1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/open.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/string.h>
9 #include <linux/mm.h>
10 #include <linux/file.h>
11 #include <linux/fdtable.h>
12 #include <linux/fsnotify.h>
13 #include <linux/module.h>
14 #include <linux/tty.h>
15 #include <linux/namei.h>
16 #include <linux/backing-dev.h>
17 #include <linux/capability.h>
18 #include <linux/securebits.h>
19 #include <linux/security.h>
20 #include <linux/mount.h>
21 #include <linux/fcntl.h>
22 #include <linux/slab.h>
23 #include <linux/uaccess.h>
24 #include <linux/fs.h>
25 #include <linux/personality.h>
26 #include <linux/pagemap.h>
27 #include <linux/syscalls.h>
28 #include <linux/rcupdate.h>
29 #include <linux/audit.h>
30 #include <linux/falloc.h>
31 #include <linux/fs_struct.h>
32 #include <linux/ima.h>
33 #include <linux/dnotify.h>
34 #include <linux/compat.h>
35 #include <linux/mnt_idmapping.h>
36 
37 #include "internal.h"
38 
do_truncate(struct user_namespace * mnt_userns,struct dentry * dentry,loff_t length,unsigned int time_attrs,struct file * filp)39 int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
40 		loff_t length, unsigned int time_attrs, struct file *filp)
41 {
42 	int ret;
43 	struct iattr newattrs;
44 
45 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46 	if (length < 0)
47 		return -EINVAL;
48 
49 	newattrs.ia_size = length;
50 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
51 	if (filp) {
52 		newattrs.ia_file = filp;
53 		newattrs.ia_valid |= ATTR_FILE;
54 	}
55 
56 	/* Remove suid, sgid, and file capabilities on truncate too */
57 	ret = dentry_needs_remove_privs(dentry);
58 	if (ret < 0)
59 		return ret;
60 	if (ret)
61 		newattrs.ia_valid |= ret | ATTR_FORCE;
62 
63 	inode_lock(dentry->d_inode);
64 	/* Note any delegations or leases have already been broken: */
65 	ret = notify_change(mnt_userns, dentry, &newattrs, NULL);
66 	inode_unlock(dentry->d_inode);
67 	return ret;
68 }
69 
vfs_truncate(const struct path * path,loff_t length)70 long vfs_truncate(const struct path *path, loff_t length)
71 {
72 	struct user_namespace *mnt_userns;
73 	struct inode *inode;
74 	long error;
75 
76 	inode = path->dentry->d_inode;
77 
78 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 	if (S_ISDIR(inode->i_mode))
80 		return -EISDIR;
81 	if (!S_ISREG(inode->i_mode))
82 		return -EINVAL;
83 
84 	error = mnt_want_write(path->mnt);
85 	if (error)
86 		goto out;
87 
88 	mnt_userns = mnt_user_ns(path->mnt);
89 	error = inode_permission(mnt_userns, 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 = security_path_truncate(path);
110 	if (!error)
111 		error = do_truncate(mnt_userns, path->dentry, length, 0, NULL);
112 
113 put_write_and_out:
114 	put_write_access(inode);
115 mnt_drop_write_and_out:
116 	mnt_drop_write(path->mnt);
117 out:
118 	return error;
119 }
120 EXPORT_SYMBOL_GPL(vfs_truncate);
121 
do_sys_truncate(const char __user * pathname,loff_t length)122 long do_sys_truncate(const char __user *pathname, loff_t length)
123 {
124 	unsigned int lookup_flags = LOOKUP_FOLLOW;
125 	struct path path;
126 	int error;
127 
128 	if (length < 0)	/* sorry, but loff_t says... */
129 		return -EINVAL;
130 
131 retry:
132 	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
133 	if (!error) {
134 		error = vfs_truncate(&path, length);
135 		path_put(&path);
136 	}
137 	if (retry_estale(error, lookup_flags)) {
138 		lookup_flags |= LOOKUP_REVAL;
139 		goto retry;
140 	}
141 	return error;
142 }
143 
SYSCALL_DEFINE2(truncate,const char __user *,path,long,length)144 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
145 {
146 	return do_sys_truncate(path, length);
147 }
148 
149 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(truncate,const char __user *,path,compat_off_t,length)150 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
151 {
152 	return do_sys_truncate(path, length);
153 }
154 #endif
155 
do_sys_ftruncate(unsigned int fd,loff_t length,int small)156 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
157 {
158 	struct inode *inode;
159 	struct dentry *dentry;
160 	struct fd f;
161 	int error;
162 
163 	error = -EINVAL;
164 	if (length < 0)
165 		goto out;
166 	error = -EBADF;
167 	f = fdget(fd);
168 	if (!f.file)
169 		goto out;
170 
171 	/* explicitly opened as large or we are on 64-bit box */
172 	if (f.file->f_flags & O_LARGEFILE)
173 		small = 0;
174 
175 	dentry = f.file->f_path.dentry;
176 	inode = dentry->d_inode;
177 	error = -EINVAL;
178 	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
179 		goto out_putf;
180 
181 	error = -EINVAL;
182 	/* Cannot ftruncate over 2^31 bytes without large file support */
183 	if (small && length > MAX_NON_LFS)
184 		goto out_putf;
185 
186 	error = -EPERM;
187 	/* Check IS_APPEND on real upper inode */
188 	if (IS_APPEND(file_inode(f.file)))
189 		goto out_putf;
190 	sb_start_write(inode->i_sb);
191 	error = security_path_truncate(&f.file->f_path);
192 	if (!error)
193 		error = do_truncate(file_mnt_user_ns(f.file), dentry, length,
194 				    ATTR_MTIME | ATTR_CTIME, f.file);
195 	sb_end_write(inode->i_sb);
196 out_putf:
197 	fdput(f);
198 out:
199 	return error;
200 }
201 
SYSCALL_DEFINE2(ftruncate,unsigned int,fd,unsigned long,length)202 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
203 {
204 	return do_sys_ftruncate(fd, length, 1);
205 }
206 
207 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(ftruncate,unsigned int,fd,compat_ulong_t,length)208 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
209 {
210 	return do_sys_ftruncate(fd, length, 1);
211 }
212 #endif
213 
214 /* LFS versions of truncate are only needed on 32 bit machines */
215 #if BITS_PER_LONG == 32
SYSCALL_DEFINE2(truncate64,const char __user *,path,loff_t,length)216 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
217 {
218 	return do_sys_truncate(path, length);
219 }
220 
SYSCALL_DEFINE2(ftruncate64,unsigned int,fd,loff_t,length)221 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
222 {
223 	return do_sys_ftruncate(fd, length, 0);
224 }
225 #endif /* BITS_PER_LONG == 32 */
226 
227 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
