1 // SPDX-License-Identifier: GPL-2.0-only
3 * Overlayfs NFS export support.
7 * Copyright (C) 2017-2018 CTERA Networks. All Rights Reserved.
11 #include <linux/cred.h>
12 #include <linux/mount.h>
13 #include <linux/namei.h>
14 #include <linux/xattr.h>
15 #include <linux/exportfs.h>
16 #include <linux/ratelimit.h>
17 #include "overlayfs.h"
19 static int ovl_encode_maybe_copy_up(struct dentry *dentry)
23 if (ovl_dentry_upper(dentry))
26 err = ovl_want_write(dentry);
28 err = ovl_copy_up(dentry);
29 ovl_drop_write(dentry);
33 pr_warn_ratelimited("failed to copy up on encode (%pd2, err=%i)\n",
41 * Before encoding a non-upper directory file handle from real layer N, we need
42 * to check if it will be possible to reconnect an overlay dentry from the real
43 * lower decoded dentry. This is done by following the overlay ancestry up to a
44 * "layer N connected" ancestor and verifying that all parents along the way are
45 * "layer N connectable". If an ancestor that is NOT "layer N connectable" is
46 * found, we need to copy up an ancestor, which is "layer N connectable", thus
47 * making that ancestor "layer N connected". For example:
52 * The overlay dentry /a is NOT "layer 2 connectable", because if dir /a is
53 * copied up and renamed, upper dir /a will be indexed by lower dir /a from
54 * layer 1. The dir /a from layer 2 will never be indexed, so the algorithm (*)
55 * in ovl_lookup_real_ancestor() will not be able to lookup a connected overlay
56 * dentry from the connected lower dentry /a/b/c.
58 * To avoid this problem on decode time, we need to copy up an ancestor of
59 * /a/b/c, which is "layer 2 connectable", on encode time. That ancestor is
60 * /a/b. After copy up (and index) of /a/b, it will become "layer 2 connected"
61 * and when the time comes to decode the file handle from lower dentry /a/b/c,
62 * ovl_lookup_real_ancestor() will find the indexed ancestor /a/b and decoding
63 * a connected overlay dentry will be accomplished.
65 * (*) the algorithm in ovl_lookup_real_ancestor() can be improved to lookup an
66 * entry /a in the lower layers above layer N and find the indexed dir /a from
67 * layer 1. If that improvement is made, then the check for "layer N connected"
68 * will need to verify there are no redirects in lower layers above N. In the
69 * example above, /a will be "layer 2 connectable". However, if layer 2 dir /a
70 * is a target of a layer 1 redirect, then /a will NOT be "layer 2 connectable":
72 * layer 1: /A (redirect = /a)
76 /* Return the lowest layer for encoding a connectable file handle */
77 static int ovl_connectable_layer(struct dentry *dentry)
79 struct ovl_entry *oe = OVL_E(dentry);
81 /* We can get overlay root from root of any layer */
82 if (dentry == dentry->d_sb->s_root)
83 return ovl_numlower(oe);
86 * If it's an unindexed merge dir, then it's not connectable with any
89 if (ovl_dentry_upper(dentry) &&
90 !ovl_test_flag(OVL_INDEX, d_inode(dentry)))
93 /* We can get upper/overlay path from indexed/lower dentry */
94 return ovl_lowerstack(oe)->layer->idx;
98 * @dentry is "connected" if all ancestors up to root or a "connected" ancestor
99 * have the same uppermost lower layer as the origin's layer. We may need to
100 * copy up a "connectable" ancestor to make it "connected". A "connected" dentry
101 * cannot become non "connected", so cache positive result in dentry flags.
103 * Return the connected origin layer or < 0 on error.
105 static int ovl_connect_layer(struct dentry *dentry)
107 struct dentry *next, *parent = NULL;
108 struct ovl_entry *oe = OVL_E(dentry);
112 if (WARN_ON(dentry == dentry->d_sb->s_root) ||
113 WARN_ON(!ovl_dentry_lower(dentry)))
116 origin_layer = ovl_lowerstack(oe)->layer->idx;
117 if (ovl_dentry_test_flag(OVL_E_CONNECTED, dentry))
120 /* Find the topmost origin layer connectable ancestor of @dentry */
123 parent = dget_parent(next);
124 if (WARN_ON(parent == next)) {
130 * If @parent is not origin layer connectable, then copy up
131 * @next which is origin layer connectable and we are done.
