xref: /openbmc/linux/fs/overlayfs/copy_up.c (revision cbdf59ad)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright (C) 2011 Novell Inc.
5  */
6 
7 #include <linux/module.h>
8 #include <linux/fs.h>
9 #include <linux/slab.h>
10 #include <linux/file.h>
11 #include <linux/splice.h>
12 #include <linux/xattr.h>
13 #include <linux/security.h>
14 #include <linux/uaccess.h>
15 #include <linux/sched/signal.h>
16 #include <linux/cred.h>
17 #include <linux/namei.h>
18 #include <linux/fdtable.h>
19 #include <linux/ratelimit.h>
20 #include <linux/exportfs.h>
21 #include "overlayfs.h"
22 
23 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
24 
25 static int ovl_ccup_set(const char *buf, const struct kernel_param *param)
26 {
27 	pr_warn("overlayfs: \"check_copy_up\" module option is obsolete\n");
28 	return 0;
29 }
30 
31 static int ovl_ccup_get(char *buf, const struct kernel_param *param)
32 {
33 	return sprintf(buf, "N\n");
34 }
35 
36 module_param_call(check_copy_up, ovl_ccup_set, ovl_ccup_get, NULL, 0644);
37 MODULE_PARM_DESC(check_copy_up, "Obsolete; does nothing");
38 
39 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
40 {
41 	ssize_t list_size, size, value_size = 0;
42 	char *buf, *name, *value = NULL;
43 	int uninitialized_var(error);
44 	size_t slen;
45 
46 	if (!(old->d_inode->i_opflags & IOP_XATTR) ||
47 	    !(new->d_inode->i_opflags & IOP_XATTR))
48 		return 0;
49 
50 	list_size = vfs_listxattr(old, NULL, 0);
51 	if (list_size <= 0) {
52 		if (list_size == -EOPNOTSUPP)
53 			return 0;
54 		return list_size;
55 	}
56 
57 	buf = kzalloc(list_size, GFP_KERNEL);
58 	if (!buf)
59 		return -ENOMEM;
60 
61 	list_size = vfs_listxattr(old, buf, list_size);
62 	if (list_size <= 0) {
63 		error = list_size;
64 		goto out;
65 	}
66 
67 	for (name = buf; list_size; name += slen) {
68 		slen = strnlen(name, list_size) + 1;
69 
70 		/* underlying fs providing us with an broken xattr list? */
71 		if (WARN_ON(slen > list_size)) {
72 			error = -EIO;
73 			break;
74 		}
75 		list_size -= slen;
76 
77 		if (ovl_is_private_xattr(name))
78 			continue;
79 retry:
80 		size = vfs_getxattr(old, name, value, value_size);
81 		if (size == -ERANGE)
82 			size = vfs_getxattr(old, name, NULL, 0);
83 
84 		if (size < 0) {
85 			error = size;
86 			break;
87 		}
88 
89 		if (size > value_size) {
90 			void *new;
91 
92 			new = krealloc(value, size, GFP_KERNEL);
93 			if (!new) {
94 				error = -ENOMEM;
95 				break;
96 			}
97 			value = new;
98 			value_size = size;
99 			goto retry;
100 		}
101 
102 		error = security_inode_copy_up_xattr(name);
103 		if (error < 0 && error != -EOPNOTSUPP)
104 			break;
105 		if (error == 1) {
106 			error = 0;
107 			continue; /* Discard */
108 		}
109 		error = vfs_setxattr(new, name, value, size, 0);
110 		if (error)
111 			break;
112 	}
113 	kfree(value);
114 out:
115 	kfree(buf);
116 	return error;
117 }
118 
119 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
120 {
121 	struct file *old_file;
122 	struct file *new_file;
123 	loff_t old_pos = 0;
124 	loff_t new_pos = 0;
125 	loff_t cloned;
126 	int error = 0;
127 
128 	if (len == 0)
129 		return 0;
130 
131 	old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
132 	if (IS_ERR(old_file))
133 		return PTR_ERR(old_file);
134 
135 	new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
136 	if (IS_ERR(new_file)) {
137 		error = PTR_ERR(new_file);
138 		goto out_fput;
139 	}
140 
141 	/* Try to use clone_file_range to clone up within the same fs */
