xref: /openbmc/linux/fs/overlayfs/copy_up.c (revision 4e3299ea)
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/fileattr.h>
12 #include <linux/splice.h>
13 #include <linux/xattr.h>
14 #include <linux/security.h>
15 #include <linux/uaccess.h>
16 #include <linux/sched/signal.h>
17 #include <linux/cred.h>
18 #include <linux/namei.h>
19 #include <linux/fdtable.h>
20 #include <linux/ratelimit.h>
21 #include <linux/exportfs.h>
22 #include "overlayfs.h"
23 
24 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
25 
26 static int ovl_ccup_set(const char *buf, const struct kernel_param *param)
27 {
28 	pr_warn("\"check_copy_up\" module option is obsolete\n");
29 	return 0;
30 }
31 
32 static int ovl_ccup_get(char *buf, const struct kernel_param *param)
33 {
34 	return sprintf(buf, "N\n");
35 }
36 
37 module_param_call(check_copy_up, ovl_ccup_set, ovl_ccup_get, NULL, 0644);
38 MODULE_PARM_DESC(check_copy_up, "Obsolete; does nothing");
39 
40 static bool ovl_must_copy_xattr(const char *name)
41 {
42 	return !strcmp(name, XATTR_POSIX_ACL_ACCESS) ||
43 	       !strcmp(name, XATTR_POSIX_ACL_DEFAULT) ||
44 	       !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN);
45 }
46 
47 int ovl_copy_xattr(struct super_block *sb, struct path *oldpath, struct dentry *new)
48 {
49 	struct dentry *old = oldpath->dentry;
50 	ssize_t list_size, size, value_size = 0;
51 	char *buf, *name, *value = NULL;
52 	int error = 0;
53 	size_t slen;
54 
55 	if (!(old->d_inode->i_opflags & IOP_XATTR) ||
56 	    !(new->d_inode->i_opflags & IOP_XATTR))
57 		return 0;
58 
59 	list_size = vfs_listxattr(old, NULL, 0);
60 	if (list_size <= 0) {
61 		if (list_size == -EOPNOTSUPP)
62 			return 0;
63 		return list_size;
64 	}
65 
66 	buf = kvzalloc(list_size, GFP_KERNEL);
67 	if (!buf)
68 		return -ENOMEM;
69 
70 	list_size = vfs_listxattr(old, buf, list_size);
71 	if (list_size <= 0) {
72 		error = list_size;
73 		goto out;
74 	}
75 
76 	for (name = buf; list_size; name += slen) {
77 		slen = strnlen(name, list_size) + 1;
78 
79 		/* underlying fs providing us with an broken xattr list? */
80 		if (WARN_ON(slen > list_size)) {
81 			error = -EIO;
82 			break;
83 		}
84 		list_size -= slen;
85 
86 		if (ovl_is_private_xattr(sb, name))
87 			continue;
88 
89 		error = security_inode_copy_up_xattr(name);
90 		if (error < 0 && error != -EOPNOTSUPP)
91 			break;
92 		if (error == 1) {
93 			error = 0;
94 			continue; /* Discard */
95 		}
96 retry:
97 		size = ovl_do_getxattr(oldpath, name, value, value_size);
98 		if (size == -ERANGE)
99 			size = ovl_do_getxattr(oldpath, name, NULL, 0);
100 
101 		if (size < 0) {
102 			error = size;
103 			break;
104 		}
105 
106 		if (size > value_size) {
107 			void *new;
108 
109 			new = kvmalloc(size, GFP_KERNEL);
110 			if (!new) {
111 				error = -ENOMEM;
112 				break;
113 			}
114 			kvfree(value);
115 			value = new;
116 			value_size = size;
117 			goto retry;
118 		}
119 
120 		error = ovl_do_setxattr(OVL_FS(sb), new, name, value, size, 0);
121 		if (error) {
122 			if (error != -EOPNOTSUPP || ovl_must_copy_xattr(name))
123 				break;
124 
125 			/* Ignore failure to copy unknown xattrs */
126 			error = 0;
127 		}
128 	}
129 	kvfree(value);
130 out:
131 	kvfree(buf);
132 	return error;
133 }
134 
135 static int ovl_copy_fileattr(struct inode *inode, struct path *old,
136 			     struct path *new)
137 {
138 	struct fileattr oldfa = { .flags_valid = true };
139 	struct fileattr newfa = { .flags_valid = true };
140 	int err;
141 
142 	err = ovl_real_fileattr_get(old, &oldfa);
143 	if (err) {
144 		/* Ntfs-3g returns -EINVAL for "no fileattr support" */
145 		if (err == -ENOTTY || err == -EINVAL)
146 			return 0;
147 		pr_warn("failed to retrieve lower fileattr (%pd2, err=%i)\n",
148 			old->dentry, err);
149 		return err;
150 	}
151 
152 	/*
153 	 * We cannot set immutable and append-only flags on upper inode,
154 	 * because we would not be able to link upper inode to upper dir
155 	 * not set overlay private xattr on upper inode.
