2 * Overlayfs NFS export support.
4 * Amir Goldstein <amir73il@gmail.com>
6 * Copyright (C) 2017-2018 CTERA Networks. All Rights Reserved.
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License version 2 as published by
10 * the Free Software Foundation.
14 #include <linux/cred.h>
15 #include <linux/mount.h>
16 #include <linux/namei.h>
17 #include <linux/xattr.h>
18 #include <linux/exportfs.h>
19 #include <linux/ratelimit.h>
20 #include "overlayfs.h"
22 static int ovl_encode_maybe_copy_up(struct dentry
*dentry
)
26 if (ovl_dentry_upper(dentry
))
29 err
= ovl_want_write(dentry
);
31 err
= ovl_copy_up(dentry
);
32 ovl_drop_write(dentry
);
36 pr_warn_ratelimited("overlayfs: failed to copy up on encode (%pd2, err=%i)\n",
44 * Before encoding a non-upper directory file handle from real layer N, we need
45 * to check if it will be possible to reconnect an overlay dentry from the real
46 * lower decoded dentry. This is done by following the overlay ancestry up to a
47 * "layer N connected" ancestor and verifying that all parents along the way are
48 * "layer N connectable". If an ancestor that is NOT "layer N connectable" is
49 * found, we need to copy up an ancestor, which is "layer N connectable", thus
50 * making that ancestor "layer N connected". For example:
55 * The overlay dentry /a is NOT "layer 2 connectable", because if dir /a is
56 * copied up and renamed, upper dir /a will be indexed by lower dir /a from
57 * layer 1. The dir /a from layer 2 will never be indexed, so the algorithm (*)
58 * in ovl_lookup_real_ancestor() will not be able to lookup a connected overlay
59 * dentry from the connected lower dentry /a/b/c.
61 * To avoid this problem on decode time, we need to copy up an ancestor of
62 * /a/b/c, which is "layer 2 connectable", on encode time. That ancestor is
63 * /a/b. After copy up (and index) of /a/b, it will become "layer 2 connected"
64 * and when the time comes to decode the file handle from lower dentry /a/b/c,
65 * ovl_lookup_real_ancestor() will find the indexed ancestor /a/b and decoding
66 * a connected overlay dentry will be accomplished.
68 * (*) the algorithm in ovl_lookup_real_ancestor() can be improved to lookup an
69 * entry /a in the lower layers above layer N and find the indexed dir /a from
70 * layer 1. If that improvement is made, then the check for "layer N connected"
71 * will need to verify there are no redirects in lower layers above N. In the
72 * example above, /a will be "layer 2 connectable". However, if layer 2 dir /a
73 * is a target of a layer 1 redirect, then /a will NOT be "layer 2 connectable":
75 * layer 1: /A (redirect = /a)
79 /* Return the lowest layer for encoding a connectable file handle */
80 static int ovl_connectable_layer(struct dentry
*dentry
)
82 struct ovl_entry
*oe
= OVL_E(dentry
);
84 /* We can get overlay root from root of any layer */
85 if (dentry
== dentry
->d_sb
->s_root
)
89 * If it's an unindexed merge dir, then it's not connectable with any
92 if (ovl_dentry_upper(dentry
) &&
93 !ovl_test_flag(OVL_INDEX
, d_inode(dentry
)))
96 /* We can get upper/overlay path from indexed/lower dentry */
97 return oe
->lowerstack
[0].layer
->idx
;
101 * @dentry is "connected" if all ancestors up to root or a "connected" ancestor
102 * have the same uppermost lower layer as the origin's layer. We may need to
103 * copy up a "connectable" ancestor to make it "connected". A "connected" dentry
104 * cannot become non "connected", so cache positive result in dentry flags.
106 * Return the connected origin layer or < 0 on error.