133 if (ovl_connectable_layer(parent) < origin_layer) {
134 err = ovl_encode_maybe_copy_up(next);
138 /* If @parent is connected or indexed we are done */
139 if (ovl_dentry_test_flag(OVL_E_CONNECTED, parent) ||
140 ovl_test_flag(OVL_INDEX, d_inode(parent)))
151 ovl_dentry_set_flag(OVL_E_CONNECTED, dentry);
153 return err ?: origin_layer;
157 * We only need to encode origin if there is a chance that the same object was
158 * encoded pre copy up and then we need to stay consistent with the same
159 * encoding also after copy up. If non-pure upper is not indexed, then it was
160 * copied up before NFS export was enabled. In that case we don't need to worry
161 * about staying consistent with pre copy up encoding and we encode an upper
162 * file handle. Overlay root dentry is a private case of non-indexed upper.
164 * The following table summarizes the different file handle encodings used for
165 * different overlay object types:
167 * Object type | Encoding
168 * --------------------------------
170 * Non-indexed upper | U
171 * Indexed upper | L (*)
174 * U = upper file handle
175 * L = lower file handle
177 * (*) Connecting an overlay dir from real lower dentry is not always
178 * possible when there are redirects in lower layers and non-indexed merge dirs.
179 * To mitigate those case, we may copy up the lower dir ancestor before encode
180 * a lower dir file handle.
182 * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error.
184 static int ovl_check_encode_origin(struct dentry *dentry)
186 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
188 /* Upper file handle for pure upper */
189 if (!ovl_dentry_lower(dentry))
193 * Upper file handle for non-indexed upper.
195 * Root is never indexed, so if there's an upper layer, encode upper for
198 if (ovl_dentry_upper(dentry) &&
199 !ovl_test_flag(OVL_INDEX, d_inode(dentry)))
203 * Decoding a merge dir, whose origin's ancestor is under a redirected
204 * lower dir or under a non-indexed upper is not always possible.
205 * ovl_connect_layer() will try to make origin's layer "connected" by
206 * copying up a "connectable" ancestor.
208 if (d_is_dir(dentry) && ovl_upper_mnt(ofs))
209 return ovl_connect_layer(dentry);
211 /* Lower file handle for indexed and non-upper dir/non-dir */
215 static int ovl_dentry_to_fid(struct ovl_fs *ofs, struct dentry *dentry,
216 u32 *fid, int buflen)
218 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(ofs, enc_lower ? ovl_dentry_lower(dentry) :
232 ovl_dentry_upper(dentry), !enc_lower);
236 len = OVL_FH_LEN(fh);
238 memcpy(fid, fh, len);
246 pr_warn_ratelimited("failed to encode file handle (%pd2, err=%i)\n",
251 static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len,
252 struct inode *parent)
254 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
255 struct dentry *dentry;
256 int bytes, buflen = *max_len << 2;
258 /* TODO: encode connectable file handles */
260 return FILEID_INVALID;
262 dentry = d_find_any_alias(inode);
264 return FILEID_INVALID;
266 bytes = ovl_dentry_to_fid(ofs, dentry, fid, buflen);
269 return FILEID_INVALID;
271 *max_len = bytes >> 2;
273 return FILEID_INVALID;
275 return OVL_FILEID_V1;
279 * Find or instantiate an overlay dentry from real dentries and index.
