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_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)
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 oe->lowerstack[0].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;
111 if (WARN_ON(dentry == dentry->d_sb->s_root) ||
112 WARN_ON(!ovl_dentry_lower(dentry)))
115 origin_layer = OVL_E(dentry)->lowerstack[0].layer->idx;
116 if (ovl_dentry_test_flag(OVL_E_CONNECTED, dentry))
119 /* Find the topmost origin layer connectable ancestor of @dentry */
122 parent = dget_parent(next);
123 if (WARN_ON(parent == next)) {
129 * If @parent is not origin layer connectable, then copy up
130 * @next which is origin layer connectable and we are done.
132 if (ovl_connectable_layer(parent) < origin_layer) {
133 err = ovl_encode_maybe_copy_up(next);
137 /* If @parent is connected or indexed we are done */
138 if (ovl_dentry_test_flag(OVL_E_CONNECTED, parent) ||
139 ovl_test_flag(OVL_INDEX, d_inode(parent)))
150 ovl_dentry_set_flag(OVL_E_CONNECTED, dentry);
152 return err ?: origin_layer;
156 * We only need to encode origin if there is a chance that the same object was
157 * encoded pre copy up and then we need to stay consistent with the same
158 * encoding also after copy up. If non-pure upper is not indexed, then it was
159 * copied up before NFS export was enabled. In that case we don't need to worry
160 * about staying consistent with pre copy up encoding and we encode an upper
161 * file handle. Overlay root dentry is a private case of non-indexed upper.
163 * The following table summarizes the different file handle encodings used for
164 * different overlay object types:
166 * Object type | Encoding
167 * --------------------------------
169 * Non-indexed upper | U
170 * Indexed upper | L (*)
173 * U = upper file handle
174 * L = lower file handle
176 * (*) Connecting an overlay dir from real lower dentry is not always
177 * possible when there are redirects in lower layers and non-indexed merge dirs.
178 * To mitigate those case, we may copy up the lower dir ancestor before encode
179 * a lower dir file handle.
181 * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error.
183 static int ovl_check_encode_origin(struct dentry *dentry)
185 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
187 /* Upper file handle for pure upper */
188 if (!ovl_dentry_lower(dentry))
192 * Upper file handle for non-indexed upper.
194 * Root is never indexed, so if there's an upper layer, encode upper for
197 if (ovl_dentry_upper(dentry) &&
198 !ovl_test_flag(OVL_INDEX, d_inode(dentry)))
202 * Decoding a merge dir, whose origin's ancestor is under a redirected
203 * lower dir or under a non-indexed upper is not always possible.
204 * ovl_connect_layer() will try to make origin's layer "connected" by
205 * copying up a "connectable" ancestor.
207 if (d_is_dir(dentry) && ovl_upper_mnt(ofs))
208 return ovl_connect_layer(dentry);
210 /* Lower file handle for indexed and non-upper dir/non-dir */
214 static int ovl_dentry_to_fid(struct dentry *dentry, u32 *fid, int buflen)
216 struct ovl_fh *fh = NULL;
221 * Check if we should encode a lower or upper file handle and maybe
222 * copy up an ancestor to make lower file handle connectable.
224 err = enc_lower = ovl_check_encode_origin(dentry);
228 /* Encode an upper or lower file handle */
229 fh = ovl_encode_real_fh(enc_lower ? ovl_dentry_lower(dentry) :
230 ovl_dentry_upper(dentry), !enc_lower);
234 len = OVL_FH_LEN(fh);
236 memcpy(fid, fh, len);
244 pr_warn_ratelimited("failed to encode file handle (%pd2, err=%i)\n",
249 static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len,
250 struct inode *parent)
252 struct dentry *dentry;
253 int bytes, buflen = *max_len << 2;
255 /* TODO: encode connectable file handles */
257 return FILEID_INVALID;
259 dentry = d_find_any_alias(inode);
261 return FILEID_INVALID;
263 bytes = ovl_dentry_to_fid(dentry, fid, buflen);
266 return FILEID_INVALID;
268 *max_len = bytes >> 2;
270 return FILEID_INVALID;
272 return OVL_FILEID_V1;
276 * Find or instantiate an overlay dentry from real dentries and index.