142 	cloned = do_clone_file_range(old_file, 0, new_file, 0, len, 0);
143 	if (cloned == len)
144 		goto out;
145 	/* Couldn't clone, so now we try to copy the data */
146 
147 	/* FIXME: copy up sparse files efficiently */
148 	while (len) {
149 		size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
150 		long bytes;
151 
152 		if (len < this_len)
153 			this_len = len;
154 
155 		if (signal_pending_state(TASK_KILLABLE, current)) {
156 			error = -EINTR;
157 			break;
158 		}
159 
160 		bytes = do_splice_direct(old_file, &old_pos,
161 					 new_file, &new_pos,
162 					 this_len, SPLICE_F_MOVE);
163 		if (bytes <= 0) {
164 			error = bytes;
165 			break;
166 		}
167 		WARN_ON(old_pos != new_pos);
168 
169 		len -= bytes;
170 	}
171 out:
172 	if (!error)
173 		error = vfs_fsync(new_file, 0);
174 	fput(new_file);
175 out_fput:
176 	fput(old_file);
177 	return error;
178 }
179 
180 static int ovl_set_size(struct dentry *upperdentry, struct kstat *stat)
181 {
182 	struct iattr attr = {
183 		.ia_valid = ATTR_SIZE,
184 		.ia_size = stat->size,
185 	};
186 
187 	return notify_change(upperdentry, &attr, NULL);
188 }
189 
190 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
191 {
192 	struct iattr attr = {
193 		.ia_valid =
194 		     ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
195 		.ia_atime = stat->atime,
196 		.ia_mtime = stat->mtime,
197 	};
198 
199 	return notify_change(upperdentry, &attr, NULL);
200 }
201 
202 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
203 {
204 	int err = 0;
205 
206 	if (!S_ISLNK(stat->mode)) {
207 		struct iattr attr = {
208 			.ia_valid = ATTR_MODE,
209 			.ia_mode = stat->mode,
210 		};
211 		err = notify_change(upperdentry, &attr, NULL);
212 	}
213 	if (!err) {
214 		struct iattr attr = {
215 			.ia_valid = ATTR_UID | ATTR_GID,
216 			.ia_uid = stat->uid,
217 			.ia_gid = stat->gid,
218 		};
219 		err = notify_change(upperdentry, &attr, NULL);
220 	}
221 	if (!err)
222 		ovl_set_timestamps(upperdentry, stat);
223 
224 	return err;
225 }
226 
227 struct ovl_fh *ovl_encode_real_fh(struct dentry *real, bool is_upper)
228 {
229 	struct ovl_fh *fh;
230 	int fh_type, fh_len, dwords;
231 	void *buf;
232 	int buflen = MAX_HANDLE_SZ;
233 	uuid_t *uuid = &real->d_sb->s_uuid;
234 
235 	buf = kmalloc(buflen, GFP_KERNEL);
236 	if (!buf)
237 		return ERR_PTR(-ENOMEM);
238 
239 	/*
240 	 * We encode a non-connectable file handle for non-dir, because we
241 	 * only need to find the lower inode number and we don't want to pay
242 	 * the price or reconnecting the dentry.
243 	 */
244 	dwords = buflen >> 2;
245 	fh_type = exportfs_encode_fh(real, buf, &dwords, 0);
246 	buflen = (dwords << 2);
247 
248 	fh = ERR_PTR(-EIO);
249 	if (WARN_ON(fh_type < 0) ||
250 	    WARN_ON(buflen > MAX_HANDLE_SZ) ||
251 	    WARN_ON(fh_type == FILEID_INVALID))
252 		goto out;
253 
254 	BUILD_BUG_ON(MAX_HANDLE_SZ + offsetof(struct ovl_fh, fid) > 255);
255 	fh_len = offsetof(struct ovl_fh, fid) + buflen;
256 	fh = kmalloc(fh_len, GFP_KERNEL);
257 	if (!fh) {
258 		fh = ERR_PTR(-ENOMEM);
259 		goto out;
260 	}
261 
262 	fh->version = OVL_FH_VERSION;
263 	fh->magic = OVL_FH_MAGIC;
264 	fh->type = fh_type;
265 	fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
266 	/*
267 	 * When we will want to decode an overlay dentry from this handle
268 	 * and all layers are on the same fs, if we get a disconncted real
269 	 * dentry when we decode fid, the only way to tell if we should assign
270 	 * it to upperdentry or to lowerstack is by checking this flag.