156 	 * Store these flags in overlay.protattr xattr instead.
157 	 */
158 	if (oldfa.flags & OVL_PROT_FS_FLAGS_MASK) {
159 		err = ovl_set_protattr(inode, new->dentry, &oldfa);
160 		if (err == -EPERM)
161 			pr_warn_once("copying fileattr: no xattr on upper\n");
162 		else if (err)
163 			return err;
164 	}
165 
166 	/* Don't bother copying flags if none are set */
167 	if (!(oldfa.flags & OVL_COPY_FS_FLAGS_MASK))
168 		return 0;
169 
170 	err = ovl_real_fileattr_get(new, &newfa);
171 	if (err) {
172 		/*
173 		 * Returning an error if upper doesn't support fileattr will
174 		 * result in a regression, so revert to the old behavior.
175 		 */
176 		if (err == -ENOTTY || err == -EINVAL) {
177 			pr_warn_once("copying fileattr: no support on upper\n");
178 			return 0;
179 		}
180 		pr_warn("failed to retrieve upper fileattr (%pd2, err=%i)\n",
181 			new->dentry, err);
182 		return err;
183 	}
184 
185 	BUILD_BUG_ON(OVL_COPY_FS_FLAGS_MASK & ~FS_COMMON_FL);
186 	newfa.flags &= ~OVL_COPY_FS_FLAGS_MASK;
187 	newfa.flags |= (oldfa.flags & OVL_COPY_FS_FLAGS_MASK);
188 
189 	BUILD_BUG_ON(OVL_COPY_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
190 	newfa.fsx_xflags &= ~OVL_COPY_FSX_FLAGS_MASK;
191 	newfa.fsx_xflags |= (oldfa.fsx_xflags & OVL_COPY_FSX_FLAGS_MASK);
192 
193 	return ovl_real_fileattr_set(new, &newfa);
194 }
195 
196 static int ovl_copy_up_data(struct ovl_fs *ofs, struct path *old,
197 			    struct path *new, loff_t len)
198 {
199 	struct file *old_file;
200 	struct file *new_file;
201 	loff_t old_pos = 0;
202 	loff_t new_pos = 0;
203 	loff_t cloned;
204 	loff_t data_pos = -1;
205 	loff_t hole_len;
206 	bool skip_hole = false;
207 	int error = 0;
208 
209 	if (len == 0)
210 		return 0;
211 
212 	old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
213 	if (IS_ERR(old_file))
214 		return PTR_ERR(old_file);
215 
216 	new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
217 	if (IS_ERR(new_file)) {
218 		error = PTR_ERR(new_file);
219 		goto out_fput;
220 	}
221 
222 	/* Try to use clone_file_range to clone up within the same fs */
223 	cloned = do_clone_file_range(old_file, 0, new_file, 0, len, 0);
224 	if (cloned == len)
225 		goto out;
226 	/* Couldn't clone, so now we try to copy the data */
227 
228 	/* Check if lower fs supports seek operation */
229 	if (old_file->f_mode & FMODE_LSEEK)
230 		skip_hole = true;
231 
232 	while (len) {
233 		size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
234 		long bytes;
235 
236 		if (len < this_len)
237 			this_len = len;
238 
239 		if (signal_pending_state(TASK_KILLABLE, current)) {
240 			error = -EINTR;
241 			break;
242 		}
243 
244 		/*
245 		 * Fill zero for hole will cost unnecessary disk space
246 		 * and meanwhile slow down the copy-up speed, so we do
247 		 * an optimization for hole during copy-up, it relies
248 		 * on SEEK_DATA implementation in lower fs so if lower
249 		 * fs does not support it, copy-up will behave as before.
250 		 *
251 		 * Detail logic of hole detection as below:
252 		 * When we detect next data position is larger than current
253 		 * position we will skip that hole, otherwise we copy
254 		 * data in the size of OVL_COPY_UP_CHUNK_SIZE. Actually,
255 		 * it may not recognize all kind of holes and sometimes
256 		 * only skips partial of hole area. However, it will be
257 		 * enough for most of the use cases.
