1 // SPDX-License-Identifier: GPL-2.0-only
3 * Overlayfs NFS export support.
5 * Amir Goldstein <amir73il@gmail.com>
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_copy_up(dentry);
28 pr_warn_ratelimited("failed to copy up on encode (%pd2, err=%i)\n",
36 * Before encoding a non-upper directory file handle from real layer N, we need
37 * to check if it will be possible to reconnect an overlay dentry from the real
38 * lower decoded dentry. This is done by following the overlay ancestry up to a
39 * "layer N connected" ancestor and verifying that all parents along the way are
40 * "layer N connectable". If an ancestor that is NOT "layer N connectable" is
41 * found, we need to copy up an ancestor, which is "layer N connectable", thus
42 * making that ancestor "layer N connected". For example:
47 * The overlay dentry /a is NOT "layer 2 connectable", because if dir /a is
48 * copied up and renamed, upper dir /a will be indexed by lower dir /a from
49 * layer 1. The dir /a from layer 2 will never be indexed, so the algorithm (*)
50 * in ovl_lookup_real_ancestor() will not be able to lookup a connected overlay
51 * dentry from the connected lower dentry /a/b/c.
53 * To avoid this problem on decode time, we need to copy up an ancestor of
54 * /a/b/c, which is "layer 2 connectable", on encode time. That ancestor is
55 * /a/b. After copy up (and index) of /a/b, it will become "layer 2 connected"
56 * and when the time comes to decode the file handle from lower dentry /a/b/c,
57 * ovl_lookup_real_ancestor() will find the indexed ancestor /a/b and decoding
58 * a connected overlay dentry will be accomplished.
60 * (*) the algorithm in ovl_lookup_real_ancestor() can be improved to lookup an
61 * entry /a in the lower layers above layer N and find the indexed dir /a from
62 * layer 1. If that improvement is made, then the check for "layer N connected"
63 * will need to verify there are no redirects in lower layers above N. In the
64 * example above, /a will be "layer 2 connectable". However, if layer 2 dir /a
65 * is a target of a layer 1 redirect, then /a will NOT be "layer 2 connectable":
67 * layer 1: /A (redirect = /a)
71 /* Return the lowest layer for encoding a connectable file handle */
72 static int ovl_connectable_layer(struct dentry *dentry)
74 struct ovl_entry *oe = OVL_E(dentry);
76 /* We can get overlay root from root of any layer */
77 if (dentry == dentry->d_sb->s_root)
78 return ovl_numlower(oe);
81 * If it's an unindexed merge dir, then it's not connectable with any
84 if (ovl_dentry_upper(dentry) &&
85 !ovl_test_flag(OVL_INDEX, d_inode(dentry)))
88 /* We can get upper/overlay path from indexed/lower dentry */
89 return ovl_lowerstack(oe)->layer->idx;
93 * @dentry is "connected" if all ancestors up to root or a "connected" ancestor
94 * have the same uppermost lower layer as the origin's layer. We may need to
95 * copy up a "connectable" ancestor to make it "connected". A "connected" dentry
96 * cannot become non "connected", so cache positive result in dentry flags.
98 * Return the connected origin layer or < 0 on error.
100 static int ovl_connect_layer(struct dentry *dentry)
102 struct dentry *next, *parent = NULL;
103 struct ovl_entry *oe = OVL_E(dentry);
107 if (WARN_ON(dentry == dentry->d_sb->s_root) ||
108 WARN_ON(!ovl_dentry_lower(dentry)))
111 origin_layer = ovl_lowerstack(oe)->layer->idx;
112 if (ovl_dentry_test_flag(OVL_E_CONNECTED, dentry))
115 /* Find the topmost origin layer connectable ancestor of @dentry */
118 parent = dget_parent(next);
119 if (WARN_ON(parent == next)) {
125 * If @parent is not origin layer connectable, then copy up
126 * @next which is origin layer connectable and we are done.