271 	 */
272 	if (is_upper)
273 		fh->flags |= OVL_FH_FLAG_PATH_UPPER;
274 	fh->len = fh_len;
275 	fh->uuid = *uuid;
276 	memcpy(fh->fid, buf, buflen);
277 
278 out:
279 	kfree(buf);
280 	return fh;
281 }
282 
283 int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
284 		   struct dentry *upper)
285 {
286 	const struct ovl_fh *fh = NULL;
287 	int err;
288 
289 	/*
290 	 * When lower layer doesn't support export operations store a 'null' fh,
291 	 * so we can use the overlay.origin xattr to distignuish between a copy
292 	 * up and a pure upper inode.
293 	 */
294 	if (ovl_can_decode_fh(lower->d_sb)) {
295 		fh = ovl_encode_real_fh(lower, false);
296 		if (IS_ERR(fh))
297 			return PTR_ERR(fh);
298 	}
299 
300 	/*
301 	 * Do not fail when upper doesn't support xattrs.
302 	 */
303 	err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
304 				 fh ? fh->len : 0, 0);
305 	kfree(fh);
306 
307 	return err;
308 }
309 
310 /* Store file handle of @upper dir in @index dir entry */
311 static int ovl_set_upper_fh(struct dentry *upper, struct dentry *index)
312 {
313 	const struct ovl_fh *fh;
314 	int err;
315 
316 	fh = ovl_encode_real_fh(upper, true);
317 	if (IS_ERR(fh))
318 		return PTR_ERR(fh);
319 
320 	err = ovl_do_setxattr(index, OVL_XATTR_UPPER, fh, fh->len, 0);
321 
322 	kfree(fh);
323 	return err;
324 }
325 
326 /*
327  * Create and install index entry.
328  *
329  * Caller must hold i_mutex on indexdir.
330  */
331 static int ovl_create_index(struct dentry *dentry, struct dentry *origin,
332 			    struct dentry *upper)
333 {
334 	struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
335 	struct inode *dir = d_inode(indexdir);
336 	struct dentry *index = NULL;
337 	struct dentry *temp = NULL;
338 	struct qstr name = { };
339 	int err;
340 
341 	/*
342 	 * For now this is only used for creating index entry for directories,
343 	 * because non-dir are copied up directly to index and then hardlinked
344 	 * to upper dir.
345 	 *
346 	 * TODO: implement create index for non-dir, so we can call it when
347 	 * encoding file handle for non-dir in case index does not exist.