258 		 */
259 
260 		if (skip_hole && data_pos < old_pos) {
261 			data_pos = vfs_llseek(old_file, old_pos, SEEK_DATA);
262 			if (data_pos > old_pos) {
263 				hole_len = data_pos - old_pos;
264 				len -= hole_len;
265 				old_pos = new_pos = data_pos;
266 				continue;
267 			} else if (data_pos == -ENXIO) {
268 				break;
269 			} else if (data_pos < 0) {
270 				skip_hole = false;
271 			}
272 		}
273 
274 		bytes = do_splice_direct(old_file, &old_pos,
275 					 new_file, &new_pos,
276 					 this_len, SPLICE_F_MOVE);
277 		if (bytes <= 0) {
278 			error = bytes;
279 			break;
280 		}
281 		WARN_ON(old_pos != new_pos);
282 
283 		len -= bytes;
284 	}
285 out:
286 	if (!error && ovl_should_sync(ofs))
287 		error = vfs_fsync(new_file, 0);
288 	fput(new_file);
289 out_fput:
290 	fput(old_file);
291 	return error;
292 }
293 
294 static int ovl_set_size(struct ovl_fs *ofs,
295 			struct dentry *upperdentry, struct kstat *stat)
296 {
297 	struct iattr attr = {
298 		.ia_valid = ATTR_SIZE,
299 		.ia_size = stat->size,
300 	};
301 
302 	return ovl_do_notify_change(ofs, upperdentry, &attr);
303 }
304 
305 static int ovl_set_timestamps(struct ovl_fs *ofs, struct dentry *upperdentry,
306 			      struct kstat *stat)
307 {
308 	struct iattr attr = {
309 		.ia_valid =
310 		     ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
311 		.ia_atime = stat->atime,
312 		.ia_mtime = stat->mtime,
313 	};
314 
315 	return ovl_do_notify_change(ofs, upperdentry, &attr);
316 }
317 
318 int ovl_set_attr(struct ovl_fs *ofs, struct dentry *upperdentry,
319 		 struct kstat *stat)
320 {
321 	int err = 0;
322 
323 	if (!S_ISLNK(stat->mode)) {
324 		struct iattr attr = {
325 			.ia_valid = ATTR_MODE,
326 			.ia_mode = stat->mode,
327 		};
328 		err = ovl_do_notify_change(ofs, upperdentry, &attr);
329 	}
330 	if (!err) {
331 		struct iattr attr = {
332 			.ia_valid = ATTR_UID | ATTR_GID,
333 			.ia_uid = stat->uid,
334 			.ia_gid = stat->gid,
335 		};
336 		err = ovl_do_notify_change(ofs, upperdentry, &attr);
337 	}
338 	if (!err)
339 		ovl_set_timestamps(ofs, upperdentry, stat);
340 
341 	return err;
342 }
343 
344 struct ovl_fh *ovl_encode_real_fh(struct ovl_fs *ofs, struct dentry *real,
345 				  bool is_upper)
346 {
347 	struct ovl_fh *fh;
348 	int fh_type, dwords;
349 	int buflen = MAX_HANDLE_SZ;
350 	uuid_t *uuid = &real->d_sb->s_uuid;
351 	int err;
352 
353 	/* Make sure the real fid stays 32bit aligned */
354 	BUILD_BUG_ON(OVL_FH_FID_OFFSET % 4);
355 	BUILD_BUG_ON(MAX_HANDLE_SZ + OVL_FH_FID_OFFSET > 255);
356 
357 	fh = kzalloc(buflen + OVL_FH_FID_OFFSET, GFP_KERNEL);
358 	if (!fh)
359 		return ERR_PTR(-ENOMEM);
360 
361 	/*
362 	 * We encode a non-connectable file handle for non-dir, because we
363 	 * only need to find the lower inode number and we don't want to pay
364 	 * the price or reconnecting the dentry.
365 	 */
366 	dwords = buflen >> 2;
367 	fh_type = exportfs_encode_fh(real, (void *)fh->fb.fid, &dwords, 0);
368 	buflen = (dwords << 2);
369 
370 	err = -EIO;
371 	if (WARN_ON(fh_type < 0) ||
372 	    WARN_ON(buflen > MAX_HANDLE_SZ) ||
373 	    WARN_ON(fh_type == FILEID_INVALID))
374 		goto out_err;
375 
376 	fh->fb.version = OVL_FH_VERSION;
377 	fh->fb.magic = OVL_FH_MAGIC;
378 	fh->fb.type = fh_type;
379 	fh->fb.flags = OVL_FH_FLAG_CPU_ENDIAN;
380 	/*
381 	 * When we will want to decode an overlay dentry from this handle
382 	 * and all layers are on the same fs, if we get a disconncted real
383 	 * dentry when we decode fid, the only way to tell if we should assign
384 	 * it to upperdentry or to lowerstack is by checking this flag.