COMPAT_SYSCALL_DEFINE3(truncate64,const char __user *,pathname,compat_arg_u64_dual (length))228 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
229 		       compat_arg_u64_dual(length))
230 {
231 	return ksys_truncate(pathname, compat_arg_u64_glue(length));
232 }
233 #endif
234 
235 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
COMPAT_SYSCALL_DEFINE3(ftruncate64,unsigned int,fd,compat_arg_u64_dual (length))236 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
237 		       compat_arg_u64_dual(length))
238 {
239 	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
240 }
241 #endif
242 
vfs_fallocate(struct file * file,int mode,loff_t offset,loff_t len)243 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
244 {
245 	struct inode *inode = file_inode(file);
246 	long ret;
247 
248 	if (offset < 0 || len <= 0)
249 		return -EINVAL;
250 
251 	/* Return error if mode is not supported */
252 	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
253 		return -EOPNOTSUPP;
254 
255 	/* Punch hole and zero range are mutually exclusive */
256 	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
257 	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
258 		return -EOPNOTSUPP;
259 
260 	/* Punch hole must have keep size set */
261 	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
262 	    !(mode & FALLOC_FL_KEEP_SIZE))
263 		return -EOPNOTSUPP;
264 
265 	/* Collapse range should only be used exclusively. */
266 	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
267 	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
268 		return -EINVAL;
269 
270 	/* Insert range should only be used exclusively. */
271 	if ((mode & FALLOC_FL_INSERT_RANGE) &&
272 	    (mode & ~FALLOC_FL_INSERT_RANGE))
273 		return -EINVAL;
274 
275 	/* Unshare range should only be used with allocate mode. */
276 	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
277 	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
278 		return -EINVAL;
279 
280 	if (!(file->f_mode & FMODE_WRITE))
281 		return -EBADF;
282 
283 	/*
284 	 * We can only allow pure fallocate on append only files
285 	 */
286 	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
287 		return -EPERM;
288 
289 	if (IS_IMMUTABLE(inode))
290 		return -EPERM;
291 
292 	/*
293 	 * We cannot allow any fallocate operation on an active swapfile
294 	 */
295 	if (IS_SWAPFILE(inode))
296 		return -ETXTBSY;
297 
298 	/*
299 	 * Revalidate the write permissions, in case security policy has
300 	 * changed since the files were opened.
301 	 */
302 	ret = security_file_permission(file, MAY_WRITE);
303 	if (ret)
304 		return ret;
305 
306 	if (S_ISFIFO(inode->i_mode))
307 		return -ESPIPE;
308 
309 	if (S_ISDIR(inode->i_mode))
310 		return -EISDIR;
311 
312 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
313 		return -ENODEV;
314 
315 	/* Check for wrap through zero too */
316 	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
317 		return -EFBIG;
318 
319 	if (!file->f_op->fallocate)
320 		return -EOPNOTSUPP;
321 
322 	file_start_write(file);
323 	ret = file->f_op->fallocate(file, mode, offset, len);
324 
325 	/*
326 	 * Create inotify and fanotify events.
327 	 *
328 	 * To keep the logic simple always create events if fallocate succeeds.
329 	 * This implies that events are even created if the file size remains
330 	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
331 	 */
332 	if (ret == 0)
333 		fsnotify_modify(file);
334 
335 	file_end_write(file);
336 	return ret;
337 }
338 EXPORT_SYMBOL_GPL(vfs_fallocate);
339 
ksys_fallocate(int fd,int mode,loff_t offset,loff_t len)340 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
341 {
342 	struct fd f = fdget(fd);
343 	int error = -EBADF;
344 
345 	if (f.file) {
346 		error = vfs_fallocate(f.file, mode, offset, len);
347 		fdput(f);
348 	}
349 	return error;
350 }
351 
SYSCALL_DEFINE4(fallocate,int,fd,int,mode,loff_t,offset,loff_t,len)352 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
353 {
354 	return ksys_fallocate(fd, mode, offset, len);
355 }
356 
357 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
COMPAT_SYSCALL_DEFINE6(fallocate,int,fd,int,mode,compat_arg_u64_dual (offset),compat_arg_u64_dual (len))358 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
359 		       compat_arg_u64_dual(len))
360 {
361 	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
362 			      compat_arg_u64_glue(len));
363 }
364 #endif
365 
366 /*
367  * access() needs to use the real uid/gid, not the effective uid/gid.
368  * We do this by temporarily clearing all FS-related capabilities and
369  * switching the fsuid/fsgid around to the real ones.
370  */
access_override_creds(void)371 static const struct cred *access_override_creds(void)
372 {
373 	const struct cred *old_cred;
374 	struct cred *override_cred;
375 
376 	override_cred = prepare_creds();
377 	if (!override_cred)
378 		return NULL;
379 
380 	override_cred->fsuid = override_cred->uid;
381 	override_cred->fsgid = override_cred->gid;
382 
383 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
384 		/* Clear the capabilities if we switch to a non-root user */
385 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
386 		if (!uid_eq(override_cred->uid, root_uid))
387 			cap_clear(override_cred->cap_effective);
388 		else
389 			override_cred->cap_effective =
390 				override_cred->cap_permitted;
391 	}
392 
393 	/*
394 	 * The new set of credentials can *only* be used in
395 	 * task-synchronous circumstances, and does not need
396 	 * RCU freeing, unless somebody then takes a separate
397 	 * reference to it.
398 	 *
399 	 * NOTE! This is _only_ true because this credential
400 	 * is used purely for override_creds() that installs
401 	 * it as the subjective cred. Other threads will be
402 	 * accessing ->real_cred, not the subjective cred.
403 	 *
404 	 * If somebody _does_ make a copy of this (using the
405 	 * 'get_current_cred()' function), that will clear the
406 	 * non_rcu field, because now that other user may be
407 	 * expecting RCU freeing. But normal thread-synchronous
408 	 * cred accesses will keep things non-RCY.