348 	 */
349 	if (WARN_ON(!d_is_dir(dentry)))
350 		return -EIO;
351 
352 	/* Directory not expected to be indexed before copy up */
353 	if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
354 		return -EIO;
355 
356 	err = ovl_get_index_name(origin, &name);
357 	if (err)
358 		return err;
359 
360 	temp = ovl_create_temp(indexdir, OVL_CATTR(S_IFDIR | 0));
361 	err = PTR_ERR(temp);
362 	if (IS_ERR(temp))
363 		goto free_name;
364 
365 	err = ovl_set_upper_fh(upper, temp);
366 	if (err)
367 		goto out;
368 
369 	index = lookup_one_len(name.name, indexdir, name.len);
370 	if (IS_ERR(index)) {
371 		err = PTR_ERR(index);
372 	} else {
373 		err = ovl_do_rename(dir, temp, dir, index, 0);
374 		dput(index);
375 	}
376 out:
377 	if (err)
378 		ovl_cleanup(dir, temp);
379 	dput(temp);
380 free_name:
381 	kfree(name.name);
382 	return err;
383 }
384 
385 struct ovl_copy_up_ctx {
386 	struct dentry *parent;
387 	struct dentry *dentry;
388 	struct path lowerpath;
389 	struct kstat stat;
390 	struct kstat pstat;
391 	const char *link;
392 	struct dentry *destdir;
393 	struct qstr destname;
394 	struct dentry *workdir;
395 	bool origin;
396 	bool indexed;
397 	bool metacopy;
398 };
399 
400 static int ovl_link_up(struct ovl_copy_up_ctx *c)
401 {
402 	int err;
403 	struct dentry *upper;
404 	struct dentry *upperdir = ovl_dentry_upper(c->parent);
405 	struct inode *udir = d_inode(upperdir);
406 
407 	/* Mark parent "impure" because it may now contain non-pure upper */
408 	err = ovl_set_impure(c->parent, upperdir);
409 	if (err)
410 		return err;
411 
412 	err = ovl_set_nlink_lower(c->dentry);
413 	if (err)
414 		return err;
415 
416 	inode_lock_nested(udir, I_MUTEX_PARENT);
417 	upper = lookup_one_len(c->dentry->d_name.name, upperdir,
418 			       c->dentry->d_name.len);
419 	err = PTR_ERR(upper);
420 	if (!IS_ERR(upper)) {
421 		err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper);
422 		dput(upper);
423 
424 		if (!err) {
425 			/* Restore timestamps on parent (best effort) */
426 			ovl_set_timestamps(upperdir, &c->pstat);
427 			ovl_dentry_set_upper_alias(c->dentry);
428 		}
429 	}
430 	inode_unlock(udir);
431 	if (err)
432 		return err;
433 
434 	err = ovl_set_nlink_upper(c->dentry);
435 
436 	return err;
437 }
438 
439 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
440 {
441 	int err;
442 
443 	/*
444 	 * Copy up data first and then xattrs. Writing data after
445 	 * xattrs will remove security.capability xattr automatically.
446 	 */
447 	if (S_ISREG(c->stat.mode) && !c->metacopy) {
448 		struct path upperpath, datapath;
449 
450 		ovl_path_upper(c->dentry, &upperpath);
451 		if (WARN_ON(upperpath.dentry != NULL))
452 			return -EIO;
453 		upperpath.dentry = temp;
454 
455 		ovl_path_lowerdata(c->dentry, &datapath);
456 		err = ovl_copy_up_data(&datapath, &upperpath, c->stat.size);
457 		if (err)
458 			return err;
459 	}
460 
461 	err = ovl_copy_xattr(c->lowerpath.dentry, temp);
462 	if (err)
463 		return err;
464 
465 	/*
466 	 * Store identifier of lower inode in upper inode xattr to
467 	 * allow lookup of the copy up origin inode.
468 	 *
469 	 * Don't set origin when we are breaking the association with a lower
470 	 * hard link.
471 	 */
472 	if (c->origin) {
473 		err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
474 		if (err)
475 			return err;
476 	}
477 
478 	if (c->metacopy) {
479 		err = ovl_check_setxattr(c->dentry, temp, OVL_XATTR_METACOPY,
480 					 NULL, 0, -EOPNOTSUPP);
481 		if (err)
482 			return err;
483 	}
484 
485 	inode_lock(temp->d_inode);
486 	if (c->metacopy)
487 		err = ovl_set_size(temp, &c->stat);
488 	if (!err)
489 		err = ovl_set_attr(temp, &c->stat);
490 	inode_unlock(temp->d_inode);
491 
492 	return err;
493 }
494 
495 struct ovl_cu_creds {
496 	const struct cred *old;
497 	struct cred *new;
498 };
499 
500 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
501 {
502 	int err;
503 
504 	cc->old = cc->new = NULL;
505 	err = security_inode_copy_up(dentry, &cc->new);
506 	if (err < 0)
507 		return err;
508 
509 	if (cc->new)
510 		cc->old = override_creds(cc->new);
511 
512 	return 0;
513 }
514 
515 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
516 {
517 	if (cc->new) {
518 		revert_creds(cc->old);
519 		put_cred(cc->new);
520 	}
521 }
522 
523 /*
524  * Copyup using workdir to prepare temp file.  Used when copying up directories,
525  * special files or when upper fs doesn't support O_TMPFILE.