108 static int ovl_connect_layer(struct dentry
*dentry
)
110 struct dentry
*next
, *parent
= NULL
;
114 if (WARN_ON(dentry
== dentry
->d_sb
->s_root
) ||
115 WARN_ON(!ovl_dentry_lower(dentry
)))
118 origin_layer
= OVL_E(dentry
)->lowerstack
[0].layer
->idx
;
119 if (ovl_dentry_test_flag(OVL_E_CONNECTED
, dentry
))
122 /* Find the topmost origin layer connectable ancestor of @dentry */
125 parent
= dget_parent(next
);
126 if (WARN_ON(parent
== next
)) {
132 * If @parent is not origin layer connectable, then copy up
133 * @next which is origin layer connectable and we are done.
135 if (ovl_connectable_layer(parent
) < origin_layer
) {
136 err
= ovl_encode_maybe_copy_up(next
);
140 /* If @parent is connected or indexed we are done */
141 if (ovl_dentry_test_flag(OVL_E_CONNECTED
, parent
) ||
142 ovl_test_flag(OVL_INDEX
, d_inode(parent
)))
153 ovl_dentry_set_flag(OVL_E_CONNECTED
, dentry
);
155 return err
?: origin_layer
;
159 * We only need to encode origin if there is a chance that the same object was
160 * encoded pre copy up and then we need to stay consistent with the same
161 * encoding also after copy up. If non-pure upper is not indexed, then it was
162 * copied up before NFS export was enabled. In that case we don't need to worry
163 * about staying consistent with pre copy up encoding and we encode an upper
164 * file handle. Overlay root dentry is a private case of non-indexed upper.
166 * The following table summarizes the different file handle encodings used for
167 * different overlay object types:
169 * Object type | Encoding
170 * --------------------------------
172 * Non-indexed upper | U
173 * Indexed upper | L (*)
176 * U = upper file handle
177 * L = lower file handle
179 * (*) Connecting an overlay dir from real lower dentry is not always
180 * possible when there are redirects in lower layers and non-indexed merge dirs.
181 * To mitigate those case, we may copy up the lower dir ancestor before encode
182 * a lower dir file handle.
184 * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error.
186 static int ovl_check_encode_origin(struct dentry
*dentry
)
188 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
190 /* Upper file handle for pure upper */
191 if (!ovl_dentry_lower(dentry
))
195 * Upper file handle for non-indexed upper.
197 * Root is never indexed, so if there's an upper layer, encode upper for
200 if (ovl_dentry_upper(dentry
) &&
201 !ovl_test_flag(OVL_INDEX
, d_inode(dentry
)))
205 * Decoding a merge dir, whose origin's ancestor is under a redirected
206 * lower dir or under a non-indexed upper is not always possible.
207 * ovl_connect_layer() will try to make origin's layer "connected" by
208 * copying up a "connectable" ancestor.
210 if (d_is_dir(dentry
) && ofs
->upper_mnt
)
211 return ovl_connect_layer(dentry
);
213 /* Lower file handle for indexed and non-upper dir/non-dir */
217 static int ovl_d_to_fh(struct dentry
*dentry
, char *buf
, int buflen
)
219 struct ovl_fh
*fh
= NULL
;
223 * Check if we should encode a lower or upper file handle and maybe
224 * copy up an ancestor to make lower file handle connectable.