128 if (ovl_connectable_layer(parent) < origin_layer) {
129 err = ovl_encode_maybe_copy_up(next);
133 /* If @parent is connected or indexed we are done */
134 if (ovl_dentry_test_flag(OVL_E_CONNECTED, parent) ||
135 ovl_test_flag(OVL_INDEX, d_inode(parent)))
146 ovl_dentry_set_flag(OVL_E_CONNECTED, dentry);
148 return err ?: origin_layer;
152 * We only need to encode origin if there is a chance that the same object was
153 * encoded pre copy up and then we need to stay consistent with the same
154 * encoding also after copy up. If non-pure upper is not indexed, then it was
155 * copied up before NFS export was enabled. In that case we don't need to worry
156 * about staying consistent with pre copy up encoding and we encode an upper
157 * file handle. Overlay root dentry is a private case of non-indexed upper.
159 * The following table summarizes the different file handle encodings used for
160 * different overlay object types:
162 * Object type | Encoding
163 * --------------------------------
165 * Non-indexed upper | U
166 * Indexed upper | L (*)
169 * U = upper file handle
170 * L = lower file handle
172 * (*) Decoding a connected overlay dir from real lower dentry is not always
173 * possible when there are redirects in lower layers and non-indexed merge dirs.
174 * To mitigate those case, we may copy up the lower dir ancestor before encode
175 * of a decodable file handle for non-upper dir.
177 * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error.
179 static int ovl_check_encode_origin(struct dentry *dentry)
181 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
182 bool decodable = ofs->config.nfs_export;
184 /* Lower file handle for non-upper non-decodable */
185 if (!ovl_dentry_upper(dentry) && !decodable)
188 /* Upper file handle for pure upper */
189 if (!ovl_dentry_lower(dentry))
193 * Root is never indexed, so if there's an upper layer, encode upper for
196 if (dentry == dentry->d_sb->s_root)
200 * Upper decodable file handle for non-indexed upper.
202 if (ovl_dentry_upper(dentry) && decodable &&
203 !ovl_test_flag(OVL_INDEX, d_inode(dentry)))
207 * Decoding a merge dir, whose origin's ancestor is under a redirected
208 * lower dir or under a non-indexed upper is not always possible.
209 * ovl_connect_layer() will try to make origin's layer "connected" by
210 * copying up a "connectable" ancestor.
212 if (d_is_dir(dentry) && ovl_upper_mnt(ofs) && decodable)
213 return ovl_connect_layer(dentry);
215 /* Lower file handle for indexed and non-upper dir/non-dir */
219 static int ovl_dentry_to_fid(struct ovl_fs *ofs, struct dentry *dentry,
220 u32 *fid, int buflen)
222 struct ovl_fh *fh = NULL;
227 * Check if we should encode a lower or upper file handle and maybe
228 * copy up an ancestor to make lower file handle connectable.
230 err = enc_lower = ovl_check_encode_origin(dentry);
234 /* Encode an upper or lower file handle */
235 fh = ovl_encode_real_fh(ofs, enc_lower ? ovl_dentry_lower(dentry) :
236 ovl_dentry_upper(dentry), !enc_lower);
240 len = OVL_FH_LEN(fh);
242 memcpy(fid, fh, len);
250 pr_warn_ratelimited("failed to encode file handle (%pd2, err=%i)\n",
255 static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len,
256 struct inode *parent)
258 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
259 struct dentry *dentry;
260 int bytes, buflen = *max_len << 2;
262 /* TODO: encode connectable file handles */
264 return FILEID_INVALID;
266 dentry = d_find_any_alias(inode);
268 return FILEID_INVALID;
270 bytes = ovl_dentry_to_fid(ofs, dentry, fid, buflen);
273 return FILEID_INVALID;
275 *max_len = bytes >> 2;
277 return FILEID_INVALID;
279 return OVL_FILEID_V1;
283 * Find or instantiate an overlay dentry from real dentries and index.
