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28 pr_warn("\"check_copy_up\" module option is obsolete\n");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 static int ovl_copy_acl(struct ovl_fs *ofs, const struct path *path,
48 struct dentry *dentry, const char *acl_name)
49 {
50 int err;
51 struct posix_acl *clone, *real_acl = NULL;
52
53 real_acl = ovl_get_acl_path(path, acl_name, false);
54 if (!real_acl)
55 return 0;
56
57 if (IS_ERR(real_acl)) {
58 err = PTR_ERR(real_acl);
59 if (err == -ENODATA || err == -EOPNOTSUPP)
60 return 0;
61 return err;
62 }
63
64 clone = posix_acl_clone(real_acl, GFP_KERNEL);
65 posix_acl_release(real_acl); /* release original acl */
66 if (!clone)
67 return -ENOMEM;
68
69 err = ovl_do_set_acl(ofs, dentry, acl_name, clone);
70
71 /* release cloned acl */
72 posix_acl_release(clone);
73 return err;
74 }
75
76 int ovl_copy_xattr(struct super_block *sb, const struct path *oldpath, struct dentry *new)
77 {
78 struct dentry *old = oldpath->dentry;
79 ssize_t list_size, size, value_size = 0;
80 char *buf, *name, *value = NULL;
81 int error = 0;
82 size_t slen;
83
84 if (!old->d_inode->i_op->listxattr || !new->d_inode->i_op->listxattr)
85 return 0;
86
87 list_size = vfs_listxattr(old, NULL, 0);
88 if (list_size <= 0) {
89 if (list_size == -EOPNOTSUPP)
90 return 0;
91 return list_size;
92 }
93
94 buf = kvzalloc(list_size, GFP_KERNEL);
95 if (!buf)
96 return -ENOMEM;
97
98 list_size = vfs_listxattr(old, buf, list_size);
99 if (list_size <= 0) {
100 error = list_size;
101 goto out;
102 }
103
104 for (name = buf; list_size; name += slen) {
105 slen = strnlen(name, list_size) + 1;
106
107 /* underlying fs providing us with an broken xattr list? */
108 if (WARN_ON(slen > list_size)) {
109 error = -EIO;
110 break;
111 }
112 list_size -= slen;
113
114 if (ovl_is_private_xattr(sb, name))
115 continue;
116
117 error = security_inode_copy_up_xattr(old, name);
118 if (error < 0 && error != -EOPNOTSUPP)
119 break;
120 if (error == 1) {
121 error = 0;
122 continue; /* Discard */
123 }
124
125 if (is_posix_acl_xattr(name)) {
126 error = ovl_copy_acl(OVL_FS(sb), oldpath, new, name);
127 if (!error)
128 continue;
129 /* POSIX ACLs must be copied. */
130 break;
131 }
132
133 retry:
134 size = ovl_do_getxattr(oldpath, name, value, value_size);
135 if (size == -ERANGE)
136 size = ovl_do_getxattr(oldpath, name, NULL, 0);
137
138 if (size < 0) {
139 error = size;
140 break;
141 }
142
143 if (size > value_size) {
144 void *new;
145
146 new = kvmalloc(size, GFP_KERNEL);
147 if (!new) {
148 error = -ENOMEM;
149 break;
150 }
151 kvfree(value);
152 value = new;
153 value_size = size;
154 goto retry;
155 }
156
157 error = ovl_do_setxattr(OVL_FS(sb), new, name, value, size, 0);
158 if (error) {
159 if (error != -EOPNOTSUPP || ovl_must_copy_xattr(name))
160 break;
161
162 /* Ignore failure to copy unknown xattrs */
163 error = 0;
164 }
165 }
166 kvfree(value);
167 out:
168 kvfree(buf);
169 return error;
170 }
171
172 static int ovl_copy_fileattr(struct inode *inode, const struct path *old,
173 const struct path *new)
174 {
175 struct fileattr oldfa = { .flags_valid = true };
176 struct fileattr newfa = { .flags_valid = true };
177 int err;
178
179 err = ovl_real_fileattr_get(old, &oldfa);
180 if (err) {
181 /* Ntfs-3g returns -EINVAL for "no fileattr support" */
182 if (err == -ENOTTY || err == -EINVAL)
183 return 0;
184 pr_warn("failed to retrieve lower fileattr (%pd2, err=%i)\n",
185 old->dentry, err);
186 return err;
187 }
188
189 /*
190 * We cannot set immutable and append-only flags on upper inode,
191 * because we would not be able to link upper inode to upper dir
192 * not set overlay private xattr on upper inode.
193 * Store these flags in overlay.protattr xattr instead.
194 */
195 if (oldfa.flags & OVL_PROT_FS_FLAGS_MASK) {
196 err = ovl_set_protattr(inode, new->dentry, &oldfa);
197 if (err == -EPERM)
198 pr_warn_once("copying fileattr: no xattr on upper\n");
199 else if (err)
200 return err;
201 }
202
203 /* Don't bother copying flags if none are set */
204 if (!(oldfa.flags & OVL_COPY_FS_FLAGS_MASK))
205 return 0;
206
207 err = ovl_real_fileattr_get(new, &newfa);
208 if (err) {
209 /*
210 * Returning an error if upper doesn't support fileattr will
211 * result in a regression, so revert to the old behavior.
