Commit | Line | Data |
---|---|---|
b8441ed2 TH |
1 | /* |
2 | * fs/kernfs/mount.c - kernfs mount implementation | |
3 | * | |
4 | * Copyright (c) 2001-3 Patrick Mochel | |
5 | * Copyright (c) 2007 SUSE Linux Products GmbH | |
6 | * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org> | |
7 | * | |
8 | * This file is released under the GPLv2. | |
9 | */ | |
fa736a95 TH |
10 | |
11 | #include <linux/fs.h> | |
12 | #include <linux/mount.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/magic.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/pagemap.h> | |
fb3c8315 | 17 | #include <linux/namei.h> |
4f41fc59 | 18 | #include <linux/seq_file.h> |
aa818825 | 19 | #include <linux/exportfs.h> |
fa736a95 TH |
20 | |
21 | #include "kernfs-internal.h" | |
22 | ||
a797bfc3 | 23 | struct kmem_cache *kernfs_node_cache; |
fa736a95 | 24 | |
6a7fed4e TH |
25 | static int kernfs_sop_remount_fs(struct super_block *sb, int *flags, char *data) |
26 | { | |
27 | struct kernfs_root *root = kernfs_info(sb)->root; | |
28 | struct kernfs_syscall_ops *scops = root->syscall_ops; | |
29 | ||
30 | if (scops && scops->remount_fs) | |
31 | return scops->remount_fs(root, flags, data); | |
32 | return 0; | |
33 | } | |
34 | ||
35 | static int kernfs_sop_show_options(struct seq_file *sf, struct dentry *dentry) | |
36 | { | |
319ba91d | 37 | struct kernfs_root *root = kernfs_root(kernfs_dentry_node(dentry)); |
6a7fed4e TH |
38 | struct kernfs_syscall_ops *scops = root->syscall_ops; |
39 | ||
40 | if (scops && scops->show_options) | |
41 | return scops->show_options(sf, root); | |
42 | return 0; | |
43 | } | |
44 | ||
4f41fc59 SH |
45 | static int kernfs_sop_show_path(struct seq_file *sf, struct dentry *dentry) |
46 | { | |
319ba91d | 47 | struct kernfs_node *node = kernfs_dentry_node(dentry); |
4f41fc59 SH |
48 | struct kernfs_root *root = kernfs_root(node); |
49 | struct kernfs_syscall_ops *scops = root->syscall_ops; | |
50 | ||
51 | if (scops && scops->show_path) | |
52 | return scops->show_path(sf, node, root); | |
53 | ||
3cc9b23c SH |
54 | seq_dentry(sf, dentry, " \t\n\\"); |
55 | return 0; | |
4f41fc59 SH |
56 | } |
57 | ||
f41c5934 | 58 | const struct super_operations kernfs_sops = { |
fa736a95 TH |
59 | .statfs = simple_statfs, |
60 | .drop_inode = generic_delete_inode, | |
c637b8ac | 61 | .evict_inode = kernfs_evict_inode, |
6a7fed4e TH |
62 | |
63 | .remount_fs = kernfs_sop_remount_fs, | |
64 | .show_options = kernfs_sop_show_options, | |
4f41fc59 | 65 | .show_path = kernfs_sop_show_path, |
fa736a95 TH |
66 | }; |
67 | ||
69fd5c39 SL |
68 | /* |
69 | * Similar to kernfs_fh_get_inode, this one gets kernfs node from inode | |
70 | * number and generation | |
71 | */ | |
72 | struct kernfs_node *kernfs_get_node_by_id(struct kernfs_root *root, | |
73 | const union kernfs_node_id *id) | |
74 | { | |
75 | struct kernfs_node *kn; | |
76 | ||
77 | kn = kernfs_find_and_get_node_by_ino(root, id->ino); | |
78 | if (!kn) | |
79 | return NULL; | |
80 | if (kn->id.generation != id->generation) { | |
81 | kernfs_put(kn); | |
82 | return NULL; | |
83 | } | |
84 | return kn; | |
