xref: /openbmc/linux/fs/open.c (revision 83635924)
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 #include <linux/filelock.h>
37 
38 #include "internal.h"
39 
do_truncate(struct mnt_idmap * idmap,struct dentry * dentry,loff_t length,unsigned int time_attrs,struct file * filp)40 int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
41 		loff_t length, unsigned int time_attrs, struct file *filp)
42 {
43 	int ret;
44 	struct iattr newattrs;
45 
46 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
47 	if (length < 0)
48 		return -EINVAL;
49 
50 	newattrs.ia_size = length;
51 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
52 	if (filp) {
53 		newattrs.ia_file = filp;
54 		newattrs.ia_valid |= ATTR_FILE;
55 	}
56 
57 	/* Remove suid, sgid, and file capabilities on truncate too */
58 	ret = dentry_needs_remove_privs(idmap, dentry);
59 	if (ret < 0)
60 		return ret;
61 	if (ret)
62 		newattrs.ia_valid |= ret | ATTR_FORCE;
63 
64 	inode_lock(dentry->d_inode);
65 	/* Note any delegations or leases have already been broken: */
66 	ret = notify_change(idmap, dentry, &newattrs, NULL);
67 	inode_unlock(dentry->d_inode);
68 	return ret;
69 }
70 
vfs_truncate(const struct path * path,loff_t length)71 long vfs_truncate(const struct path *path, loff_t length)
72 {
73 	struct mnt_idmap *idmap;
74 	struct inode *inode;
75 	long error;
76 
77 	inode = path->dentry->d_inode;
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 	idmap = mnt_idmap(path->mnt);
90 	error = inode_permission(idmap, inode, MAY_WRITE);
91 	if (error)
92 		goto mnt_drop_write_and_out;
93 
94 	error = -EPERM;
95 	if (IS_APPEND(inode))
96 		goto mnt_drop_write_and_out;
97 
98 	error = get_write_access(inode);
99 	if (error)
100 		goto mnt_drop_write_and_out;
101 
102 	/*
103 	 * Make sure that there are no leases.  get_write_access() protects
104 	 * against the truncate racing with a lease-granting setlease().
105 	 */
106 	error = break_lease(inode, O_WRONLY);
107 	if (error)
108 		goto put_write_and_out;
109 
110 	error = security_path_truncate(path);
111 	if (!error)
112 		error = do_truncate(idmap, path->dentry, length, 0, NULL);
113 
114 put_write_and_out:
115 	put_write_access(inode);
116 mnt_drop_write_and_out:
117 	mnt_drop_write(path->mnt);
118 out:
119 	return error;
120 }
121 EXPORT_SYMBOL_GPL(vfs_truncate);
122 
do_sys_truncate(const char __user * pathname,loff_t length)123 long do_sys_truncate(const char __user *pathname, loff_t length)
124 {
125 	unsigned int lookup_flags = LOOKUP_FOLLOW;
126 	struct path path;
127 	int error;
128 
129 	if (length < 0)	/* sorry, but loff_t says... */
130 		return -EINVAL;
131 
132 retry:
133 	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
134 	if (!error) {
135 		error = vfs_truncate(&path, length);
136 		path_put(&path);
137 	}
138 	if (retry_estale(error, lookup_flags)) {
139 		lookup_flags |= LOOKUP_REVAL;
140 		goto retry;
141 	}
142 	return error;
143 }
144 
SYSCALL_DEFINE2(truncate,const char __user *,path,long,length)145 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
146 {
147 	return do_sys_truncate(path, length);
148 }
149 
150 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(truncate,const char __user *,path,compat_off_t,length)151 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
152 {
153 	return do_sys_truncate(path, length);
154 }
155 #endif
156 
do_sys_ftruncate(unsigned int fd,loff_t length,int small)157 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
158 {
159 	struct inode *inode;
160 	struct dentry *dentry;
161 	struct fd f;
162 	int error;
163 
164 	error = -EINVAL;
165 	if (length < 0)
166 		goto out;
167 	error = -EBADF;
168 	f = fdget(fd);
169 	if (!f.file)
170 		goto out;
171 
172 	/* explicitly opened as large or we are on 64-bit box */
173 	if (f.file->f_flags & O_LARGEFILE)
174 		small = 0;
175 
176 	dentry = f.file->f_path.dentry;
177 	inode = dentry->d_inode;
178 	error = -EINVAL;
179 	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
180 		goto out_putf;
181 
182 	error = -EINVAL;
183 	/* Cannot ftruncate over 2^31 bytes without large file support */
184 	if (small && length > MAX_NON_LFS)
185 		goto out_putf;
186 
187 	error = -EPERM;
188 	/* Check IS_APPEND on real upper inode */
189 	if (IS_APPEND(file_inode(f.file)))
190 		goto out_putf;
191 	sb_start_write(inode->i_sb);
192 	error = security_file_truncate(f.file);
193 	if (!error)
194 		error = do_truncate(file_mnt_idmap(f.file), dentry, length,
195 				    ATTR_MTIME | ATTR_CTIME, f.file);
196 	sb_end_write(inode->i_sb);
197 out_putf:
198 	fdput(f);
199 out:
200 	return error;
201 }
202 
SYSCALL_DEFINE2(ftruncate,unsigned int,fd,off_t,length)203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
204 {
205 	return do_sys_ftruncate(fd, length, 1);
206 }
207 
208 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(ftruncate,unsigned int,fd,compat_off_t,length)209 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
210 {
211 	return do_sys_ftruncate(fd, length, 1);
212 }
213 #endif
214 
215 /* LFS versions of truncate are only needed on 32 bit machines */
216 #if BITS_PER_LONG == 32
SYSCALL_DEFINE2(truncate64,const char __user *,path,loff_t,length)217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218 {
219 	return do_sys_truncate(path, length);
220 }
221 
SYSCALL_DEFINE2(ftruncate64,unsigned int,fd,loff_t,length)222 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
223 {
224 	return do_sys_ftruncate(fd, length, 0);
225 }
226 #endif /* BITS_PER_LONG == 32 */
227 
228 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
COMPAT_SYSCALL_DEFINE3(truncate64,const char __user *,pathname,compat_arg_u64_dual (length))229 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
230 		       compat_arg_u64_dual(length))
231 {
232 	return ksys_truncate(pathname, compat_arg_u64_glue(length));
233 }
234 #endif
235 
236 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
COMPAT_SYSCALL_DEFINE3(ftruncate64,unsigned int,fd,compat_arg_u64_dual (length))237 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
238 		       compat_arg_u64_dual(length))
239 {
240 	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
241 }
242 #endif
243 
vfs_fallocate(struct file * file,int mode,loff_t offset,loff_t len)244 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
245 {
246 	struct inode *inode = file_inode(file);
247 	long ret;
248 
249 	if (offset < 0 || len <= 0)
250 		return -EINVAL;
251 
252 	/* Return error if mode is not supported */
253 	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
254 		return -EOPNOTSUPP;
255 
256 	/* Punch hole and zero range are mutually exclusive */
257 	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
258 	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
259 		return -EOPNOTSUPP;
260 
261 	/* Punch hole must have keep size set */
262 	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
263 	    !(mode & FALLOC_FL_KEEP_SIZE))
264 		return -EOPNOTSUPP;
265 
266 	/* Collapse range should only be used exclusively. */
267 	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
268 	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
269 		return -EINVAL;
270 
271 	/* Insert range should only be used exclusively. */
272 	if ((mode & FALLOC_FL_INSERT_RANGE) &&
273 	    (mode & ~FALLOC_FL_INSERT_RANGE))
274 		return -EINVAL;
275 
276 	/* Unshare range should only be used with allocate mode. */
277 	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
278 	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
279 		return -EINVAL;
280 
281 	if (!(file->f_mode & FMODE_WRITE))
282 		return -EBADF;
283 
284 	/*
285 	 * We can only allow pure fallocate on append only files
286 	 */
287 	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
288 		return -EPERM;
289 
290 	if (IS_IMMUTABLE(inode))
291 		return -EPERM;
292 
293 	/*
294 	 * We cannot allow any fallocate operation on an active swapfile
295 	 */
296 	if (IS_SWAPFILE(inode))
297 		return -ETXTBSY;
298 
299 	/*
300 	 * Revalidate the write permissions, in case security policy has
301 	 * changed since the files were opened.