526  */
527 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
528 {
529 	struct inode *inode;
530 	struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
531 	struct dentry *temp, *upper;
532 	struct ovl_cu_creds cc;
533 	int err;
534 	struct ovl_cattr cattr = {
535 		/* Can't properly set mode on creation because of the umask */
536 		.mode = c->stat.mode & S_IFMT,
537 		.rdev = c->stat.rdev,
538 		.link = c->link
539 	};
540 
541 	err = ovl_lock_rename_workdir(c->workdir, c->destdir);
542 	if (err)
543 		return err;
544 
545 	err = ovl_prep_cu_creds(c->dentry, &cc);
546 	if (err)
547 		goto unlock;
548 
549 	temp = ovl_create_temp(c->workdir, &cattr);
550 	ovl_revert_cu_creds(&cc);
551 
552 	err = PTR_ERR(temp);
553 	if (IS_ERR(temp))
554 		goto unlock;
555 
556 	err = ovl_copy_up_inode(c, temp);
557 	if (err)
558 		goto cleanup;
559 
560 	if (S_ISDIR(c->stat.mode) && c->indexed) {
561 		err = ovl_create_index(c->dentry, c->lowerpath.dentry, temp);
562 		if (err)
563 			goto cleanup;
564 	}
565 
566 	upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
567 	err = PTR_ERR(upper);
568 	if (IS_ERR(upper))
569 		goto cleanup;
570 
571 	err = ovl_do_rename(wdir, temp, udir, upper, 0);
572 	dput(upper);
573 	if (err)
574 		goto cleanup;
575 
576 	if (!c->metacopy)
577 		ovl_set_upperdata(d_inode(c->dentry));
578 	inode = d_inode(c->dentry);
579 	ovl_inode_update(inode, temp);
580 	if (S_ISDIR(inode->i_mode))
581 		ovl_set_flag(OVL_WHITEOUTS, inode);
582 unlock:
583 	unlock_rename(c->workdir, c->destdir);
584 
585 	return err;
586 
587 cleanup:
588 	ovl_cleanup(wdir, temp);
589 	dput(temp);
590 	goto unlock;
591 }
592 
593 /* Copyup using O_TMPFILE which does not require cross dir locking */
594 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
595 {
596 	struct inode *udir = d_inode(c->destdir);
597 	struct dentry *temp, *upper;
598 	struct ovl_cu_creds cc;
599 	int err;
600 
601 	err = ovl_prep_cu_creds(c->dentry, &cc);
602 	if (err)
603 		return err;
604 
605 	temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
606 	ovl_revert_cu_creds(&cc);
607 
608 	if (IS_ERR(temp))
609 		return PTR_ERR(temp);
610 
611 	err = ovl_copy_up_inode(c, temp);
612 	if (err)
613 		goto out_dput;
614 
615 	inode_lock_nested(udir, I_MUTEX_PARENT);
616 
617 	upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
618 	err = PTR_ERR(upper);
619 	if (!IS_ERR(upper)) {
620 		err = ovl_do_link(temp, udir, upper);
621 		dput(upper);
622 	}
623 	inode_unlock(udir);
624 
625 	if (err)
626 		goto out_dput;
627 
628 	if (!c->metacopy)
629 		ovl_set_upperdata(d_inode(c->dentry));
630 	ovl_inode_update(d_inode(c->dentry), temp);
631 
632 	return 0;
633 
634 out_dput:
635 	dput(temp);
636 	return err;
637 }
638 
639 /*
640  * Copy up a single dentry
641  *
642  * All renames start with copy up of source if necessary.  The actual
643  * rename will only proceed once the copy up was successful.  Copy up uses
644  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
645  * is possible that the copy up will lock the old parent.  At that point
646  * the file will have already been copied up anyway.
647  */
648 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
649 {
650 	int err;
651 	struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
652 	bool to_index = false;
653 
654 	/*
655 	 * Indexed non-dir is copied up directly to the index entry and then
656 	 * hardlinked to upper dir. Indexed dir is copied up to indexdir,
657 	 * then index entry is created and then copied up dir installed.