212 */
213 if (err == -ENOTTY || err == -EINVAL) {
214 pr_warn_once("copying fileattr: no support on upper\n");
215 return 0;
216 }
217 pr_warn("failed to retrieve upper fileattr (%pd2, err=%i)\n",
218 new->dentry, err);
219 return err;
220 }
221
222 BUILD_BUG_ON(OVL_COPY_FS_FLAGS_MASK & ~FS_COMMON_FL);
223 newfa.flags &= ~OVL_COPY_FS_FLAGS_MASK;
224 newfa.flags |= (oldfa.flags & OVL_COPY_FS_FLAGS_MASK);
225
226 BUILD_BUG_ON(OVL_COPY_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
227 newfa.fsx_xflags &= ~OVL_COPY_FSX_FLAGS_MASK;
228 newfa.fsx_xflags |= (oldfa.fsx_xflags & OVL_COPY_FSX_FLAGS_MASK);
229
230 return ovl_real_fileattr_set(new, &newfa);
231 }
232
233 static int ovl_verify_area(loff_t pos, loff_t pos2, loff_t len, loff_t totlen)
234 {
235 loff_t tmp;
236
237 if (pos != pos2)
238 return -EIO;
239 if (pos < 0 || len < 0 || totlen < 0)
240 return -EIO;
241 if (check_add_overflow(pos, len, &tmp))
242 return -EIO;
243 return 0;
244 }
245
246 static int ovl_sync_file(struct path *path)
247 {
248 struct file *new_file;
249 int err;
250
251 new_file = ovl_path_open(path, O_LARGEFILE | O_RDONLY);
252 if (IS_ERR(new_file))
253 return PTR_ERR(new_file);
254
255 err = vfs_fsync(new_file, 0);
256 fput(new_file);
257
258 return err;
259 }
260
261 static int ovl_copy_up_file(struct ovl_fs *ofs, struct dentry *dentry,
262 struct file *new_file, loff_t len,
263 bool datasync)
264 {
265 struct path datapath;
266 struct file *old_file;
267 loff_t old_pos = 0;
268 loff_t new_pos = 0;
269 loff_t cloned;
270 loff_t data_pos = -1;
271 loff_t hole_len;
272 bool skip_hole = false;
273 int error = 0;
274
275 ovl_path_lowerdata(dentry, &datapath);
276 if (WARN_ON_ONCE(datapath.dentry == NULL) ||
277 WARN_ON_ONCE(len < 0))
278 return -EIO;
279
280 old_file = ovl_path_open(&datapath, O_LARGEFILE | O_RDONLY);
281 if (IS_ERR(old_file))
282 return PTR_ERR(old_file);
283
284 /* Try to use clone_file_range to clone up within the same fs */
285 cloned = vfs_clone_file_range(old_file, 0, new_file, 0, len, 0);
286 if (cloned == len)
287 goto out_fput;
288
289 /* Couldn't clone, so now we try to copy the data */
290 error = rw_verify_area(READ, old_file, &old_pos, len);
291 if (!error)
292 error = rw_verify_area(WRITE, new_file, &new_pos, len);
293 if (error)
294 goto out_fput;
295
296 /* Check if lower fs supports seek operation */
297 if (old_file->f_mode & FMODE_LSEEK)
298 skip_hole = true;
299
300 while (len) {
301 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
302 ssize_t bytes;
303
304 if (len < this_len)
305 this_len = len;
306
307 if (signal_pending_state(TASK_KILLABLE, current)) {
308 error = -EINTR;
309 break;
310 }
311
312 /*
313 * Fill zero for hole will cost unnecessary disk space
314 * and meanwhile slow down the copy-up speed, so we do
315 * an optimization for hole during copy-up, it relies
316 * on SEEK_DATA implementation in lower fs so if lower
317 * fs does not support it, copy-up will behave as before.
318 *
319 * Detail logic of hole detection as below:
320 * When we detect next data position is larger than current
321 * position we will skip that hole, otherwise we copy
322 * data in the size of OVL_COPY_UP_CHUNK_SIZE. Actually,
323 * it may not recognize all kind of holes and sometimes
324 * only skips partial of hole area. However, it will be
325 * enough for most of the use cases.