226 err
= enc_lower
= ovl_check_encode_origin(dentry
);
230 /* Encode an upper or lower file handle */
231 fh
= ovl_encode_real_fh(enc_lower
? ovl_dentry_lower(dentry
) :
232 ovl_dentry_upper(dentry
), !enc_lower
);
238 if (fh
->len
> buflen
)
241 memcpy(buf
, (char *)fh
, fh
->len
);
249 pr_warn_ratelimited("overlayfs: failed to encode file handle (%pd2, err=%i, buflen=%d, len=%d, type=%d)\n",
250 dentry
, err
, buflen
, fh
? (int)fh
->len
: 0,
255 static int ovl_dentry_to_fh(struct dentry
*dentry
, u32
*fid
, int *max_len
)
257 int res
, len
= *max_len
<< 2;
259 res
= ovl_d_to_fh(dentry
, (char *)fid
, len
);
261 return FILEID_INVALID
;
265 /* Round up to dwords */
266 *max_len
= (len
+ 3) >> 2;
270 static int ovl_encode_fh(struct inode
*inode
, u32
*fid
, int *max_len
,
271 struct inode
*parent
)
273 struct dentry
*dentry
;
276 /* TODO: encode connectable file handles */
278 return FILEID_INVALID
;
280 dentry
= d_find_any_alias(inode
);
281 if (WARN_ON(!dentry
))
282 return FILEID_INVALID
;
284 type
= ovl_dentry_to_fh(dentry
, fid
, max_len
);
291 * Find or instantiate an overlay dentry from real dentries and index.
293 static struct dentry
*ovl_obtain_alias(struct super_block
*sb
,
294 struct dentry
*upper_alias
,
295 struct ovl_path
*lowerpath
,
296 struct dentry
*index
)
298 struct dentry
*lower
= lowerpath
? lowerpath
->dentry
: NULL
;
299 struct dentry
*upper
= upper_alias
?: index
;
300 struct dentry
*dentry
;
302 struct ovl_entry
*oe
;
303 struct ovl_inode_params oip
= {
304 .lowerpath
= lowerpath
,
309 /* We get overlay directory dentries with ovl_lookup_real() */
310 if (d_is_dir(upper
?: lower
))
311 return ERR_PTR(-EIO
);
313 oip
.upperdentry
= dget(upper
);
314 inode
= ovl_get_inode(sb
, &oip
);
317 return ERR_CAST(inode
);
321 ovl_set_flag(OVL_UPPERDATA
, inode
);
323 dentry
= d_find_any_alias(inode
);
325 dentry
= d_alloc_anon(inode
->i_sb
);
328 oe
= ovl_alloc_entry(lower
? 1 : 0);
333 oe
->lowerstack
->dentry
= dget(lower
);
334 oe
->lowerstack
->layer
= lowerpath
->layer
;
336 dentry
->d_fsdata
= oe
;
338 ovl_dentry_set_upper_alias(dentry
);
341 return d_instantiate_anon(dentry
, inode
);
346 return ERR_PTR(-ENOMEM
);
349 /* Get the upper or lower dentry in stach whose on layer @idx */
350 static struct dentry
*ovl_dentry_real_at(struct dentry
*dentry
, int idx
)
352 struct ovl_entry
*oe
= dentry
->d_fsdata
;
356 return ovl_dentry_upper(dentry
);
358 for (i
= 0; i
< oe
->numlower
; i
++) {
359 if (oe
->lowerstack
[i
].layer
->idx
== idx
)
360 return oe
->lowerstack
[i
].dentry
;
367 * Lookup a child overlay dentry to get a connected overlay dentry whose real
368 * dentry is @real. If @real is on upper layer, we lookup a child overlay
369 * dentry with the same name as the real dentry. Otherwise, we need to consult
372 static struct dentry
*ovl_lookup_real_one(struct dentry
*connected
,
374 struct ovl_layer
*layer
)
376 struct inode
*dir
= d_inode(connected
);
377 struct dentry
*this, *parent
= NULL
;
378 struct name_snapshot name
;
382 * Lookup child overlay dentry by real name. The dir mutex protects us
383 * from racing with overlay rename. If the overlay dentry that is above
384 * real has already been moved to a parent that is not under the
385 * connected overlay dir, we return -ECHILD and restart the lookup of
386 * connected real path from the top.
388 inode_lock_nested(dir
, I_MUTEX_PARENT
);
390 parent
= dget_parent(real
);
391 if (ovl_dentry_real_at(connected
, layer
->idx
) != parent
)
395 * We also need to take a snapshot of real dentry name to protect us
396 * from racing with underlying layer rename. In this case, we don't
397 * care about returning ESTALE, only from dereferencing a free name
398 * pointer because we hold no lock on the real dentry.
