);
287 sb_end_write(inode->i_sb);
288 return ret;
289 }
290
SYSCALL_DEFINE4(fallocate,int,fd,int,mode,loff_t,offset,loff_t,len)291 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
292 {
293 struct fd f = fdget(fd);
294 int error = -EBADF;
295
296 if (f.file) {
297 error = do_fallocate(f.file, mode, offset, len);
298 fdput(f);
299 }
300 return error;
301 }
302
303 /*
304 * access() needs to use the real uid/gid, not the effective uid/gid.
305 * We do this by temporarily clearing all FS-related capabilities and
306 * switching the fsuid/fsgid around to the real ones.
307 */
SYSCALL_DEFINE3(faccessat,int,dfd,const char __user *,filename,int,mode)308 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
309 {
310 const struct cred *old_cred;
311 struct cred *override_cred;
312 struct path path;
313 struct inode *inode;
314 struct vfsmount *mnt;
315 int res;
316 unsigned int lookup_flags = LOOKUP_FOLLOW;
317
318 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
319 return -EINVAL;
320
321 override_cred = prepare_creds();
322 if (!override_cred)
323 return -ENOMEM;
324
325 override_cred->fsuid = override_cred->uid;
326 override_cred->fsgid = override_cred->gid;
327
328 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
329 /* Clear the capabilities if we switch to a non-root user */
330 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
331 if (!uid_eq(override_cred->uid, root_uid))
332 cap_clear(override_cred->cap_effective);
333 else
334 override_cred->cap_effective =
335 override_cred->cap_permitted;
336 }
337
338 old_cred = override_creds(override_cred);
339 retry:
340 res = user_path_at(dfd, filename, lookup_flags, &path);
341 if (res)
342 goto out;
343
344 inode = path.dentry->d_inode;
345 mnt = path.mnt;
346
347 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
348 /*
349 * MAY_EXEC on regular files is denied if the fs is mounted
350 * with the "noexec" flag.
351 */
352 res = -EACCES;
353 if (path.mnt->mnt_flags & MNT_NOEXEC)
354 goto out_path_release;
355 }
356
357 res = inode_permission2(mnt, inode, mode | MAY_ACCESS);
358 /* SuS v2 requires we report a read only fs too */
359 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
360 goto out_path_release;
361 /*
362 * This is a rare case where using __mnt_is_readonly()
363 * is OK without a mnt_want/drop_write() pair. Since
364 * no actual write to the fs is performed here, we do
365 * not need to telegraph to that to anyone.
366 *
367 * By doing this, we accept that this access is
368 * inherently racy and know that the fs may change
369 * state before we even see this result.
370 */
371 if (__mnt_is_readonly(path.mnt))
372 res = -EROFS;
373
374 out_path_release:
375 path_put(&path);
376 if (retry_estale(res, lookup_flags)) {
377 lookup_flags |= LOOKUP_REVAL;
378 goto retry;
379 }
380 out:
381 revert_creds(old_cred);
382 put_cred(override_cred);
383 return res;
384 }
385
SYSCALL_DEFINE2(access,const char __user *,filename,int,mode)386 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
387 {
388 return sys_faccessat(AT_FDCWD, filename, mode);
389 }
390
SYSCALL_DEFINE1(chdir,const char __user *,filename)391 SYSCALL_DEFINE1(chdir, const char __user *, filename)
392 {
393 struct path path;
394 int error;
395 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
396 retry:
397 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
398 if (error)
399 goto out;
400
401 error = inode_permission2(path.mnt, path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
402 if (error)
403 goto dput_and_out;
404
405 set_fs_pwd(current->fs, &path);
406
407 dput_and_out:
408 path_put(&path);
409 if (retry_estale(error, lookup_flags)) {
410 lookup_flags |= LOOKUP_REVAL;
411 goto retry;
412 }
413 out:
414 return error;
415 }
416
SYSCALL_DEFINE1(fchdir,unsigned int,fd)417 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
418 {
419 struct fd f = fdget_raw(fd);
420 struct inode *inode;
421 struct vfsmount *mnt;
422 int error = -EBADF;
423
424 error = -EBADF;
425 if (!f.file)
426 goto out;
427
428 inode = file_inode(f.file);
429 mnt = f.file->f_path.mnt;
430
431 error = -ENOTDIR;
432 if (!S_ISDIR(inode->i_mode))
433 goto out_putf;
434
435 error = inode_permission2(mnt, inode, MAY_EXEC | MAY_CHDIR);
436 if (!error)