326 *
327 * We do not hold upper sb_writers throughout the loop to avert
328 * lockdep warning with llseek of lower file in nested overlay:
329 * - upper sb_writers
330 * -- lower ovl_inode_lock (ovl_llseek)
331 */
332 if (skip_hole && data_pos < old_pos) {
333 data_pos = vfs_llseek(old_file, old_pos, SEEK_DATA);
334 if (data_pos > old_pos) {
335 hole_len = data_pos - old_pos;
336 len -= hole_len;
337 old_pos = new_pos = data_pos;
338 continue;
339 } else if (data_pos == -ENXIO) {
340 break;
341 } else if (data_pos < 0) {
342 skip_hole = false;
343 }
344 }
345
346 error = ovl_verify_area(old_pos, new_pos, this_len, len);
347 if (error)
348 break;
349
350 bytes = do_splice_direct(old_file, &old_pos,
351 new_file, &new_pos,
352 this_len, SPLICE_F_MOVE);
353 if (bytes <= 0) {
354 error = bytes;
355 break;
356 }
357 WARN_ON(old_pos != new_pos);
358
359 len -= bytes;
360 }
361 /* call fsync once, either now or later along with metadata */
362 if (!error && ovl_should_sync(ofs) && datasync)
363 error = vfs_fsync(new_file, 0);
364 out_fput:
365 fput(old_file);
366 return error;
367 }
368
369 static int ovl_set_size(struct ovl_fs *ofs,
370 struct dentry *upperdentry, struct kstat *stat)
371 {
372 struct iattr attr = {
373 .ia_valid = ATTR_SIZE,
374 .ia_size = stat->size,
375 };
376
377 return ovl_do_notify_change(ofs, upperdentry, &attr);
378 }
379
380 static int ovl_set_timestamps(struct ovl_fs *ofs, struct dentry *upperdentry,
381 struct kstat *stat)
382 {
383 struct iattr attr = {
384 .ia_valid =
385 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_CTIME,
386 .ia_atime = stat->atime,
387 .ia_mtime = stat->mtime,
388 };
389
390 return ovl_do_notify_change(ofs, upperdentry, &attr);
391 }
392
393 int ovl_set_attr(struct ovl_fs *ofs, struct dentry *upperdentry,
394 struct kstat *stat)
395 {
396 int err = 0;
397
398 if (!S_ISLNK(stat->mode)) {
399 struct iattr attr = {
400 .ia_valid = ATTR_MODE,
401 .ia_mode = stat->mode,
402 };
403 err = ovl_do_notify_change(ofs, upperdentry, &attr);
404 }
405 if (!err) {
406 struct iattr attr = {
407 .ia_valid = ATTR_UID | ATTR_GID,
408 .ia_vfsuid = VFSUIDT_INIT(stat->uid),
409 .ia_vfsgid = VFSGIDT_INIT(stat->gid),
410 };
411 err = ovl_do_notify_change(ofs, upperdentry, &attr);
412 }
413 if (!err)
414 ovl_set_timestamps(ofs, upperdentry, stat);
415
416 return err;
417 }
418
419 struct ovl_fh *ovl_encode_real_fh(struct ovl_fs *ofs, struct dentry *real,
420 bool is_upper)
421 {
422 struct ovl_fh *fh;
423 int fh_type, dwords;
424 int buflen = MAX_HANDLE_SZ;
425 uuid_t *uuid = &real->d_sb->s_uuid;
426 int err;
427
428 /* Make sure the real fid stays 32bit aligned */
429 BUILD_BUG_ON(OVL_FH_FID_OFFSET % 4);
430 BUILD_BUG_ON(MAX_HANDLE_SZ + OVL_FH_FID_OFFSET > 255);
431
432 fh = kzalloc(buflen + OVL_FH_FID_OFFSET, GFP_KERNEL);
433 if (!fh)
434 return ERR_PTR(-ENOMEM);
435
436 /*
437 * We encode a non-connectable file handle for non-dir, because we
438 * only need to find the lower inode number and we don't want to pay
439 * the price or reconnecting the dentry.
440 */
441 dwords = buflen >> 2;
442 fh_type = exportfs_encode_fh(real, (void *)fh->fb.fid, &dwords, 0);
443 buflen = (dwords << 2);
444
445 err = -EIO;
446 if (WARN_ON(fh_type < 0) ||
447 WARN_ON(buflen > MAX_HANDLE_SZ) ||
448 WARN_ON(fh_type == FILEID_INVALID))
449 goto out_err;
450
451 fh->fb.version = OVL_FH_VERSION;
452 fh->fb.magic = OVL_FH_MAGIC;
453 fh->fb.type = fh_type;
454 fh->fb.flags = OVL_FH_FLAG_CPU_ENDIAN;
455 /*
456 * When we will want to decode an overlay dentry from this handle
457 * and all layers are on the same fs, if we get a disconncted real
458 * dentry when we decode fid, the only way to tell if we should assign
459 * it to upperdentry or to lowerstack is by checking this flag.
460 */
461 if (is_upper)
462 fh->fb.flags |= OVL_FH_FLAG_PATH_UPPER;
463 fh->fb.len = sizeof(fh->fb) + buflen;
464 if (ovl_origin_uuid(ofs))
465 fh->fb.uuid = *uuid;
466
467 return fh;
468
469 out_err:
470 kfree(fh);
471 return ERR_PTR(err);
472 }
473
474 struct ovl_fh *ovl_get_origin_fh(struct ovl_fs *ofs, struct dentry *origin)
475 {
476 /*
477 * When lower layer doesn't support export operations store a 'null' fh,
478 * so we can use the overlay.origin xattr to distignuish between a copy
479 * up and a pure upper inode.
480 */
481 if (!ovl_can_decode_fh(origin->d_sb))
482 return NULL;
483
484 return ovl_encode_real_fh(ofs, origin, false);
485 }
486
487 int ovl_set_origin_fh(struct ovl_fs *ofs, const struct ovl_fh *fh,
488 struct dentry *upper)
489 {
490 int err;
491
492 /*
493 * Do not fail when upper doesn't support xattrs.