400 take_dentry_name_snapshot(&name
, real
);
401 this = lookup_one_len(name
.name
, connected
, strlen(name
.name
));
405 } else if (!this || !this->d_inode
) {
409 } else if (ovl_dentry_real_at(this, layer
->idx
) != real
) {
416 release_dentry_name_snapshot(&name
);
422 pr_warn_ratelimited("overlayfs: failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
423 real
, layer
->idx
, connected
, err
);
428 static struct dentry
*ovl_lookup_real(struct super_block
*sb
,
430 struct ovl_layer
*layer
);
433 * Lookup an indexed or hashed overlay dentry by real inode.
435 static struct dentry
*ovl_lookup_real_inode(struct super_block
*sb
,
437 struct ovl_layer
*layer
)
439 struct ovl_fs
*ofs
= sb
->s_fs_info
;
440 struct ovl_layer upper_layer
= { .mnt
= ofs
->upper_mnt
};
441 struct dentry
*index
= NULL
;
442 struct dentry
*this = NULL
;
446 * Decoding upper dir from index is expensive, so first try to lookup
447 * overlay dentry in inode/dcache.
449 inode
= ovl_lookup_inode(sb
, real
, !layer
->idx
);
451 return ERR_CAST(inode
);
453 this = d_find_any_alias(inode
);
458 * For decoded lower dir file handle, lookup index by origin to check
459 * if lower dir was copied up and and/or removed.
461 if (!this && layer
->idx
&& ofs
->indexdir
&& !WARN_ON(!d_is_dir(real
))) {
462 index
= ovl_lookup_index(ofs
, NULL
, real
, false);
467 /* Get connected upper overlay dir from index */
469 struct dentry
*upper
= ovl_index_upper(ofs
, index
);
472 if (IS_ERR_OR_NULL(upper
))
476 * ovl_lookup_real() in lower layer may call recursively once to
477 * ovl_lookup_real() in upper layer. The first level call walks
478 * back lower parents to the topmost indexed parent. The second
479 * recursive call walks back from indexed upper to the topmost
480 * connected/hashed upper parent (or up to root).
482 this = ovl_lookup_real(sb
, upper
, &upper_layer
);
486 if (IS_ERR_OR_NULL(this))
489 if (WARN_ON(ovl_dentry_real_at(this, layer
->idx
) != real
)) {
491 this = ERR_PTR(-EIO
);
498 * Lookup an indexed or hashed overlay dentry, whose real dentry is an
501 static struct dentry
*ovl_lookup_real_ancestor(struct super_block
*sb
,
503 struct ovl_layer
*layer
)
505 struct dentry
*next
, *parent
= NULL
;
506 struct dentry
*ancestor
= ERR_PTR(-EIO
);
508 if (real
== layer
->mnt
->mnt_root
)
509 return dget(sb
->s_root
);
511 /* Find the topmost indexed or hashed ancestor */
514 parent
= dget_parent(next
);
517 * Lookup a matching overlay dentry in inode/dentry
518 * cache or in index by real inode.
520 ancestor
= ovl_lookup_real_inode(sb
, next
, layer
);
524 if (parent
== layer
->mnt
->mnt_root
) {
525 ancestor
= dget(sb
->s_root
);
530 * If @real has been moved out of the layer root directory,
531 * we will eventully hit the real fs root. This cannot happen
532 * by legit overlay rename, so we return error in that case.
534 if (parent
== next
) {
535 ancestor
= ERR_PTR(-EXDEV
);
550 * Lookup a connected overlay dentry whose real dentry is @real.
551 * If @real is on upper layer, we lookup a child overlay dentry with the same
552 * path the real dentry. Otherwise, we need to consult index for lookup.