437 set_fs_pwd(current->fs, &f.file->f_path);
438 out_putf:
439 fdput(f);
440 out:
441 return error;
442 }
443
SYSCALL_DEFINE1(chroot,const char __user *,filename)444 SYSCALL_DEFINE1(chroot, const char __user *, filename)
445 {
446 struct path path;
447 int error;
448 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
449 retry:
450 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
451 if (error)
452 goto out;
453
454 error = inode_permission2(path.mnt, path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
455 if (error)
456 goto dput_and_out;
457
458 error = -EPERM;
459 if (!nsown_capable(CAP_SYS_CHROOT))
460 goto dput_and_out;
461 error = security_path_chroot(&path);
462 if (error)
463 goto dput_and_out;
464
465 set_fs_root(current->fs, &path);
466 error = 0;
467 dput_and_out:
468 path_put(&path);
469 if (retry_estale(error, lookup_flags)) {
470 lookup_flags |= LOOKUP_REVAL;
471 goto retry;
472 }
473 out:
474 return error;
475 }
476
chmod_common(struct path * path,umode_t mode)477 static int chmod_common(struct path *path, umode_t mode)
478 {
479 struct inode *inode = path->dentry->d_inode;
480 struct iattr newattrs;
481 int error;
482
483 error = mnt_want_write(path->mnt);
484 if (error)
485 return error;
486 mutex_lock(&inode->i_mutex);
487 error = security_path_chmod(path, mode);
488 if (error)
489 goto out_unlock;
490 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
491 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
492 error = notify_change2(path->mnt, path->dentry, &newattrs);
493 out_unlock:
494 mutex_unlock(&inode->i_mutex);
495 mnt_drop_write(path->mnt);
496 return error;
497 }
498
SYSCALL_DEFINE2(fchmod,unsigned int,fd,umode_t,mode)499 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
500 {
501 struct file * file;
502 int err = -EBADF;
503
504 file = fget(fd);
505 if (file) {
506 audit_inode(NULL, file->f_path.dentry, 0);
507 err = chmod_common(&file->f_path, mode);
508 fput(file);
509 }
510 return err;
511 }
512
SYSCALL_DEFINE3(fchmodat,int,dfd,const char __user *,filename,umode_t,mode)513 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
514 {
515 struct path path;
516 int error;
517 unsigned int lookup_flags = LOOKUP_FOLLOW;
518 retry:
519 error = user_path_at(dfd, filename, lookup_flags, &path);
520 if (!error) {
521 error = chmod_common(&path, mode);
522 path_put(&path);
523 if (retry_estale(error, lookup_flags)) {
524 lookup_flags |= LOOKUP_REVAL;
525 goto retry;
526 }
527 }
528 return error;
529 }
530
SYSCALL_DEFINE2(chmod,const char __user *,filename,umode_t,mode)531 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
532 {
533 return sys_fchmodat(AT_FDCWD, filename, mode);
534 }
535
chown_common(struct path * path,uid_t user,gid_t group)536 static int chown_common(struct path *path, uid_t user, gid_t group)
537 {
538 struct inode *inode = path->dentry->d_inode;
539 int error;
540 struct iattr newattrs;
541 kuid_t uid;
542 kgid_t gid;
543
544 uid = make_kuid(current_user_ns(), user);
545 gid = make_kgid(current_user_ns(), group);
546
547 newattrs.ia_valid = ATTR_CTIME;
548 if (user != (uid_t) -1) {
549 if (!uid_valid(uid))
550 return -EINVAL;
551 newattrs.ia_valid |= ATTR_UID;
552 newattrs.ia_uid = uid;
553 }
554 if (group != (gid_t) -1) {
555 if (!gid_valid(gid))
556 return -EINVAL;
557 newattrs.ia_valid |= ATTR_GID;
558 newattrs.ia_gid = gid;
559 }
560 if (!S_ISDIR(inode->i_mode))
561 newattrs.ia_valid |=
562 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
563 mutex_lock(&inode->i_mutex);
564 error = security_path_chown(path, uid, gid);
565 if (!error)
566 error = notify_change2(path->mnt, path->dentry, &newattrs);
567 mutex_unlock(&inode->i_mutex);
568
569 return error;
570 }
571
SYSCALL_DEFINE5(fchownat,int,dfd,const char __user *,filename,uid_t,user,gid_t,group,int,flag)572 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
573 gid_t, group, int, flag)
574 {
575 struct path path;
576 int error = -EINVAL;
577 int lookup_flags;
578
579 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
580 goto out;
581
582 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
583 if (flag & AT_EMPTY_PATH)
584 lookup_flags |= LOOKUP_EMPTY;
585 retry:
586 error = user_path_at(dfd, filename, lookup_flags, &path);
587 if (error)
588 goto out;
