1// SPDX-License-Identifier: GPL-2.0-or-later
2/* AFS dynamic root handling
3 *
4 * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/fs.h>
9#include <linux/namei.h>
10#include <linux/dns_resolver.h>
11#include "internal.h"
12
13static atomic_t afs_autocell_ino;
14
15/*
16 * iget5() comparator for inode created by autocell operations
17 *
18 * These pseudo inodes don't match anything.
19 */
20static int afs_iget5_pseudo_test(struct inode *inode, void *opaque)
21{
22	return 0;
23}
24
25/*
26 * iget5() inode initialiser
27 */
28static int afs_iget5_pseudo_set(struct inode *inode, void *opaque)
29{
30	struct afs_super_info *as = AFS_FS_S(inode->i_sb);
31	struct afs_vnode *vnode = AFS_FS_I(inode);
32	struct afs_fid *fid = opaque;
33
34	vnode->volume		= as->volume;
35	vnode->fid		= *fid;
36	inode->i_ino		= fid->vnode;
37	inode->i_generation	= fid->unique;
38	return 0;
39}
40
41/*
42 * Create an inode for a dynamic root directory or an autocell dynamic
43 * automount dir.
44 */
45struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
46{
47	struct afs_super_info *as = AFS_FS_S(sb);
48	struct afs_vnode *vnode;
49	struct inode *inode;
50	struct afs_fid fid = {};
51
52	_enter("");
53
54	if (as->volume)
55		fid.vid = as->volume->vid;
56	if (root) {
57		fid.vnode = 1;
58		fid.unique = 1;
59	} else {
60		fid.vnode = atomic_inc_return(&afs_autocell_ino);
61		fid.unique = 0;
62	}
63
64	inode = iget5_locked(sb, fid.vnode,
65			     afs_iget5_pseudo_test, afs_iget5_pseudo_set, &fid);
66	if (!inode) {
67		_leave(" = -ENOMEM");
68		return ERR_PTR(-ENOMEM);
69	}
70
71	_debug("GOT INODE %p { ino=%lu, vl=%llx, vn=%llx, u=%x }",
72	       inode, inode->i_ino, fid.vid, fid.vnode, fid.unique);
73
74	vnode = AFS_FS_I(inode);
75
76	/* there shouldn't be an existing inode */
77	BUG_ON(!(inode->i_state & I_NEW));
78
79	netfs_inode_init(&vnode->netfs, NULL, false);
80	inode->i_size		= 0;
81	inode->i_mode		= S_IFDIR | S_IRUGO | S_IXUGO;
82	if (root) {
83		inode->i_op	= &afs_dynroot_inode_operations;
84		inode->i_fop	= &simple_dir_operations;
85	} else {
86		inode->i_op	= &afs_autocell_inode_operations;
87	}
88	set_nlink(inode, 2);
89	inode->i_uid		= GLOBAL_ROOT_UID;
90	inode->i_gid		= GLOBAL_ROOT_GID;
91	simple_inode_init_ts(inode);
92	inode->i_blocks		= 0;
93	inode->i_generation	= 0;
94
95	set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
96	if (!root) {
97		set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
98		inode->i_flags |= S_AUTOMOUNT;
99	}
100
101	inode->i_flags |= S_NOATIME;
102	unlock_new_inode(inode);
103	_leave(" = %p", inode);
104	return inode;
105}
106
107/*
108 * Probe to see if a cell may exist.  This prevents positive dentries from
109 * being created unnecessarily.