385 	 */
386 	if (is_upper)
387 		fh->fb.flags |= OVL_FH_FLAG_PATH_UPPER;
388 	fh->fb.len = sizeof(fh->fb) + buflen;
389 	if (ofs->config.uuid)
390 		fh->fb.uuid = *uuid;
391 
392 	return fh;
393 
394 out_err:
395 	kfree(fh);
396 	return ERR_PTR(err);
397 }
398 
399 int ovl_set_origin(struct ovl_fs *ofs, struct dentry *lower,
400 		   struct dentry *upper)
401 {
402 	const struct ovl_fh *fh = NULL;
403 	int err;
404 
405 	/*
406 	 * When lower layer doesn't support export operations store a 'null' fh,
407 	 * so we can use the overlay.origin xattr to distignuish between a copy
408 	 * up and a pure upper inode.
409 	 */
410 	if (ovl_can_decode_fh(lower->d_sb)) {
411 		fh = ovl_encode_real_fh(ofs, lower, false);
412 		if (IS_ERR(fh))
413 			return PTR_ERR(fh);
414 	}
415 
416 	/*
417 	 * Do not fail when upper doesn't support xattrs.
418 	 */
419 	err = ovl_check_setxattr(ofs, upper, OVL_XATTR_ORIGIN, fh->buf,
420 				 fh ? fh->fb.len : 0, 0);
421 	kfree(fh);
422 
423 	/* Ignore -EPERM from setting "user.*" on symlink/special */
424 	return err == -EPERM ? 0 : err;
425 }
426 
427 /* Store file handle of @upper dir in @index dir entry */
428 static int ovl_set_upper_fh(struct ovl_fs *ofs, struct dentry *upper,
429 			    struct dentry *index)
430 {
431 	const struct ovl_fh *fh;
432 	int err;
433 
434 	fh = ovl_encode_real_fh(ofs, upper, true);
435 	if (IS_ERR(fh))
436 		return PTR_ERR(fh);
437 
438 	err = ovl_setxattr(ofs, index, OVL_XATTR_UPPER, fh->buf, fh->fb.len);
439 
440 	kfree(fh);
441 	return err;
442 }
443 
444 /*
445  * Create and install index entry.
446  *
447  * Caller must hold i_mutex on indexdir.
448  */
449 static int ovl_create_index(struct dentry *dentry, struct dentry *origin,
450 			    struct dentry *upper)
451 {
452 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
453 	struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
454 	struct inode *dir = d_inode(indexdir);
455 	struct dentry *index = NULL;
456 	struct dentry *temp = NULL;
457 	struct qstr name = { };
458 	int err;
459 
460 	/*
461 	 * For now this is only used for creating index entry for directories,
462 	 * because non-dir are copied up directly to index and then hardlinked
463 	 * to upper dir.
464 	 *
465 	 * TODO: implement create index for non-dir, so we can call it when
466 	 * encoding file handle for non-dir in case index does not exist.
467 	 */
468 	if (WARN_ON(!d_is_dir(dentry)))
469 		return -EIO;
470 
471 	/* Directory not expected to be indexed before copy up */
472 	if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
473 		return -EIO;
474 
475 	err = ovl_get_index_name(ofs, origin, &name);
476 	if (err)
477 		return err;
478 
479 	temp = ovl_create_temp(ofs, indexdir, OVL_CATTR(S_IFDIR | 0));
480 	err = PTR_ERR(temp);
481 	if (IS_ERR(temp))
482 		goto free_name;
483 
484 	err = ovl_set_upper_fh(ofs, upper, temp);
485 	if (err)
486 		goto out;
487 
488 	index = ovl_lookup_upper(ofs, name.name, indexdir, name.len);
489 	if (IS_ERR(index)) {
490 		err = PTR_ERR(index);
491 	} else {
492 		err = ovl_do_rename(ofs, dir, temp, dir, index, 0);
493 		dput(index);
494 	}
495 out:
496 	if (err)
497 		ovl_cleanup(ofs, dir, temp);
498 	dput(temp);
499 free_name:
500 	kfree(name.name);
501 	return err;
502 }
503 
504 struct ovl_copy_up_ctx {
505 	struct dentry *parent;
506 	struct dentry *dentry;
507 	struct path lowerpath;
508 	struct kstat stat;
509 	struct kstat pstat;
510 	const char *link;
511 	struct dentry *destdir;
512 	struct qstr destname;
513 	struct dentry *workdir;
514 	bool origin;
515 	bool indexed;
516 	bool metacopy;
517 };
518 
519 static int ovl_link_up(struct ovl_copy_up_ctx *c)
520 {
