projects/linux·blob master · refs

482 lines · 12.2 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * fs/kernfs/mount.c - kernfs mount implementation
4 *
5 * Copyright (c) 2001-3 Patrick Mochel
6 * Copyright (c) 2007 SUSE Linux Products GmbH
7 * Copyright (c) 2007, 2013 Tejun Heo <[email protected]>
8 */
9
10#include <linux/fs.h>
11#include <linux/mount.h>
12#include <linux/init.h>
13#include <linux/magic.h>
14#include <linux/slab.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
17#include <linux/seq_file.h>
18#include <linux/exportfs.h>
19#include <linux/uuid.h>
20#include <linux/statfs.h>
21
22#include "kernfs-internal.h"
23
24struct kmem_cache *kernfs_node_cache __ro_after_init;
25struct kmem_cache *kernfs_iattrs_cache __ro_after_init;
26struct kernfs_global_locks *kernfs_locks __ro_after_init;
27
28static int kernfs_sop_show_options(struct seq_file *sf, struct dentry *dentry)
29{
30 struct kernfs_root *root = kernfs_root(kernfs_dentry_node(dentry));
31 struct kernfs_syscall_ops *scops = root->syscall_ops;
32
33 if (scops && scops->show_options)
34 return scops->show_options(sf, root);
35 return 0;
36}
37
38static int kernfs_sop_show_path(struct seq_file *sf, struct dentry *dentry)
39{
40 struct kernfs_node *node = kernfs_dentry_node(dentry);
41 struct kernfs_root *root = kernfs_root(node);
42 struct kernfs_syscall_ops *scops = root->syscall_ops;
43
44 if (scops && scops->show_path)
45 return scops->show_path(sf, node, root);
46
47 seq_dentry(sf, dentry, " \t\n\\");
48 return 0;
49}
50
51static int kernfs_statfs(struct dentry *dentry, struct kstatfs *buf)
52{
53 simple_statfs(dentry, buf);
54 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
55 return 0;
56}
57
58const struct super_operations kernfs_sops = {
59 .statfs = kernfs_statfs,
60 .drop_inode = inode_just_drop,
61 .evict_inode = kernfs_evict_inode,
62
63 .show_options = kernfs_sop_show_options,
64 .show_path = kernfs_sop_show_path,
65
66 /*
67 * sysfs is built on top of kernfs and sysfs provides the power
68 * management infrastructure to support suspend/hibernate by
69 * writing to various files in /sys/power/. As filesystems may
70 * be automatically frozen during suspend/hibernate implementing
71 * freeze/thaw support for kernfs generically will cause
72 * deadlocks as the suspending/hibernation initiating task will
73 * hold a VFS lock that it will then wait upon to be released.
74 * If freeze/thaw for kernfs is needed talk to the VFS.
75 */
76 .freeze_fs = NULL,
77 .unfreeze_fs = NULL,
78 .freeze_super = NULL,
79 .thaw_super = NULL,
80};
81
82static int kernfs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
83 struct inode *parent)
84{
85 struct kernfs_node *kn = inode->i_private;
86
87 if (*max_len < 2) {
88 *max_len = 2;
89 return FILEID_INVALID;
90 }
91
92 *max_len = 2;
93 *(u64 *)fh = kn->id;
94 return FILEID_KERNFS;
95}
96
97static struct dentry *__kernfs_fh_to_dentry(struct super_block *sb,
98 struct fid *fid, int fh_len,
99 int fh_type, bool get_parent)
100{
101 struct kernfs_super_info *info = kernfs_info(sb);
102 struct kernfs_node *kn;
103 struct inode *inode;
104 u64 id;
105
106 if (fh_len < 2)
107 return NULL;
108
109 switch (fh_type) {
110 case FILEID_KERNFS:
111 id = *(u64 *)fid;
112 break;
113 case FILEID_INO32_GEN:
114 case FILEID_INO32_GEN_PARENT:
115 /*
116 * blk_log_action() exposes "LOW32,HIGH32" pair without
117 * type and userland can call us with generic fid
118 * constructed from them. Combine it back to ID. See
119 * blk_log_action().