302 	 */
303 	ret = security_file_permission(file, MAY_WRITE);
304 	if (ret)
305 		return ret;
306 
307 	if (S_ISFIFO(inode->i_mode))
308 		return -ESPIPE;
309 
310 	if (S_ISDIR(inode->i_mode))
311 		return -EISDIR;
312 
313 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
314 		return -ENODEV;
315 
316 	/* Check for wrap through zero too */
317 	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
318 		return -EFBIG;
319 
320 	if (!file->f_op->fallocate)
321 		return -EOPNOTSUPP;
322 
323 	file_start_write(file);
324 	ret = file->f_op->fallocate(file, mode, offset, len);
325 
326 	/*
327 	 * Create inotify and fanotify events.
328 	 *
329 	 * To keep the logic simple always create events if fallocate succeeds.
330 	 * This implies that events are even created if the file size remains
331 	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
332 	 */
333 	if (ret == 0)
334 		fsnotify_modify(file);
335 
336 	file_end_write(file);
337 	return ret;
338 }
339 EXPORT_SYMBOL_GPL(vfs_fallocate);
340 
ksys_fallocate(int fd,int mode,loff_t offset,loff_t len)341 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
342 {
343 	struct fd f = fdget(fd);
344 	int error = -EBADF;
345 
346 	if (f.file) {
347 		error = vfs_fallocate(f.file, mode, offset, len);
348 		fdput(f);
349 	}
350 	return error;
351 }
352 
SYSCALL_DEFINE4(fallocate,int,fd,int,mode,loff_t,offset,loff_t,len)353 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
354 {
355 	return ksys_fallocate(fd, mode, offset, len);
356 }
357 
358 #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))359 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
360 		       compat_arg_u64_dual(len))
361 {
362 	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
363 			      compat_arg_u64_glue(len));
364 }
365 #endif
366 
367 /*
368  * access() needs to use the real uid/gid, not the effective uid/gid.
369  * We do this by temporarily clearing all FS-related capabilities and
370  * switching the fsuid/fsgid around to the real ones.
371  *
372  * Creating new credentials is expensive, so we try to skip doing it,
373  * which we can if the result would match what we already got.
374  */
access_need_override_creds(int flags)375 static bool access_need_override_creds(int flags)
376 {
377 	const struct cred *cred;
378 
379 	if (flags & AT_EACCESS)
380 		return false;
381 
382 	cred = current_cred();
383 	if (!uid_eq(cred->fsuid, cred->uid) ||
384 	    !gid_eq(cred->fsgid, cred->gid))
385 		return true;
386 
387 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
388 		kuid_t root_uid = make_kuid(cred->user_ns, 0);
389 		if (!uid_eq(cred->uid, root_uid)) {
390 			if (!cap_isclear(cred->cap_effective))
391 				return true;
392 		} else {
393 			if (!cap_isidentical(cred->cap_effective,
394 			    cred->cap_permitted))
395 				return true;
396 		}
397 	}
398 
399 	return false;
400 }
401 
access_override_creds(void)402 static const struct cred *access_override_creds(void)
403 {
404 	const struct cred *old_cred;
405 	struct cred *override_cred;
406 
407 	override_cred = prepare_creds();
408 	if (!override_cred)
409 		return NULL;
410 
411 	/*
412 	 * XXX access_need_override_creds performs checks in hopes of skipping
413 	 * this work. Make sure it stays in sync if making any changes in this
414 	 * routine.
415 	 */
416 
417 	override_cred->fsuid = override_cred->uid;
418 	override_cred->fsgid = override_cred->gid;
419 
420 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
421 		/* Clear the capabilities if we switch to a non-root user */
422 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
423 		if (!uid_eq(override_cred->uid, root_uid))
424 			cap_clear(override_cred->cap_effective);
425 		else
426 			override_cred->cap_effective =
427 				override_cred->cap_permitted;
428 	}
429 
430 	/*
431 	 * The new set of credentials can *only* be used in
432 	 * task-synchronous circumstances, and does not need
433 	 * RCU freeing, unless somebody then takes a separate
434 	 * reference to it.
435 	 *
436 	 * NOTE! This is _only_ true because this credential
437 	 * is used purely for override_creds() that installs
438 	 * it as the subjective cred. Other threads will be
439 	 * accessing ->real_cred, not the subjective cred.
440 	 *
441 	 * If somebody _does_ make a copy of this (using the
442 	 * 'get_current_cred()' function), that will clear the
443 	 * non_rcu field, because now that other user may be
444 	 * expecting RCU freeing. But normal thread-synchronous
445 	 * cred accesses will keep things non-RCY.