589 error = mnt_want_write(path.mnt);
590 if (error)
591 goto out_release;
592 error = chown_common(&path, user, group);
593 mnt_drop_write(path.mnt);
594 out_release:
595 path_put(&path);
596 if (retry_estale(error, lookup_flags)) {
597 lookup_flags |= LOOKUP_REVAL;
598 goto retry;
599 }
600 out:
601 return error;
602 }
603
SYSCALL_DEFINE3(chown,const char __user *,filename,uid_t,user,gid_t,group)604 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
605 {
606 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
607 }
608
SYSCALL_DEFINE3(lchown,const char __user *,filename,uid_t,user,gid_t,group)609 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
610 {
611 return sys_fchownat(AT_FDCWD, filename, user, group,
612 AT_SYMLINK_NOFOLLOW);
613 }
614
SYSCALL_DEFINE3(fchown,unsigned int,fd,uid_t,user,gid_t,group)615 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
616 {
617 struct fd f = fdget(fd);
618 int error = -EBADF;
619
620 if (!f.file)
621 goto out;
622
623 error = mnt_want_write_file(f.file);
624 if (error)
625 goto out_fput;
626 audit_inode(NULL, f.file->f_path.dentry, 0);
627 error = chown_common(&f.file->f_path, user, group);
628 mnt_drop_write_file(f.file);
629 out_fput:
630 fdput(f);
631 out:
632 return error;
633 }
634
635 /*
636 * You have to be very careful that these write
637 * counts get cleaned up in error cases and
638 * upon __fput(). This should probably never
639 * be called outside of __dentry_open().
640 */
__get_file_write_access(struct inode * inode,struct vfsmount * mnt)641 static inline int __get_file_write_access(struct inode *inode,
642 struct vfsmount *mnt)
643 {
644 int error;
645 error = get_write_access(inode);
646 if (error)
647 return error;
648 /*
649 * Do not take mount writer counts on
650 * special files since no writes to
651 * the mount itself will occur.
652 */
653 if (!special_file(inode->i_mode)) {
654 /*
655 * Balanced in __fput()
656 */
657 error = __mnt_want_write(mnt);
658 if (error)
659 put_write_access(inode);
660 }
661 return error;
662 }
663
open_check_o_direct(struct file * f)664 int open_check_o_direct(struct file *f)
665 {
666 /* NB: we're sure to have correct a_ops only after f_op->open */
667 if (f->f_flags & O_DIRECT) {
668 if (!f->f_mapping->a_ops ||
669 ((!f->f_mapping->a_ops->direct_IO) &&
670 (!f->f_mapping->a_ops->get_xip_mem))) {
671 return -EINVAL;
672 }
673 }
674 return 0;
675 }
676
do_dentry_open(struct file * f,int (* open)(struct inode *,struct file *),const struct cred * cred)677 static int do_dentry_open(struct file *f,
678 int (*open)(struct inode *, struct file *),
679 const struct cred *cred)
680 {
681 static const struct file_operations empty_fops = {};
682 struct inode *inode;
683 int error;
684
685 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
686 FMODE_PREAD | FMODE_PWRITE;
687
688 if (unlikely(f->f_flags & O_PATH))
689 f->f_mode = FMODE_PATH;
690
691 path_get(&f->f_path);
692 inode = f->f_inode = f->f_path.dentry->d_inode;
693 if (f->f_mode & FMODE_WRITE) {
694 error = __get_file_write_access(inode, f->f_path.mnt);
695 if (error)
696 goto cleanup_file;
697 if (!special_file(inode->i_mode))
698 file_take_write(f);
699 }
700
701 f->f_mapping = inode->i_mapping;
702 file_sb_list_add(f, inode->i_sb);
703
704 if (unlikely(f->f_mode & FMODE_PATH)) {
705 f->f_op = &empty_fops;
706 return 0;
707 }
708
709 f->f_op = fops_get(inode->i_fop);
710
711 error = security_file_open(f, cred);
712 if (error)
713 goto cleanup_all;
714
715 error = break_lease(inode, f->f_flags);
716 if (error)
717 goto cleanup_all;
718
719 if (!open && f->f_op)
720 open = f->f_op->open;
721 if (open) {
722 error = open(inode, f);
723 if (error)
724 goto cleanup_all;
725 }
726 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
727 i_readcount_inc(inode);
728
729 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
730
731 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
732
733 return 0;
734
735 cleanup_all:
736 fops_put(f->f_op);
737 file_sb_list_del(f);
738 if (f->f_mode & FMODE_WRITE) {
739 put_write_access(inode);
740 if (!special_file(inode->i_mode)) {
741 /*
742 * We don't consider this a real
743 * mnt_want/drop_write() pair
744 * because it all happenend right
745 * here, so just reset the state.
