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27 pr_warn("\"check_copy_up\" module option is obsolete\n");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