85 | } | |
86 | ||
aa818825 SL |
87 | static struct inode *kernfs_fh_get_inode(struct super_block *sb, |
88 | u64 ino, u32 generation) | |
89 | { | |
90 | struct kernfs_super_info *info = kernfs_info(sb); | |
91 | struct inode *inode; | |
92 | struct kernfs_node *kn; | |
93 | ||
94 | if (ino == 0) | |
95 | return ERR_PTR(-ESTALE); | |
96 | ||
97 | kn = kernfs_find_and_get_node_by_ino(info->root, ino); | |
98 | if (!kn) | |
99 | return ERR_PTR(-ESTALE); | |
100 | inode = kernfs_get_inode(sb, kn); | |
101 | kernfs_put(kn); | |
ef13ecbc DC |
102 | if (!inode) |
103 | return ERR_PTR(-ESTALE); | |
aa818825 SL |
104 | |
105 | if (generation && inode->i_generation != generation) { | |
106 | /* we didn't find the right inode.. */ | |
107 | iput(inode); | |
108 | return ERR_PTR(-ESTALE); | |
109 | } | |
110 | return inode; | |
111 | } | |
112 | ||
113 | static struct dentry *kernfs_fh_to_dentry(struct super_block *sb, struct fid *fid, | |
114 | int fh_len, int fh_type) | |
115 | { | |
116 | return generic_fh_to_dentry(sb, fid, fh_len, fh_type, | |
117 | kernfs_fh_get_inode); | |
118 | } | |
119 | ||
120 | static struct dentry *kernfs_fh_to_parent(struct super_block *sb, struct fid *fid, | |
121 | int fh_len, int fh_type) | |
122 | { | |
123 | return generic_fh_to_parent(sb, fid, fh_len, fh_type, | |
124 | kernfs_fh_get_inode); | |
125 | } | |
126 | ||
127 | static struct dentry *kernfs_get_parent_dentry(struct dentry *child) | |
128 | { | |
129 | struct kernfs_node *kn = kernfs_dentry_node(child); | |
130 | ||
131 | return d_obtain_alias(kernfs_get_inode(child->d_sb, kn->parent)); | |
132 | } | |
133 | ||
134 | static const struct export_operations kernfs_export_ops = { | |
135 | .fh_to_dentry = kernfs_fh_to_dentry, | |
136 | .fh_to_parent = kernfs_fh_to_parent, | |
137 | .get_parent = kernfs_get_parent_dentry, | |
138 | }; | |
139 | ||
0c23b225 TH |
140 | /** |
141 | * kernfs_root_from_sb - determine kernfs_root associated with a super_block | |
142 | * @sb: the super_block in question | |
143 | * | |
144 | * Return the kernfs_root associated with @sb. If @sb is not a kernfs one, | |
145 | * %NULL is returned. | |
146 | */ | |
147 | struct kernfs_root *kernfs_root_from_sb(struct super_block *sb) | |
148 | { | |
149 | if (sb->s_op == &kernfs_sops) | |
150 | return kernfs_info(sb)->root; | |
151 | return NULL; | |
152 | } | |
153 | ||
fb3c8315 AK |
154 | /* |
155 | * find the next ancestor in the path down to @child, where @parent was the | |
156 | * ancestor whose descendant we want to find. | |
157 | * | |
158 | * Say the path is /a/b/c/d. @child is d, @parent is NULL. We return the root | |
159 | * node. If @parent is b, then we return the node for c. | |
160 | * Passing in d as @parent is not ok. | |
161 | */ | |
162 | static struct kernfs_node *find_next_ancestor(struct kernfs_node *child, | |
163 | struct kernfs_node *parent) | |
164 | { | |
165 | if (child == parent) { | |
166 | pr_crit_once("BUG in find_next_ancestor: called with parent == child"); | |
167 | return NULL; | |
168 | } | |
169 | ||
170 | while (child->parent != parent) { | |