110 */
111static int afs_probe_cell_name(struct dentry *dentry)
112{
113	struct afs_cell *cell;
114	struct afs_net *net = afs_d2net(dentry);
115	const char *name = dentry->d_name.name;
116	size_t len = dentry->d_name.len;
117	char *result = NULL;
118	int ret;
119
120	/* Names prefixed with a dot are R/W mounts. */
121	if (name[0] == '.') {
122		if (len == 1)
123			return -EINVAL;
124		name++;
125		len--;
126	}
127
128	cell = afs_find_cell(net, name, len, afs_cell_trace_use_probe);
129	if (!IS_ERR(cell)) {
130		afs_unuse_cell(net, cell, afs_cell_trace_unuse_probe);
131		return 0;
132	}
133
134	ret = dns_query(net->net, "afsdb", name, len, "srv=1",
135			&result, NULL, false);
136	if (ret == -ENODATA || ret == -ENOKEY || ret == 0)
137		ret = -ENOENT;
138	if (ret > 0 && ret >= sizeof(struct dns_server_list_v1_header)) {
139		struct dns_server_list_v1_header *v1 = (void *)result;
140
141		if (v1->hdr.zero == 0 &&
142		    v1->hdr.content == DNS_PAYLOAD_IS_SERVER_LIST &&
143		    v1->hdr.version == 1 &&
144		    (v1->status != DNS_LOOKUP_GOOD &&
145		     v1->status != DNS_LOOKUP_GOOD_WITH_BAD))
146			return -ENOENT;
147
148	}
149
150	kfree(result);
151	return ret;
152}
153
154/*
155 * Try to auto mount the mountpoint with pseudo directory, if the autocell
156 * operation is setted.
157 */
158struct inode *afs_try_auto_mntpt(struct dentry *dentry, struct inode *dir)
159{
160	struct afs_vnode *vnode = AFS_FS_I(dir);
161	struct inode *inode;
162	int ret = -ENOENT;
163
164	_enter("%p{%pd}, {%llx:%llu}",
165	       dentry, dentry, vnode->fid.vid, vnode->fid.vnode);
166
167	if (!test_bit(AFS_VNODE_AUTOCELL, &vnode->flags))
168		goto out;
169
170	ret = afs_probe_cell_name(dentry);
171	if (ret < 0)
172		goto out;
173
174	inode = afs_iget_pseudo_dir(dir->i_sb, false);
175	if (IS_ERR(inode)) {
176		ret = PTR_ERR(inode);
177		goto out;
178	}
179
180	_leave("= %p", inode);
181	return inode;
182
183out:
184	_leave("= %d", ret);
185	return ret == -ENOENT ? NULL : ERR_PTR(ret);
186}
187
188/*
189 * Look up @cell in a dynroot directory.  This is a substitution for the
190 * local cell name for the net namespace.
191 */
192static struct dentry *afs_lookup_atcell(struct dentry *dentry)
193{
194	struct afs_cell *cell;
195	struct afs_net *net = afs_d2net(dentry);
196	struct dentry *ret;
197	char *name;
198	int len;
199
200	if (!net->ws_cell)
201		return ERR_PTR(-ENOENT);
202
203	ret = ERR_PTR(-ENOMEM);
204	name = kmalloc(AFS_MAXCELLNAME + 1, GFP_KERNEL);
205	if (!name)
206		goto out_p;
207
208	down_read(&net->cells_lock);
209	cell = net->ws_cell;
210	if (cell) {
211		len = cell->name_len;
212		memcpy(name, cell->name, len + 1);
213	}
214	up_read(&net->cells_lock);
215
216	ret = ERR_PTR(-ENOENT);
217	if (!cell)
218		goto out_n;
219
220	ret = lookup_one_len(name, dentry->d_parent, len);
221
222	/* We don't want to d_add() the @cell dentry here as we don't want to
223	 * the cached dentry to hide changes to the local cell name.
224	 */
225
226out_n:
227	kfree(name);
228out_p:
229	return ret;
230}
231
232/*
233 * Look up an entry in a dynroot directory.
234 */
235static struct dentry *afs_dynroot_lookup(struct inode *dir, struct dentry *dentry,
236					 unsigned int flags)
237{
238	_enter("%pd", dentry);
239
240	ASSERTCMP(d_inode(dentry), ==, NULL);
241
242	if (flags & LOOKUP_CREATE)
243		return ERR_PTR(-EOPNOTSUPP);
244
245	if (dentry->d_name.len >= AFSNAMEMAX) {
246		_leave(" = -ENAMETOOLONG");
247		return ERR_PTR(-ENAMETOOLONG);
248	}
249
250	if (dentry->d_name.len == 5 &&
251	    memcmp(dentry->d_name.name, "@cell", 5) == 0)
252		return afs_lookup_atcell(dentry);
253
254	return d_splice_alias(afs_try_auto_mntpt(dentry, dir), dentry);
255}
256
257const struct inode_operations afs_dynroot_inode_operations = {
258	.lookup		= afs_dynroot_lookup,
259};
260
261const struct dentry_operations afs_dynroot_dentry_operations = {
262	.d_delete	= always_delete_dentry,
263	.d_release	= afs_d_release,
264	.d_automount	= afs_d_automount,
265};
266
267/*
268 * Create a manually added cell mount directory.