554 static struct dentry
*ovl_lookup_real(struct super_block
*sb
,
556 struct ovl_layer
*layer
)
558 struct dentry
*connected
;
561 connected
= ovl_lookup_real_ancestor(sb
, real
, layer
);
562 if (IS_ERR(connected
))
566 struct dentry
*next
, *this;
567 struct dentry
*parent
= NULL
;
568 struct dentry
*real_connected
= ovl_dentry_real_at(connected
,
571 if (real_connected
== real
)
574 /* Find the topmost dentry not yet connected */
577 parent
= dget_parent(next
);
579 if (parent
== real_connected
)
583 * If real has been moved out of 'real_connected',
584 * we will not find 'real_connected' and hit the layer
585 * root. In that case, we need to restart connecting.
586 * This game can go on forever in the worst case. We
587 * may want to consider taking s_vfs_rename_mutex if
588 * this happens more than once.
590 if (parent
== layer
->mnt
->mnt_root
) {
592 connected
= dget(sb
->s_root
);
597 * If real file has been moved out of the layer root
598 * directory, we will eventully hit the real fs root.
599 * This cannot happen by legit overlay rename, so we
600 * return error in that case.
602 if (parent
== next
) {
612 this = ovl_lookup_real_one(connected
, next
, layer
);
617 * Lookup of child in overlay can fail when racing with
618 * overlay rename of child away from 'connected' parent.
619 * In this case, we need to restart the lookup from the
620 * top, because we cannot trust that 'real_connected' is
621 * still an ancestor of 'real'. There is a good chance
622 * that the renamed overlay ancestor is now in cache, so
623 * ovl_lookup_real_ancestor() will find it and we can
624 * continue to connect exactly from where lookup failed.
626 if (err
== -ECHILD
) {
627 this = ovl_lookup_real_ancestor(sb
, real
,
629 err
= PTR_ERR_OR_ZERO(this);
647 pr_warn_ratelimited("overlayfs: failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
648 real
, layer
->idx
, connected
, err
);
654 * Get an overlay dentry from upper/lower real dentries and index.
656 static struct dentry
*ovl_get_dentry(struct super_block
*sb
,
657 struct dentry
*upper
,
658 struct ovl_path
*lowerpath
,
659 struct dentry
*index
)
661 struct ovl_fs
*ofs
= sb
->s_fs_info
;
662 struct ovl_layer upper_layer
= { .mnt
= ofs
->upper_mnt
};
663 struct ovl_layer
*layer
= upper
? &upper_layer
: lowerpath
->layer
;
664 struct dentry
*real
= upper
?: (index
?: lowerpath
->dentry
);
667 * Obtain a disconnected overlay dentry from a non-dir real dentry
671 return ovl_obtain_alias(sb
, upper
, lowerpath
, index
);
673 /* Removed empty directory? */
674 if ((real
->d_flags
& DCACHE_DISCONNECTED
) || d_unhashed(real
))
675 return ERR_PTR(-ENOENT
);
678 * If real dentry is connected and hashed, get a connected overlay
679 * dentry whose real dentry is @real.