494 */
495 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_ORIGIN, fh->buf,
496 fh ? fh->fb.len : 0, 0);
497
498 /* Ignore -EPERM from setting "user.*" on symlink/special */
499 return err == -EPERM ? 0 : err;
500 }
501
502 /* Store file handle of @upper dir in @index dir entry */
503 static int ovl_set_upper_fh(struct ovl_fs *ofs, struct dentry *upper,
504 struct dentry *index)
505 {
506 const struct ovl_fh *fh;
507 int err;
508
509 fh = ovl_encode_real_fh(ofs, upper, true);
510 if (IS_ERR(fh))
511 return PTR_ERR(fh);
512
513 err = ovl_setxattr(ofs, index, OVL_XATTR_UPPER, fh->buf, fh->fb.len);
514
515 kfree(fh);
516 return err;
517 }
518
519 /*
520 * Create and install index entry.
521 *
522 * Caller must hold i_mutex on indexdir.
523 */
524 static int ovl_create_index(struct dentry *dentry, const struct ovl_fh *fh,
525 struct dentry *upper)
526 {
527 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
528 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
529 struct inode *dir = d_inode(indexdir);
530 struct dentry *index = NULL;
531 struct dentry *temp = NULL;
532 struct qstr name = { };
533 int err;
534
535 /*
536 * For now this is only used for creating index entry for directories,
537 * because non-dir are copied up directly to index and then hardlinked
538 * to upper dir.
539 *
540 * TODO: implement create index for non-dir, so we can call it when
541 * encoding file handle for non-dir in case index does not exist.
542 */
543 if (WARN_ON(!d_is_dir(dentry)))
544 return -EIO;
545
546 /* Directory not expected to be indexed before copy up */
547 if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
548 return -EIO;
549
550 err = ovl_get_index_name_fh(fh, &name);
551 if (err)
552 return err;
553
554 temp = ovl_create_temp(ofs, indexdir, OVL_CATTR(S_IFDIR | 0));
555 err = PTR_ERR(temp);
556 if (IS_ERR(temp))
557 goto free_name;
558
559 err = ovl_set_upper_fh(ofs, upper, temp);
560 if (err)
561 goto out;
562
563 index = ovl_lookup_upper(ofs, name.name, indexdir, name.len);
564 if (IS_ERR(index)) {
565 err = PTR_ERR(index);
566 } else {
567 err = ovl_do_rename(ofs, dir, temp, dir, index, 0);
568 dput(index);
569 }
570 out:
571 if (err)
572 ovl_cleanup(ofs, dir, temp);
573 dput(temp);
574 free_name:
575 kfree(name.name);
576 return err;
577 }
578
579 struct ovl_copy_up_ctx {
580 struct dentry *parent;
581 struct dentry *dentry;
582 struct path lowerpath;
583 struct kstat stat;
584 struct kstat pstat;
585 const char *link;
586 struct dentry *destdir;
587 struct qstr destname;
588 struct dentry *workdir;
589 const struct ovl_fh *origin_fh;
590 bool origin;
591 bool indexed;
592 bool metacopy;
593 bool metacopy_digest;
594 bool metadata_fsync;
595 };
596
597 static int ovl_link_up(struct ovl_copy_up_ctx *c)
598 {
599 int err;
600 struct dentry *upper;
601 struct dentry *upperdir = ovl_dentry_upper(c->parent);
602 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
603 struct inode *udir = d_inode(upperdir);
604
605 ovl_start_write(c->dentry);
606
607 /* Mark parent "impure" because it may now contain non-pure upper */
608 err = ovl_set_impure(c->parent, upperdir);
609 if (err)
610 goto out;
611
612 err = ovl_set_nlink_lower(c->dentry);
613 if (err)
614 goto out;
615
616 inode_lock_nested(udir, I_MUTEX_PARENT);
617 upper = ovl_lookup_upper(ofs, c->dentry->d_name.name, upperdir,
618 c->dentry->d_name.len);
619 err = PTR_ERR(upper);
620 if (!IS_ERR(upper)) {
621 err = ovl_do_link(ofs, ovl_dentry_upper(c->dentry), udir, upper);
622 dput(upper);
623
624 if (!err) {
625 /* Restore timestamps on parent (best effort) */
626 ovl_set_timestamps(ofs, upperdir, &c->pstat);
627 ovl_dentry_set_upper_alias(c->dentry);
628 ovl_dentry_update_reval(c->dentry, upper);
629 }
630 }
631 inode_unlock(udir);
632 if (err)
633 goto out;
634
635 err = ovl_set_nlink_upper(c->dentry);
636
637 out:
638 ovl_end_write(c->dentry);
639 return err;
640 }
641
642 static int ovl_copy_up_data(struct ovl_copy_up_ctx *c, const struct path *temp)
643 {
644 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
645 struct file *new_file;
646 int err;
647
648 if (!S_ISREG(c->stat.mode) || c->metacopy || !c->stat.size)
649 return 0;
650
651 new_file = ovl_path_open(temp, O_LARGEFILE | O_WRONLY);
652 if (IS_ERR(new_file))
653 return PTR_ERR(new_file);
654
655 err = ovl_copy_up_file(ofs, c->dentry, new_file, c->stat.size,
656 !c->metadata_fsync);
657 fput(new_file);
658
659 return err;
660 }
661
662 static int ovl_copy_up_metadata(struct ovl_copy_up_ctx *c, struct dentry *temp)
663 {
664 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
665 struct inode *inode = d_inode(c->dentry);
666 struct path upperpath = { .mnt = ovl_upper_mnt(ofs), .dentry = temp };
667 int err;
668
669 err = ovl_copy_xattr(c->dentry->d_sb, &c->lowerpath, temp);
670 if (err)
671 return err;
672
673 if (inode->i_flags & OVL_COPY_I_FLAGS_MASK &&
674 (S_ISREG(c->stat.mode) || S_ISDIR(c->stat.mode))) {
675 /*
676 * Copy the fileattr inode flags that are the source of already
677 * copied i_flags
678 */
679 err = ovl_copy_fileattr(inode, &c->lowerpath, &upperpath);
680 if (err)
681 return err;
682 }
683
684 /*
685 * Store identifier of lower inode in upper inode xattr to
686 * allow lookup of the copy up origin inode.