658 	 * Copying dir up to indexdir instead of workdir simplifies locking.
659 	 */
660 	if (ovl_need_index(c->dentry)) {
661 		c->indexed = true;
662 		if (S_ISDIR(c->stat.mode))
663 			c->workdir = ovl_indexdir(c->dentry->d_sb);
664 		else
665 			to_index = true;
666 	}
667 
668 	if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index)
669 		c->origin = true;
670 
671 	if (to_index) {
672 		c->destdir = ovl_indexdir(c->dentry->d_sb);
673 		err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
674 		if (err)
675 			return err;
676 	} else if (WARN_ON(!c->parent)) {
677 		/* Disconnected dentry must be copied up to index dir */
678 		return -EIO;
679 	} else {
680 		/*
681 		 * Mark parent "impure" because it may now contain non-pure
682 		 * upper
683 		 */
684 		err = ovl_set_impure(c->parent, c->destdir);
685 		if (err)
686 			return err;
687 	}
688 
689 	/* Should we copyup with O_TMPFILE or with workdir? */
690 	if (S_ISREG(c->stat.mode) && ofs->tmpfile)
691 		err = ovl_copy_up_tmpfile(c);
692 	else
693 		err = ovl_copy_up_workdir(c);
694 	if (err)
695 		goto out;
696 
697 	if (c->indexed)
698 		ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
699 
700 	if (to_index) {
701 		/* Initialize nlink for copy up of disconnected dentry */
702 		err = ovl_set_nlink_upper(c->dentry);
703 	} else {
704 		struct inode *udir = d_inode(c->destdir);
705 
706 		/* Restore timestamps on parent (best effort) */
707 		inode_lock(udir);
708 		ovl_set_timestamps(c->destdir, &c->pstat);
709 		inode_unlock(udir);
710 
711 		ovl_dentry_set_upper_alias(c->dentry);
712 	}
713 
714 out:
715 	if (to_index)
716 		kfree(c->destname.name);
717 	return err;
718 }
719 
720 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
721 				  int flags)
722 {
723 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
724 
725 	if (!ofs->config.metacopy)
726 		return false;
727 
728 	if (!S_ISREG(mode))
729 		return false;
730 
731 	if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
732 		return false;
733 
734 	return true;
735 }
736 
737 /* Copy up data of an inode which was copied up metadata only in the past. */
738 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
739 {
740 	struct path upperpath, datapath;
741 	int err;
742 	char *capability = NULL;
743 	ssize_t uninitialized_var(cap_size);
744 
745 	ovl_path_upper(c->dentry, &upperpath);
746 	if (WARN_ON(upperpath.dentry == NULL))
747 		return -EIO;
748 
749 	ovl_path_lowerdata(c->dentry, &datapath);
750 	if (WARN_ON(datapath.dentry == NULL))
751 		return -EIO;
752 
753 	if (c->stat.size) {
754 		err = cap_size = ovl_getxattr(upperpath.dentry, XATTR_NAME_CAPS,
755 					      &capability, 0);
756 		if (err < 0 && err != -ENODATA)
757 			goto out;
758 	}
759 
760 	err = ovl_copy_up_data(&datapath, &upperpath, c->stat.size);
761 	if (err)
762 		goto out_free;
763 
764 	/*
765 	 * Writing to upper file will clear security.capability xattr. We
766 	 * don't want that to happen for normal copy-up operation.