269 * - The caller must hold net->proc_cells_lock
270 */
271int afs_dynroot_mkdir(struct afs_net *net, struct afs_cell *cell)
272{
273	struct super_block *sb = net->dynroot_sb;
274	struct dentry *root, *subdir;
275	int ret;
276
277	if (!sb || atomic_read(&sb->s_active) == 0)
278		return 0;
279
280	/* Let the ->lookup op do the creation */
281	root = sb->s_root;
282	inode_lock(root->d_inode);
283	subdir = lookup_one_len(cell->name, root, cell->name_len);
284	if (IS_ERR(subdir)) {
285		ret = PTR_ERR(subdir);
286		goto unlock;
287	}
288
289	/* Note that we're retaining an extra ref on the dentry */
290	subdir->d_fsdata = (void *)1UL;
291	ret = 0;
292unlock:
293	inode_unlock(root->d_inode);
294	return ret;
295}
296
297/*
298 * Remove a manually added cell mount directory.
299 * - The caller must hold net->proc_cells_lock
300 */
301void afs_dynroot_rmdir(struct afs_net *net, struct afs_cell *cell)
302{
303	struct super_block *sb = net->dynroot_sb;
304	struct dentry *root, *subdir;
305
306	if (!sb || atomic_read(&sb->s_active) == 0)
307		return;
308
309	root = sb->s_root;
310	inode_lock(root->d_inode);
311
312	/* Don't want to trigger a lookup call, which will re-add the cell */
313	subdir = try_lookup_one_len(cell->name, root, cell->name_len);
314	if (IS_ERR_OR_NULL(subdir)) {
315		_debug("lookup %ld", PTR_ERR(subdir));
316		goto no_dentry;
317	}
318
319	_debug("rmdir %pd %u", subdir, d_count(subdir));
320
321	if (subdir->d_fsdata) {
322		_debug("unpin %u", d_count(subdir));
323		subdir->d_fsdata = NULL;
324		dput(subdir);
325	}
326	dput(subdir);
327no_dentry:
328	inode_unlock(root->d_inode);
329	_leave("");
330}
331
332/*
333 * Populate a newly created dynamic root with cell names.
334 */
335int afs_dynroot_populate(struct super_block *sb)
336{
337	struct afs_cell *cell;
338	struct afs_net *net = afs_sb2net(sb);
339	int ret;
340
341	mutex_lock(&net->proc_cells_lock);
342
343	net->dynroot_sb = sb;
344	hlist_for_each_entry(cell, &net->proc_cells, proc_link) {
345		ret = afs_dynroot_mkdir(net, cell);
346		if (ret < 0)
347			goto error;
348	}
349
350	ret = 0;
351out:
352	mutex_unlock(&net->proc_cells_lock);
353	return ret;
354
355error:
356	net->dynroot_sb = NULL;
357	goto out;
358}
359
360/*
361 * When a dynamic root that's in the process of being destroyed, depopulate it
362 * of pinned directories.
363 */
364void afs_dynroot_depopulate(struct super_block *sb)
365{
366	struct afs_net *net = afs_sb2net(sb);
367	struct dentry *root = sb->s_root, *subdir;
368
369	/* Prevent more subdirs from being created */
370	mutex_lock(&net->proc_cells_lock);
371	if (net->dynroot_sb == sb)
372		net->dynroot_sb = NULL;
373	mutex_unlock(&net->proc_cells_lock);
374
375	if (root) {
376		struct hlist_node *n;
377		inode_lock(root->d_inode);
378
379		/* Remove all the pins for dirs created for manually added cells */
380		hlist_for_each_entry_safe(subdir, n, &root->d_children, d_sib) {
381			if (subdir->d_fsdata) {
382				subdir->d_fsdata = NULL;
383				dput(subdir);
384			}
385		}
386
387		inode_unlock(root->d_inode);
388	}
389}
390