687 *
688 * Don't set origin when we are breaking the association with a lower
689 * hard link.
690 */
691 if (c->origin) {
692 err = ovl_set_origin_fh(ofs, c->origin_fh, temp);
693 if (err)
694 return err;
695 }
696
697 if (c->metacopy) {
698 struct path lowerdatapath;
699 struct ovl_metacopy metacopy_data = OVL_METACOPY_INIT;
700
701 ovl_path_lowerdata(c->dentry, &lowerdatapath);
702 if (WARN_ON_ONCE(lowerdatapath.dentry == NULL))
703 return -EIO;
704 err = ovl_get_verity_digest(ofs, &lowerdatapath, &metacopy_data);
705 if (err)
706 return err;
707
708 if (metacopy_data.digest_algo)
709 c->metacopy_digest = true;
710
711 err = ovl_set_metacopy_xattr(ofs, temp, &metacopy_data);
712 if (err)
713 return err;
714 }
715
716 inode_lock(temp->d_inode);
717 if (S_ISREG(c->stat.mode))
718 err = ovl_set_size(ofs, temp, &c->stat);
719 if (!err)
720 err = ovl_set_attr(ofs, temp, &c->stat);
721 inode_unlock(temp->d_inode);
722
723 /* fsync metadata before moving it into upper dir */
724 if (!err && ovl_should_sync(ofs) && c->metadata_fsync)
725 err = ovl_sync_file(&upperpath);
726
727 return err;
728 }
729
730 struct ovl_cu_creds {
731 const struct cred *old;
732 struct cred *new;
733 };
734
735 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
736 {
737 int err;
738
739 cc->old = cc->new = NULL;
740 err = security_inode_copy_up(dentry, &cc->new);
741 if (err < 0)
742 return err;
743
744 if (cc->new)
745 cc->old = override_creds(cc->new);
746
747 return 0;
748 }
749
750 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
751 {
752 if (cc->new) {
753 revert_creds(cc->old);
754 put_cred(cc->new);
755 }
756 }
757
758 /*
759 * Copyup using workdir to prepare temp file. Used when copying up directories,
760 * special files or when upper fs doesn't support O_TMPFILE.
761 */
762 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
763 {
764 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
765 struct inode *inode;
766 struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
767 struct path path = { .mnt = ovl_upper_mnt(ofs) };
768 struct dentry *temp, *upper, *trap;
769 struct ovl_cu_creds cc;
770 int err;
771 struct ovl_cattr cattr = {
772 /* Can't properly set mode on creation because of the umask */
773 .mode = c->stat.mode & S_IFMT,
774 .rdev = c->stat.rdev,
775 .link = c->link
776 };
777
778 err = ovl_prep_cu_creds(c->dentry, &cc);
779 if (err)
780 return err;
781
782 ovl_start_write(c->dentry);
783 inode_lock(wdir);
784 temp = ovl_create_temp(ofs, c->workdir, &cattr);
785 inode_unlock(wdir);
786 ovl_end_write(c->dentry);
787 ovl_revert_cu_creds(&cc);
788
789 if (IS_ERR(temp))
790 return PTR_ERR(temp);
791
792 /*
793 * Copy up data first and then xattrs. Writing data after
794 * xattrs will remove security.capability xattr automatically.
795 */
796 path.dentry = temp;
797 err = ovl_copy_up_data(c, &path);
798 /*
799 * We cannot hold lock_rename() throughout this helper, because of
800 * lock ordering with sb_writers, which shouldn't be held when calling
801 * ovl_copy_up_data(), so lock workdir and destdir and make sure that
802 * temp wasn't moved before copy up completion or cleanup.