446 	 */
447 	override_cred->non_rcu = 1;
448 
449 	old_cred = override_creds(override_cred);
450 
451 	/* override_cred() gets its own ref */
452 	put_cred(override_cred);
453 
454 	return old_cred;
455 }
456 
do_faccessat(int dfd,const char __user * filename,int mode,int flags)457 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
458 {
459 	struct path path;
460 	struct inode *inode;
461 	int res;
462 	unsigned int lookup_flags = LOOKUP_FOLLOW;
463 	const struct cred *old_cred = NULL;
464 
465 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
466 		return -EINVAL;
467 
468 	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
469 		return -EINVAL;
470 
471 	if (flags & AT_SYMLINK_NOFOLLOW)
472 		lookup_flags &= ~LOOKUP_FOLLOW;
473 	if (flags & AT_EMPTY_PATH)
474 		lookup_flags |= LOOKUP_EMPTY;
475 
476 	if (access_need_override_creds(flags)) {
477 		old_cred = access_override_creds();
478 		if (!old_cred)
479 			return -ENOMEM;
480 	}
481 
482 retry:
483 	res = user_path_at(dfd, filename, lookup_flags, &path);
484 	if (res)
485 		goto out;
486 
487 	inode = d_backing_inode(path.dentry);
488 
489 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
490 		/*
491 		 * MAY_EXEC on regular files is denied if the fs is mounted
492 		 * with the "noexec" flag.
493 		 */
494 		res = -EACCES;
495 		if (path_noexec(&path))
496 			goto out_path_release;
497 	}
498 
499 	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
500 	/* SuS v2 requires we report a read only fs too */
501 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
502 		goto out_path_release;
503 	/*
504 	 * This is a rare case where using __mnt_is_readonly()
505 	 * is OK without a mnt_want/drop_write() pair.  Since
506 	 * no actual write to the fs is performed here, we do
507 	 * not need to telegraph to that to anyone.
508 	 *
509 	 * By doing this, we accept that this access is
510 	 * inherently racy and know that the fs may change
511 	 * state before we even see this result.
512 	 */
513 	if (__mnt_is_readonly(path.mnt))
514 		res = -EROFS;
515 
516 out_path_release:
517 	path_put(&path);
518 	if (retry_estale(res, lookup_flags)) {
519 		lookup_flags |= LOOKUP_REVAL;
520 		goto retry;
521 	}
522 out:
523 	if (old_cred)
524 		revert_creds(old_cred);
525 
526 	return res;
527 }
528 
SYSCALL_DEFINE3(faccessat,int,dfd,const char __user *,filename,int,mode)529 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
530 {
531 	return do_faccessat(dfd, filename, mode, 0);
532 }
533 
SYSCALL_DEFINE4(faccessat2,int,dfd,const char __user *,filename,int,mode,int,flags)534 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
535 		int, flags)
536 {
537 	return do_faccessat(dfd, filename, mode, flags);
538 }
539 
SYSCALL_DEFINE2(access,const char __user *,filename,int,mode)540 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
541 {
542 	return do_faccessat(AT_FDCWD, filename, mode, 0);
543 }
544 
SYSCALL_DEFINE1(chdir,const char __user *,filename)545 SYSCALL_DEFINE1(chdir, const char __user *, filename)
546 {
547 	struct path path;
548 	int error;
549 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
550 retry:
551 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
552 	if (error)
553 		goto out;
554 
555 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
556 	if (error)
557 		goto dput_and_out;
558 
559 	set_fs_pwd(current->fs, &path);
560 
561 dput_and_out:
562 	path_put(&path);
563 	if (retry_estale(error, lookup_flags)) {
564 		lookup_flags |= LOOKUP_REVAL;
565 		goto retry;
566 	}
567 out:
568 	return error;
569 }
570 
SYSCALL_DEFINE1(fchdir,unsigned int,fd)571 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
572 {
573 	struct fd f = fdget_raw(fd);
574 	int error;
575 
576 	error = -EBADF;
577 	if (!f.file)
578 		goto out;
579 
580 	error = -ENOTDIR;
581 	if (!d_can_lookup(f.file->f_path.dentry))
582 		goto out_putf;
583 
584 	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
585 	if (!error)
586 		set_fs_pwd(current->fs, &f.file->f_path);
587 out_putf:
588 	fdput(f);
589 out:
590 	return error;
591 }
592 
SYSCALL_DEFINE1(chroot,const char __user *,filename)593 SYSCALL_DEFINE1(chroot, const char __user *, filename)
594 {
595 	struct path path;
596 	int error;
597 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
598 retry:
599 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
600 	if (error)
601 		goto out;
602 
603 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
604 	if (error)
605 		goto dput_and_out;
606 
607 	error = -EPERM;
608 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
609 		goto dput_and_out;
610 	error = security_path_chroot(&path);
611 	if (error)
612 		goto dput_and_out;
613 
614 	set_fs_root(current->fs, &path);
615 	error = 0;
616 dput_and_out:
617 	path_put(&path);
618 	if (retry_estale(error, lookup_flags)) {
619 		lookup_flags |= LOOKUP_REVAL;
620 		goto retry;
621 	}
622 out:
623 	return error;
624 }
625 
chmod_common(const struct path * path,umode_t mode)626 int chmod_common(const struct path *path, umode_t mode)
627 {
628 	struct inode *inode = path->dentry->d_inode;
629 	struct inode *delegated_inode = NULL;
630 	struct iattr newattrs;
631 	int error;
632 
633 	error = mnt_want_write(path->mnt);
634 	if (error)
635 		return error;
636 retry_deleg:
637 	inode_lock(inode);
638 	error = security_path_chmod(path, mode);
639 	if (error)
640 		goto out_unlock;
641 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
642 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
643 	error = notify_change(mnt_idmap(path->mnt), path->dentry,
644 			      &newattrs, &delegated_inode);
645 out_unlock:
646 	inode_unlock(inode);
647 	if (delegated_inode) {
648 		error = break_deleg_wait(&delegated_inode);
649 		if (!error)
650 			goto retry_deleg;
651 	}
652 	mnt_drop_write(path->mnt);
653 	return error;
654 }
655 
vfs_fchmod(struct file * file,umode_t mode)656 int vfs_fchmod(struct file *file, umode_t mode)
657 {
658 	audit_file(file);
659 	return chmod_common(&file->f_path, mode);
660 }
661 
SYSCALL_DEFINE2(fchmod,unsigned int,fd,umode_t,mode)662 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
663 {
664 	struct fd f = fdget(fd);
665 	int err = -EBADF;
666 
667 	if (f.file) {
668 		err = vfs_fchmod(f.file, mode);
669 		fdput(f);
670 	}
671 	return err;
672 }
673 
do_fchmodat(int dfd,const char __user * filename,umode_t mode,unsigned int flags)674 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
675 		       unsigned int flags)
676 {
677 	struct path path;
678 	int error;
679 	unsigned int lookup_flags;
680 
681 	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
682 		return -EINVAL;
683 
684 	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
685 	if (flags & AT_EMPTY_PATH)
686 		lookup_flags |= LOOKUP_EMPTY;
687 
688 retry:
689 	error = user_path_at(dfd, filename, lookup_flags, &path);
690 	if (!error) {
691 		error = chmod_common(&path, mode);
692 		path_put(&path);
693 		if (retry_estale(error, lookup_flags)) {
694 			lookup_flags |= LOOKUP_REVAL;
695 			goto retry;
696 		}
697 	}
698 	return error;
699 }
700 
SYSCALL_DEFINE4(fchmodat2,int,dfd,const char __user *,filename,umode_t,mode,unsigned int,flags)701 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
702 		umode_t, mode, unsigned int, flags)
703 {
704 	return do_fchmodat(dfd, filename, mode, flags);
705 }
706 
SYSCALL_DEFINE3(fchmodat,int,dfd,const char __user *,filename,umode_t,mode)707 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
708 		umode_t, mode)
709 {
710 	return do_fchmodat(dfd, filename, mode, 0);
711 }
712 
SYSCALL_DEFINE2(chmod,const char __user *,filename,umode_t,mode)713 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
714 {
715 	return do_fchmodat(AT_FDCWD, filename, mode, 0);
716 }
717 
718 /*
719  * Check whether @kuid is valid and if so generate and set vfsuid_t in
720  * ia_vfsuid.