120 */
121 id = ((u64)fid->i32.gen << 32) | fid->i32.ino;
122 break;
123 default:
124 return NULL;
125 }
126
127 /*
128 * Hold kernfs_rwsem across the lookup as well as kernfs_get_inode().
129 * __kernfs_remove() deactivates the subtree and clears i_nlink on its
130 * inodes under the write lock, so under the read lock either
131 * kernfs_find_and_get_node_by_id() refuses the node, or the inode is
132 * in the inode hash before the ilookup() pass goes looking for it.
133 */
134 scoped_guard(rwsem_read, &info->root->kernfs_rwsem) {
135 kn = kernfs_find_and_get_node_by_id(info->root, id);
136 if (!kn)
137 return ERR_PTR(-ESTALE);
138
139 if (get_parent) {
140 struct kernfs_node *parent;
141
142 parent = kernfs_get_parent(kn);
143 kernfs_put(kn);
144 kn = parent;
145 if (!kn)
146 return ERR_PTR(-ESTALE);
147 }
148
149 inode = kernfs_get_inode(sb, kn);
150 kernfs_put(kn);
151 }
152
153 return d_obtain_alias(inode);
154}
155
156static struct dentry *kernfs_fh_to_dentry(struct super_block *sb,
157 struct fid *fid, int fh_len,
158 int fh_type)
159{
160 return __kernfs_fh_to_dentry(sb, fid, fh_len, fh_type, false);
161}
162
163static struct dentry *kernfs_fh_to_parent(struct super_block *sb,
164 struct fid *fid, int fh_len,
165 int fh_type)
166{
167 return __kernfs_fh_to_dentry(sb, fid, fh_len, fh_type, true);
168}
169
170static struct dentry *kernfs_get_parent_dentry(struct dentry *child)
171{
172 struct kernfs_node *kn = kernfs_dentry_node(child);
173 struct kernfs_root *root = kernfs_root(kn);
174
175 guard(rwsem_read)(&root->kernfs_rwsem);
176 return d_obtain_alias(kernfs_get_inode(child->d_sb, kernfs_parent(kn)));
177}
178
179static const struct export_operations kernfs_export_ops = {
180 .encode_fh = kernfs_encode_fh,
181 .fh_to_dentry = kernfs_fh_to_dentry,
182 .fh_to_parent = kernfs_fh_to_parent,
183 .get_parent = kernfs_get_parent_dentry,
184};
185
186/**
187 * kernfs_root_from_sb - determine kernfs_root associated with a super_block
188 * @sb: the super_block in question
189 *
190 * Return: the kernfs_root associated with @sb. If @sb is not a kernfs one,
191 * %NULL is returned.
192 */
193struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
194{
195 if (sb->s_op == &kernfs_sops)
196 return kernfs_info(sb)->root;
197 return NULL;
198}
199
200/*
201 * find the next ancestor in the path down to @child, where @parent was the
202 * ancestor whose descendant we want to find.
203 *
204 * Say the path is /a/b/c/d. @child is d, @parent is %NULL. We return the root
205 * node. If @parent is b, then we return the node for c.
206 * Passing in d as @parent is not ok.
207 */
208static struct kernfs_node *find_next_ancestor(struct kernfs_node *child,
209 struct kernfs_node *parent)
210{
211 if (child == parent) {
212 pr_crit_once("BUG in find_next_ancestor: called with parent == child");
213 return NULL;
214 }
215
216 while (kernfs_parent(child) != parent) {
217 child = kernfs_parent(child);
218 if (!child)
219 return NULL;
220 }
221
222 return child;
223}
224
225/**
226 * kernfs_node_dentry - get a dentry for the given kernfs_node
227 * @kn: kernfs_node for which a dentry is needed
228 * @sb: the kernfs super_block
229 *
230 * Return: the dentry pointer
231 */
232struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
233 struct super_block *sb)
234{
235 struct dentry *dentry;
236 struct kernfs_node *knparent;
237 struct kernfs_root *root;
238
239 BUG_ON(sb->s_op != &kernfs_sops);
240
241 dentry = dget(sb->s_root);
242
243 /* Check if this is the root kernfs_node */
244 if (!rcu_access_pointer(kn->__parent))
245 return dentry;
246
247 root = kernfs_root(kn);
248 /*
249 * As long as kn is valid, its parent can not vanish. This is cgroup's
250 * kn so it can't have its parent replaced. Therefore it is safe to use
251 * the ancestor node outside of the RCU or locked section.