521 	int err;
522 	struct dentry *upper;
523 	struct dentry *upperdir = ovl_dentry_upper(c->parent);
524 	struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
525 	struct inode *udir = d_inode(upperdir);
526 
527 	/* Mark parent "impure" because it may now contain non-pure upper */
528 	err = ovl_set_impure(c->parent, upperdir);
529 	if (err)
530 		return err;
531 
532 	err = ovl_set_nlink_lower(c->dentry);
533 	if (err)
534 		return err;
535 
536 	inode_lock_nested(udir, I_MUTEX_PARENT);
537 	upper = ovl_lookup_upper(ofs, c->dentry->d_name.name, upperdir,
538 				 c->dentry->d_name.len);
539 	err = PTR_ERR(upper);
540 	if (!IS_ERR(upper)) {
541 		err = ovl_do_link(ofs, ovl_dentry_upper(c->dentry), udir, upper);
542 		dput(upper);
543 
544 		if (!err) {
545 			/* Restore timestamps on parent (best effort) */
546 			ovl_set_timestamps(ofs, upperdir, &c->pstat);
547 			ovl_dentry_set_upper_alias(c->dentry);
548 		}
549 	}
550 	inode_unlock(udir);
551 	if (err)
552 		return err;
553 
554 	err = ovl_set_nlink_upper(c->dentry);
555 
556 	return err;
557 }
558 
559 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
560 {
561 	struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
562 	struct inode *inode = d_inode(c->dentry);
563 	struct path upperpath, datapath;
564 	int err;
565 
566 	ovl_path_upper(c->dentry, &upperpath);
567 	if (WARN_ON(upperpath.dentry != NULL))
568 		return -EIO;
569 
570 	upperpath.dentry = temp;
571 
572 	/*
573 	 * Copy up data first and then xattrs. Writing data after
574 	 * xattrs will remove security.capability xattr automatically.
575 	 */
576 	if (S_ISREG(c->stat.mode) && !c->metacopy) {
577 		ovl_path_lowerdata(c->dentry, &datapath);
578 		err = ovl_copy_up_data(ofs, &datapath, &upperpath,
579 				       c->stat.size);
580 		if (err)
581 			return err;
582 	}
583 
584 	err = ovl_copy_xattr(c->dentry->d_sb, &c->lowerpath, temp);
585 	if (err)
586 		return err;
587 
588 	if (inode->i_flags & OVL_COPY_I_FLAGS_MASK) {
589 		/*
590 		 * Copy the fileattr inode flags that are the source of already
591 		 * copied i_flags
592 		 */
593 		err = ovl_copy_fileattr(inode, &c->lowerpath, &upperpath);
594 		if (err)
595 			return err;
596 	}
597 
598 	/*
599 	 * Store identifier of lower inode in upper inode xattr to
600 	 * allow lookup of the copy up origin inode.
601 	 *
602 	 * Don't set origin when we are breaking the association with a lower
603 	 * hard link.
604 	 */
605 	if (c->origin) {
606 		err = ovl_set_origin(ofs, c->lowerpath.dentry, temp);
607 		if (err)
608 			return err;
609 	}
610 
611 	if (c->metacopy) {
612 		err = ovl_check_setxattr(ofs, temp, OVL_XATTR_METACOPY,
613 					 NULL, 0, -EOPNOTSUPP);
614 		if (err)
615 			return err;
616 	}
617 
618 	inode_lock(temp->d_inode);
619 	if (S_ISREG(c->stat.mode))
620 		err = ovl_set_size(ofs, temp, &c->stat);
621 	if (!err)
622 		err = ovl_set_attr(ofs, temp, &c->stat);
623 	inode_unlock(temp->d_inode);
624 
625 	return err;
626 }
627 
628 struct ovl_cu_creds {
629 	const struct cred *old;
630 	struct cred *new;
631 };
632 
633 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
634 {
635 	int err;
636 
637 	cc->old = cc->new = NULL;
638 	err = security_inode_copy_up(dentry, &cc->new);
639 	if (err < 0)
640 		return err;
641 
642 	if (cc->new)
643 		cc->old = override_creds(cc->new);
644 
645 	return 0;
646 }
647 
648 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
649 {
650 	if (cc->new) {
651 		revert_creds(cc->old);
652 		put_cred(cc->new);
653 	}
654 }
655 
656 /*
657  * Copyup using workdir to prepare temp file.  Used when copying up directories,
658  * special files or when upper fs doesn't support O_TMPFILE.