281 static struct dentry *ovl_obtain_alias(struct super_block *sb,
282 struct dentry *upper_alias,
283 struct ovl_path *lowerpath,
284 struct dentry *index)
286 struct dentry *lower = lowerpath ? lowerpath->dentry : NULL;
287 struct dentry *upper = upper_alias ?: index;
288 struct dentry *dentry;
289 struct inode *inode = NULL;
290 struct ovl_entry *oe;
291 struct ovl_inode_params oip = {
295 /* We get overlay directory dentries with ovl_lookup_real() */
296 if (d_is_dir(upper ?: lower))
297 return ERR_PTR(-EIO);
299 oe = ovl_alloc_entry(!!lower);
301 return ERR_PTR(-ENOMEM);
303 oip.upperdentry = dget(upper);
305 ovl_lowerstack(oe)->dentry = dget(lower);
306 ovl_lowerstack(oe)->layer = lowerpath->layer;
309 inode = ovl_get_inode(sb, &oip);
313 return ERR_CAST(inode);
317 ovl_set_flag(OVL_UPPERDATA, inode);
319 dentry = d_find_any_alias(inode);
323 dentry = d_alloc_anon(inode->i_sb);
324 if (unlikely(!dentry))
328 ovl_dentry_set_upper_alias(dentry);
330 ovl_dentry_init_reval(dentry, upper, OVL_I_E(inode));
332 return d_instantiate_anon(dentry, inode);
336 dentry = ERR_PTR(-ENOMEM);
342 /* Get the upper or lower dentry in stack whose on layer @idx */
343 static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx)
345 struct ovl_entry *oe = OVL_E(dentry);
346 struct ovl_path *lowerstack = ovl_lowerstack(oe);
350 return ovl_dentry_upper(dentry);
352 for (i = 0; i < ovl_numlower(oe); i++) {
353 if (lowerstack[i].layer->idx == idx)
354 return lowerstack[i].dentry;
361 * Lookup a child overlay dentry to get a connected overlay dentry whose real
362 * dentry is @real. If @real is on upper layer, we lookup a child overlay
363 * dentry with the same name as the real dentry. Otherwise, we need to consult
366 static struct dentry *ovl_lookup_real_one(struct dentry *connected,
368 const struct ovl_layer *layer)
370 struct inode *dir = d_inode(connected);
371 struct dentry *this, *parent = NULL;
372 struct name_snapshot name;
376 * Lookup child overlay dentry by real name. The dir mutex protects us
377 * from racing with overlay rename. If the overlay dentry that is above
378 * real has already been moved to a parent that is not under the
379 * connected overlay dir, we return -ECHILD and restart the lookup of
380 * connected real path from the top.
382 inode_lock_nested(dir, I_MUTEX_PARENT);
384 parent = dget_parent(real);
385 if (ovl_dentry_real_at(connected, layer->idx) != parent)
389 * We also need to take a snapshot of real dentry name to protect us
390 * from racing with underlying layer rename. In this case, we don't
391 * care about returning ESTALE, only from dereferencing a free name
392 * pointer because we hold no lock on the real dentry.
394 take_dentry_name_snapshot(&name, real);
396 * No idmap handling here: it's an internal lookup. Could skip
397 * permission checking altogether, but for now just use non-idmap
400 this = lookup_one_len(name.name.name, connected, name.name.len);
401 release_dentry_name_snapshot(&name);
405 } else if (!this || !this->d_inode) {
409 } else if (ovl_dentry_real_at(this, layer->idx) != real) {
421 pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
422 real, layer->idx, connected, err);
427 static struct dentry *ovl_lookup_real(struct super_block *sb,
429 const struct ovl_layer *layer);
432 * Lookup an indexed or hashed overlay dentry by real inode.
434 static struct dentry *ovl_lookup_real_inode(struct super_block *sb,
436 const struct ovl_layer *layer)
438 struct ovl_fs *ofs = sb->s_fs_info;
439 struct dentry *index = NULL;
440 struct dentry *this = NULL;
444 * Decoding upper dir from index is expensive, so first try to lookup
445 * overlay dentry in inode/dcache.
447 inode = ovl_lookup_inode(sb, real, !layer->idx);
449 return ERR_CAST(inode);
451 this = d_find_any_alias(inode);
456 * For decoded lower dir file handle, lookup index by origin to check
457 * if lower dir was copied up and and/or removed.
459 if (!this && layer->idx && ofs->indexdir && !WARN_ON(!d_is_dir(real))) {
460 index = ovl_lookup_index(ofs, NULL, real, false);
465 /* Get connected upper overlay dir from index */
467 struct dentry *upper = ovl_index_upper(ofs, index, true);
470 if (IS_ERR_OR_NULL(upper))
474 * ovl_lookup_real() in lower layer may call recursively once to
475 * ovl_lookup_real() in upper layer. The first level call walks
476 * back lower parents to the topmost indexed parent. The second
477 * recursive call walks back from indexed upper to the topmost
478 * connected/hashed upper parent (or up to root).