681 return ovl_lookup_real(sb
, real
, layer
);
684 static struct dentry
*ovl_upper_fh_to_d(struct super_block
*sb
,
687 struct ovl_fs
*ofs
= sb
->s_fs_info
;
688 struct dentry
*dentry
;
689 struct dentry
*upper
;
692 return ERR_PTR(-EACCES
);
694 upper
= ovl_decode_real_fh(fh
, ofs
->upper_mnt
, true);
695 if (IS_ERR_OR_NULL(upper
))
698 dentry
= ovl_get_dentry(sb
, upper
, NULL
, NULL
);
704 static struct dentry
*ovl_lower_fh_to_d(struct super_block
*sb
,
707 struct ovl_fs
*ofs
= sb
->s_fs_info
;
708 struct ovl_path origin
= { };
709 struct ovl_path
*stack
= &origin
;
710 struct dentry
*dentry
= NULL
;
711 struct dentry
*index
= NULL
;
715 /* First lookup overlay inode in inode cache by origin fh */
716 err
= ovl_check_origin_fh(ofs
, fh
, false, NULL
, &stack
);
720 if (!d_is_dir(origin
.dentry
) ||
721 !(origin
.dentry
->d_flags
& DCACHE_DISCONNECTED
)) {
722 inode
= ovl_lookup_inode(sb
, origin
.dentry
, false);
723 err
= PTR_ERR(inode
);
727 dentry
= d_find_any_alias(inode
);
734 /* Then lookup indexed upper/whiteout by origin fh */
736 index
= ovl_get_index_fh(ofs
, fh
);
737 err
= PTR_ERR(index
);
744 /* Then try to get a connected upper dir by index */
745 if (index
&& d_is_dir(index
)) {
746 struct dentry
*upper
= ovl_index_upper(ofs
, index
);
748 err
= PTR_ERR(upper
);
749 if (IS_ERR_OR_NULL(upper
))
752 dentry
= ovl_get_dentry(sb
, upper
, NULL
, NULL
);
757 /* Otherwise, get a connected non-upper dir or disconnected non-dir */
758 if (d_is_dir(origin
.dentry
) &&
759 (origin
.dentry
->d_flags
& DCACHE_DISCONNECTED
)) {
761 origin
.dentry
= NULL
;
762 err
= ovl_check_origin_fh(ofs
, fh
, true, NULL
, &stack
);
767 err
= ovl_verify_origin(index
, origin
.dentry
, false);
772 dentry
= ovl_get_dentry(sb
, NULL
, &origin
, index
);
780 dentry
= ERR_PTR(err
);
784 static struct dentry
*ovl_fh_to_dentry(struct super_block
*sb
, struct fid
*fid
,
785 int fh_len
, int fh_type
)
787 struct dentry
*dentry
= NULL
;
788 struct ovl_fh
*fh
= (struct ovl_fh
*) fid
;
789 int len
= fh_len
<< 2;
790 unsigned int flags
= 0;
794 if (fh_type
!= OVL_FILEID
)
797 err
= ovl_check_fh_len(fh
, len
);
802 dentry
= (flags
& OVL_FH_FLAG_PATH_UPPER
) ?
803 ovl_upper_fh_to_d(sb
, fh
) :
804 ovl_lower_fh_to_d(sb
, fh
);
805 err
= PTR_ERR(dentry
);
806 if (IS_ERR(dentry
) && err
!= -ESTALE
)
812 pr_warn_ratelimited("overlayfs: failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n",
813 len
, fh_type
, flags
, err
);
817 static struct dentry
*ovl_fh_to_parent(struct super_block
*sb
, struct fid
*fid
,
818 int fh_len
, int fh_type
)
820 pr_warn_ratelimited("overlayfs: connectable file handles not supported; use 'no_subtree_check' exportfs option.\n");
821 return ERR_PTR(-EACCES
);
824 static int ovl_get_name(struct dentry
*parent
, char *name
,
825 struct dentry
*child
)
828 * ovl_fh_to_dentry() returns connected dir overlay dentries and
829 * ovl_fh_to_parent() is not implemented, so we should not get here.
835 static struct dentry
*ovl_get_parent(struct dentry
*dentry
)
838 * ovl_fh_to_dentry() returns connected dir overlay dentries, so we
839 * should not get here.
842 return ERR_PTR(-EIO
);
845 const struct export_operations ovl_export_operations
= {
846 .encode_fh
= ovl_encode_fh
,
847 .fh_to_dentry
= ovl_fh_to_dentry
,
848 .fh_to_parent
= ovl_fh_to_parent
,
849 .get_name
= ovl_get_name
,
850 .get_parent
= ovl_get_parent
,