278 static struct dentry *ovl_obtain_alias(struct super_block *sb,
279 struct dentry *upper_alias,
280 struct ovl_path *lowerpath,
281 struct dentry *index)
283 struct dentry *lower = lowerpath ? lowerpath->dentry : NULL;
284 struct dentry *upper = upper_alias ?: index;
285 struct dentry *dentry;
287 struct ovl_entry *oe;
288 struct ovl_inode_params oip = {
289 .lowerpath = lowerpath,
294 /* We get overlay directory dentries with ovl_lookup_real() */
295 if (d_is_dir(upper ?: lower))
296 return ERR_PTR(-EIO);
298 oip.upperdentry = dget(upper);
299 inode = ovl_get_inode(sb, &oip);
302 return ERR_CAST(inode);
306 ovl_set_flag(OVL_UPPERDATA, inode);
308 dentry = d_find_any_alias(inode);
312 dentry = d_alloc_anon(inode->i_sb);
313 if (unlikely(!dentry))
315 oe = ovl_alloc_entry(lower ? 1 : 0);
320 oe->lowerstack->dentry = dget(lower);
321 oe->lowerstack->layer = lowerpath->layer;
323 dentry->d_fsdata = oe;
325 ovl_dentry_set_upper_alias(dentry);
327 ovl_dentry_init_reval(dentry, upper);
329 return d_instantiate_anon(dentry, inode);
333 dentry = ERR_PTR(-ENOMEM);
339 /* Get the upper or lower dentry in stach whose on layer @idx */
340 static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx)
342 struct ovl_entry *oe = dentry->d_fsdata;
346 return ovl_dentry_upper(dentry);
348 for (i = 0; i < oe->numlower; i++) {
349 if (oe->lowerstack[i].layer->idx == idx)
350 return oe->lowerstack[i].dentry;
357 * Lookup a child overlay dentry to get a connected overlay dentry whose real
358 * dentry is @real. If @real is on upper layer, we lookup a child overlay
359 * dentry with the same name as the real dentry. Otherwise, we need to consult
362 static struct dentry *ovl_lookup_real_one(struct dentry *connected,
364 const struct ovl_layer *layer)
366 struct inode *dir = d_inode(connected);
367 struct dentry *this, *parent = NULL;
368 struct name_snapshot name;
372 * Lookup child overlay dentry by real name. The dir mutex protects us
373 * from racing with overlay rename. If the overlay dentry that is above
374 * real has already been moved to a parent that is not under the
375 * connected overlay dir, we return -ECHILD and restart the lookup of
376 * connected real path from the top.
378 inode_lock_nested(dir, I_MUTEX_PARENT);
380 parent = dget_parent(real);
381 if (ovl_dentry_real_at(connected, layer->idx) != parent)
385 * We also need to take a snapshot of real dentry name to protect us
386 * from racing with underlying layer rename. In this case, we don't
387 * care about returning ESTALE, only from dereferencing a free name
388 * pointer because we hold no lock on the real dentry.
390 take_dentry_name_snapshot(&name, real);
391 this = lookup_one_len(name.name.name, connected, name.name.len);
392 release_dentry_name_snapshot(&name);
396 } else if (!this || !this->d_inode) {
400 } else if (ovl_dentry_real_at(this, layer->idx) != real) {
412 pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
413 real, layer->idx, connected, err);
418 static struct dentry *ovl_lookup_real(struct super_block *sb,
420 const struct ovl_layer *layer);
423 * Lookup an indexed or hashed overlay dentry by real inode.
425 static struct dentry *ovl_lookup_real_inode(struct super_block *sb,
427 const struct ovl_layer *layer)
429 struct ovl_fs *ofs = sb->s_fs_info;
430 struct dentry *index = NULL;
431 struct dentry *this = NULL;
435 * Decoding upper dir from index is expensive, so first try to lookup
436 * overlay dentry in inode/dcache.
438 inode = ovl_lookup_inode(sb, real, !layer->idx);
440 return ERR_CAST(inode);
442 this = d_find_any_alias(inode);
447 * For decoded lower dir file handle, lookup index by origin to check
448 * if lower dir was copied up and and/or removed.
450 if (!this && layer->idx && ofs->indexdir && !WARN_ON(!d_is_dir(real))) {
451 index = ovl_lookup_index(ofs, NULL, real, false);
456 /* Get connected upper overlay dir from index */
458 struct dentry *upper = ovl_index_upper(ofs, index);
461 if (IS_ERR_OR_NULL(upper))
465 * ovl_lookup_real() in lower layer may call recursively once to
466 * ovl_lookup_real() in upper layer. The first level call walks
467 * back lower parents to the topmost indexed parent. The second
468 * recursive call walks back from indexed upper to the topmost
469 * connected/hashed upper parent (or up to root).