480 this = ovl_lookup_real(sb, upper, &ofs->layers[0]);
484 if (IS_ERR_OR_NULL(this))
487 if (ovl_dentry_real_at(this, layer->idx) != real) {
489 this = ERR_PTR(-EIO);
496 * Lookup an indexed or hashed overlay dentry, whose real dentry is an
499 static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb,
501 const struct ovl_layer *layer)
503 struct dentry *next, *parent = NULL;
504 struct dentry *ancestor = ERR_PTR(-EIO);
506 if (real == layer->mnt->mnt_root)
507 return dget(sb->s_root);
509 /* Find the topmost indexed or hashed ancestor */
512 parent = dget_parent(next);
515 * Lookup a matching overlay dentry in inode/dentry
516 * cache or in index by real inode.
518 ancestor = ovl_lookup_real_inode(sb, next, layer);
522 if (parent == layer->mnt->mnt_root) {
523 ancestor = dget(sb->s_root);
528 * If @real has been moved out of the layer root directory,
529 * we will eventully hit the real fs root. This cannot happen
530 * by legit overlay rename, so we return error in that case.
532 if (parent == next) {
533 ancestor = ERR_PTR(-EXDEV);
548 * Lookup a connected overlay dentry whose real dentry is @real.
549 * If @real is on upper layer, we lookup a child overlay dentry with the same
550 * path the real dentry. Otherwise, we need to consult index for lookup.
552 static struct dentry *ovl_lookup_real(struct super_block *sb,
554 const struct ovl_layer *layer)
556 struct dentry *connected;
559 connected = ovl_lookup_real_ancestor(sb, real, layer);
560 if (IS_ERR(connected))
564 struct dentry *next, *this;
565 struct dentry *parent = NULL;
566 struct dentry *real_connected = ovl_dentry_real_at(connected,
569 if (real_connected == real)
572 /* Find the topmost dentry not yet connected */
575 parent = dget_parent(next);
577 if (parent == real_connected)
581 * If real has been moved out of 'real_connected',
582 * we will not find 'real_connected' and hit the layer
583 * root. In that case, we need to restart connecting.
584 * This game can go on forever in the worst case. We
585 * may want to consider taking s_vfs_rename_mutex if
586 * this happens more than once.
588 if (parent == layer->mnt->mnt_root) {
590 connected = dget(sb->s_root);
595 * If real file has been moved out of the layer root
596 * directory, we will eventully hit the real fs root.
597 * This cannot happen by legit overlay rename, so we
598 * return error in that case.
600 if (parent == next) {
610 this = ovl_lookup_real_one(connected, next, layer);
615 * Lookup of child in overlay can fail when racing with
616 * overlay rename of child away from 'connected' parent.
617 * In this case, we need to restart the lookup from the
618 * top, because we cannot trust that 'real_connected' is
619 * still an ancestor of 'real'. There is a good chance
620 * that the renamed overlay ancestor is now in cache, so
621 * ovl_lookup_real_ancestor() will find it and we can
622 * continue to connect exactly from where lookup failed.
624 if (err == -ECHILD) {
625 this = ovl_lookup_real_ancestor(sb, real,
627 err = PTR_ERR_OR_ZERO(this);
645 pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
646 real, layer->idx, connected, err);
652 * Get an overlay dentry from upper/lower real dentries and index.
654 static struct dentry *ovl_get_dentry(struct super_block *sb,
655 struct dentry *upper,
656 struct ovl_path *lowerpath,
657 struct dentry *index)
659 struct ovl_fs *ofs = sb->s_fs_info;
660 const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer;
661 struct dentry *real = upper ?: (index ?: lowerpath->dentry);
664 * Obtain a disconnected overlay dentry from a non-dir real dentry
668 return ovl_obtain_alias(sb, upper, lowerpath, index);
670 /* Removed empty directory? */
671 if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(real))
672 return ERR_PTR(-ENOENT);
675 * If real dentry is connected and hashed, get a connected overlay
676 * dentry whose real dentry is @real.