171 | if (!child->parent) | |
172 | return NULL; | |
173 | child = child->parent; | |
174 | } | |
175 | ||
176 | return child; | |
177 | } | |
178 | ||
179 | /** | |
180 | * kernfs_node_dentry - get a dentry for the given kernfs_node | |
181 | * @kn: kernfs_node for which a dentry is needed | |
182 | * @sb: the kernfs super_block | |
183 | */ | |
184 | struct dentry *kernfs_node_dentry(struct kernfs_node *kn, | |
185 | struct super_block *sb) | |
186 | { | |
187 | struct dentry *dentry; | |
188 | struct kernfs_node *knparent = NULL; | |
189 | ||
190 | BUG_ON(sb->s_op != &kernfs_sops); | |
191 | ||
192 | dentry = dget(sb->s_root); | |
193 | ||
194 | /* Check if this is the root kernfs_node */ | |
195 | if (!kn->parent) | |
196 | return dentry; | |
197 | ||
198 | knparent = find_next_ancestor(kn, NULL); | |
199 | if (WARN_ON(!knparent)) | |
200 | return ERR_PTR(-EINVAL); | |
201 | ||
202 | do { | |
203 | struct dentry *dtmp; | |
204 | struct kernfs_node *kntmp; | |
205 | ||
206 | if (kn == knparent) | |
207 | return dentry; | |
208 | kntmp = find_next_ancestor(kn, knparent); | |
209 | if (WARN_ON(!kntmp)) | |
210 | return ERR_PTR(-EINVAL); | |
779b8391 AV |
211 | dtmp = lookup_one_len_unlocked(kntmp->name, dentry, |
212 | strlen(kntmp->name)); | |
fb3c8315 AK |
213 | dput(dentry); |
214 | if (IS_ERR(dtmp)) | |
215 | return dtmp; | |
216 | knparent = kntmp; | |
217 | dentry = dtmp; | |
218 | } while (true); | |
219 | } | |
220 | ||
26fc9cd2 | 221 | static int kernfs_fill_super(struct super_block *sb, unsigned long magic) |
fa736a95 | 222 | { |
c525aadd | 223 | struct kernfs_super_info *info = kernfs_info(sb); |
fa736a95 TH |
224 | struct inode *inode; |
225 | struct dentry *root; | |
226 | ||
7d568a83 | 227 | info->sb = sb; |
a2982cc9 EB |
228 | /* Userspace would break if executables or devices appear on sysfs */ |
229 | sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV; | |
09cbfeaf KS |
230 | sb->s_blocksize = PAGE_SIZE; |
231 | sb->s_blocksize_bits = PAGE_SHIFT; | |
26fc9cd2 | 232 | sb->s_magic = magic; |
a797bfc3 | 233 | sb->s_op = &kernfs_sops; |
e72a1a8b | 234 | sb->s_xattr = kernfs_xattr_handlers; |
aa818825 SL |
235 | if (info->root->flags & KERNFS_ROOT_SUPPORT_EXPORTOP) |
236 | sb->s_export_op = &kernfs_export_ops; | |
fa736a95 TH |
237 | sb->s_time_gran = 1; |
238 | ||
4b85afbd JW |
239 | /* sysfs dentries and inodes don't require IO to create */ |
240 | sb->s_shrink.seeks = 0; | |
241 | ||
fa736a95 | 242 | /* get root inode, initialize and unlock it */ |
a797bfc3 | 243 | mutex_lock(&kernfs_mutex); |
c637b8ac | 244 | inode = kernfs_get_inode(sb, info->root->kn); |
a797bfc3 | 245 | mutex_unlock(&kernfs_mutex); |
fa736a95 | 246 | if (!inode) { |
c637b8ac | 247 | pr_debug("kernfs: could not get root inode\n"); |
fa736a95 TH |
248 | return -ENOMEM; |
249 | } | |
250 | ||
251 | /* instantiate and link root dentry */ | |
252 | root = d_make_root(inode); | |
253 | if (!root) { | |
254 | pr_debug("%s: could not get root dentry!\n", __func__); | |
255 | return -ENOMEM; | |
256 | } | |
fa736a95 | 257 | sb->s_root = root; |