721  *
722  * Return: true if @kuid is valid, false if not.
723  */
setattr_vfsuid(struct iattr * attr,kuid_t kuid)724 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
725 {
726 	if (!uid_valid(kuid))
727 		return false;
728 	attr->ia_valid |= ATTR_UID;
729 	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
730 	return true;
731 }
732 
733 /*
734  * Check whether @kgid is valid and if so generate and set vfsgid_t in
735  * ia_vfsgid.
736  *
737  * Return: true if @kgid is valid, false if not.
738  */
setattr_vfsgid(struct iattr * attr,kgid_t kgid)739 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
740 {
741 	if (!gid_valid(kgid))
742 		return false;
743 	attr->ia_valid |= ATTR_GID;
744 	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
745 	return true;
746 }
747 
chown_common(const struct path * path,uid_t user,gid_t group)748 int chown_common(const struct path *path, uid_t user, gid_t group)
749 {
750 	struct mnt_idmap *idmap;
751 	struct user_namespace *fs_userns;
752 	struct inode *inode = path->dentry->d_inode;
753 	struct inode *delegated_inode = NULL;
754 	int error;
755 	struct iattr newattrs;
756 	kuid_t uid;
757 	kgid_t gid;
758 
759 	uid = make_kuid(current_user_ns(), user);
760 	gid = make_kgid(current_user_ns(), group);
761 
762 	idmap = mnt_idmap(path->mnt);
763 	fs_userns = i_user_ns(inode);
764 
765 retry_deleg:
766 	newattrs.ia_vfsuid = INVALID_VFSUID;
767 	newattrs.ia_vfsgid = INVALID_VFSGID;
768 	newattrs.ia_valid =  ATTR_CTIME;
769 	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
770 		return -EINVAL;
771 	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
772 		return -EINVAL;
773 	inode_lock(inode);
774 	if (!S_ISDIR(inode->i_mode))
775 		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
776 				     setattr_should_drop_sgid(idmap, inode);
777 	/* Continue to send actual fs values, not the mount values. */
778 	error = security_path_chown(
779 		path,
780 		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
781 		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
782 	if (!error)
783 		error = notify_change(idmap, path->dentry, &newattrs,
784 				      &delegated_inode);
785 	inode_unlock(inode);
786 	if (delegated_inode) {
787 		error = break_deleg_wait(&delegated_inode);
788 		if (!error)
789 			goto retry_deleg;
790 	}
791 	return error;
792 }
793 
do_fchownat(int dfd,const char __user * filename,uid_t user,gid_t group,int flag)794 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
795 		int flag)
796 {
797 	struct path path;
798 	int error = -EINVAL;
799 	int lookup_flags;
800 
801 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
802 		goto out;
803 
804 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
805 	if (flag & AT_EMPTY_PATH)
806 		lookup_flags |= LOOKUP_EMPTY;
807 retry:
808 	error = user_path_at(dfd, filename, lookup_flags, &path);
809 	if (error)
810 		goto out;
811 	error = mnt_want_write(path.mnt);
812 	if (error)
813 		goto out_release;
814 	error = chown_common(&path, user, group);
815 	mnt_drop_write(path.mnt);
816 out_release:
817 	path_put(&path);
818 	if (retry_estale(error, lookup_flags)) {
819 		lookup_flags |= LOOKUP_REVAL;
820 		goto retry;
821 	}
822 out:
823 	return error;
824 }
825 
SYSCALL_DEFINE5(fchownat,int,dfd,const char __user *,filename,uid_t,user,gid_t,group,int,flag)826 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
827 		gid_t, group, int, flag)
828 {
829 	return do_fchownat(dfd, filename, user, group, flag);
830 }
831 
SYSCALL_DEFINE3(chown,const char __user *,filename,uid_t,user,gid_t,group)832 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
833 {
834 	return do_fchownat(AT_FDCWD, filename, user, group, 0);
835 }
836 
SYSCALL_DEFINE3(lchown,const char __user *,filename,uid_t,user,gid_t,group)837 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
838 {
839 	return do_fchownat(AT_FDCWD, filename, user, group,
840 			   AT_SYMLINK_NOFOLLOW);
841 }
842 
vfs_fchown(struct file * file,uid_t user,gid_t group)843 int vfs_fchown(struct file *file, uid_t user, gid_t group)
844 {
845 	int error;
846 
847 	error = mnt_want_write_file(file);
848 	if (error)
849 		return error;
850 	audit_file(file);
851 	error = chown_common(&file->f_path, user, group);
852 	mnt_drop_write_file(file);
853 	return error;
854 }
855 
ksys_fchown(unsigned int fd,uid_t user,gid_t group)856 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
857 {
858 	struct fd f = fdget(fd);
859 	int error = -EBADF;
860 
861 	if (f.file) {
862 		error = vfs_fchown(f.file, user, group);
863 		fdput(f);
864 	}
865 	return error;
866 }
867 
SYSCALL_DEFINE3(fchown,unsigned int,fd,uid_t,user,gid_t,group)868 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
869 {
870 	return ksys_fchown(fd, user, group);
871 }
872 
do_dentry_open(struct file * f,struct inode * inode,int (* open)(struct inode *,struct file *))873 static int do_dentry_open(struct file *f,
874 			  struct inode *inode,
875 			  int (*open)(struct inode *, struct file *))
876 {
877 	static const struct file_operations empty_fops = {};
878 	int error;
879 
880 	path_get(&f->f_path);
881 	f->f_inode = inode;
882 	f->f_mapping = inode->i_mapping;
883 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
884 	f->f_sb_err = file_sample_sb_err(f);
885 
886 	if (unlikely(f->f_flags & O_PATH)) {
887 		f->f_mode = FMODE_PATH | FMODE_OPENED;
888 		f->f_op = &empty_fops;
889 		return 0;
890 	}
891 
892 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
893 		i_readcount_inc(inode);
894 	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
895 		error = get_write_access(inode);
896 		if (unlikely(error))
897 			goto cleanup_file;
898 		error = __mnt_want_write(f->f_path.mnt);
899 		if (unlikely(error)) {
900 			put_write_access(inode);
901 			goto cleanup_file;
902 		}
903 		f->f_mode |= FMODE_WRITER;
904 	}
905 
906 	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
907 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
908 		f->f_mode |= FMODE_ATOMIC_POS;
909 
910 	f->f_op = fops_get(inode->i_fop);
911 	if (WARN_ON(!f->f_op)) {
912 		error = -ENODEV;
913 		goto cleanup_all;
914 	}
915 
916 	error = security_file_open(f);
917 	if (error)
918 		goto cleanup_all;
919 
920 	error = break_lease(file_inode(f), f->f_flags);
921 	if (error)
922 		goto cleanup_all;
923 
924 	/* normally all 3 are set; ->open() can clear them if needed */
925 	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
926 	if (!open)
927 		open = f->f_op->open;
928 	if (open) {
929 		error = open(inode, f);
930 		if (error)
931 			goto cleanup_all;
932 	}
933 	f->f_mode |= FMODE_OPENED;
934 	if ((f->f_mode & FMODE_READ) &&
935 	     likely(f->f_op->read || f->f_op->read_iter))
936 		f->f_mode |= FMODE_CAN_READ;
937 	if ((f->f_mode & FMODE_WRITE) &&
938 	     likely(f->f_op->write || f->f_op->write_iter))
939 		f->f_mode |= FMODE_CAN_WRITE;
940 	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
941 		f->f_mode &= ~FMODE_LSEEK;
942 	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
943 		f->f_mode |= FMODE_CAN_ODIRECT;
944 
945 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
946 	f->f_iocb_flags = iocb_flags(f);
947 
948 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
949 
950 	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
951 		return -EINVAL;
952 
953 	/*
954 	 * XXX: Huge page cache doesn't support writing yet. Drop all page
955 	 * cache for this file before processing writes.