746 */
747 file_reset_write(f);
748 __mnt_drop_write(f->f_path.mnt);
749 }
750 }
751 cleanup_file:
752 path_put(&f->f_path);
753 f->f_path.mnt = NULL;
754 f->f_path.dentry = NULL;
755 f->f_inode = NULL;
756 return error;
757 }
758
759 /**
760 * finish_open - finish opening a file
761 * @od: opaque open data
762 * @dentry: pointer to dentry
763 * @open: open callback
764 *
765 * This can be used to finish opening a file passed to i_op->atomic_open().
766 *
767 * If the open callback is set to NULL, then the standard f_op->open()
768 * filesystem callback is substituted.
769 */
finish_open(struct file * file,struct dentry * dentry,int (* open)(struct inode *,struct file *),int * opened)770 int finish_open(struct file *file, struct dentry *dentry,
771 int (*open)(struct inode *, struct file *),
772 int *opened)
773 {
774 int error;
775 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
776
777 file->f_path.dentry = dentry;
778 error = do_dentry_open(file, open, current_cred());
779 if (!error)
780 *opened |= FILE_OPENED;
781
782 return error;
783 }
784 EXPORT_SYMBOL(finish_open);
785
786 /**
787 * finish_no_open - finish ->atomic_open() without opening the file
788 *
789 * @od: opaque open data
790 * @dentry: dentry or NULL (as returned from ->lookup())
791 *
792 * This can be used to set the result of a successful lookup in ->atomic_open().
793 * The filesystem's atomic_open() method shall return NULL after calling this.
794 */
finish_no_open(struct file * file,struct dentry * dentry)795 int finish_no_open(struct file *file, struct dentry *dentry)
796 {
797 file->f_path.dentry = dentry;
798 return 1;
799 }
800 EXPORT_SYMBOL(finish_no_open);
801
dentry_open(const struct path * path,int flags,const struct cred * cred)802 struct file *dentry_open(const struct path *path, int flags,
803 const struct cred *cred)
804 {
805 int error;
806 struct file *f;
807
808 validate_creds(cred);
809
810 /* We must always pass in a valid mount pointer. */
811 BUG_ON(!path->mnt);
812
813 f = get_empty_filp();
814 if (!IS_ERR(f)) {
815 f->f_flags = flags;
816 f->f_path = *path;
817 error = do_dentry_open(f, NULL, cred);
818 if (!error) {
819 /* from now on we need fput() to dispose of f */
820 error = open_check_o_direct(f);
821 if (error) {
822 fput(f);
823 f = ERR_PTR(error);
824 }
825 } else {
826 put_filp(f);
827 f = ERR_PTR(error);
828 }
829 }
830 return f;
831 }
832 EXPORT_SYMBOL(dentry_open);
833
build_open_flags(int flags,umode_t mode,struct open_flags * op)834 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
835 {
836 int lookup_flags = 0;
837 int acc_mode;
838
839 if (flags & O_CREAT)
840 op->mode = (mode & S_IALLUGO) | S_IFREG;
841 else
842 op->mode = 0;
843
844 /* Must never be set by userspace */
845 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
846
847 /*
848 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
849 * check for O_DSYNC if the need any syncing at all we enforce it's
850 * always set instead of having to deal with possibly weird behaviour
851 * for malicious applications setting only __O_SYNC.