678 return ovl_lookup_real(sb, real, layer);
681 static struct dentry *ovl_upper_fh_to_d(struct super_block *sb,
684 struct ovl_fs *ofs = sb->s_fs_info;
685 struct dentry *dentry;
686 struct dentry *upper;
688 if (!ovl_upper_mnt(ofs))
689 return ERR_PTR(-EACCES);
691 upper = ovl_decode_real_fh(ofs, fh, ovl_upper_mnt(ofs), true);
692 if (IS_ERR_OR_NULL(upper))
695 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
701 static struct dentry *ovl_lower_fh_to_d(struct super_block *sb,
704 struct ovl_fs *ofs = sb->s_fs_info;
705 struct ovl_path origin = { };
706 struct ovl_path *stack = &origin;
707 struct dentry *dentry = NULL;
708 struct dentry *index = NULL;
712 /* First lookup overlay inode in inode cache by origin fh */
713 err = ovl_check_origin_fh(ofs, fh, false, NULL, &stack);
717 if (!d_is_dir(origin.dentry) ||
718 !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) {
719 inode = ovl_lookup_inode(sb, origin.dentry, false);
720 err = PTR_ERR(inode);
724 dentry = d_find_any_alias(inode);
731 /* Then lookup indexed upper/whiteout by origin fh */
733 index = ovl_get_index_fh(ofs, fh);
734 err = PTR_ERR(index);
741 /* Then try to get a connected upper dir by index */
742 if (index && d_is_dir(index)) {
743 struct dentry *upper = ovl_index_upper(ofs, index, true);
745 err = PTR_ERR(upper);
746 if (IS_ERR_OR_NULL(upper))
749 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
754 /* Find origin.dentry again with ovl_acceptable() layer check */
755 if (d_is_dir(origin.dentry)) {
757 origin.dentry = NULL;
758 err = ovl_check_origin_fh(ofs, fh, true, NULL, &stack);
763 err = ovl_verify_origin(ofs, index, origin.dentry, false);
768 /* Get a connected non-upper dir or disconnected non-dir */
769 dentry = ovl_get_dentry(sb, NULL, &origin, index);
777 dentry = ERR_PTR(err);
781 static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type)
785 /* If on-wire inner fid is aligned - nothing to do */
786 if (fh_type == OVL_FILEID_V1)
787 return (struct ovl_fh *)fid;
789 if (fh_type != OVL_FILEID_V0)
790 return ERR_PTR(-EINVAL);
792 if (buflen <= OVL_FH_WIRE_OFFSET)
793 return ERR_PTR(-EINVAL);
795 fh = kzalloc(buflen, GFP_KERNEL);
797 return ERR_PTR(-ENOMEM);
799 /* Copy unaligned inner fh into aligned buffer */
800 memcpy(fh->buf, fid, buflen - OVL_FH_WIRE_OFFSET);
804 static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid,
805 int fh_len, int fh_type)
807 struct dentry *dentry = NULL;
808 struct ovl_fh *fh = NULL;
809 int len = fh_len << 2;
810 unsigned int flags = 0;
813 fh = ovl_fid_to_fh(fid, len, fh_type);
818 err = ovl_check_fh_len(fh, len);
822 flags = fh->fb.flags;
823 dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ?
824 ovl_upper_fh_to_d(sb, fh) :
825 ovl_lower_fh_to_d(sb, fh);
826 err = PTR_ERR(dentry);
827 if (IS_ERR(dentry) && err != -ESTALE)
831 /* We may have needed to re-align OVL_FILEID_V0 */
832 if (!IS_ERR_OR_NULL(fh) && fh != (void *)fid)
838 pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n",
839 fh_len, fh_type, flags, err);
840 dentry = ERR_PTR(err);
844 static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid,
845 int fh_len, int fh_type)
847 pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n");
848 return ERR_PTR(-EACCES);
851 static int ovl_get_name(struct dentry *parent, char *name,
852 struct dentry *child)
855 * ovl_fh_to_dentry() returns connected dir overlay dentries and
856 * ovl_fh_to_parent() is not implemented, so we should not get here.
862 static struct dentry *ovl_get_parent(struct dentry *dentry)
865 * ovl_fh_to_dentry() returns connected dir overlay dentries, so we
866 * should not get here.
869 return ERR_PTR(-EIO);
872 const struct export_operations ovl_export_operations = {
873 .encode_fh = ovl_encode_fh,
874 .fh_to_dentry = ovl_fh_to_dentry,
875 .fh_to_parent = ovl_fh_to_parent,
876 .get_name = ovl_get_name,
877 .get_parent = ovl_get_parent,