803 */
804 ovl_start_write(c->dentry);
805 trap = lock_rename(c->workdir, c->destdir);
806 if (trap || temp->d_parent != c->workdir) {
807 /* temp or workdir moved underneath us? abort without cleanup */
808 dput(temp);
809 err = -EIO;
810 if (IS_ERR(trap))
811 goto out;
812 goto unlock;
813 } else if (err) {
814 goto cleanup;
815 }
816
817 err = ovl_copy_up_metadata(c, temp);
818 if (err)
819 goto cleanup;
820
821 if (S_ISDIR(c->stat.mode) && c->indexed) {
822 err = ovl_create_index(c->dentry, c->origin_fh, temp);
823 if (err)
824 goto cleanup;
825 }
826
827 upper = ovl_lookup_upper(ofs, c->destname.name, c->destdir,
828 c->destname.len);
829 err = PTR_ERR(upper);
830 if (IS_ERR(upper))
831 goto cleanup;
832
833 err = ovl_do_rename(ofs, wdir, temp, udir, upper, 0);
834 dput(upper);
835 if (err)
836 goto cleanup;
837
838 inode = d_inode(c->dentry);
839 if (c->metacopy_digest)
840 ovl_set_flag(OVL_HAS_DIGEST, inode);
841 else
842 ovl_clear_flag(OVL_HAS_DIGEST, inode);
843 ovl_clear_flag(OVL_VERIFIED_DIGEST, inode);
844
845 if (!c->metacopy)
846 ovl_set_upperdata(inode);
847 ovl_inode_update(inode, temp);
848 if (S_ISDIR(inode->i_mode))
849 ovl_set_flag(OVL_WHITEOUTS, inode);
850 unlock:
851 unlock_rename(c->workdir, c->destdir);
852 out:
853 ovl_end_write(c->dentry);
854
855 return err;
856
857 cleanup:
858 ovl_cleanup(ofs, wdir, temp);
859 dput(temp);
860 goto unlock;
861 }
862
863 /* Copyup using O_TMPFILE which does not require cross dir locking */
864 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
865 {
866 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
867 struct inode *udir = d_inode(c->destdir);
868 struct dentry *temp, *upper;
869 struct file *tmpfile;
870 struct ovl_cu_creds cc;
871 int err;
872
873 err = ovl_prep_cu_creds(c->dentry, &cc);
874 if (err)
875 return err;
876
877 ovl_start_write(c->dentry);
878 tmpfile = ovl_do_tmpfile(ofs, c->workdir, c->stat.mode);
879 ovl_end_write(c->dentry);
880 ovl_revert_cu_creds(&cc);
881 if (IS_ERR(tmpfile))
882 return PTR_ERR(tmpfile);
883
884 temp = tmpfile->f_path.dentry;
885 if (!c->metacopy && c->stat.size) {
886 err = ovl_copy_up_file(ofs, c->dentry, tmpfile, c->stat.size,
887 !c->metadata_fsync);
888 if (err)
889 goto out_fput;
890 }
891
892 ovl_start_write(c->dentry);
893
894 err = ovl_copy_up_metadata(c, temp);
895 if (err)
896 goto out;
897
898 inode_lock_nested(udir, I_MUTEX_PARENT);
899
900 upper = ovl_lookup_upper(ofs, c->destname.name, c->destdir,
901 c->destname.len);
902 err = PTR_ERR(upper);
903 if (!IS_ERR(upper)) {
904 err = ovl_do_link(ofs, temp, udir, upper);
905 dput(upper);
906 }
907 inode_unlock(udir);
908
909 if (err)
910 goto out;
911
912 if (c->metacopy_digest)
913 ovl_set_flag(OVL_HAS_DIGEST, d_inode(c->dentry));
914 else
915 ovl_clear_flag(OVL_HAS_DIGEST, d_inode(c->dentry));
916 ovl_clear_flag(OVL_VERIFIED_DIGEST, d_inode(c->dentry));
917
918 if (!c->metacopy)
919 ovl_set_upperdata(d_inode(c->dentry));
920 ovl_inode_update(d_inode(c->dentry), dget(temp));
921
922 out:
923 ovl_end_write(c->dentry);
924 out_fput:
925 fput(tmpfile);
926 return err;
927 }
928
929 /*
930 * Copy up a single dentry
931 *
932 * All renames start with copy up of source if necessary. The actual
933 * rename will only proceed once the copy up was successful. Copy up uses
934 * upper parent i_mutex for exclusion. Since rename can change d_parent it
935 * is possible that the copy up will lock the old parent. At that point
936 * the file will have already been copied up anyway.
937 */
938 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
939 {
940 int err;
941 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
942 struct dentry *origin = c->lowerpath.dentry;
943 struct ovl_fh *fh = NULL;
944 bool to_index = false;
945
946 /*
947 * Indexed non-dir is copied up directly to the index entry and then
948 * hardlinked to upper dir. Indexed dir is copied up to indexdir,
949 * then index entry is created and then copied up dir installed.
950 * Copying dir up to indexdir instead of workdir simplifies locking.
951 */
952 if (ovl_need_index(c->dentry)) {
953 c->indexed = true;
954 if (S_ISDIR(c->stat.mode))
955 c->workdir = ovl_indexdir(c->dentry->d_sb);
956 else
957 to_index = true;
958 }
959
960 if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index) {
961 fh = ovl_get_origin_fh(ofs, origin);
962 if (IS_ERR(fh))
963 return PTR_ERR(fh);
964
965 /* origin_fh may be NULL */
966 c->origin_fh = fh;
967 c->origin = true;
968 }
969
970 if (to_index) {
971 c->destdir = ovl_indexdir(c->dentry->d_sb);
972 err = ovl_get_index_name(ofs, origin, &c->destname);
973 if (err)
974 goto out_free_fh;
975 } else if (WARN_ON(!c->parent)) {
976 /* Disconnected dentry must be copied up to index dir */
977 err = -EIO;
978 goto out_free_fh;
979 } else {
980 /*
981 * c->dentry->d_name is stabilzed by ovl_copy_up_start(),
982 * because if we got here, it means that c->dentry has no upper
983 * alias and changing ->d_name means going through ovl_rename()
984 * that will call ovl_copy_up() on source and target dentry.