852 */
853 if (flags & __O_SYNC)
854 flags |= O_DSYNC;
855
856 /*
857 * If we have O_PATH in the open flag. Then we
858 * cannot have anything other than the below set of flags
859 */
860 if (flags & O_PATH) {
861 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
862 acc_mode = 0;
863 } else {
864 acc_mode = MAY_OPEN | ACC_MODE(flags);
865 }
866
867 op->open_flag = flags;
868
869 /* O_TRUNC implies we need access checks for write permissions */
870 if (flags & O_TRUNC)
871 acc_mode |= MAY_WRITE;
872
873 /* Allow the LSM permission hook to distinguish append
874 access from general write access. */
875 if (flags & O_APPEND)
876 acc_mode |= MAY_APPEND;
877
878 op->acc_mode = acc_mode;
879
880 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
881
882 if (flags & O_CREAT) {
883 op->intent |= LOOKUP_CREATE;
884 if (flags & O_EXCL)
885 op->intent |= LOOKUP_EXCL;
886 }
887
888 if (flags & O_DIRECTORY)
889 lookup_flags |= LOOKUP_DIRECTORY;
890 if (!(flags & O_NOFOLLOW))
891 lookup_flags |= LOOKUP_FOLLOW;
892 return lookup_flags;
893 }
894
895 /**
896 * file_open_name - open file and return file pointer
897 *
898 * @name: struct filename containing path to open
899 * @flags: open flags as per the open(2) second argument
900 * @mode: mode for the new file if O_CREAT is set, else ignored
901 *
902 * This is the helper to open a file from kernelspace if you really
903 * have to. But in generally you should not do this, so please move
904 * along, nothing to see here..
905 */
file_open_name(struct filename * name,int flags,umode_t mode)906 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
907 {
908 struct open_flags op;
909 int lookup = build_open_flags(flags, mode, &op);
910 return do_filp_open(AT_FDCWD, name, &op, lookup);
911 }
912
913 /**
914 * filp_open - open file and return file pointer
915 *
916 * @filename: path to open
917 * @flags: open flags as per the open(2) second argument
918 * @mode: mode for the new file if O_CREAT is set, else ignored
919 *
920 * This is the helper to open a file from kernelspace if you really
921 * have to. But in generally you should not do this, so please move
922 * along, nothing to see here..
923 */
filp_open(const char * filename,int flags,umode_t mode)924 struct file *filp_open(const char *filename, int flags, umode_t mode)
925 {
926 struct filename name = {.name = filename};
927 return file_open_name(&name, flags, mode);
928 }
929 EXPORT_SYMBOL(filp_open);
930
file_open_root(struct dentry * dentry,struct vfsmount * mnt,const char * filename,int flags)931 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
932 const char *filename, int flags)
933 {
934 struct open_flags op;
935 int lookup = build_open_flags(flags, 0, &op);
936 if (flags & O_CREAT)
937 return ERR_PTR(-EINVAL);
938 if (!filename && (flags & O_DIRECTORY))
939 if (!dentry->d_inode->i_op->lookup)
940 return ERR_PTR(-ENOTDIR);
941 return do_file_open_root(dentry, mnt, filename, &op, lookup);
942 }
943 EXPORT_SYMBOL(file_open_root);
944
do_sys_open(int dfd,const char __user * filename,int flags,umode_t mode)945 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
946 {
947 struct open_flags op;
948 int lookup = build_open_flags(flags, mode, &op);
949 struct filename *tmp = getname(filename);
950 int fd = PTR_ERR(tmp);
951
952 if (!IS_ERR(tmp)) {
953 fd = get_unused_fd_flags(flags);
954 if (fd >= 0) {
955 struct file *f = do_filp_open(dfd, tmp, &op, lookup);
956 if (IS_ERR(f)) {
957 put_unused_fd(fd);
958 fd = PTR_ERR(f);
959 } else {
960 fsnotify_open(f);
961 fd_install(fd, f);
962 }
963 }
964 putname(tmp);
965 }
966 return fd;
967 }
968
SYSCALL_DEFINE3(open,const char __user *,filename,int,flags,umode_t,mode)969 SYSCALL_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
SYSCALL_DEFINE4(openat,int,dfd,const char __user *,filename,int,flags,umode_t,mode)977 SYSCALL_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 */
SYSCALL_DEFINE2(creat,const char __user *,pathname,umode_t,mode)992 SYSCALL_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 */
filp_close(struct file * filp,fl_owner_t id)1003 int 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
1023 EXPORT_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 */
SYSCALL_DEFINE1(close,unsigned int,fd)1030 SYSCALL_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 }
1043 EXPORT_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 */
SYSCALL_DEFINE0(vhangup)1049 SYSCALL_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 */
generic_file_open(struct inode * inode,struct file * filp)1064 int 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
1071 EXPORT_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 */
nonseekable_open(struct inode * inode,struct file * filp)1079 int nonseekable_open(struct inode *inode, struct file *filp)
1080 {
1081 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1082 return 0;
1083 }
1084
1085 EXPORT_SYMBOL(nonseekable_open);
1086