471 this = ovl_lookup_real(sb, upper, &ofs->layers[0]);
475 if (IS_ERR_OR_NULL(this))
478 if (ovl_dentry_real_at(this, layer->idx) != real) {
480 this = ERR_PTR(-EIO);
487 * Lookup an indexed or hashed overlay dentry, whose real dentry is an
490 static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb,
492 const struct ovl_layer *layer)
494 struct dentry *next, *parent = NULL;
495 struct dentry *ancestor = ERR_PTR(-EIO);
497 if (real == layer->mnt->mnt_root)
498 return dget(sb->s_root);
500 /* Find the topmost indexed or hashed ancestor */
503 parent = dget_parent(next);
506 * Lookup a matching overlay dentry in inode/dentry
507 * cache or in index by real inode.
509 ancestor = ovl_lookup_real_inode(sb, next, layer);
513 if (parent == layer->mnt->mnt_root) {
514 ancestor = dget(sb->s_root);
519 * If @real has been moved out of the layer root directory,
520 * we will eventully hit the real fs root. This cannot happen
521 * by legit overlay rename, so we return error in that case.
523 if (parent == next) {
524 ancestor = ERR_PTR(-EXDEV);
539 * Lookup a connected overlay dentry whose real dentry is @real.
540 * If @real is on upper layer, we lookup a child overlay dentry with the same
541 * path the real dentry. Otherwise, we need to consult index for lookup.
543 static struct dentry *ovl_lookup_real(struct super_block *sb,
545 const struct ovl_layer *layer)
547 struct dentry *connected;
550 connected = ovl_lookup_real_ancestor(sb, real, layer);
551 if (IS_ERR(connected))
555 struct dentry *next, *this;
556 struct dentry *parent = NULL;
557 struct dentry *real_connected = ovl_dentry_real_at(connected,
560 if (real_connected == real)
563 /* Find the topmost dentry not yet connected */
566 parent = dget_parent(next);
568 if (parent == real_connected)
572 * If real has been moved out of 'real_connected',
573 * we will not find 'real_connected' and hit the layer
574 * root. In that case, we need to restart connecting.
575 * This game can go on forever in the worst case. We
576 * may want to consider taking s_vfs_rename_mutex if
577 * this happens more than once.
579 if (parent == layer->mnt->mnt_root) {
581 connected = dget(sb->s_root);
586 * If real file has been moved out of the layer root
587 * directory, we will eventully hit the real fs root.
588 * This cannot happen by legit overlay rename, so we
589 * return error in that case.
591 if (parent == next) {
601 this = ovl_lookup_real_one(connected, next, layer);
606 * Lookup of child in overlay can fail when racing with
607 * overlay rename of child away from 'connected' parent.
608 * In this case, we need to restart the lookup from the
609 * top, because we cannot trust that 'real_connected' is
610 * still an ancestor of 'real'. There is a good chance
611 * that the renamed overlay ancestor is now in cache, so
612 * ovl_lookup_real_ancestor() will find it and we can
613 * continue to connect exactly from where lookup failed.
615 if (err == -ECHILD) {
616 this = ovl_lookup_real_ancestor(sb, real,
618 err = PTR_ERR_OR_ZERO(this);
636 pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
637 real, layer->idx, connected, err);
643 * Get an overlay dentry from upper/lower real dentries and index.
645 static struct dentry *ovl_get_dentry(struct super_block *sb,
646 struct dentry *upper,
647 struct ovl_path *lowerpath,
648 struct dentry *index)
650 struct ovl_fs *ofs = sb->s_fs_info;
651 const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer;
652 struct dentry *real = upper ?: (index ?: lowerpath->dentry);
655 * Obtain a disconnected overlay dentry from a non-dir real dentry
659 return ovl_obtain_alias(sb, upper, lowerpath, index);
661 /* Removed empty directory? */
662 if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(real))
663 return ERR_PTR(-ENOENT);
666 * If real dentry is connected and hashed, get a connected overlay
667 * dentry whose real dentry is @real.