285 static struct dentry *ovl_obtain_alias(struct super_block *sb,
286 struct dentry *upper_alias,
287 struct ovl_path *lowerpath,
288 struct dentry *index)
290 struct dentry *lower = lowerpath ? lowerpath->dentry : NULL;
291 struct dentry *upper = upper_alias ?: index;
292 struct inode *inode = NULL;
293 struct ovl_entry *oe;
294 struct ovl_inode_params oip = {
298 /* We get overlay directory dentries with ovl_lookup_real() */
299 if (d_is_dir(upper ?: lower))
300 return ERR_PTR(-EIO);
302 oe = ovl_alloc_entry(!!lower);
304 return ERR_PTR(-ENOMEM);
306 oip.upperdentry = dget(upper);
308 ovl_lowerstack(oe)->dentry = dget(lower);
309 ovl_lowerstack(oe)->layer = lowerpath->layer;
312 inode = ovl_get_inode(sb, &oip);
316 return ERR_CAST(inode);
320 ovl_set_flag(OVL_UPPERDATA, inode);
322 return d_obtain_alias(inode);
325 /* Get the upper or lower dentry in stack whose on layer @idx */
326 static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx)
328 struct ovl_entry *oe = OVL_E(dentry);
329 struct ovl_path *lowerstack = ovl_lowerstack(oe);
333 return ovl_dentry_upper(dentry);
335 for (i = 0; i < ovl_numlower(oe); i++) {
336 if (lowerstack[i].layer->idx == idx)
337 return lowerstack[i].dentry;
344 * Lookup a child overlay dentry to get a connected overlay dentry whose real
345 * dentry is @real. If @real is on upper layer, we lookup a child overlay
346 * dentry with the same name as the real dentry. Otherwise, we need to consult
349 static struct dentry *ovl_lookup_real_one(struct dentry *connected,
351 const struct ovl_layer *layer)
353 struct inode *dir = d_inode(connected);
354 struct dentry *this, *parent = NULL;
355 struct name_snapshot name;
359 * Lookup child overlay dentry by real name. The dir mutex protects us
360 * from racing with overlay rename. If the overlay dentry that is above
361 * real has already been moved to a parent that is not under the
362 * connected overlay dir, we return -ECHILD and restart the lookup of
363 * connected real path from the top.
365 inode_lock_nested(dir, I_MUTEX_PARENT);
367 parent = dget_parent(real);
368 if (ovl_dentry_real_at(connected, layer->idx) != parent)
372 * We also need to take a snapshot of real dentry name to protect us
373 * from racing with underlying layer rename. In this case, we don't
374 * care about returning ESTALE, only from dereferencing a free name
375 * pointer because we hold no lock on the real dentry.
377 take_dentry_name_snapshot(&name, real);
379 * No idmap handling here: it's an internal lookup. Could skip
380 * permission checking altogether, but for now just use non-idmap
383 this = lookup_one_len(name.name.name, connected, name.name.len);
384 release_dentry_name_snapshot(&name);
388 } else if (!this || !this->d_inode) {
392 } else if (ovl_dentry_real_at(this, layer->idx) != real) {
404 pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
405 real, layer->idx, connected, err);
410 static struct dentry *ovl_lookup_real(struct super_block *sb,
412 const struct ovl_layer *layer);
415 * Lookup an indexed or hashed overlay dentry by real inode.
417 static struct dentry *ovl_lookup_real_inode(struct super_block *sb,
419 const struct ovl_layer *layer)
421 struct ovl_fs *ofs = OVL_FS(sb);
422 struct dentry *index = NULL;
423 struct dentry *this = NULL;
427 * Decoding upper dir from index is expensive, so first try to lookup
428 * overlay dentry in inode/dcache.
430 inode = ovl_lookup_inode(sb, real, !layer->idx);
432 return ERR_CAST(inode);
434 this = d_find_any_alias(inode);
439 * For decoded lower dir file handle, lookup index by origin to check
440 * if lower dir was copied up and and/or removed.
442 if (!this && layer->idx && ovl_indexdir(sb) && !WARN_ON(!d_is_dir(real))) {
443 index = ovl_lookup_index(ofs, NULL, real, false);
448 /* Get connected upper overlay dir from index */
450 struct dentry *upper = ovl_index_upper(ofs, index, true);
453 if (IS_ERR_OR_NULL(upper))
457 * ovl_lookup_real() in lower layer may call recursively once to
458 * ovl_lookup_real() in upper layer. The first level call walks
459 * back lower parents to the topmost indexed parent. The second
460 * recursive call walks back from indexed upper to the topmost
461 * connected/hashed upper parent (or up to root).