659  */
660 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
661 {
662 	struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
663 	struct inode *inode;
664 	struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
665 	struct dentry *temp, *upper;
666 	struct ovl_cu_creds cc;
667 	int err;
668 	struct ovl_cattr cattr = {
669 		/* Can't properly set mode on creation because of the umask */
670 		.mode = c->stat.mode & S_IFMT,
671 		.rdev = c->stat.rdev,
672 		.link = c->link
673 	};
674 
675 	/* workdir and destdir could be the same when copying up to indexdir */
676 	err = -EIO;
677 	if (lock_rename(c->workdir, c->destdir) != NULL)
678 		goto unlock;
679 
680 	err = ovl_prep_cu_creds(c->dentry, &cc);
681 	if (err)
682 		goto unlock;
683 
684 	temp = ovl_create_temp(ofs, c->workdir, &cattr);
685 	ovl_revert_cu_creds(&cc);
686 
687 	err = PTR_ERR(temp);
688 	if (IS_ERR(temp))
689 		goto unlock;
690 
691 	err = ovl_copy_up_inode(c, temp);
692 	if (err)
693 		goto cleanup;
694 
695 	if (S_ISDIR(c->stat.mode) && c->indexed) {
696 		err = ovl_create_index(c->dentry, c->lowerpath.dentry, temp);
697 		if (err)
698 			goto cleanup;
699 	}
700 
701 	upper = ovl_lookup_upper(ofs, c->destname.name, c->destdir,
702 				 c->destname.len);
703 	err = PTR_ERR(upper);
704 	if (IS_ERR(upper))
705 		goto cleanup;
706 
707 	err = ovl_do_rename(ofs, wdir, temp, udir, upper, 0);
708 	dput(upper);
709 	if (err)
710 		goto cleanup;
711 
712 	if (!c->metacopy)
713 		ovl_set_upperdata(d_inode(c->dentry));
714 	inode = d_inode(c->dentry);
715 	ovl_inode_update(inode, temp);
716 	if (S_ISDIR(inode->i_mode))
717 		ovl_set_flag(OVL_WHITEOUTS, inode);
718 unlock:
719 	unlock_rename(c->workdir, c->destdir);
720 
721 	return err;
722 
723 cleanup:
724 	ovl_cleanup(ofs, wdir, temp);
725 	dput(temp);
726 	goto unlock;
727 }
728 
729 /* Copyup using O_TMPFILE which does not require cross dir locking */
730 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
731 {
732 	struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
733 	struct inode *udir = d_inode(c->destdir);
734 	struct dentry *temp, *upper;
735 	struct ovl_cu_creds cc;
736 	int err;
737 
738 	err = ovl_prep_cu_creds(c->dentry, &cc);
739 	if (err)
740 		return err;
741 
742 	temp = ovl_do_tmpfile(ofs, c->workdir, c->stat.mode);
743 	ovl_revert_cu_creds(&cc);
744 
745 	if (IS_ERR(temp))
746 		return PTR_ERR(temp);
747 
748 	err = ovl_copy_up_inode(c, temp);
749 	if (err)
750 		goto out_dput;
751 
752 	inode_lock_nested(udir, I_MUTEX_PARENT);
753 
754 	upper = ovl_lookup_upper(ofs, c->destname.name, c->destdir,
755 				 c->destname.len);
756 	err = PTR_ERR(upper);
757 	if (!IS_ERR(upper)) {
758 		err = ovl_do_link(ofs, temp, udir, upper);
759 		dput(upper);
760 	}
761 	inode_unlock(udir);
762 
763 	if (err)
764 		goto out_dput;
765 
766 	if (!c->metacopy)
767 		ovl_set_upperdata(d_inode(c->dentry));
768 	ovl_inode_update(d_inode(c->dentry), temp);
769 
770 	return 0;
771 
772 out_dput:
773 	dput(temp);
774 	return err;
775 }
776 
777 /*
778  * Copy up a single dentry
779  *
780  * All renames start with copy up of source if necessary.  The actual
781  * rename will only proceed once the copy up was successful.  Copy up uses
782  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
783  * is possible that the copy up will lock the old parent.  At that point
784  * the file will have already been copied up anyway.
785  */
786 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
787 {
788 	int err;
789 	struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
790 	bool to_index = false;
791 
792 	/*
793 	 * Indexed non-dir is copied up directly to the index entry and then
794 	 * hardlinked to upper dir. Indexed dir is copied up to indexdir,
795 	 * then index entry is created and then copied up dir installed.
796 	 * Copying dir up to indexdir instead of workdir simplifies locking.