252 */
253 if (WARN_ON_ONCE(!(root->flags & KERNFS_ROOT_INVARIANT_PARENT)))
254 return ERR_PTR(-EINVAL);
255 scoped_guard(rcu) {
256 knparent = find_next_ancestor(kn, NULL);
257 }
258 if (WARN_ON(!knparent)) {
259 dput(dentry);
260 return ERR_PTR(-EINVAL);
261 }
262
263 do {
264 struct dentry *dtmp;
265 struct kernfs_node *kntmp;
266 const char *name;
267
268 if (kn == knparent)
269 return dentry;
270
271 scoped_guard(rwsem_read, &root->kernfs_rwsem) {
272 kntmp = find_next_ancestor(kn, knparent);
273 if (WARN_ON(!kntmp)) {
274 dput(dentry);
275 return ERR_PTR(-EINVAL);
276 }
277 name = kstrdup(kernfs_rcu_name(kntmp), GFP_KERNEL);
278 }
279 if (!name) {
280 dput(dentry);
281 return ERR_PTR(-ENOMEM);
282 }
283 dtmp = lookup_noperm_positive_unlocked(&QSTR(name), dentry);
284 dput(dentry);
285 kfree(name);
286 if (IS_ERR(dtmp))
287 return dtmp;
288 knparent = kntmp;
289 dentry = dtmp;
290 } while (true);
291}
292
293static int kernfs_fill_super(struct super_block *sb, struct kernfs_fs_context *kfc)
294{
295 struct kernfs_super_info *info = kernfs_info(sb);
296 struct kernfs_root *kf_root = kfc->root;
297 struct inode *inode;
298 struct dentry *root;
299
300 info->sb = sb;
301 /* Userspace would break if executables or devices appear on sysfs */
302 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
303 sb->s_blocksize = PAGE_SIZE;
304 sb->s_blocksize_bits = PAGE_SHIFT;
305 sb->s_magic = kfc->magic;
306 sb->s_op = &kernfs_sops;
307 sb->s_xattr = kernfs_xattr_handlers;
308 if (info->root->flags & KERNFS_ROOT_SUPPORT_EXPORTOP)
309 sb->s_export_op = &kernfs_export_ops;
310 sb->s_time_gran = 1;
311 sb->s_maxbytes = MAX_LFS_FILESIZE;
312
313 /* sysfs dentries and inodes don't require IO to create */
314 sb->s_shrink->seeks = 0;
315
316 /* get root inode, initialize and unlock it */
317 down_read(&kf_root->kernfs_rwsem);
318 inode = kernfs_get_inode(sb, info->root->kn);
319 up_read(&kf_root->kernfs_rwsem);
320 if (!inode) {
321 pr_debug("kernfs: could not get root inode\n");
322 return -ENOMEM;
323 }
324
325 /* instantiate and link root dentry */
326 root = d_make_root(inode);
327 if (!root) {
328 pr_debug("%s: could not get root dentry!\n", __func__);
329 return -ENOMEM;
330 }
331 sb->s_root = root;
332 set_default_d_op(sb, &kernfs_dops);
333 return 0;
334}
335
336static int kernfs_test_super(struct super_block *sb, struct fs_context *fc)
337{
338 struct kernfs_super_info *sb_info = kernfs_info(sb);
339 struct kernfs_super_info *info = fc->s_fs_info;
340
341 return sb_info->root == info->root && sb_info->ns == info->ns;
342}
343
344static int kernfs_set_super(struct super_block *sb, struct fs_context *fc)
345{
346 struct kernfs_fs_context *kfc = fc->fs_private;
347
348 kfc->ns_tag = NULL;
349 return set_anon_super_fc(sb, fc);
350}
351
352/**
353 * kernfs_super_ns - determine the namespace tag of a kernfs super_block
354 * @sb: super_block of interest
355 *
356 * Return: the namespace tag associated with kernfs super_block @sb.