409 	 */
410 	override_cred->non_rcu = 1;
411 
412 	old_cred = override_creds(override_cred);
413 
414 	/* override_cred() gets its own ref */
415 	put_cred(override_cred);
416 
417 	return old_cred;
418 }
419 
do_faccessat(int dfd,const char __user * filename,int mode,int flags)420 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
421 {
422 	struct path path;
423 	struct inode *inode;
424 	int res;
425 	unsigned int lookup_flags = LOOKUP_FOLLOW;
426 	const struct cred *old_cred = NULL;
427 
428 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
429 		return -EINVAL;
430 
431 	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
432 		return -EINVAL;
433 
434 	if (flags & AT_SYMLINK_NOFOLLOW)
435 		lookup_flags &= ~LOOKUP_FOLLOW;
436 	if (flags & AT_EMPTY_PATH)
437 		lookup_flags |= LOOKUP_EMPTY;
438 
439 	if (!(flags & AT_EACCESS)) {
440 		old_cred = access_override_creds();
441 		if (!old_cred)
442 			return -ENOMEM;
443 	}
444 
445 retry:
446 	res = user_path_at(dfd, filename, lookup_flags, &path);
447 	if (res)
448 		goto out;
449 
450 	inode = d_backing_inode(path.dentry);
451 
452 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
453 		/*
454 		 * MAY_EXEC on regular files is denied if the fs is mounted
455 		 * with the "noexec" flag.
456 		 */
457 		res = -EACCES;
458 		if (path_noexec(&path))
459 			goto out_path_release;
460 	}
461 
462 	res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
463 	/* SuS v2 requires we report a read only fs too */
464 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
465 		goto out_path_release;
466 	/*
467 	 * This is a rare case where using __mnt_is_readonly()
468 	 * is OK without a mnt_want/drop_write() pair.  Since
469 	 * no actual write to the fs is performed here, we do
470 	 * not need to telegraph to that to anyone.
471 	 *
472 	 * By doing this, we accept that this access is
473 	 * inherently racy and know that the fs may change
474 	 * state before we even see this result.
475 	 */
476 	if (__mnt_is_readonly(path.mnt))
477 		res = -EROFS;
478 
479 out_path_release:
480 	path_put(&path);
481 	if (retry_estale(res, lookup_flags)) {
482 		lookup_flags |= LOOKUP_REVAL;
483 		goto retry;
484 	}
485 out:
486 	if (old_cred)
487 		revert_creds(old_cred);
488 
489 	return res;
490 }
491 
SYSCALL_DEFINE3(faccessat,int,dfd,const char __user *,filename,int,mode)492 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
493 {
494 	return do_faccessat(dfd, filename, mode, 0);
495 }
496 
SYSCALL_DEFINE4(faccessat2,int,dfd,const char __user *,filename,int,mode,int,flags)497 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
498 		int, flags)
499 {
500 	return do_faccessat(dfd, filename, mode, flags);
501 }
502 
SYSCALL_DEFINE2(access,const char __user *,filename,int,mode)503 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
504 {
505 	return do_faccessat(AT_FDCWD, filename, mode, 0);
506 }
507 
SYSCALL_DEFINE1(chdir,const char __user *,filename)508 SYSCALL_DEFINE1(chdir, const char __user *, filename)
509 {
510 	struct path path;
511 	int error;
512 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
513 retry:
514 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
515 	if (error)
516 		goto out;
517 
518 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
519 	if (error)
520 		goto dput_and_out;
521 
522 	set_fs_pwd(current->fs, &path);
523 
524 dput_and_out:
525 	path_put(&path);
526 	if (retry_estale(error, lookup_flags)) {
527 		lookup_flags |= LOOKUP_REVAL;
528 		goto retry;
529 	}
530 out:
531 	return error;
532 }
533 
SYSCALL_DEFINE1(fchdir,unsigned int,fd)534 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
535 {
536 	struct fd f = fdget_raw(fd);
537 	int error;
538 
539 	error = -EBADF;
540 	if (!f.file)
541 		goto out;
542 
543 	error = -ENOTDIR;
544 	if (!d_can_lookup(f.file->f_path.dentry))
545 		goto out_putf;
546 
547 	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
548 	if (!error)
549 		set_fs_pwd(current->fs, &f.file->f_path);
550 out_putf:
551 	fdput(f);
552 out:
553 	return error;
554 }
555 
SYSCALL_DEFINE1(chroot,const char __user *,filename)556 SYSCALL_DEFINE1(chroot, const char __user *, filename)
557 {
558 	struct path path;
559 	int error;
560 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
561 retry:
562 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
563 	if (error)
564 		goto out;
565 
566 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
567 	if (error)
568 		goto dput_and_out;
569 
570 	error = -EPERM;
571 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
572 		goto dput_and_out;
573 	error = security_path_chroot(&path);
574 	if (error)
575 		goto dput_and_out;
576 
577 	set_fs_root(current->fs, &path);
578 	error = 0;
579 dput_and_out:
580 	path_put(&path);
581 	if (retry_estale(error, lookup_flags)) {
582 		lookup_flags |= LOOKUP_REVAL;
583 		goto retry;
584 	}
585 out:
586 	return error;
587 }
588 
chmod_common(const struct path * path,umode_t mode)589 int chmod_common(const struct path *path, umode_t mode)
590 {
591 	struct inode *inode = path->dentry->d_inode;
592 	struct inode *delegated_inode = NULL;
593 	struct iattr newattrs;
594 	int error;
595 
596 	error = mnt_want_write(path->mnt);
597 	if (error)
598 		return error;
599 retry_deleg:
600 	inode_lock(inode);
601 	error = security_path_chmod(path, mode);
602 	if (error)
603 		goto out_unlock;
604 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
605 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