a797bfc3 | 258 | sb->s_d_op = &kernfs_dops; |
fa736a95 TH |
259 | return 0; |
260 | } | |
261 | ||
c637b8ac | 262 | static int kernfs_test_super(struct super_block *sb, void *data) |
fa736a95 | 263 | { |
c525aadd TH |
264 | struct kernfs_super_info *sb_info = kernfs_info(sb); |
265 | struct kernfs_super_info *info = data; | |
fa736a95 TH |
266 | |
267 | return sb_info->root == info->root && sb_info->ns == info->ns; | |
268 | } | |
269 | ||
c637b8ac | 270 | static int kernfs_set_super(struct super_block *sb, void *data) |
fa736a95 TH |
271 | { |
272 | int error; | |
273 | error = set_anon_super(sb, data); | |
274 | if (!error) | |
275 | sb->s_fs_info = data; | |
276 | return error; | |
277 | } | |
278 | ||
279 | /** | |
280 | * kernfs_super_ns - determine the namespace tag of a kernfs super_block | |
281 | * @sb: super_block of interest | |
282 | * | |
283 | * Return the namespace tag associated with kernfs super_block @sb. | |
284 | */ | |
285 | const void *kernfs_super_ns(struct super_block *sb) | |
286 | { | |
c525aadd | 287 | struct kernfs_super_info *info = kernfs_info(sb); |
fa736a95 TH |
288 | |
289 | return info->ns; | |
290 | } | |
291 | ||
292 | /** | |
293 | * kernfs_mount_ns - kernfs mount helper | |
294 | * @fs_type: file_system_type of the fs being mounted | |
295 | * @flags: mount flags specified for the mount | |
296 | * @root: kernfs_root of the hierarchy being mounted | |
26fc9cd2 | 297 | * @magic: file system specific magic number |
fed95bab | 298 | * @new_sb_created: tell the caller if we allocated a new superblock |
fa736a95 TH |
299 | * @ns: optional namespace tag of the mount |
300 | * | |
301 | * This is to be called from each kernfs user's file_system_type->mount() | |
302 | * implementation, which should pass through the specified @fs_type and | |
303 | * @flags, and specify the hierarchy and namespace tag to mount via @root | |
304 | * and @ns, respectively. | |
305 | * | |
306 | * The return value can be passed to the vfs layer verbatim. | |
307 | */ | |
308 | struct dentry *kernfs_mount_ns(struct file_system_type *fs_type, int flags, | |
26fc9cd2 JZ |
309 | struct kernfs_root *root, unsigned long magic, |
310 | bool *new_sb_created, const void *ns) | |
fa736a95 TH |
311 | { |
312 | struct super_block *sb; | |
c525aadd | 313 | struct kernfs_super_info *info; |
fa736a95 TH |
314 | int error; |
315 | ||
316 | info = kzalloc(sizeof(*info), GFP_KERNEL); | |
317 | if (!info) | |
318 | return ERR_PTR(-ENOMEM); | |
319 | ||
320 | info->root = root; | |
321 | info->ns = ns; | |
82382ace | 322 | INIT_LIST_HEAD(&info->node); |
fa736a95 | 323 | |
6e4eab57 EB |
324 | sb = sget_userns(fs_type, kernfs_test_super, kernfs_set_super, flags, |
325 | &init_user_ns, info); | |
fa736a95 TH |
326 | if (IS_ERR(sb) || sb->s_fs_info != info) |
327 | kfree(info); | |
328 | if (IS_ERR(sb)) | |
329 | return ERR_CAST(sb); | |
fed95bab LZ |
330 | |
331 | if (new_sb_created) | |
332 | *new_sb_created = !sb->s_root; | |
333 | ||
fa736a95 | 334 | if (!sb->s_root) { |
7d568a83 TH |
335 | struct kernfs_super_info *info = kernfs_info(sb); |
336 | ||