956 	 */
957 	if (f->f_mode & FMODE_WRITE) {
958 		/*
959 		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
960 		 * is up to date and the update to i_writecount by
961 		 * get_write_access() is visible. Ensures subsequent insertion
962 		 * of THPs into the page cache will fail.
963 		 */
964 		smp_mb();
965 		if (filemap_nr_thps(inode->i_mapping)) {
966 			struct address_space *mapping = inode->i_mapping;
967 
968 			filemap_invalidate_lock(inode->i_mapping);
969 			/*
970 			 * unmap_mapping_range just need to be called once
971 			 * here, because the private pages is not need to be
972 			 * unmapped mapping (e.g. data segment of dynamic
973 			 * shared libraries here).
974 			 */
975 			unmap_mapping_range(mapping, 0, 0, 0);
976 			truncate_inode_pages(mapping, 0);
977 			filemap_invalidate_unlock(inode->i_mapping);
978 		}
979 	}
980 
981 	/*
982 	 * Once we return a file with FMODE_OPENED, __fput() will call
983 	 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
984 	 */
985 	fsnotify_open(f);
986 	return 0;
987 
988 cleanup_all:
989 	if (WARN_ON_ONCE(error > 0))
990 		error = -EINVAL;
991 	fops_put(f->f_op);
992 	put_file_access(f);
993 cleanup_file:
994 	path_put(&f->f_path);
995 	f->f_path.mnt = NULL;
996 	f->f_path.dentry = NULL;
997 	f->f_inode = NULL;
998 	return error;
999 }
1000 
1001 /**
1002  * finish_open - finish opening a file
1003  * @file: file pointer
1004  * @dentry: pointer to dentry
1005  * @open: open callback
1006  *
1007  * This can be used to finish opening a file passed to i_op->atomic_open().
1008  *
1009  * If the open callback is set to NULL, then the standard f_op->open()
1010  * filesystem callback is substituted.
1011  *
1012  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1013  * the return value of d_splice_alias(), then the caller needs to perform dput()
1014  * on it after finish_open().
1015  *
1016  * Returns zero on success or -errno if the open failed.
1017  */
finish_open(struct file * file,struct dentry * dentry,int (* open)(struct inode *,struct file *))1018 int finish_open(struct file *file, struct dentry *dentry,
1019 		int (*open)(struct inode *, struct file *))
1020 {
1021 	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1022 
1023 	file->f_path.dentry = dentry;
1024 	return do_dentry_open(file, d_backing_inode(dentry), open);
1025 }
1026 EXPORT_SYMBOL(finish_open);
1027 
1028 /**
1029  * finish_no_open - finish ->atomic_open() without opening the file
1030  *
1031  * @file: file pointer
1032  * @dentry: dentry or NULL (as returned from ->lookup())
1033  *
1034  * This can be used to set the result of a successful lookup in ->atomic_open().
1035  *
1036  * NB: unlike finish_open() this function does consume the dentry reference and
1037  * the caller need not dput() it.
1038  *
1039  * Returns "0" which must be the return value of ->atomic_open() after having
1040  * called this function.
1041  */
finish_no_open(struct file * file,struct dentry * dentry)1042 int finish_no_open(struct file *file, struct dentry *dentry)
1043 {
1044 	file->f_path.dentry = dentry;
1045 	return 0;
1046 }
1047 EXPORT_SYMBOL(finish_no_open);
1048 
file_path(struct file * filp,char * buf,int buflen)1049 char *file_path(struct file *filp, char *buf, int buflen)
1050 {
1051 	return d_path(&filp->f_path, buf, buflen);
1052 }
1053 EXPORT_SYMBOL(file_path);
1054 
1055 /**
1056  * vfs_open - open the file at the given path
1057  * @path: path to open
1058  * @file: newly allocated file with f_flag initialized
1059  */
vfs_open(const struct path * path,struct file * file)1060 int vfs_open(const struct path *path, struct file *file)
1061 {
1062 	file->f_path = *path;
1063 	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1064 }
1065 
dentry_open(const struct path * path,int flags,const struct cred * cred)1066 struct file *dentry_open(const struct path *path, int flags,
1067 			 const struct cred *cred)
1068 {
1069 	int error;
1070 	struct file *f;
1071 
1072 	/* We must always pass in a valid mount pointer. */
1073 	BUG_ON(!path->mnt);
1074 
1075 	f = alloc_empty_file(flags, cred);
1076 	if (!IS_ERR(f)) {
1077 		error = vfs_open(path, f);
1078 		if (error) {
1079 			fput(f);
1080 			f = ERR_PTR(error);
1081 		}
1082 	}
1083 	return f;
1084 }
1085 EXPORT_SYMBOL(dentry_open);
1086 
1087 /**
1088  * dentry_create - Create and open a file
1089  * @path: path to create
1090  * @flags: O_ flags
1091  * @mode: mode bits for new file
1092  * @cred: credentials to use
1093  *
1094  * Caller must hold the parent directory's lock, and have prepared
1095  * a negative dentry, placed in @path->dentry, for the new file.