797 	 */
798 	if (ovl_need_index(c->dentry)) {
799 		c->indexed = true;
800 		if (S_ISDIR(c->stat.mode))
801 			c->workdir = ovl_indexdir(c->dentry->d_sb);
802 		else
803 			to_index = true;
804 	}
805 
806 	if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index)
807 		c->origin = true;
808 
809 	if (to_index) {
810 		c->destdir = ovl_indexdir(c->dentry->d_sb);
811 		err = ovl_get_index_name(ofs, c->lowerpath.dentry, &c->destname);
812 		if (err)
813 			return err;
814 	} else if (WARN_ON(!c->parent)) {
815 		/* Disconnected dentry must be copied up to index dir */
816 		return -EIO;
817 	} else {
818 		/*
819 		 * Mark parent "impure" because it may now contain non-pure
820 		 * upper
821 		 */
822 		err = ovl_set_impure(c->parent, c->destdir);
823 		if (err)
824 			return err;
825 	}
826 
827 	/* Should we copyup with O_TMPFILE or with workdir? */
828 	if (S_ISREG(c->stat.mode) && ofs->tmpfile)
829 		err = ovl_copy_up_tmpfile(c);
830 	else
831 		err = ovl_copy_up_workdir(c);
832 	if (err)
833 		goto out;
834 
835 	if (c->indexed)
836 		ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
837 
838 	if (to_index) {
839 		/* Initialize nlink for copy up of disconnected dentry */
840 		err = ovl_set_nlink_upper(c->dentry);
841 	} else {
842 		struct inode *udir = d_inode(c->destdir);
843 
844 		/* Restore timestamps on parent (best effort) */
845 		inode_lock(udir);
846 		ovl_set_timestamps(ofs, c->destdir, &c->pstat);
847 		inode_unlock(udir);
848 
849 		ovl_dentry_set_upper_alias(c->dentry);
850 	}
851 
852 out:
853 	if (to_index)
854 		kfree(c->destname.name);
855 	return err;
856 }
857 
858 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
859 				  int flags)
860 {
861 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
862 
863 	if (!ofs->config.metacopy)
864 		return false;
865 
866 	if (!S_ISREG(mode))
867 		return false;
868 
869 	if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
870 		return false;
871 
872 	return true;
873 }
874 
875 static ssize_t ovl_getxattr_value(struct path *path, char *name, char **value)
876 {
877 	ssize_t res;
878 	char *buf;
879 
880 	res = ovl_do_getxattr(path, name, NULL, 0);
881 	if (res == -ENODATA || res == -EOPNOTSUPP)
882 		res = 0;
883 
884 	if (res > 0) {
885 		buf = kzalloc(res, GFP_KERNEL);
886 		if (!buf)
887 			return -ENOMEM;
888 
889 		res = ovl_do_getxattr(path, name, buf, res);
890 		if (res < 0)
891 			kfree(buf);
892 		else
893 			*value = buf;
894 	}
895 	return res;
896 }
897 
898 /* Copy up data of an inode which was copied up metadata only in the past. */
899 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
900 {
901 	struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
902 	struct path upperpath, datapath;
903 	int err;
904 	char *capability = NULL;
905 	ssize_t cap_size;
906 
907 	ovl_path_upper(c->dentry, &upperpath);
908 	if (WARN_ON(upperpath.dentry == NULL))
909 		return -EIO;
910 
911 	ovl_path_lowerdata(c->dentry, &datapath);
912 	if (WARN_ON(datapath.dentry == NULL))
913 		return -EIO;
914 
915 	if (c->stat.size) {
916 		err = cap_size = ovl_getxattr_value(&upperpath, XATTR_NAME_CAPS,
917 						    &capability);
918 		if (cap_size < 0)
919 			goto out;
920 	}
921 
922 	err = ovl_copy_up_data(ofs, &datapath, &upperpath, c->stat.size);
923 	if (err)
924 		goto out_free;
925 
926 	/*
927 	 * Writing to upper file will clear security.capability xattr. We
928 	 * don't want that to happen for normal copy-up operation.