606 	error = notify_change(mnt_user_ns(path->mnt), path->dentry,
607 			      &newattrs, &delegated_inode);
608 out_unlock:
609 	inode_unlock(inode);
610 	if (delegated_inode) {
611 		error = break_deleg_wait(&delegated_inode);
612 		if (!error)
613 			goto retry_deleg;
614 	}
615 	mnt_drop_write(path->mnt);
616 	return error;
617 }
618 
vfs_fchmod(struct file * file,umode_t mode)619 int vfs_fchmod(struct file *file, umode_t mode)
620 {
621 	audit_file(file);
622 	return chmod_common(&file->f_path, mode);
623 }
624 
SYSCALL_DEFINE2(fchmod,unsigned int,fd,umode_t,mode)625 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
626 {
627 	struct fd f = fdget(fd);
628 	int err = -EBADF;
629 
630 	if (f.file) {
631 		err = vfs_fchmod(f.file, mode);
632 		fdput(f);
633 	}
634 	return err;
635 }
636 
do_fchmodat(int dfd,const char __user * filename,umode_t mode)637 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
638 {
639 	struct path path;
640 	int error;
641 	unsigned int lookup_flags = LOOKUP_FOLLOW;
642 retry:
643 	error = user_path_at(dfd, filename, lookup_flags, &path);
644 	if (!error) {
645 		error = chmod_common(&path, mode);
646 		path_put(&path);
647 		if (retry_estale(error, lookup_flags)) {
648 			lookup_flags |= LOOKUP_REVAL;
649 			goto retry;
650 		}
651 	}
652 	return error;
653 }
654 
SYSCALL_DEFINE3(fchmodat,int,dfd,const char __user *,filename,umode_t,mode)655 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
656 		umode_t, mode)
657 {
658 	return do_fchmodat(dfd, filename, mode);
659 }
660 
SYSCALL_DEFINE2(chmod,const char __user *,filename,umode_t,mode)661 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
662 {
663 	return do_fchmodat(AT_FDCWD, filename, mode);
664 }
665 
chown_common(const struct path * path,uid_t user,gid_t group)666 int chown_common(const struct path *path, uid_t user, gid_t group)
667 {
668 	struct user_namespace *mnt_userns, *fs_userns;
669 	struct inode *inode = path->dentry->d_inode;
670 	struct inode *delegated_inode = NULL;
671 	int error;
672 	struct iattr newattrs;
673 	kuid_t uid;
674 	kgid_t gid;
675 
676 	uid = make_kuid(current_user_ns(), user);
677 	gid = make_kgid(current_user_ns(), group);
678 
679 	mnt_userns = mnt_user_ns(path->mnt);
680 	fs_userns = i_user_ns(inode);
681 	uid = mapped_kuid_user(mnt_userns, fs_userns, uid);
682 	gid = mapped_kgid_user(mnt_userns, fs_userns, gid);
683 
684 retry_deleg:
685 	newattrs.ia_valid =  ATTR_CTIME;
686 	if (user != (uid_t) -1) {
687 		if (!uid_valid(uid))
688 			return -EINVAL;
689 		newattrs.ia_valid |= ATTR_UID;
690 		newattrs.ia_uid = uid;
691 	}
692 	if (group != (gid_t) -1) {
693 		if (!gid_valid(gid))
694 			return -EINVAL;
695 		newattrs.ia_valid |= ATTR_GID;
696 		newattrs.ia_gid = gid;
697 	}
698 	if (!S_ISDIR(inode->i_mode))
699 		newattrs.ia_valid |=
700 			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
701 	inode_lock(inode);
702 	error = security_path_chown(path, uid, gid);
703 	if (!error)
704 		error = notify_change(mnt_userns, path->dentry, &newattrs,
705 				      &delegated_inode);
706 	inode_unlock(inode);
707 	if (delegated_inode) {
708 		error = break_deleg_wait(&delegated_inode);
709 		if (!error)
710 			goto retry_deleg;
711 	}
712 	return error;
713 }
714 
do_fchownat(int dfd,const char __user * filename,uid_t user,gid_t group,int flag)715 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
716 		int flag)
717 {
718 	struct path path;
719 	int error = -EINVAL;
720 	int lookup_flags;
721 
722 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
723 		goto out;
724 
725 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
726 	if (flag & AT_EMPTY_PATH)
727 		lookup_flags |= LOOKUP_EMPTY;
728 retry:
729 	error = user_path_at(dfd, filename, lookup_flags, &path);
730 	if (error)
731 		goto out;
732 	error = mnt_want_write(path.mnt);
733 	if (error)
734 		goto out_release;
735 	error = chown_common(&path, user, group);
736 	mnt_drop_write(path.mnt);
737 out_release:
738 	path_put(&path);
739 	if (retry_estale(error, lookup_flags)) {
740 		lookup_flags |= LOOKUP_REVAL;
741 		goto retry;
742 	}
743 out:
744 	return error;
745 }
746 
SYSCALL_DEFINE5(fchownat,int,dfd,const char __user *,filename,uid_t,user,gid_t,group,int,flag)747 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
748 		gid_t, group, int, flag)
749 {
750 	return do_fchownat(dfd, filename, user, group, flag);
751 }
752 
SYSCALL_DEFINE3(chown,const char __user *,filename,uid_t,user,gid_t,group)753 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
754 {
755 	return do_fchownat(AT_FDCWD, filename, user, group, 0);
756 }
757 
SYSCALL_DEFINE3(lchown,const char __user *,filename,uid_t,user,gid_t,group)758 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
759 {
760 	return do_fchownat(AT_FDCWD, filename, user, group,
761 			   AT_SYMLINK_NOFOLLOW);
762 }
763 
vfs_fchown(struct file * file,uid_t user,gid_t group)764 int vfs_fchown(struct file *file, uid_t user, gid_t group)
765 {
766 	int error;
767 
768 	error = mnt_want_write_file(file);
769 	if (error)
770 		return error;
771 	audit_file(file);
772 	error = chown_common(&file->f_path, user, group);
773 	mnt_drop_write_file(file);
774 	return error;
775 }
776 
ksys_fchown(unsigned int fd,uid_t user,gid_t group)777 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
778 {
779 	struct fd f = fdget(fd);
780 	int error = -EBADF;
781 
782 	if (f.file) {
783 		error = vfs_fchown(f.file, user, group);
784 		fdput(f);
785 	}
786 	return error;
787 }
788 