463 this = ovl_lookup_real(sb, upper, &ofs->layers[0]);
467 if (IS_ERR_OR_NULL(this))
470 if (ovl_dentry_real_at(this, layer->idx) != real) {
472 this = ERR_PTR(-EIO);
479 * Lookup an indexed or hashed overlay dentry, whose real dentry is an
482 static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb,
484 const struct ovl_layer *layer)
486 struct dentry *next, *parent = NULL;
487 struct dentry *ancestor = ERR_PTR(-EIO);
489 if (real == layer->mnt->mnt_root)
490 return dget(sb->s_root);
492 /* Find the topmost indexed or hashed ancestor */
495 parent = dget_parent(next);
498 * Lookup a matching overlay dentry in inode/dentry
499 * cache or in index by real inode.
501 ancestor = ovl_lookup_real_inode(sb, next, layer);
505 if (parent == layer->mnt->mnt_root) {
506 ancestor = dget(sb->s_root);
511 * If @real has been moved out of the layer root directory,
512 * we will eventully hit the real fs root. This cannot happen
513 * by legit overlay rename, so we return error in that case.
515 if (parent == next) {
516 ancestor = ERR_PTR(-EXDEV);
531 * Lookup a connected overlay dentry whose real dentry is @real.
532 * If @real is on upper layer, we lookup a child overlay dentry with the same
533 * path the real dentry. Otherwise, we need to consult index for lookup.
535 static struct dentry *ovl_lookup_real(struct super_block *sb,
537 const struct ovl_layer *layer)
539 struct dentry *connected;
542 connected = ovl_lookup_real_ancestor(sb, real, layer);
543 if (IS_ERR(connected))
547 struct dentry *next, *this;
548 struct dentry *parent = NULL;
549 struct dentry *real_connected = ovl_dentry_real_at(connected,
552 if (real_connected == real)
555 /* Find the topmost dentry not yet connected */
558 parent = dget_parent(next);
560 if (parent == real_connected)
564 * If real has been moved out of 'real_connected',
565 * we will not find 'real_connected' and hit the layer
566 * root. In that case, we need to restart connecting.
567 * This game can go on forever in the worst case. We
568 * may want to consider taking s_vfs_rename_mutex if
569 * this happens more than once.
571 if (parent == layer->mnt->mnt_root) {
573 connected = dget(sb->s_root);
578 * If real file has been moved out of the layer root
579 * directory, we will eventully hit the real fs root.
580 * This cannot happen by legit overlay rename, so we
581 * return error in that case.
583 if (parent == next) {
593 this = ovl_lookup_real_one(connected, next, layer);
598 * Lookup of child in overlay can fail when racing with
599 * overlay rename of child away from 'connected' parent.
600 * In this case, we need to restart the lookup from the
601 * top, because we cannot trust that 'real_connected' is
602 * still an ancestor of 'real'. There is a good chance
603 * that the renamed overlay ancestor is now in cache, so
604 * ovl_lookup_real_ancestor() will find it and we can
605 * continue to connect exactly from where lookup failed.
607 if (err == -ECHILD) {
608 this = ovl_lookup_real_ancestor(sb, real,
610 err = PTR_ERR_OR_ZERO(this);
628 pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
629 real, layer->idx, connected, err);
635 * Get an overlay dentry from upper/lower real dentries and index.
637 static struct dentry *ovl_get_dentry(struct super_block *sb,
638 struct dentry *upper,
639 struct ovl_path *lowerpath,
640 struct dentry *index)
642 struct ovl_fs *ofs = OVL_FS(sb);
643 const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer;
644 struct dentry *real = upper ?: (index ?: lowerpath->dentry);
647 * Obtain a disconnected overlay dentry from a non-dir real dentry
651 return ovl_obtain_alias(sb, upper, lowerpath, index);
653 /* Removed empty directory? */
654 if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(real))
655 return ERR_PTR(-ENOENT);
658 * If real dentry is connected and hashed, get a connected overlay
659 * dentry whose real dentry is @real.