929 	 */
930 	if (capability) {
931 		err = ovl_do_setxattr(ofs, upperpath.dentry, XATTR_NAME_CAPS,
932 				      capability, cap_size, 0);
933 		if (err)
934 			goto out_free;
935 	}
936 
937 
938 	err = ovl_removexattr(ofs, upperpath.dentry, OVL_XATTR_METACOPY);
939 	if (err)
940 		goto out_free;
941 
942 	ovl_set_upperdata(d_inode(c->dentry));
943 out_free:
944 	kfree(capability);
945 out:
946 	return err;
947 }
948 
949 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
950 			   int flags)
951 {
952 	int err;
953 	DEFINE_DELAYED_CALL(done);
954 	struct path parentpath;
955 	struct ovl_copy_up_ctx ctx = {
956 		.parent = parent,
957 		.dentry = dentry,
958 		.workdir = ovl_workdir(dentry),
959 	};
960 
961 	if (WARN_ON(!ctx.workdir))
962 		return -EROFS;
963 
964 	ovl_path_lower(dentry, &ctx.lowerpath);
965 	err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
966 			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
967 	if (err)
968 		return err;
969 
970 	ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
971 
972 	if (parent) {
973 		ovl_path_upper(parent, &parentpath);
974 		ctx.destdir = parentpath.dentry;
975 		ctx.destname = dentry->d_name;
976 
977 		err = vfs_getattr(&parentpath, &ctx.pstat,
978 				  STATX_ATIME | STATX_MTIME,
979 				  AT_STATX_SYNC_AS_STAT);
980 		if (err)
981 			return err;
982 	}
983 
984 	/* maybe truncate regular file. this has no effect on dirs */
985 	if (flags & O_TRUNC)
986 		ctx.stat.size = 0;
987 
988 	if (S_ISLNK(ctx.stat.mode)) {
989 		ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
990 		if (IS_ERR(ctx.link))
991 			return PTR_ERR(ctx.link);
992 	}
993 
994 	err = ovl_copy_up_start(dentry, flags);
995 	/* err < 0: interrupted, err > 0: raced with another copy-up */
996 	if (unlikely(err)) {
997 		if (err > 0)
998 			err = 0;
999 	} else {
1000 		if (!ovl_dentry_upper(dentry))
1001 			err = ovl_do_copy_up(&ctx);
1002 		if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
1003 			err = ovl_link_up(&ctx);
1004 		if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
1005 			err = ovl_copy_up_meta_inode_data(&ctx);
1006 		ovl_copy_up_end(dentry);
1007 	}
1008 	do_delayed_call(&done);
1009 
1010 	return err;
1011 }
1012 
1013 static int ovl_copy_up_flags(struct dentry *dentry, int flags)
1014 {
1015 	int err = 0;
1016 	const struct cred *old_cred;
1017 	bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
1018 
1019 	/*
1020 	 * With NFS export, copy up can get called for a disconnected non-dir.
1021 	 * In this case, we will copy up lower inode to index dir without
1022 	 * linking it to upper dir.
1023 	 */
1024 	if (WARN_ON(disconnected && d_is_dir(dentry)))
1025 		return -EIO;
1026 
1027 	old_cred = ovl_override_creds(dentry->d_sb);
1028 	while (!err) {
1029 		struct dentry *next;
1030 		struct dentry *parent = NULL;
1031 
1032 		if (ovl_already_copied_up(dentry, flags))
1033 			break;
1034 
1035 		next = dget(dentry);
1036 		/* find the topmost dentry not yet copied up */
1037 		for (; !disconnected;) {
1038 			parent = dget_parent(next);
1039 
1040 			if (ovl_dentry_upper(parent))
1041 				break;
1042 
1043 			dput(next);
1044 			next = parent;
1045 		}
1046 
1047 		err = ovl_copy_up_one(parent, next, flags);
1048 
1049 		dput(parent);
1050 		dput(next);
1051 	}
1052 	revert_creds(old_cred);
1053 
1054 	return err;
1055 }
1056 
1057 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
1058 {
1059 	/* Copy up of disconnected dentry does not set upper alias */
1060 	if (ovl_already_copied_up(dentry, flags))
1061 		return false;
1062 
1063 	if (special_file(d_inode(dentry)->i_mode))
1064 		return false;
1065 
1066 	if (!ovl_open_flags_need_copy_up(flags))
1067 		return false;
1068 
1069 	return true;
1070 }
1071 
1072 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
1073 {
1074 	int err = 0;
1075 
1076 	if (ovl_open_need_copy_up(dentry, flags)) {
1077 		err = ovl_want_write(dentry);
1078 		if (!err) {
1079 			err = ovl_copy_up_flags(dentry, flags);
1080 			ovl_drop_write(dentry);
1081 		}
1082 	}
1083 
1084 	return err;
1085 }
1086 
1087 int ovl_copy_up_with_data(struct dentry *dentry)
1088 {
1089 	return ovl_copy_up_flags(dentry, O_WRONLY);
1090 }
1091 
1092 int ovl_copy_up(struct dentry *dentry)
1093 {
1094 	return ovl_copy_up_flags(dentry, 0);
1095 }
1096