SYSCALL_DEFINE3(fchown,unsigned int,fd,uid_t,user,gid_t,group)789 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
790 {
791 	return ksys_fchown(fd, user, group);
792 }
793 
do_dentry_open(struct file * f,struct inode * inode,int (* open)(struct inode *,struct file *))794 static int do_dentry_open(struct file *f,
795 			  struct inode *inode,
796 			  int (*open)(struct inode *, struct file *))
797 {
798 	static const struct file_operations empty_fops = {};
799 	int error;
800 
801 	path_get(&f->f_path);
802 	f->f_inode = inode;
803 	f->f_mapping = inode->i_mapping;
804 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
805 	f->f_sb_err = file_sample_sb_err(f);
806 
807 	if (unlikely(f->f_flags & O_PATH)) {
808 		f->f_mode = FMODE_PATH | FMODE_OPENED;
809 		f->f_op = &empty_fops;
810 		return 0;
811 	}
812 
813 	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
814 		error = get_write_access(inode);
815 		if (unlikely(error))
816 			goto cleanup_file;
817 		error = __mnt_want_write(f->f_path.mnt);
818 		if (unlikely(error)) {
819 			put_write_access(inode);
820 			goto cleanup_file;
821 		}
822 		f->f_mode |= FMODE_WRITER;
823 	}
824 
825 	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
826 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
827 		f->f_mode |= FMODE_ATOMIC_POS;
828 
829 	f->f_op = fops_get(inode->i_fop);
830 	if (WARN_ON(!f->f_op)) {
831 		error = -ENODEV;
832 		goto cleanup_all;
833 	}
834 
835 	error = security_file_open(f);
836 	if (error)
837 		goto cleanup_all;
838 
839 	error = break_lease(locks_inode(f), f->f_flags);
840 	if (error)
841 		goto cleanup_all;
842 
843 	/* normally all 3 are set; ->open() can clear them if needed */
844 	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
845 	if (!open)
846 		open = f->f_op->open;
847 	if (open) {
848 		error = open(inode, f);
849 		if (error)
850 			goto cleanup_all;
851 	}
852 	f->f_mode |= FMODE_OPENED;
853 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
854 		i_readcount_inc(inode);
855 	if ((f->f_mode & FMODE_READ) &&
856 	     likely(f->f_op->read || f->f_op->read_iter))
857 		f->f_mode |= FMODE_CAN_READ;
858 	if ((f->f_mode & FMODE_WRITE) &&
859 	     likely(f->f_op->write || f->f_op->write_iter))
860 		f->f_mode |= FMODE_CAN_WRITE;
861 	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
862 		f->f_mode |= FMODE_CAN_ODIRECT;
863 
864 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
865 
866 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
867 
868 	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
869 		return -EINVAL;
870 
871 	/*
872 	 * XXX: Huge page cache doesn't support writing yet. Drop all page
873 	 * cache for this file before processing writes.
874 	 */
875 	if (f->f_mode & FMODE_WRITE) {
876 		/*
877 		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
878 		 * is up to date and the update to i_writecount by
879 		 * get_write_access() is visible. Ensures subsequent insertion
880 		 * of THPs into the page cache will fail.
881 		 */
882 		smp_mb();
883 		if (filemap_nr_thps(inode->i_mapping)) {
884 			struct address_space *mapping = inode->i_mapping;
885 
886 			filemap_invalidate_lock(inode->i_mapping);
887 			/*
888 			 * unmap_mapping_range just need to be called once
889 			 * here, because the private pages is not need to be
890 			 * unmapped mapping (e.g. data segment of dynamic
891 			 * shared libraries here).
892 			 */
893 			unmap_mapping_range(mapping, 0, 0, 0);
894 			truncate_inode_pages(mapping, 0);
895 			filemap_invalidate_unlock(inode->i_mapping);
896 		}
897 	}
898 
899 	return 0;
900 
901 cleanup_all:
902 	if (WARN_ON_ONCE(error > 0))
903 		error = -EINVAL;
904 	fops_put(f->f_op);
905 	if (f->f_mode & FMODE_WRITER) {
906 		put_write_access(inode);
907 		__mnt_drop_write(f->f_path.mnt);
908 	}
909 cleanup_file:
910 	path_put(&f->f_path);
911 	f->f_path.mnt = NULL;
912 	f->f_path.dentry = NULL;
913 	f->f_inode = NULL;
914 	return error;
915 }
916 
917 /**
918  * finish_open - finish opening a file
919  * @file: file pointer
920  * @dentry: pointer to dentry
921  * @open: open callback
922  * @opened: state of open
923  *
924  * This can be used to finish opening a file passed to i_op->atomic_open().
925  *
926  * If the open callback is set to NULL, then the standard f_op->open()
927  * filesystem callback is substituted.
928  *
929  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
930  * the return value of d_splice_alias(), then the caller needs to perform dput()
931  * on it after finish_open().
932  *
933  * Returns zero on success or -errno if the open failed.
934  */
finish_open(struct file * file,struct dentry * dentry,int (* open)(struct inode *,struct file *))935 int finish_open(struct file *file, struct dentry *dentry,
936 		int (*open)(struct inode *, struct file *))
937 {
938 	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
939 
940 	file->f_path.dentry = dentry;
941 	return do_dentry_open(file, d_backing_inode(dentry), open);
942 }
943 EXPORT_SYMBOL(finish_open);
944 
945 /**
946  * finish_no_open - finish ->atomic_open() without opening the file
947  *
948  * @file: file pointer
949  * @dentry: dentry or NULL (as returned from ->lookup())
950  *
951  * This can be used to set the result of a successful lookup in ->atomic_open().
952  *
953  * NB: unlike finish_open() this function does consume the dentry reference and
954  * the caller need not dput() it.
955  *
956  * Returns "0" which must be the return value of ->atomic_open() after having
957  * called this function.