669 return ovl_lookup_real(sb, real, layer);
672 static struct dentry *ovl_upper_fh_to_d(struct super_block *sb,
675 struct ovl_fs *ofs = sb->s_fs_info;
676 struct dentry *dentry;
677 struct dentry *upper;
679 if (!ovl_upper_mnt(ofs))
680 return ERR_PTR(-EACCES);
682 upper = ovl_decode_real_fh(fh, ovl_upper_mnt(ofs), true);
683 if (IS_ERR_OR_NULL(upper))
686 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
692 static struct dentry *ovl_lower_fh_to_d(struct super_block *sb,
695 struct ovl_fs *ofs = sb->s_fs_info;
696 struct ovl_path origin = { };
697 struct ovl_path *stack = &origin;
698 struct dentry *dentry = NULL;
699 struct dentry *index = NULL;
703 /* First lookup overlay inode in inode cache by origin fh */
704 err = ovl_check_origin_fh(ofs, fh, false, NULL, &stack);
708 if (!d_is_dir(origin.dentry) ||
709 !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) {
710 inode = ovl_lookup_inode(sb, origin.dentry, false);
711 err = PTR_ERR(inode);
715 dentry = d_find_any_alias(inode);
722 /* Then lookup indexed upper/whiteout by origin fh */
724 index = ovl_get_index_fh(ofs, fh);
725 err = PTR_ERR(index);
732 /* Then try to get a connected upper dir by index */
733 if (index && d_is_dir(index)) {
734 struct dentry *upper = ovl_index_upper(ofs, index);
736 err = PTR_ERR(upper);
737 if (IS_ERR_OR_NULL(upper))
740 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
745 /* Find origin.dentry again with ovl_acceptable() layer check */
746 if (d_is_dir(origin.dentry)) {
748 origin.dentry = NULL;
749 err = ovl_check_origin_fh(ofs, fh, true, NULL, &stack);
754 err = ovl_verify_origin(ofs, index, origin.dentry, false);
759 /* Get a connected non-upper dir or disconnected non-dir */
760 dentry = ovl_get_dentry(sb, NULL, &origin, index);
768 dentry = ERR_PTR(err);
772 static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type)
776 /* If on-wire inner fid is aligned - nothing to do */
777 if (fh_type == OVL_FILEID_V1)
778 return (struct ovl_fh *)fid;
780 if (fh_type != OVL_FILEID_V0)
781 return ERR_PTR(-EINVAL);
783 if (buflen <= OVL_FH_WIRE_OFFSET)
784 return ERR_PTR(-EINVAL);
786 fh = kzalloc(buflen, GFP_KERNEL);
788 return ERR_PTR(-ENOMEM);
790 /* Copy unaligned inner fh into aligned buffer */
791 memcpy(&fh->fb, fid, buflen - OVL_FH_WIRE_OFFSET);
795 static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid,
796 int fh_len, int fh_type)
798 struct dentry *dentry = NULL;
799 struct ovl_fh *fh = NULL;
800 int len = fh_len << 2;
801 unsigned int flags = 0;
804 fh = ovl_fid_to_fh(fid, len, fh_type);
809 err = ovl_check_fh_len(fh, len);
813 flags = fh->fb.flags;
814 dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ?
815 ovl_upper_fh_to_d(sb, fh) :
816 ovl_lower_fh_to_d(sb, fh);
817 err = PTR_ERR(dentry);
818 if (IS_ERR(dentry) && err != -ESTALE)
822 /* We may have needed to re-align OVL_FILEID_V0 */
823 if (!IS_ERR_OR_NULL(fh) && fh != (void *)fid)
829 pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n",
830 fh_len, fh_type, flags, err);
831 dentry = ERR_PTR(err);
835 static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid,
836 int fh_len, int fh_type)
838 pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n");
839 return ERR_PTR(-EACCES);
842 static int ovl_get_name(struct dentry *parent, char *name,
843 struct dentry *child)
846 * ovl_fh_to_dentry() returns connected dir overlay dentries and
847 * ovl_fh_to_parent() is not implemented, so we should not get here.
853 static struct dentry *ovl_get_parent(struct dentry *dentry)
856 * ovl_fh_to_dentry() returns connected dir overlay dentries, so we
857 * should not get here.
860 return ERR_PTR(-EIO);
863 const struct export_operations ovl_export_operations = {
864 .encode_fh = ovl_encode_fh,
865 .fh_to_dentry = ovl_fh_to_dentry,
866 .fh_to_parent = ovl_fh_to_parent,
867 .get_name = ovl_get_name,
868 .get_parent = ovl_get_parent,