661 return ovl_lookup_real(sb, real, layer);
664 static struct dentry *ovl_upper_fh_to_d(struct super_block *sb,
667 struct ovl_fs *ofs = OVL_FS(sb);
668 struct dentry *dentry;
669 struct dentry *upper;
671 if (!ovl_upper_mnt(ofs))
672 return ERR_PTR(-EACCES);
674 upper = ovl_decode_real_fh(ofs, fh, ovl_upper_mnt(ofs), true);
675 if (IS_ERR_OR_NULL(upper))
678 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
684 static struct dentry *ovl_lower_fh_to_d(struct super_block *sb,
687 struct ovl_fs *ofs = OVL_FS(sb);
688 struct ovl_path origin = { };
689 struct ovl_path *stack = &origin;
690 struct dentry *dentry = NULL;
691 struct dentry *index = NULL;
695 /* First lookup overlay inode in inode cache by origin fh */
696 err = ovl_check_origin_fh(ofs, fh, false, NULL, &stack);
700 if (!d_is_dir(origin.dentry) ||
701 !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) {
702 inode = ovl_lookup_inode(sb, origin.dentry, false);
703 err = PTR_ERR(inode);
707 dentry = d_find_any_alias(inode);
714 /* Then lookup indexed upper/whiteout by origin fh */
715 if (ovl_indexdir(sb)) {
716 index = ovl_get_index_fh(ofs, fh);
717 err = PTR_ERR(index);
724 /* Then try to get a connected upper dir by index */
725 if (index && d_is_dir(index)) {
726 struct dentry *upper = ovl_index_upper(ofs, index, true);
728 err = PTR_ERR(upper);
729 if (IS_ERR_OR_NULL(upper))
732 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
737 /* Find origin.dentry again with ovl_acceptable() layer check */
738 if (d_is_dir(origin.dentry)) {
740 origin.dentry = NULL;
741 err = ovl_check_origin_fh(ofs, fh, true, NULL, &stack);
746 err = ovl_verify_origin(ofs, index, origin.dentry, false);
751 /* Get a connected non-upper dir or disconnected non-dir */
752 dentry = ovl_get_dentry(sb, NULL, &origin, index);
760 dentry = ERR_PTR(err);
764 static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type)
768 /* If on-wire inner fid is aligned - nothing to do */
769 if (fh_type == OVL_FILEID_V1)
770 return (struct ovl_fh *)fid;
772 if (fh_type != OVL_FILEID_V0)
773 return ERR_PTR(-EINVAL);
775 if (buflen <= OVL_FH_WIRE_OFFSET)
776 return ERR_PTR(-EINVAL);
778 fh = kzalloc(buflen, GFP_KERNEL);
780 return ERR_PTR(-ENOMEM);
782 /* Copy unaligned inner fh into aligned buffer */
783 memcpy(fh->buf, fid, buflen - OVL_FH_WIRE_OFFSET);
787 static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid,
788 int fh_len, int fh_type)
790 struct dentry *dentry = NULL;
791 struct ovl_fh *fh = NULL;
792 int len = fh_len << 2;
793 unsigned int flags = 0;
796 fh = ovl_fid_to_fh(fid, len, fh_type);
801 err = ovl_check_fh_len(fh, len);
805 flags = fh->fb.flags;
806 dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ?
807 ovl_upper_fh_to_d(sb, fh) :
808 ovl_lower_fh_to_d(sb, fh);
809 err = PTR_ERR(dentry);
810 if (IS_ERR(dentry) && err != -ESTALE)
814 /* We may have needed to re-align OVL_FILEID_V0 */
815 if (!IS_ERR_OR_NULL(fh) && fh != (void *)fid)
821 pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n",
822 fh_len, fh_type, flags, err);
823 dentry = ERR_PTR(err);
827 static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid,
828 int fh_len, int fh_type)
830 pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n");
831 return ERR_PTR(-EACCES);
834 static int ovl_get_name(struct dentry *parent, char *name,
835 struct dentry *child)
838 * ovl_fh_to_dentry() returns connected dir overlay dentries and
839 * ovl_fh_to_parent() is not implemented, so we should not get here.
845 static struct dentry *ovl_get_parent(struct dentry *dentry)
848 * ovl_fh_to_dentry() returns connected dir overlay dentries, so we
849 * should not get here.
852 return ERR_PTR(-EIO);
855 const struct export_operations ovl_export_operations = {
856 .encode_fh = ovl_encode_fh,
857 .fh_to_dentry = ovl_fh_to_dentry,
858 .fh_to_parent = ovl_fh_to_parent,
859 .get_name = ovl_get_name,
860 .get_parent = ovl_get_parent,
863 /* encode_fh() encodes non-decodable file handles with nfs_export=off */
864 const struct export_operations ovl_export_fid_operations = {
865 .encode_fh = ovl_encode_fh,