958  */
finish_no_open(struct file * file,struct dentry * dentry)959 int finish_no_open(struct file *file, struct dentry *dentry)
960 {
961 	file->f_path.dentry = dentry;
962 	return 0;
963 }
964 EXPORT_SYMBOL(finish_no_open);
965 
file_path(struct file * filp,char * buf,int buflen)966 char *file_path(struct file *filp, char *buf, int buflen)
967 {
968 	return d_path(&filp->f_path, buf, buflen);
969 }
970 EXPORT_SYMBOL(file_path);
971 
972 /**
973  * vfs_open - open the file at the given path
974  * @path: path to open
975  * @file: newly allocated file with f_flag initialized
976  * @cred: credentials to use
977  */
vfs_open(const struct path * path,struct file * file)978 int vfs_open(const struct path *path, struct file *file)
979 {
980 	file->f_path = *path;
981 	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
982 }
983 
dentry_open(const struct path * path,int flags,const struct cred * cred)984 struct file *dentry_open(const struct path *path, int flags,
985 			 const struct cred *cred)
986 {
987 	int error;
988 	struct file *f;
989 
990 	validate_creds(cred);
991 
992 	/* We must always pass in a valid mount pointer. */
993 	BUG_ON(!path->mnt);
994 
995 	f = alloc_empty_file(flags, cred);
996 	if (!IS_ERR(f)) {
997 		error = vfs_open(path, f);
998 		if (error) {
999 			fput(f);
1000 			f = ERR_PTR(error);
1001 		}
1002 	}
1003 	return f;
1004 }
1005 EXPORT_SYMBOL(dentry_open);
1006 
1007 /**
1008  * dentry_create - Create and open a file
1009  * @path: path to create
1010  * @flags: O_ flags
1011  * @mode: mode bits for new file
1012  * @cred: credentials to use
1013  *
1014  * Caller must hold the parent directory's lock, and have prepared
1015  * a negative dentry, placed in @path->dentry, for the new file.
1016  *
1017  * Caller sets @path->mnt to the vfsmount of the filesystem where
1018  * the new file is to be created. The parent directory and the
1019  * negative dentry must reside on the same filesystem instance.
1020  *
1021  * On success, returns a "struct file *". Otherwise a ERR_PTR
1022  * is returned.
1023  */
dentry_create(const struct path * path,int flags,umode_t mode,const struct cred * cred)1024 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1025 			   const struct cred *cred)
1026 {
1027 	struct file *f;
1028 	int error;
1029 
1030 	validate_creds(cred);
1031 	f = alloc_empty_file(flags, cred);
1032 	if (IS_ERR(f))
1033 		return f;
1034 
1035 	error = vfs_create(mnt_user_ns(path->mnt),
1036 			   d_inode(path->dentry->d_parent),
1037 			   path->dentry, mode, true);
1038 	if (!error)
1039 		error = vfs_open(path, f);
1040 
1041 	if (unlikely(error)) {
1042 		fput(f);
1043 		return ERR_PTR(error);
1044 	}
1045 	return f;
1046 }
1047 EXPORT_SYMBOL(dentry_create);
1048 
open_with_fake_path(const struct path * path,int flags,struct inode * inode,const struct cred * cred)1049 struct file *open_with_fake_path(const struct path *path, int flags,
1050 				struct inode *inode, const struct cred *cred)
1051 {
1052 	struct file *f = alloc_empty_file_noaccount(flags, cred);
1053 	if (!IS_ERR(f)) {
1054 		int error;
1055 
1056 		f->f_path = *path;
1057 		error = do_dentry_open(f, inode, NULL);
1058 		if (error) {
1059 			fput(f);
1060 			f = ERR_PTR(error);
1061 		}
1062 	}
1063 	return f;
1064 }
1065 EXPORT_SYMBOL(open_with_fake_path);
1066 
1067 #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1068 #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1069 
build_open_how(int flags,umode_t mode)1070 inline struct open_how build_open_how(int flags, umode_t mode)
1071 {
1072 	struct open_how how = {
1073 		.flags = flags & VALID_OPEN_FLAGS,
1074 		.mode = mode & S_IALLUGO,
1075 	};
1076 
1077 	/* O_PATH beats everything else. */
1078 	if (how.flags & O_PATH)
1079 		how.flags &= O_PATH_FLAGS;
1080 	/* Modes should only be set for create-like flags. */
1081 	if (!WILL_CREATE(how.flags))
1082 		how.mode = 0;
1083 	return how;
1084 }
1085 
build_open_flags(const struct open_how * how,struct open_flags * op)1086 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1087 {
1088 	u64 flags = how->flags;
1089 	u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1090 	int lookup_flags = 0;
1091 	int acc_mode = ACC_MODE(flags);
1092 
1093 	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1094 			 "struct open_flags doesn't yet handle flags > 32 bits");
1095 
1096 	/*
1097 	 * Strip flags that either shouldn't be set by userspace like
1098 	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1099 	 * open_flags like O_CLOEXEC.
1100 	 */
1101 	flags &= ~strip;
1102 
1103 	/*
1104 	 * Older syscalls implicitly clear all of the invalid flags or argument
1105 	 * values before calling build_open_flags(), but openat2(2) checks all
1106 	 * of its arguments.
1107 	 */
1108 	if (flags & ~VALID_OPEN_FLAGS)
1109 		return -EINVAL;
1110 	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1111 		return -EINVAL;
1112 
1113 	/* Scoping flags are mutually exclusive. */
1114 	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1115 		return -EINVAL;
1116 
1117 	/* Deal with the mode. */
1118 	if (WILL_CREATE(flags)) {
1119 		if (how->mode & ~S_IALLUGO)
1120 			return -EINVAL;
1121 		op->mode = how->mode | S_IFREG;
1122 	} else {
1123 		if (how->mode != 0)
1124 			return -EINVAL;
1125 		op->mode = 0;
1126 	}
1127 
1128 	/*
1129 	 * In order to ensure programs get explicit errors when trying to use
1130 	 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1131 	 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1132 	 * have to require userspace to explicitly set it.
1133 	 */
1134 	if (flags & __O_TMPFILE) {
1135 		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1136 			return -EINVAL;
1137 		if (!(acc_mode & MAY_WRITE))
1138 			return -EINVAL;
1139 	}
1140 	if (flags & O_PATH) {
1141 		/* O_PATH only permits certain other flags to be set. */
1142 		if (flags & ~O_PATH_FLAGS)
1143 			return -EINVAL;
1144 		acc_mode = 0;
1145 	}
1146 
1147 	/*
1148 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1149 	 * check for O_DSYNC if the need any syncing at all we enforce it's
1150 	 * always set instead of having to deal with possibly weird behaviour
1151 	 * for malicious applications setting only __O_SYNC.
