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 ovl_fs *ofs, struct dentry *dentry,
215 u32 *fid, int buflen)
217 struct ovl_fh *fh = NULL;
222 * Check if we should encode a lower or upper file handle and maybe
223 * copy up an ancestor to make lower file handle connectable.
225 err = enc_lower = ovl_check_encode_origin(dentry);
229 /* Encode an upper or lower file handle */
230 fh = ovl_encode_real_fh(ofs, enc_lower ? ovl_dentry_lower(dentry) :
231 ovl_dentry_upper(dentry), !enc_lower);
235 len = OVL_FH_LEN(fh);
237 memcpy(fid, fh, len);
245 pr_warn_ratelimited("failed to encode file handle (%pd2, err=%i)\n",
250 static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len,
251 struct inode *parent)
253 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
254 struct dentry *dentry;
255 int bytes, buflen = *max_len << 2;
257 /* TODO: encode connectable file handles */
259 return FILEID_INVALID;
261 dentry = d_find_any_alias(inode);
263 return FILEID_INVALID;
265 bytes = ovl_dentry_to_fid(ofs, dentry, fid, buflen);
268 return FILEID_INVALID;
270 *max_len = bytes >> 2;
272 return FILEID_INVALID;
274 return OVL_FILEID_V1;
278 * Find or instantiate an overlay dentry from real dentries and index.
280 static struct dentry *ovl_obtain_alias(struct super_block *sb,
281 struct dentry *upper_alias,
282 struct ovl_path *lowerpath,
283 struct dentry *index)
285 struct dentry *lower = lowerpath ? lowerpath->dentry : NULL;
286 struct dentry *upper = upper_alias ?: index;
287 struct dentry *dentry;
289 struct ovl_entry *oe;
290 struct ovl_inode_params oip = {
291 .lowerpath = lowerpath,
296 /* We get overlay directory dentries with ovl_lookup_real() */
297 if (d_is_dir(upper ?: lower))
298 return ERR_PTR(-EIO);
300 oip.upperdentry = dget(upper);
301 inode = ovl_get_inode(sb, &oip);
304 return ERR_CAST(inode);
308 ovl_set_flag(OVL_UPPERDATA, inode);
310 dentry = d_find_any_alias(inode);
314 dentry = d_alloc_anon(inode->i_sb);
315 if (unlikely(!dentry))
317 oe = ovl_alloc_entry(lower ? 1 : 0);
322 oe->lowerstack->dentry = dget(lower);
323 oe->lowerstack->layer = lowerpath->layer;
325 dentry->d_fsdata = oe;
327 ovl_dentry_set_upper_alias(dentry);
329 ovl_dentry_update_reval(dentry, upper,
330 DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
332 return d_instantiate_anon(dentry, inode);
336 dentry = ERR_PTR(-ENOMEM);
342 /* Get the upper or lower dentry in stach whose on layer @idx */
343 static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx)
345 struct ovl_entry *oe = dentry->d_fsdata;
349 return ovl_dentry_upper(dentry);
351 for (i = 0; i < oe->numlower; i++) {
352 if (oe->lowerstack[i].layer->idx == idx)
353 return oe->lowerstack[i].dentry;
360 * Lookup a child overlay dentry to get a connected overlay dentry whose real
361 * dentry is @real. If @real is on upper layer, we lookup a child overlay
362 * dentry with the same name as the real dentry. Otherwise, we need to consult
365 static struct dentry *ovl_lookup_real_one(struct dentry *connected,
367 const struct ovl_layer *layer)
369 struct inode *dir = d_inode(connected);
370 struct dentry *this, *parent = NULL;
371 struct name_snapshot name;
375 * Lookup child overlay dentry by real name. The dir mutex protects us
376 * from racing with overlay rename. If the overlay dentry that is above
377 * real has already been moved to a parent that is not under the
378 * connected overlay dir, we return -ECHILD and restart the lookup of
379 * connected real path from the top.
381 inode_lock_nested(dir, I_MUTEX_PARENT);
383 parent = dget_parent(real);
384 if (ovl_dentry_real_at(connected, layer->idx) != parent)
388 * We also need to take a snapshot of real dentry name to protect us
389 * from racing with underlying layer rename. In this case, we don't
390 * care about returning ESTALE, only from dereferencing a free name
391 * pointer because we hold no lock on the real dentry.
393 take_dentry_name_snapshot(&name, real);
395 * No mnt_userns handling here: it's an internal lookup. Could skip
396 * permission checking altogether, but for now just use non-mnt_userns
399 this = lookup_one_len(name.name.name, connected, name.name.len);
400 release_dentry_name_snapshot(&name);
404 } else if (!this || !this->d_inode) {
408 } else if (ovl_dentry_real_at(this, layer->idx) != real) {
420 pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
421 real, layer->idx, connected, err);
426 static struct dentry *ovl_lookup_real(struct super_block *sb,
428 const struct ovl_layer *layer);
431 * Lookup an indexed or hashed overlay dentry by real inode.