26fc9cd2 | 337 | error = kernfs_fill_super(sb, magic); |
fa736a95 TH |
338 | if (error) { |
339 | deactivate_locked_super(sb); | |
340 | return ERR_PTR(error); | |
341 | } | |
1751e8a6 | 342 | sb->s_flags |= SB_ACTIVE; |
7d568a83 TH |
343 | |
344 | mutex_lock(&kernfs_mutex); | |
345 | list_add(&info->node, &root->supers); | |
346 | mutex_unlock(&kernfs_mutex); | |
fa736a95 TH |
347 | } |
348 | ||
349 | return dget(sb->s_root); | |
350 | } | |
351 | ||
352 | /** | |
353 | * kernfs_kill_sb - kill_sb for kernfs | |
354 | * @sb: super_block being killed | |
355 | * | |
356 | * This can be used directly for file_system_type->kill_sb(). If a kernfs | |
357 | * user needs extra cleanup, it can implement its own kill_sb() and call | |
358 | * this function at the end. | |
359 | */ | |
360 | void kernfs_kill_sb(struct super_block *sb) | |
361 | { | |
c525aadd | 362 | struct kernfs_super_info *info = kernfs_info(sb); |
fa736a95 | 363 | |
7d568a83 TH |
364 | mutex_lock(&kernfs_mutex); |
365 | list_del(&info->node); | |
366 | mutex_unlock(&kernfs_mutex); | |
367 | ||
fa736a95 TH |
368 | /* |
369 | * Remove the superblock from fs_supers/s_instances | |
c525aadd | 370 | * so we can't find it, before freeing kernfs_super_info. |
fa736a95 TH |
371 | */ |
372 | kill_anon_super(sb); | |
373 | kfree(info); | |
fa736a95 TH |
374 | } |
375 | ||
4e26445f LZ |
376 | /** |
377 | * kernfs_pin_sb: try to pin the superblock associated with a kernfs_root | |
378 | * @kernfs_root: the kernfs_root in question | |
379 | * @ns: the namespace tag | |
380 | * | |
381 | * Pin the superblock so the superblock won't be destroyed in subsequent | |
382 | * operations. This can be used to block ->kill_sb() which may be useful | |
383 | * for kernfs users which dynamically manage superblocks. | |
384 | * | |
385 | * Returns NULL if there's no superblock associated to this kernfs_root, or | |
386 | * -EINVAL if the superblock is being freed. | |
387 | */ | |
388 | struct super_block *kernfs_pin_sb(struct kernfs_root *root, const void *ns) | |
389 | { | |
390 | struct kernfs_super_info *info; | |
391 | struct super_block *sb = NULL; | |
392 | ||
393 | mutex_lock(&kernfs_mutex); | |
394 | list_for_each_entry(info, &root->supers, node) { | |
395 | if (info->ns == ns) { | |
396 | sb = info->sb; | |
397 | if (!atomic_inc_not_zero(&info->sb->s_active)) | |
398 | sb = ERR_PTR(-EINVAL); | |
399 | break; | |
400 | } | |
401 | } | |
402 | mutex_unlock(&kernfs_mutex); | |
403 | return sb; | |
404 | } | |
405 | ||
fa736a95 TH |
406 | void __init kernfs_init(void) |
407 | { | |
ba16b284 SL |
408 | |
409 | /* | |
410 | * the slab is freed in RCU context, so kernfs_find_and_get_node_by_ino | |
411 | * can access the slab lock free. This could introduce stale nodes, | |
412 | * please see how kernfs_find_and_get_node_by_ino filters out stale | |
413 | * nodes. | |
414 | */ | |
a797bfc3 | 415 | kernfs_node_cache = kmem_cache_create("kernfs_node_cache", |
324a56e1 | 416 | sizeof(struct kernfs_node), |
ba16b284 SL |
417 | 0, |
418 | SLAB_PANIC | SLAB_TYPESAFE_BY_RCU, | |
419 | NULL); | |
fa736a95 | 420 | } |