1152 	 */
1153 	if (flags & __O_SYNC)
1154 		flags |= O_DSYNC;
1155 
1156 	op->open_flag = flags;
1157 
1158 	/* O_TRUNC implies we need access checks for write permissions */
1159 	if (flags & O_TRUNC)
1160 		acc_mode |= MAY_WRITE;
1161 
1162 	/* Allow the LSM permission hook to distinguish append
1163 	   access from general write access. */
1164 	if (flags & O_APPEND)
1165 		acc_mode |= MAY_APPEND;
1166 
1167 	op->acc_mode = acc_mode;
1168 
1169 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1170 
1171 	if (flags & O_CREAT) {
1172 		op->intent |= LOOKUP_CREATE;
1173 		if (flags & O_EXCL) {
1174 			op->intent |= LOOKUP_EXCL;
1175 			flags |= O_NOFOLLOW;
1176 		}
1177 	}
1178 
1179 	if (flags & O_DIRECTORY)
1180 		lookup_flags |= LOOKUP_DIRECTORY;
1181 	if (!(flags & O_NOFOLLOW))
1182 		lookup_flags |= LOOKUP_FOLLOW;
1183 
1184 	if (how->resolve & RESOLVE_NO_XDEV)
1185 		lookup_flags |= LOOKUP_NO_XDEV;
1186 	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1187 		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1188 	if (how->resolve & RESOLVE_NO_SYMLINKS)
1189 		lookup_flags |= LOOKUP_NO_SYMLINKS;
1190 	if (how->resolve & RESOLVE_BENEATH)
1191 		lookup_flags |= LOOKUP_BENEATH;
1192 	if (how->resolve & RESOLVE_IN_ROOT)
1193 		lookup_flags |= LOOKUP_IN_ROOT;
1194 	if (how->resolve & RESOLVE_CACHED) {
1195 		/* Don't bother even trying for create/truncate/tmpfile open */
1196 		if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1197 			return -EAGAIN;
1198 		lookup_flags |= LOOKUP_CACHED;
1199 	}
1200 
1201 	op->lookup_flags = lookup_flags;
1202 	return 0;
1203 }
1204 
1205 /**
1206  * file_open_name - open file and return file pointer
1207  *
1208  * @name:	struct filename containing path to open
1209  * @flags:	open flags as per the open(2) second argument
1210  * @mode:	mode for the new file if O_CREAT is set, else ignored
1211  *
1212  * This is the helper to open a file from kernelspace if you really
1213  * have to.  But in generally you should not do this, so please move
1214  * along, nothing to see here..
1215  */
file_open_name(struct filename * name,int flags,umode_t mode)1216 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1217 {
1218 	struct open_flags op;
1219 	struct open_how how = build_open_how(flags, mode);
1220 	int err = build_open_flags(&how, &op);
1221 	if (err)
1222 		return ERR_PTR(err);
1223 	return do_filp_open(AT_FDCWD, name, &op);
1224 }
1225 
1226 /**
1227  * filp_open - open file and return file pointer
1228  *
1229  * @filename:	path to open
1230  * @flags:	open flags as per the open(2) second argument
1231  * @mode:	mode for the new file if O_CREAT is set, else ignored
1232  *
1233  * This is the helper to open a file from kernelspace if you really
1234  * have to.  But in generally you should not do this, so please move
1235  * along, nothing to see here..
1236  */
filp_open(const char * filename,int flags,umode_t mode)1237 struct file *filp_open(const char *filename, int flags, umode_t mode)
1238 {
1239 	struct filename *name = getname_kernel(filename);
1240 	struct file *file = ERR_CAST(name);
1241 
1242 	if (!IS_ERR(name)) {
1243 		file = file_open_name(name, flags, mode);
1244 		putname(name);
1245 	}
1246 	return file;
1247 }
1248 EXPORT_SYMBOL(filp_open);
1249 
file_open_root(const struct path * root,const char * filename,int flags,umode_t mode)1250 struct file *file_open_root(const struct path *root,
1251 			    const char *filename, int flags, umode_t mode)
1252 {
1253 	struct open_flags op;
1254 	struct open_how how = build_open_how(flags, mode);
1255 	int err = build_open_flags(&how, &op);
1256 	if (err)
1257 		return ERR_PTR(err);
1258 	return do_file_open_root(root, filename, &op);
1259 }
1260 EXPORT_SYMBOL(file_open_root);
1261 
do_sys_openat2(int dfd,const char __user * filename,struct open_how * how)1262 static long do_sys_openat2(int dfd, const char __user *filename,
1263 			   struct open_how *how)
1264 {
1265 	struct open_flags op;
1266 	int fd = build_open_flags(how, &op);
1267 	struct filename *tmp;
1268 
1269 	if (fd)
1270 		return fd;
1271 
1272 	tmp = getname(filename);
1273 	if (IS_ERR(tmp))
1274 		return PTR_ERR(tmp);
1275 
1276 	fd = get_unused_fd_flags(how->flags);
1277 	if (fd >= 0) {
1278 		struct file *f = do_filp_open(dfd, tmp, &op);
1279 		if (IS_ERR(f)) {
1280 			put_unused_fd(fd);
1281 			fd = PTR_ERR(f);
1282 		} else {
1283 			fsnotify_open(f);
1284 			fd_install(fd, f);
1285 		}
1286 	}
1287 	putname(tmp);
1288 	return fd;
1289 }
1290 
do_sys_open(int dfd,const char __user * filename,int flags,umode_t mode)1291 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1292 {
1293 	struct open_how how = build_open_how(flags, mode);
1294 	return do_sys_openat2(dfd, filename, &how);
1295 }
1296 
1297 
SYSCALL_DEFINE3(open,const char __user *,filename,int,flags,umode_t,mode)1298 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1299 {
1300 	if (force_o_largefile())
1301 		flags |= O_LARGEFILE;
1302 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1303 }
1304 
SYSCALL_DEFINE4(openat,int,dfd,const char __user *,filename,int,flags,umode_t,mode)1305 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1306 		umode_t, mode)
1307 {
1308 	if (force_o_largefile())
1309 		flags |= O_LARGEFILE;
1310 	return do_sys_open(dfd, filename, flags, mode);
1311 }
1312 
SYSCALL_DEFINE4(openat2,int,dfd,const char __user *,filename,struct open_how __user *,how,size_t,usize)1313 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1314 		struct open_how __user *, how, size_t, usize)
1315 {
1316 	int err;
1317 	struct open_how tmp;
1318 
1319 	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1320 	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1321 
1322 	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1323 		return -EINVAL;
1324 
1325 	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1326 	if (err)
1327 		return err;
1328 
1329 	audit_openat2_how(&tmp);
1330 
1331 	/* O_LARGEFILE is only allowed for non-O_PATH. */
1332 	if (!(tmp.flags & O_PATH) && force_o_largefile())
1333 		tmp.flags |= O_LARGEFILE;
1334 
1335 	return do_sys_openat2(dfd, filename, &tmp);
1336 }
1337 
1338 #ifdef CONFIG_COMPAT
1339 /*
1340  * Exactly like sys_open(), except that it doesn't set the
1341  * O_LARGEFILE flag.