985 */
986 c->destname = c->dentry->d_name;
987 /*
988 * Mark parent "impure" because it may now contain non-pure
989 * upper
990 */
991 ovl_start_write(c->dentry);
992 err = ovl_set_impure(c->parent, c->destdir);
993 ovl_end_write(c->dentry);
994 if (err)
995 goto out_free_fh;
996 }
997
998 /* Should we copyup with O_TMPFILE or with workdir? */
999 if (S_ISREG(c->stat.mode) && ofs->tmpfile)
1000 err = ovl_copy_up_tmpfile(c);
1001 else
1002 err = ovl_copy_up_workdir(c);
1003 if (err)
1004 goto out;
1005
1006 if (c->indexed)
1007 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
1008
1009 ovl_start_write(c->dentry);
1010 if (to_index) {
1011 /* Initialize nlink for copy up of disconnected dentry */
1012 err = ovl_set_nlink_upper(c->dentry);
1013 } else {
1014 struct inode *udir = d_inode(c->destdir);
1015
1016 /* Restore timestamps on parent (best effort) */
1017 inode_lock(udir);
1018 ovl_set_timestamps(ofs, c->destdir, &c->pstat);
1019 inode_unlock(udir);
1020
1021 ovl_dentry_set_upper_alias(c->dentry);
1022 ovl_dentry_update_reval(c->dentry, ovl_dentry_upper(c->dentry));
1023 }
1024 ovl_end_write(c->dentry);
1025
1026 out:
1027 if (to_index)
1028 kfree(c->destname.name);
1029 out_free_fh:
1030 kfree(fh);
1031 return err;
1032 }
1033
1034 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
1035 int flags)
1036 {
1037 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1038
1039 if (!ofs->config.metacopy)
1040 return false;
1041
1042 if (!S_ISREG(mode))
1043 return false;
1044
1045 if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
1046 return false;
1047
1048 /* Fall back to full copy if no fsverity on source data and we require verity */
1049 if (ofs->config.verity_mode == OVL_VERITY_REQUIRE) {
1050 struct path lowerdata;
1051
1052 ovl_path_lowerdata(dentry, &lowerdata);
1053
1054 if (WARN_ON_ONCE(lowerdata.dentry == NULL) ||
1055 ovl_ensure_verity_loaded(&lowerdata) ||
1056 !fsverity_active(d_inode(lowerdata.dentry))) {
1057 return false;
1058 }
1059 }
1060
1061 return true;
1062 }
1063
1064 static ssize_t ovl_getxattr_value(const struct path *path, char *name, char **value)
1065 {
1066 ssize_t res;
1067 char *buf;
1068
1069 res = ovl_do_getxattr(path, name, NULL, 0);
1070 if (res == -ENODATA || res == -EOPNOTSUPP)
1071 res = 0;
1072
1073 if (res > 0) {
1074 buf = kzalloc(res, GFP_KERNEL);
1075 if (!buf)
1076 return -ENOMEM;
1077
1078 res = ovl_do_getxattr(path, name, buf, res);
1079 if (res < 0)
1080 kfree(buf);
1081 else
1082 *value = buf;
1083 }
1084 return res;
1085 }
1086
1087 /* Copy up data of an inode which was copied up metadata only in the past. */
1088 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
1089 {
1090 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
1091 struct path upperpath;
1092 int err;
1093 char *capability = NULL;
1094 ssize_t cap_size;
1095
1096 ovl_path_upper(c->dentry, &upperpath);
1097 if (WARN_ON(upperpath.dentry == NULL))
1098 return -EIO;
1099
1100 if (c->stat.size) {
1101 err = cap_size = ovl_getxattr_value(&upperpath, XATTR_NAME_CAPS,
1102 &capability);
1103 if (cap_size < 0)
1104 goto out;
1105 }
1106
1107 err = ovl_copy_up_data(c, &upperpath);
1108 if (err)
1109 goto out_free;
1110
1111 /*
1112 * Writing to upper file will clear security.capability xattr. We
1113 * don't want that to happen for normal copy-up operation.
1114 */
1115 ovl_start_write(c->dentry);
1116 if (capability) {
1117 err = ovl_do_setxattr(ofs, upperpath.dentry, XATTR_NAME_CAPS,
1118 capability, cap_size, 0);
1119 }
1120 if (!err) {
1121 err = ovl_removexattr(ofs, upperpath.dentry,
1122 OVL_XATTR_METACOPY);
1123 }
1124 ovl_end_write(c->dentry);
1125 if (err)
1126 goto out_free;
1127
1128 ovl_clear_flag(OVL_HAS_DIGEST, d_inode(c->dentry));
1129 ovl_clear_flag(OVL_VERIFIED_DIGEST, d_inode(c->dentry));
1130 ovl_set_upperdata(d_inode(c->dentry));
1131 out_free:
1132 kfree(capability);
1133 out:
1134 return err;
1135 }
1136
1137 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
1138 int flags)
1139 {
1140 int err;
1141 DEFINE_DELAYED_CALL(done);
1142 struct path parentpath;
1143 struct ovl_copy_up_ctx ctx = {
1144 .parent = parent,
1145 .dentry = dentry,
1146 .workdir = ovl_workdir(dentry),
1147 };
1148
1149 if (WARN_ON(!ctx.workdir))
1150 return -EROFS;
1151
1152 ovl_path_lower(dentry, &ctx.lowerpath);
1153 err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
1154 STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
1155 if (err)
1156 return err;
1157
1158 if (!kuid_has_mapping(current_user_ns(), ctx.stat.uid) ||
1159 !kgid_has_mapping(current_user_ns(), ctx.stat.gid))
1160 return -EOVERFLOW;
1161
1162 /*
1163 * With metacopy disabled, we fsync after final metadata copyup, for
1164 * both regular files and directories to get atomic copyup semantics
1165 * on filesystems that do not use strict metadata ordering (e.g. ubifs).