1096  *
1097  * Caller sets @path->mnt to the vfsmount of the filesystem where
1098  * the new file is to be created. The parent directory and the
1099  * negative dentry must reside on the same filesystem instance.
1100  *
1101  * On success, returns a "struct file *". Otherwise a ERR_PTR
1102  * is returned.
1103  */
dentry_create(const struct path * path,int flags,umode_t mode,const struct cred * cred)1104 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1105 			   const struct cred *cred)
1106 {
1107 	struct file *f;
1108 	int error;
1109 
1110 	f = alloc_empty_file(flags, cred);
1111 	if (IS_ERR(f))
1112 		return f;
1113 
1114 	error = vfs_create(mnt_idmap(path->mnt),
1115 			   d_inode(path->dentry->d_parent),
1116 			   path->dentry, mode, true);
1117 	if (!error)
1118 		error = vfs_open(path, f);
1119 
1120 	if (unlikely(error)) {
1121 		fput(f);
1122 		return ERR_PTR(error);
1123 	}
1124 	return f;
1125 }
1126 EXPORT_SYMBOL(dentry_create);
1127 
1128 /**
1129  * kernel_file_open - open a file for kernel internal use
1130  * @path:	path of the file to open
1131  * @flags:	open flags
1132  * @inode:	the inode
1133  * @cred:	credentials for open
1134  *
1135  * Open a file for use by in-kernel consumers. The file is not accounted
1136  * against nr_files and must not be installed into the file descriptor
1137  * table.
1138  *
1139  * Return: Opened file on success, an error pointer on failure.
1140  */
kernel_file_open(const struct path * path,int flags,struct inode * inode,const struct cred * cred)1141 struct file *kernel_file_open(const struct path *path, int flags,
1142 				struct inode *inode, const struct cred *cred)
1143 {
1144 	struct file *f;
1145 	int error;
1146 
1147 	f = alloc_empty_file_noaccount(flags, cred);
1148 	if (IS_ERR(f))
1149 		return f;
1150 
1151 	f->f_path = *path;
1152 	error = do_dentry_open(f, inode, NULL);
1153 	if (error) {
1154 		fput(f);
1155 		f = ERR_PTR(error);
1156 	}
1157 	return f;
1158 }
1159 EXPORT_SYMBOL_GPL(kernel_file_open);
1160 
1161 /**
1162  * backing_file_open - open a backing file for kernel internal use
1163  * @path:	path of the file to open
1164  * @flags:	open flags
1165  * @real_path:	path of the backing file
1166  * @cred:	credentials for open
1167  *
1168  * Open a backing file for a stackable filesystem (e.g., overlayfs).
1169  * @path may be on the stackable filesystem and backing inode on the
1170  * underlying filesystem. In this case, we want to be able to return
1171  * the @real_path of the backing inode. This is done by embedding the
1172  * returned file into a container structure that also stores the path of
1173  * the backing inode on the underlying filesystem, which can be
1174  * retrieved using backing_file_real_path().
1175  */
backing_file_open(const struct path * path,int flags,const struct path * real_path,const struct cred * cred)1176 struct file *backing_file_open(const struct path *path, int flags,
1177 			       const struct path *real_path,
1178 			       const struct cred *cred)
1179 {
1180 	struct file *f;
1181 	int error;
1182 
1183 	f = alloc_empty_backing_file(flags, cred);
1184 	if (IS_ERR(f))
1185 		return f;
1186 
1187 	f->f_path = *path;
1188 	path_get(real_path);
1189 	*backing_file_real_path(f) = *real_path;
1190 	error = do_dentry_open(f, d_inode(real_path->dentry), NULL);
1191 	if (error) {
1192 		fput(f);
1193 		f = ERR_PTR(error);
1194 	}
1195 
1196 	return f;
1197 }
1198 EXPORT_SYMBOL_GPL(backing_file_open);
1199 
1200 #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1201 #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1202 
build_open_how(int flags,umode_t mode)1203 inline struct open_how build_open_how(int flags, umode_t mode)
1204 {
1205 	struct open_how how = {
1206 		.flags = flags & VALID_OPEN_FLAGS,
1207 		.mode = mode & S_IALLUGO,
1208 	};
1209 
1210 	/* O_PATH beats everything else. */
1211 	if (how.flags & O_PATH)
1212 		how.flags &= O_PATH_FLAGS;
1213 	/* Modes should only be set for create-like flags. */
1214 	if (!WILL_CREATE(how.flags))
1215 		how.mode = 0;
1216 	return how;
1217 }
1218 
build_open_flags(const struct open_how * how,struct open_flags * op)1219 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1220 {
1221 	u64 flags = how->flags;
1222 	u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1223 	int lookup_flags = 0;
1224 	int acc_mode = ACC_MODE(flags);
1225 
1226 	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1227 			 "struct open_flags doesn't yet handle flags > 32 bits");
1228 
1229 	/*
1230 	 * Strip flags that either shouldn't be set by userspace like
1231 	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1232 	 * open_flags like O_CLOEXEC.
1233 	 */
1234 	flags &= ~strip;
1235 
1236 	/*
1237 	 * Older syscalls implicitly clear all of the invalid flags or argument
1238 	 * values before calling build_open_flags(), but openat2(2) checks all
1239 	 * of its arguments.
1240 	 */
1241 	if (flags & ~VALID_OPEN_FLAGS)
1242 		return -EINVAL;
1243 	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1244 		return -EINVAL;
1245 
1246 	/* Scoping flags are mutually exclusive. */
1247 	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1248 		return -EINVAL;
1249 
1250 	/* Deal with the mode. */
1251 	if (WILL_CREATE(flags)) {
1252 		if (how->mode & ~S_IALLUGO)
1253 			return -EINVAL;
1254 		op->mode = how->mode | S_IFREG;
1255 	} else {
1256 		if (how->mode != 0)
1257 			return -EINVAL;
1258 		op->mode = 0;
1259 	}
1260 
1261 	/*
1262 	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1263 	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1264 	 * O_TMPFILE below which requires O_DIRECTORY being raised.
1265 	 */
1266 	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1267 		return -EINVAL;
1268 
1269 	/* Now handle the creative implementation of O_TMPFILE. */
1270 	if (flags & __O_TMPFILE) {
1271 		/*
1272 		 * In order to ensure programs get explicit errors when trying
1273 		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1274 		 * is raised alongside __O_TMPFILE.