357 */
358const struct ns_common *kernfs_super_ns(struct super_block *sb)
359{
360 struct kernfs_super_info *info = kernfs_info(sb);
361
362 return info->ns;
363}
364
365/**
366 * kernfs_get_tree - kernfs filesystem access/retrieval helper
367 * @fc: The filesystem context.
368 *
369 * This is to be called from each kernfs user's fs_context->ops->get_tree()
370 * implementation, which should set the specified ->@fs_type and ->@flags, and
371 * specify the hierarchy and namespace tag to mount via ->@root and ->@ns,
372 * respectively.
373 *
374 * Return: %0 on success, -errno on failure.
375 */
376int kernfs_get_tree(struct fs_context *fc)
377{
378 struct kernfs_fs_context *kfc = fc->fs_private;
379 struct super_block *sb;
380 struct kernfs_super_info *info;
381 int error;
382
383 info = kzalloc_obj(*info);
384 if (!info)
385 return -ENOMEM;
386
387 info->root = kfc->root;
388 info->ns = kfc->ns_tag;
389 INIT_LIST_HEAD(&info->node);
390
391 fc->s_fs_info = info;
392 sb = sget_fc(fc, kernfs_test_super, kernfs_set_super);
393 if (IS_ERR(sb))
394 return PTR_ERR(sb);
395
396 if (!sb->s_root) {
397 struct kernfs_super_info *info = kernfs_info(sb);
398 struct kernfs_root *root = kfc->root;
399
400 kfc->new_sb_created = true;
401
402 error = kernfs_fill_super(sb, kfc);
403 if (error) {
404 deactivate_locked_super(sb);
405 return error;
406 }
407 sb->s_flags |= SB_ACTIVE;
408
409 uuid_t uuid;
410 uuid_gen(&uuid);
411 super_set_uuid(sb, uuid.b, sizeof(uuid));
412
413 down_write(&root->kernfs_supers_rwsem);
414 list_add(&info->node, &info->root->supers);
415 up_write(&root->kernfs_supers_rwsem);
416 }
417
418 fc->root = dget(sb->s_root);
419 return 0;
420}
421
422void kernfs_free_fs_context(struct fs_context *fc)
423{
424 /* Note that we don't deal with kfc->ns_tag here. */
425 kfree(fc->s_fs_info);
426 fc->s_fs_info = NULL;
427}
428
429/**
430 * kernfs_kill_sb - kill_sb for kernfs
431 * @sb: super_block being killed
432 *
433 * This can be used directly for file_system_type->kill_sb(). If a kernfs
434 * user needs extra cleanup, it can implement its own kill_sb() and call
435 * this function at the end.
436 */
437void kernfs_kill_sb(struct super_block *sb)
438{
439 struct kernfs_super_info *info = kernfs_info(sb);
440 struct kernfs_root *root = info->root;
441
442 down_write(&root->kernfs_supers_rwsem);
443 list_del(&info->node);
444 up_write(&root->kernfs_supers_rwsem);
445
446 /*
447 * Mark the superblock dead so sget_fc() can't find it,
448 * before freeing kernfs_super_info.
449 */
450 kill_anon_super(sb);
451 kfree(info);
452}
453
454static void __init kernfs_mutex_init(void)
455{
456 int count;
457
458 for (count = 0; count < NR_KERNFS_LOCKS; count++)
459 mutex_init(&kernfs_locks->node_mutex[count]);
460}
461
462static void __init kernfs_lock_init(void)
463{
464 kernfs_locks = kmalloc_obj(struct kernfs_global_locks);
465 WARN_ON(!kernfs_locks);
466
467 kernfs_mutex_init();
468}
469
470void __init kernfs_init(void)
471{
472 kernfs_node_cache = kmem_cache_create("kernfs_node_cache",
473 sizeof(struct kernfs_node),
474 0, SLAB_PANIC, NULL);
475
476 /* Creates slab cache for kernfs inode attributes */
477 kernfs_iattrs_cache = kmem_cache_create("kernfs_iattrs_cache",
478 sizeof(struct kernfs_iattrs),
479 0, SLAB_PANIC, NULL);
480
481 kernfs_lock_init();
482}