1166 *
1167 * With metacopy enabled we want to avoid fsync on all meta copyup
1168 * that will hurt performance of workloads such as chown -R, so we
1169 * only fsync on data copyup as legacy behavior.
1170 */
1171 ctx.metadata_fsync = !OVL_FS(dentry->d_sb)->config.metacopy &&
1172 (S_ISREG(ctx.stat.mode) || S_ISDIR(ctx.stat.mode));
1173 ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
1174
1175 if (parent) {
1176 ovl_path_upper(parent, &parentpath);
1177 ctx.destdir = parentpath.dentry;
1178
1179 err = vfs_getattr(&parentpath, &ctx.pstat,
1180 STATX_ATIME | STATX_MTIME,
1181 AT_STATX_SYNC_AS_STAT);
1182 if (err)
1183 return err;
1184 }
1185
1186 /* maybe truncate regular file. this has no effect on dirs */
1187 if (flags & O_TRUNC)
1188 ctx.stat.size = 0;
1189
1190 if (S_ISLNK(ctx.stat.mode)) {
1191 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
1192 if (IS_ERR(ctx.link))
1193 return PTR_ERR(ctx.link);
1194 }
1195
1196 err = ovl_copy_up_start(dentry, flags);
1197 /* err < 0: interrupted, err > 0: raced with another copy-up */
1198 if (unlikely(err)) {
1199 if (err > 0)
1200 err = 0;
1201 } else {
1202 if (!ovl_dentry_upper(dentry))
1203 err = ovl_do_copy_up(&ctx);
1204 if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
1205 err = ovl_link_up(&ctx);
1206 if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
1207 err = ovl_copy_up_meta_inode_data(&ctx);
1208 ovl_copy_up_end(dentry);
1209 }
1210 do_delayed_call(&done);
1211
1212 return err;
1213 }
1214
1215 static int ovl_copy_up_flags(struct dentry *dentry, int flags)
1216 {
1217 int err = 0;
1218 const struct cred *old_cred;
1219 bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
1220
1221 /*
1222 * With NFS export, copy up can get called for a disconnected non-dir.
1223 * In this case, we will copy up lower inode to index dir without
1224 * linking it to upper dir.
1225 */
1226 if (WARN_ON(disconnected && d_is_dir(dentry)))
1227 return -EIO;
1228
1229 /*
1230 * We may not need lowerdata if we are only doing metacopy up, but it is
1231 * not very important to optimize this case, so do lazy lowerdata lookup
1232 * before any copy up, so we can do it before taking ovl_inode_lock().
1233 */
1234 err = ovl_verify_lowerdata(dentry);
1235 if (err)
1236 return err;
1237
1238 old_cred = ovl_override_creds(dentry->d_sb);
1239 while (!err) {
1240 struct dentry *next;
1241 struct dentry *parent = NULL;
1242
1243 if (ovl_already_copied_up(dentry, flags))
1244 break;
1245
1246 next = dget(dentry);
1247 /* find the topmost dentry not yet copied up */
1248 for (; !disconnected;) {
1249 parent = dget_parent(next);
1250
1251 if (ovl_dentry_upper(parent))
1252 break;
1253
1254 dput(next);
1255 next = parent;
1256 }
1257
1258 err = ovl_copy_up_one(parent, next, flags);
1259
1260 dput(parent);
1261 dput(next);
1262 }
1263 revert_creds(old_cred);
1264
1265 return err;
1266 }
1267
1268 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
1269 {
1270 /* Copy up of disconnected dentry does not set upper alias */
1271 if (ovl_already_copied_up(dentry, flags))
1272 return false;
1273
1274 if (special_file(d_inode(dentry)->i_mode))
1275 return false;
1276
1277 if (!ovl_open_flags_need_copy_up(flags))
1278 return false;
1279
1280 return true;
1281 }
1282
1283 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
1284 {
1285 if (!ovl_open_need_copy_up(dentry, flags))
1286 return 0;
1287
1288 return ovl_copy_up_flags(dentry, flags);
1289 }
1290
1291 int ovl_copy_up_with_data(struct dentry *dentry)
1292 {
1293 return ovl_copy_up_flags(dentry, O_WRONLY);
1294 }
1295
1296 int ovl_copy_up(struct dentry *dentry)
1297 {
1298 return ovl_copy_up_flags(dentry, 0);
1299 }
1300
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.