433 static struct dentry *ovl_lookup_real_inode(struct super_block *sb,
435 const struct ovl_layer *layer)
437 struct ovl_fs *ofs = sb->s_fs_info;
438 struct dentry *index = NULL;
439 struct dentry *this = NULL;
443 * Decoding upper dir from index is expensive, so first try to lookup
444 * overlay dentry in inode/dcache.
446 inode = ovl_lookup_inode(sb, real, !layer->idx);
448 return ERR_CAST(inode);
450 this = d_find_any_alias(inode);
455 * For decoded lower dir file handle, lookup index by origin to check
456 * if lower dir was copied up and and/or removed.
458 if (!this && layer->idx && ofs->indexdir && !WARN_ON(!d_is_dir(real))) {
459 index = ovl_lookup_index(ofs, NULL, real, false);
464 /* Get connected upper overlay dir from index */
466 struct dentry *upper = ovl_index_upper(ofs, index);
469 if (IS_ERR_OR_NULL(upper))
473 * ovl_lookup_real() in lower layer may call recursively once to
474 * ovl_lookup_real() in upper layer. The first level call walks
475 * back lower parents to the topmost indexed parent. The second
476 * recursive call walks back from indexed upper to the topmost
477 * connected/hashed upper parent (or up to root).
479 this = ovl_lookup_real(sb, upper, &ofs->layers[0]);
483 if (IS_ERR_OR_NULL(this))
486 if (ovl_dentry_real_at(this, layer->idx) != real) {
488 this = ERR_PTR(-EIO);
495 * Lookup an indexed or hashed overlay dentry, whose real dentry is an
498 static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb,
500 const struct ovl_layer *layer)
502 struct dentry *next, *parent = NULL;
503 struct dentry *ancestor = ERR_PTR(-EIO);
505 if (real == layer->mnt->mnt_root)
506 return dget(sb->s_root);
508 /* Find the topmost indexed or hashed ancestor */
511 parent = dget_parent(next);
514 * Lookup a matching overlay dentry in inode/dentry
515 * cache or in index by real inode.
517 ancestor = ovl_lookup_real_inode(sb, next, layer);
521 if (parent == layer->mnt->mnt_root) {
522 ancestor = dget(sb->s_root);
527 * If @real has been moved out of the layer root directory,
528 * we will eventully hit the real fs root. This cannot happen
529 * by legit overlay rename, so we return error in that case.
531 if (parent == next) {
532 ancestor = ERR_PTR(-EXDEV);
547 * Lookup a connected overlay dentry whose real dentry is @real.
548 * If @real is on upper layer, we lookup a child overlay dentry with the same
549 * path the real dentry. Otherwise, we need to consult index for lookup.
551 static struct dentry *ovl_lookup_real(struct super_block *sb,
553 const struct ovl_layer *layer)
555 struct dentry *connected;
558 connected = ovl_lookup_real_ancestor(sb, real, layer);
559 if (IS_ERR(connected))
563 struct dentry *next, *this;
564 struct dentry *parent = NULL;
565 struct dentry *real_connected = ovl_dentry_real_at(connected,
568 if (real_connected == real)
571 /* Find the topmost dentry not yet connected */
574 parent = dget_parent(next);
576 if (parent == real_connected)
580 * If real has been moved out of 'real_connected',
581 * we will not find 'real_connected' and hit the layer
582 * root. In that case, we need to restart connecting.
583 * This game can go on forever in the worst case. We
584 * may want to consider taking s_vfs_rename_mutex if
585 * this happens more than once.
587 if (parent == layer->mnt->mnt_root) {
589 connected = dget(sb->s_root);
594 * If real file has been moved out of the layer root
595 * directory, we will eventully hit the real fs root.
596 * This cannot happen by legit overlay rename, so we
597 * return error in that case.
599 if (parent == next) {
609 this = ovl_lookup_real_one(connected, next, layer);
614 * Lookup of child in overlay can fail when racing with
615 * overlay rename of child away from 'connected' parent.
616 * In this case, we need to restart the lookup from the
617 * top, because we cannot trust that 'real_connected' is
618 * still an ancestor of 'real'. There is a good chance
619 * that the renamed overlay ancestor is now in cache, so
620 * ovl_lookup_real_ancestor() will find it and we can
621 * continue to connect exactly from where lookup failed.
623 if (err == -ECHILD) {
624 this = ovl_lookup_real_ancestor(sb, real,
626 err = PTR_ERR_OR_ZERO(this);
644 pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n",
645 real, layer->idx, connected, err);
651 * Get an overlay dentry from upper/lower real dentries and index.
653 static struct dentry *ovl_get_dentry(struct super_block *sb,
654 struct dentry *upper,
655 struct ovl_path *lowerpath,
656 struct dentry *index)
658 struct ovl_fs *ofs = sb->s_fs_info;
659 const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer;
660 struct dentry *real = upper ?: (index ?: lowerpath->dentry);
663 * Obtain a disconnected overlay dentry from a non-dir real dentry
667 return ovl_obtain_alias(sb, upper, lowerpath, index);
669 /* Removed empty directory? */
670 if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(real))
671 return ERR_PTR(-ENOENT);
674 * If real dentry is connected and hashed, get a connected overlay
675 * dentry whose real dentry is @real.