1275 		 */
1276 		if (!(flags & O_DIRECTORY))
1277 			return -EINVAL;
1278 		if (!(acc_mode & MAY_WRITE))
1279 			return -EINVAL;
1280 	}
1281 	if (flags & O_PATH) {
1282 		/* O_PATH only permits certain other flags to be set. */
1283 		if (flags & ~O_PATH_FLAGS)
1284 			return -EINVAL;
1285 		acc_mode = 0;
1286 	}
1287 
1288 	/*
1289 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1290 	 * check for O_DSYNC if the need any syncing at all we enforce it's
1291 	 * always set instead of having to deal with possibly weird behaviour
1292 	 * for malicious applications setting only __O_SYNC.
1293 	 */
1294 	if (flags & __O_SYNC)
1295 		flags |= O_DSYNC;
1296 
1297 	op->open_flag = flags;
1298 
1299 	/* O_TRUNC implies we need access checks for write permissions */
1300 	if (flags & O_TRUNC)
1301 		acc_mode |= MAY_WRITE;
1302 
1303 	/* Allow the LSM permission hook to distinguish append
1304 	   access from general write access. */
1305 	if (flags & O_APPEND)
1306 		acc_mode |= MAY_APPEND;
1307 
1308 	op->acc_mode = acc_mode;
1309 
1310 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1311 
1312 	if (flags & O_CREAT) {
1313 		op->intent |= LOOKUP_CREATE;
1314 		if (flags & O_EXCL) {
1315 			op->intent |= LOOKUP_EXCL;
1316 			flags |= O_NOFOLLOW;
1317 		}
1318 	}
1319 
1320 	if (flags & O_DIRECTORY)
1321 		lookup_flags |= LOOKUP_DIRECTORY;
1322 	if (!(flags & O_NOFOLLOW))
1323 		lookup_flags |= LOOKUP_FOLLOW;
1324 
1325 	if (how->resolve & RESOLVE_NO_XDEV)
1326 		lookup_flags |= LOOKUP_NO_XDEV;
1327 	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1328 		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1329 	if (how->resolve & RESOLVE_NO_SYMLINKS)
1330 		lookup_flags |= LOOKUP_NO_SYMLINKS;
1331 	if (how->resolve & RESOLVE_BENEATH)
1332 		lookup_flags |= LOOKUP_BENEATH;
1333 	if (how->resolve & RESOLVE_IN_ROOT)
1334 		lookup_flags |= LOOKUP_IN_ROOT;
1335 	if (how->resolve & RESOLVE_CACHED) {
1336 		/* Don't bother even trying for create/truncate/tmpfile open */
1337 		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1338 			return -EAGAIN;
1339 		lookup_flags |= LOOKUP_CACHED;
1340 	}
1341 
1342 	op->lookup_flags = lookup_flags;
1343 	return 0;
1344 }
1345 
1346 /**
1347  * file_open_name - open file and return file pointer
1348  *
1349  * @name:	struct filename containing path to open
1350  * @flags:	open flags as per the open(2) second argument
1351  * @mode:	mode for the new file if O_CREAT is set, else ignored
1352  *
1353  * This is the helper to open a file from kernelspace if you really
1354  * have to.  But in generally you should not do this, so please move
1355  * along, nothing to see here..
1356  */
file_open_name(struct filename * name,int flags,umode_t mode)1357 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1358 {
1359 	struct open_flags op;
1360 	struct open_how how = build_open_how(flags, mode);
1361 	int err = build_open_flags(&how, &op);
1362 	if (err)
1363 		return ERR_PTR(err);
1364 	return do_filp_open(AT_FDCWD, name, &op);
1365 }
1366 
1367 /**
1368  * filp_open - open file and return file pointer
1369  *
1370  * @filename:	path to open
1371  * @flags:	open flags as per the open(2) second argument
1372  * @mode:	mode for the new file if O_CREAT is set, else ignored
1373  *
1374  * This is the helper to open a file from kernelspace if you really
1375  * have to.  But in generally you should not do this, so please move
1376  * along, nothing to see here..
1377  */
filp_open(const char * filename,int flags,umode_t mode)1378 struct file *filp_open(const char *filename, int flags, umode_t mode)
1379 {
1380 	struct filename *name = getname_kernel(filename);
1381 	struct file *file = ERR_CAST(name);
1382 
1383 	if (!IS_ERR(name)) {
1384 		file = file_open_name(name, flags, mode);
1385 		putname(name);
1386 	}
1387 	return file;
1388 }
1389 EXPORT_SYMBOL(filp_open);
1390 
file_open_root(const struct path * root,const char * filename,int flags,umode_t mode)1391 struct file *file_open_root(const struct path *root,
1392 			    const char *filename, int flags, umode_t mode)
1393 {
1394 	struct open_flags op;
1395 	struct open_how how = build_open_how(flags, mode);
1396 	int err = build_open_flags(&how, &op);
1397 	if (err)
1398 		return ERR_PTR(err);
1399 	return do_file_open_root(root, filename, &op);
1400 }
1401 EXPORT_SYMBOL(file_open_root);
1402 
do_sys_openat2(int dfd,const char __user * filename,struct open_how * how)1403 static long do_sys_openat2(int dfd, const char __user *filename,
1404 			   struct open_how *how)
1405 {
1406 	struct open_flags op;
1407 	int fd = build_open_flags(how, &op);
1408 	struct filename *tmp;
1409 
1410 	if (fd)
1411 		return fd;
1412 
1413 	tmp = getname(filename);
1414 	if (IS_ERR(tmp))
1415 		return PTR_ERR(tmp);
1416 
1417 	fd = get_unused_fd_flags(how->flags);
1418 	if (fd >= 0) {
1419 		struct file *f = do_filp_open(dfd, tmp, &op);
1420 		if (IS_ERR(f)) {
1421 			put_unused_fd(fd);
1422 			fd = PTR_ERR(f);
1423 		} else {
1424 			fd_install(fd, f);
1425 		}
1426 	}
1427 	putname(tmp);
1428 	return fd;
1429 }
1430 
do_sys_open(int dfd,const char __user * filename,int flags,umode_t mode)1431 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1432 {
1433 	struct open_how how = build_open_how(flags, mode);
1434 	return do_sys_openat2(dfd, filename, &how);
1435 }
1436 
1437 
SYSCALL_DEFINE3(open,const char __user *,filename,int,flags,umode_t,mode)1438 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1439 {
1440 	if (force_o_largefile())
1441 		flags |= O_LARGEFILE;
1442 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1443 }
1444 
SYSCALL_DEFINE4(openat,int,dfd,const char __user *,filename,int,flags,umode_t,mode)1445 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1446 		umode_t, mode)
1447 {
1448 	if (force_o_largefile())
1449 		flags |= O_LARGEFILE;
1450 	return do_sys_open(dfd, filename, flags, mode);
1451 }
1452 
SYSCALL_DEFINE4(openat2,int,dfd,const char __user *,filename,struct open_how __user *,how,size_t,usize)1453 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1454 		struct open_how __user *, how, size_t, usize)
1455 {
1456 	int err;
1457 	struct open_how tmp;
1458 
1459 	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1460 	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1461 
1462 	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1463 		return -EINVAL;
1464 
1465 	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1466 	if (err)
1467 		return err;
1468 
1469 	audit_openat2_how(&tmp);
1470 
1471 	/* O_LARGEFILE is only allowed for non-O_PATH. */
1472 	if (!(tmp.flags & O_PATH) && force_o_largefile())
1473 		tmp.flags |= O_LARGEFILE;
1474 
1475 	return do_sys_openat2(dfd, filename, &tmp);
1476 }
1477 
1478 #ifdef CONFIG_COMPAT
1479 /*
1480  * Exactly like sys_open(), except that it doesn't set the
1481  * O_LARGEFILE flag.