1342  */
COMPAT_SYSCALL_DEFINE3(open,const char __user *,filename,int,flags,umode_t,mode)1343 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1344 {
1345 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1346 }
1347 
1348 /*
1349  * Exactly like sys_openat(), except that it doesn't set the
1350  * O_LARGEFILE flag.
1351  */
COMPAT_SYSCALL_DEFINE4(openat,int,dfd,const char __user *,filename,int,flags,umode_t,mode)1352 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1353 {
1354 	return do_sys_open(dfd, filename, flags, mode);
1355 }
1356 #endif
1357 
1358 #ifndef __alpha__
1359 
1360 /*
1361  * For backward compatibility?  Maybe this should be moved
1362  * into arch/i386 instead?
1363  */
SYSCALL_DEFINE2(creat,const char __user *,pathname,umode_t,mode)1364 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1365 {
1366 	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1367 
1368 	if (force_o_largefile())
1369 		flags |= O_LARGEFILE;
1370 	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1371 }
1372 #endif
1373 
1374 /*
1375  * "id" is the POSIX thread ID. We use the
1376  * files pointer for this..
1377  */
filp_close(struct file * filp,fl_owner_t id)1378 int filp_close(struct file *filp, fl_owner_t id)
1379 {
1380 	int retval = 0;
1381 
1382 	if (!file_count(filp)) {
1383 		printk(KERN_ERR "VFS: Close: file count is 0\n");
1384 		return 0;
1385 	}
1386 
1387 	if (filp->f_op->flush)
1388 		retval = filp->f_op->flush(filp, id);
1389 
1390 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1391 		dnotify_flush(filp, id);
1392 		locks_remove_posix(filp, id);
1393 	}
1394 	fput(filp);
1395 	return retval;
1396 }
1397 
1398 EXPORT_SYMBOL(filp_close);
1399 
1400 /*
1401  * Careful here! We test whether the file pointer is NULL before
1402  * releasing the fd. This ensures that one clone task can't release
1403  * an fd while another clone is opening it.
1404  */
SYSCALL_DEFINE1(close,unsigned int,fd)1405 SYSCALL_DEFINE1(close, unsigned int, fd)
1406 {
1407 	int retval = close_fd(fd);
1408 
1409 	/* can't restart close syscall because file table entry was cleared */
1410 	if (unlikely(retval == -ERESTARTSYS ||
1411 		     retval == -ERESTARTNOINTR ||
1412 		     retval == -ERESTARTNOHAND ||
1413 		     retval == -ERESTART_RESTARTBLOCK))
1414 		retval = -EINTR;
1415 
1416 	return retval;
1417 }
1418 
1419 /**
1420  * close_range() - Close all file descriptors in a given range.
1421  *
1422  * @fd:     starting file descriptor to close
1423  * @max_fd: last file descriptor to close
1424  * @flags:  reserved for future extensions
1425  *
1426  * This closes a range of file descriptors. All file descriptors
1427  * from @fd up to and including @max_fd are closed.
1428  * Currently, errors to close a given file descriptor are ignored.
1429  */
SYSCALL_DEFINE3(close_range,unsigned int,fd,unsigned int,max_fd,unsigned int,flags)1430 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1431 		unsigned int, flags)
1432 {
1433 	return __close_range(fd, max_fd, flags);
1434 }
1435 
1436 /*
1437  * This routine simulates a hangup on the tty, to arrange that users
1438  * are given clean terminals at login time.
1439  */
SYSCALL_DEFINE0(vhangup)1440 SYSCALL_DEFINE0(vhangup)
1441 {
1442 	if (capable(CAP_SYS_TTY_CONFIG)) {
1443 		tty_vhangup_self();
1444 		return 0;
1445 	}
1446 	return -EPERM;
1447 }
1448 
1449 /*
1450  * Called when an inode is about to be open.
1451  * We use this to disallow opening large files on 32bit systems if
1452  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1453  * on this flag in sys_open.
1454  */
generic_file_open(struct inode * inode,struct file * filp)1455 int generic_file_open(struct inode * inode, struct file * filp)
1456 {
1457 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1458 		return -EOVERFLOW;
1459 	return 0;
1460 }
1461 
1462 EXPORT_SYMBOL(generic_file_open);
1463 
1464 /*
1465  * This is used by subsystems that don't want seekable
1466  * file descriptors. The function is not supposed to ever fail, the only
1467  * reason it returns an 'int' and not 'void' is so that it can be plugged
1468  * directly into file_operations structure.
1469  */
nonseekable_open(struct inode * inode,struct file * filp)1470 int nonseekable_open(struct inode *inode, struct file *filp)
1471 {
1472 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1473 	return 0;
1474 }
1475 
1476 EXPORT_SYMBOL(nonseekable_open);
1477 
1478 /*
1479  * stream_open is used by subsystems that want stream-like file descriptors.
1480  * Such file descriptors are not seekable and don't have notion of position
1481  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1482  * Contrary to file descriptors of other regular files, .read() and .write()
1483  * can run simultaneously.
1484  *
1485  * stream_open never fails and is marked to return int so that it could be
1486  * directly used as file_operations.open .
1487  */
stream_open(struct inode * inode,struct file * filp)1488 int stream_open(struct inode *inode, struct file *filp)
1489 {
1490 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1491 	filp->f_mode |= FMODE_STREAM;
1492 	return 0;
1493 }
1494 
1495 EXPORT_SYMBOL(stream_open);
1496