677 return ovl_lookup_real(sb, real, layer);
680 static struct dentry *ovl_upper_fh_to_d(struct super_block *sb,
683 struct ovl_fs *ofs = sb->s_fs_info;
684 struct dentry *dentry;
685 struct dentry *upper;
687 if (!ovl_upper_mnt(ofs))
688 return ERR_PTR(-EACCES);
690 upper = ovl_decode_real_fh(ofs, fh, ovl_upper_mnt(ofs), true);
691 if (IS_ERR_OR_NULL(upper))
694 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
700 static struct dentry *ovl_lower_fh_to_d(struct super_block *sb,
703 struct ovl_fs *ofs = sb->s_fs_info;
704 struct ovl_path origin = { };
705 struct ovl_path *stack = &origin;
706 struct dentry *dentry = NULL;
707 struct dentry *index = NULL;
711 /* First lookup overlay inode in inode cache by origin fh */
712 err = ovl_check_origin_fh(ofs, fh, false, NULL, &stack);
716 if (!d_is_dir(origin.dentry) ||
717 !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) {
718 inode = ovl_lookup_inode(sb, origin.dentry, false);
719 err = PTR_ERR(inode);
723 dentry = d_find_any_alias(inode);
730 /* Then lookup indexed upper/whiteout by origin fh */
732 index = ovl_get_index_fh(ofs, fh);
733 err = PTR_ERR(index);
740 /* Then try to get a connected upper dir by index */
741 if (index && d_is_dir(index)) {
742 struct dentry *upper = ovl_index_upper(ofs, index);
744 err = PTR_ERR(upper);
745 if (IS_ERR_OR_NULL(upper))
748 dentry = ovl_get_dentry(sb, upper, NULL, NULL);
753 /* Find origin.dentry again with ovl_acceptable() layer check */
754 if (d_is_dir(origin.dentry)) {
756 origin.dentry = NULL;
757 err = ovl_check_origin_fh(ofs, fh, true, NULL, &stack);
762 err = ovl_verify_origin(ofs, index, origin.dentry, false);
767 /* Get a connected non-upper dir or disconnected non-dir */
768 dentry = ovl_get_dentry(sb, NULL, &origin, index);
776 dentry = ERR_PTR(err);
780 static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type)
784 /* If on-wire inner fid is aligned - nothing to do */
785 if (fh_type == OVL_FILEID_V1)
786 return (struct ovl_fh *)fid;
788 if (fh_type != OVL_FILEID_V0)
789 return ERR_PTR(-EINVAL);
791 if (buflen <= OVL_FH_WIRE_OFFSET)
792 return ERR_PTR(-EINVAL);
794 fh = kzalloc(buflen, GFP_KERNEL);
796 return ERR_PTR(-ENOMEM);
798 /* Copy unaligned inner fh into aligned buffer */
799 memcpy(&fh->fb, fid, buflen - OVL_FH_WIRE_OFFSET);
803 static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid,
804 int fh_len, int fh_type)
806 struct dentry *dentry = NULL;
807 struct ovl_fh *fh = NULL;
808 int len = fh_len << 2;
809 unsigned int flags = 0;
812 fh = ovl_fid_to_fh(fid, len, fh_type);
817 err = ovl_check_fh_len(fh, len);
821 flags = fh->fb.flags;
822 dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ?
823 ovl_upper_fh_to_d(sb, fh) :
824 ovl_lower_fh_to_d(sb, fh);
825 err = PTR_ERR(dentry);
826 if (IS_ERR(dentry) && err != -ESTALE)
830 /* We may have needed to re-align OVL_FILEID_V0 */
831 if (!IS_ERR_OR_NULL(fh) && fh != (void *)fid)
837 pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n",
838 fh_len, fh_type, flags, err);
839 dentry = ERR_PTR(err);
843 static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid,
844 int fh_len, int fh_type)
846 pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n");
847 return ERR_PTR(-EACCES);
850 static int ovl_get_name(struct dentry *parent, char *name,
851 struct dentry *child)
854 * ovl_fh_to_dentry() returns connected dir overlay dentries and
855 * ovl_fh_to_parent() is not implemented, so we should not get here.
861 static struct dentry *ovl_get_parent(struct dentry *dentry)
864 * ovl_fh_to_dentry() returns connected dir overlay dentries, so we
865 * should not get here.
868 return ERR_PTR(-EIO);
871 const struct export_operations ovl_export_operations = {
872 .encode_fh = ovl_encode_fh,
873 .fh_to_dentry = ovl_fh_to_dentry,
874 .fh_to_parent = ovl_fh_to_parent,
875 .get_name = ovl_get_name,
876 .get_parent = ovl_get_parent,