767 	 */
768 	if (capability) {
769 		err = ovl_do_setxattr(upperpath.dentry, XATTR_NAME_CAPS,
770 				      capability, cap_size, 0);
771 		if (err)
772 			goto out_free;
773 	}
774 
775 
776 	err = vfs_removexattr(upperpath.dentry, OVL_XATTR_METACOPY);
777 	if (err)
778 		goto out_free;
779 
780 	ovl_set_upperdata(d_inode(c->dentry));
781 out_free:
782 	kfree(capability);
783 out:
784 	return err;
785 }
786 
787 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
788 			   int flags)
789 {
790 	int err;
791 	DEFINE_DELAYED_CALL(done);
792 	struct path parentpath;
793 	struct ovl_copy_up_ctx ctx = {
794 		.parent = parent,
795 		.dentry = dentry,
796 		.workdir = ovl_workdir(dentry),
797 	};
798 
799 	if (WARN_ON(!ctx.workdir))
800 		return -EROFS;
801 
802 	ovl_path_lower(dentry, &ctx.lowerpath);
803 	err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
804 			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
805 	if (err)
806 		return err;
807 
808 	ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
809 
810 	if (parent) {
811 		ovl_path_upper(parent, &parentpath);
812 		ctx.destdir = parentpath.dentry;
813 		ctx.destname = dentry->d_name;
814 
815 		err = vfs_getattr(&parentpath, &ctx.pstat,
816 				  STATX_ATIME | STATX_MTIME,
817 				  AT_STATX_SYNC_AS_STAT);
818 		if (err)
819 			return err;
820 	}
821 
822 	/* maybe truncate regular file. this has no effect on dirs */
823 	if (flags & O_TRUNC)
824 		ctx.stat.size = 0;
825 
826 	if (S_ISLNK(ctx.stat.mode)) {
827 		ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
828 		if (IS_ERR(ctx.link))
829 			return PTR_ERR(ctx.link);
830 	}
831 
832 	err = ovl_copy_up_start(dentry, flags);
833 	/* err < 0: interrupted, err > 0: raced with another copy-up */
834 	if (unlikely(err)) {
835 		if (err > 0)
836 			err = 0;
837 	} else {
838 		if (!ovl_dentry_upper(dentry))
839 			err = ovl_do_copy_up(&ctx);
840 		if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
841 			err = ovl_link_up(&ctx);
842 		if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
843 			err = ovl_copy_up_meta_inode_data(&ctx);
844 		ovl_copy_up_end(dentry);
845 	}
846 	do_delayed_call(&done);
847 
848 	return err;
849 }
850 
851 int ovl_copy_up_flags(struct dentry *dentry, int flags)
852 {
853 	int err = 0;
854 	const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
855 	bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
856 
857 	/*
858 	 * With NFS export, copy up can get called for a disconnected non-dir.
859 	 * In this case, we will copy up lower inode to index dir without
860 	 * linking it to upper dir.
861 	 */
862 	if (WARN_ON(disconnected && d_is_dir(dentry)))
863 		return -EIO;
864 
865 	while (!err) {
866 		struct dentry *next;
867 		struct dentry *parent = NULL;
868 
869 		if (ovl_already_copied_up(dentry, flags))
870 			break;
871 
872 		next = dget(dentry);
873 		/* find the topmost dentry not yet copied up */
874 		for (; !disconnected;) {
875 			parent = dget_parent(next);
876 
877 			if (ovl_dentry_upper(parent))
878 				break;
879 
880 			dput(next);
881 			next = parent;
882 		}
883 
884 		err = ovl_copy_up_one(parent, next, flags);
885 
886 		dput(parent);
887 		dput(next);
888 	}
889 	revert_creds(old_cred);
890 
891 	return err;
892 }
893 
894 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
895 {
896 	/* Copy up of disconnected dentry does not set upper alias */
897 	if (ovl_already_copied_up(dentry, flags))
898 		return false;
899 
900 	if (special_file(d_inode(dentry)->i_mode))
901 		return false;
902 
903 	if (!ovl_open_flags_need_copy_up(flags))
904 		return false;
905 
906 	return true;
907 }
908 
909 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
910 {
911 	int err = 0;
912 
913 	if (ovl_open_need_copy_up(dentry, flags)) {
914 		err = ovl_want_write(dentry);
915 		if (!err) {
916 			err = ovl_copy_up_flags(dentry, flags);
917 			ovl_drop_write(dentry);
918 		}
919 	}
920 
921 	return err;
922 }
923 
924 int ovl_copy_up_with_data(struct dentry *dentry)
925 {
926 	return ovl_copy_up_flags(dentry, O_WRONLY);
927 }
928 
929 int ovl_copy_up(struct dentry *dentry)
930 {
931 	return ovl_copy_up_flags(dentry, 0);
932 }
933