1482  */
COMPAT_SYSCALL_DEFINE3(open,const char __user *,filename,int,flags,umode_t,mode)1483 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1484 {
1485 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1486 }
1487 
1488 /*
1489  * Exactly like sys_openat(), except that it doesn't set the
1490  * O_LARGEFILE flag.
1491  */
COMPAT_SYSCALL_DEFINE4(openat,int,dfd,const char __user *,filename,int,flags,umode_t,mode)1492 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1493 {
1494 	return do_sys_open(dfd, filename, flags, mode);
1495 }
1496 #endif
1497 
1498 #ifndef __alpha__
1499 
1500 /*
1501  * For backward compatibility?  Maybe this should be moved
1502  * into arch/i386 instead?
1503  */
SYSCALL_DEFINE2(creat,const char __user *,pathname,umode_t,mode)1504 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1505 {
1506 	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1507 
1508 	if (force_o_largefile())
1509 		flags |= O_LARGEFILE;
1510 	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1511 }
1512 #endif
1513 
1514 /*
1515  * "id" is the POSIX thread ID. We use the
1516  * files pointer for this..
1517  */
filp_flush(struct file * filp,fl_owner_t id)1518 static int filp_flush(struct file *filp, fl_owner_t id)
1519 {
1520 	int retval = 0;
1521 
1522 	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1523 			"VFS: Close: file count is 0 (f_op=%ps)",
1524 			filp->f_op)) {
1525 		return 0;
1526 	}
1527 
1528 	if (filp->f_op->flush)
1529 		retval = filp->f_op->flush(filp, id);
1530 
1531 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1532 		dnotify_flush(filp, id);
1533 		locks_remove_posix(filp, id);
1534 	}
1535 	return retval;
1536 }
1537 
filp_close(struct file * filp,fl_owner_t id)1538 int filp_close(struct file *filp, fl_owner_t id)
1539 {
1540 	int retval;
1541 
1542 	retval = filp_flush(filp, id);
1543 	fput(filp);
1544 
1545 	return retval;
1546 }
1547 EXPORT_SYMBOL(filp_close);
1548 
1549 /*
1550  * Careful here! We test whether the file pointer is NULL before
1551  * releasing the fd. This ensures that one clone task can't release
1552  * an fd while another clone is opening it.
1553  */
SYSCALL_DEFINE1(close,unsigned int,fd)1554 SYSCALL_DEFINE1(close, unsigned int, fd)
1555 {
1556 	int retval;
1557 	struct file *file;
1558 
1559 	file = close_fd_get_file(fd);
1560 	if (!file)
1561 		return -EBADF;
1562 
1563 	retval = filp_flush(file, current->files);
1564 
1565 	/*
1566 	 * We're returning to user space. Don't bother
1567 	 * with any delayed fput() cases.
1568 	 */
1569 	__fput_sync(file);
1570 
1571 	/* can't restart close syscall because file table entry was cleared */
1572 	if (unlikely(retval == -ERESTARTSYS ||
1573 		     retval == -ERESTARTNOINTR ||
1574 		     retval == -ERESTARTNOHAND ||
1575 		     retval == -ERESTART_RESTARTBLOCK))
1576 		retval = -EINTR;
1577 
1578 	return retval;
1579 }
1580 
1581 /**
1582  * sys_close_range() - Close all file descriptors in a given range.
1583  *
1584  * @fd:     starting file descriptor to close
1585  * @max_fd: last file descriptor to close
1586  * @flags:  reserved for future extensions
1587  *
1588  * This closes a range of file descriptors. All file descriptors
1589  * from @fd up to and including @max_fd are closed.
1590  * Currently, errors to close a given file descriptor are ignored.
1591  */
SYSCALL_DEFINE3(close_range,unsigned int,fd,unsigned int,max_fd,unsigned int,flags)1592 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1593 		unsigned int, flags)
1594 {
1595 	return __close_range(fd, max_fd, flags);
1596 }
1597 
1598 /*
1599  * This routine simulates a hangup on the tty, to arrange that users
1600  * are given clean terminals at login time.
1601  */
SYSCALL_DEFINE0(vhangup)1602 SYSCALL_DEFINE0(vhangup)
1603 {
1604 	if (capable(CAP_SYS_TTY_CONFIG)) {
1605 		tty_vhangup_self();
1606 		return 0;
1607 	}
1608 	return -EPERM;
1609 }
1610 
1611 /*
1612  * Called when an inode is about to be open.
1613  * We use this to disallow opening large files on 32bit systems if
1614  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1615  * on this flag in sys_open.
1616  */
generic_file_open(struct inode * inode,struct file * filp)1617 int generic_file_open(struct inode * inode, struct file * filp)
1618 {
1619 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1620 		return -EOVERFLOW;
1621 	return 0;
1622 }
1623 
1624 EXPORT_SYMBOL(generic_file_open);
1625 
1626 /*
1627  * This is used by subsystems that don't want seekable
1628  * file descriptors. The function is not supposed to ever fail, the only
1629  * reason it returns an 'int' and not 'void' is so that it can be plugged
1630  * directly into file_operations structure.
1631  */
nonseekable_open(struct inode * inode,struct file * filp)1632 int nonseekable_open(struct inode *inode, struct file *filp)
1633 {
1634 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1635 	return 0;
1636 }
1637 
1638 EXPORT_SYMBOL(nonseekable_open);
1639 
1640 /*
1641  * stream_open is used by subsystems that want stream-like file descriptors.
1642  * Such file descriptors are not seekable and don't have notion of position
1643  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1644  * Contrary to file descriptors of other regular files, .read() and .write()
1645  * can run simultaneously.
1646  *
1647  * stream_open never fails and is marked to return int so that it could be
1648  * directly used as file_operations.open .
1649  */
stream_open(struct inode * inode,struct file * filp)1650 int stream_open(struct inode *inode, struct file *filp)
1651 {
1652 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1653 	filp->f_mode |= FMODE_STREAM;
1654 	return 0;
1655 }
1656 
1657 EXPORT_SYMBOL(stream_open);
1658