Commit | Line | Data |
---|---|---|
b8441ed2 TH |
1 | /* |
2 | * fs/kernfs/dir.c - kernfs directory 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 | */ | |
fd7b9f7b | 10 | |
abd54f02 | 11 | #include <linux/sched.h> |
fd7b9f7b TH |
12 | #include <linux/fs.h> |
13 | #include <linux/namei.h> | |
14 | #include <linux/idr.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/security.h> | |
17 | #include <linux/hash.h> | |
18 | ||
19 | #include "kernfs-internal.h" | |
20 | ||
a797bfc3 | 21 | DEFINE_MUTEX(kernfs_mutex); |
fd7b9f7b | 22 | |
adc5e8b5 | 23 | #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb) |
fd7b9f7b | 24 | |
81c173cb TH |
25 | static bool kernfs_active(struct kernfs_node *kn) |
26 | { | |
27 | lockdep_assert_held(&kernfs_mutex); | |
28 | return atomic_read(&kn->active) >= 0; | |
29 | } | |
30 | ||
182fd64b TH |
31 | static bool kernfs_lockdep(struct kernfs_node *kn) |
32 | { | |
33 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
34 | return kn->flags & KERNFS_LOCKDEP; | |
35 | #else | |
36 | return false; | |
37 | #endif | |
38 | } | |
39 | ||
fd7b9f7b | 40 | /** |
c637b8ac | 41 | * kernfs_name_hash |
fd7b9f7b TH |
42 | * @name: Null terminated string to hash |
43 | * @ns: Namespace tag to hash | |
44 | * | |
45 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
46 | */ | |
c637b8ac | 47 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b TH |
48 | { |
49 | unsigned long hash = init_name_hash(); | |
50 | unsigned int len = strlen(name); | |
51 | while (len--) | |
52 | hash = partial_name_hash(*name++, hash); | |
53 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); | |
54 | hash &= 0x7fffffffU; | |
55 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
56 | if (hash < 1) | |
57 | hash += 2; | |
58 | if (hash >= INT_MAX) | |
59 | hash = INT_MAX - 1; | |
60 | return hash; | |
61 | } | |
62 | ||
c637b8ac TH |
63 | static int kernfs_name_compare(unsigned int hash, const char *name, |
64 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 65 | { |
adc5e8b5 TH |
66 | if (hash != kn->hash) |
67 | return hash - kn->hash; | |
68 | if (ns != kn->ns) | |
69 | return ns - kn->ns; | |
70 | return strcmp(name, kn->name); | |
fd7b9f7b TH |
71 | } |
72 | ||
c637b8ac TH |
73 | static int kernfs_sd_compare(const struct kernfs_node *left, |
74 | const struct kernfs_node *right) | |
fd7b9f7b | 75 | { |
c637b8ac | 76 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
77 | } |
78 | ||
79 | /** | |
c637b8ac | 80 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 81 | * @kn: kernfs_node of interest |
fd7b9f7b | 82 | * |
324a56e1 | 83 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 84 | * @kn->parent->dir.children. |
fd7b9f7b TH |
85 | * |
86 | * Locking: | |
a797bfc3 | 87 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
88 | * |
89 | * RETURNS: | |
90 | * 0 on susccess -EEXIST on failure. | |
91 | */ | |
c637b8ac | 92 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 93 | { |
adc5e8b5 | 94 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
95 | struct rb_node *parent = NULL; |
96 | ||
df23fc39 | 97 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 98 | kn->parent->dir.subdirs++; |
fd7b9f7b TH |
99 | |
100 | while (*node) { | |
324a56e1 | 101 | struct kernfs_node *pos; |
fd7b9f7b TH |
102 | int result; |
103 | ||
324a56e1 | 104 | pos = rb_to_kn(*node); |
fd7b9f7b | 105 | parent = *node; |
c637b8ac | 106 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 107 | if (result < 0) |
adc5e8b5 | 108 | node = &pos->rb.rb_left; |
fd7b9f7b | 109 | else if (result > 0) |
adc5e8b5 | 110 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
111 | else |
112 | return -EEXIST; | |
113 | } | |
114 | /* add new node and rebalance the tree */ | |
adc5e8b5 TH |
115 | rb_link_node(&kn->rb, parent, node); |
116 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
fd7b9f7b TH |
117 | return 0; |
118 | } | |
119 | ||
120 | /** | |
c637b8ac | 121 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 122 | * @kn: kernfs_node of interest |
fd7b9f7b | 123 | * |
35beab06 TH |
124 | * Try to unlink @kn from its sibling rbtree which starts from |
125 | * kn->parent->dir.children. Returns %true if @kn was actually | |
126 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
127 | * |
128 | * Locking: | |
a797bfc3 | 129 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 130 | */ |
35beab06 | 131 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 132 | { |
35beab06 TH |
133 | if (RB_EMPTY_NODE(&kn->rb)) |
134 | return false; | |
135 | ||
df23fc39 | 136 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 137 | kn->parent->dir.subdirs--; |
fd7b9f7b | 138 | |
adc5e8b5 | 139 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
140 | RB_CLEAR_NODE(&kn->rb); |
141 | return true; | |
fd7b9f7b TH |
142 | } |
143 | ||
144 | /** | |
c637b8ac | 145 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 146 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 147 | * |
324a56e1 | 148 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
149 | * is NULL. |
150 | * | |
151 | * RETURNS: | |
324a56e1 | 152 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 153 | */ |
c637b8ac | 154 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 155 | { |
324a56e1 | 156 | if (unlikely(!kn)) |
fd7b9f7b TH |
157 | return NULL; |
158 | ||
f4b3e631 GKH |
159 | if (!atomic_inc_unless_negative(&kn->active)) |
160 | return NULL; | |
895a068a | 161 | |
182fd64b | 162 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
163 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
164 | return kn; | |
fd7b9f7b TH |
165 | } |
166 | ||
167 | /** | |
c637b8ac | 168 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 169 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 170 | * |
324a56e1 | 171 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
172 | * is NULL. |
173 | */ | |
c637b8ac | 174 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 175 | { |
abd54f02 | 176 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
177 | int v; |
178 | ||
324a56e1 | 179 | if (unlikely(!kn)) |
fd7b9f7b TH |
180 | return; |
181 | ||
182fd64b | 182 | if (kernfs_lockdep(kn)) |
324a56e1 | 183 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 184 | v = atomic_dec_return(&kn->active); |
df23fc39 | 185 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
186 | return; |
187 | ||
abd54f02 | 188 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
189 | } |
190 | ||
191 | /** | |
81c173cb TH |
192 | * kernfs_drain - drain kernfs_node |
193 | * @kn: kernfs_node to drain | |
fd7b9f7b | 194 | * |
81c173cb TH |
195 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
196 | * removers may invoke this function concurrently on @kn and all will | |
197 | * return after draining is complete. | |
fd7b9f7b | 198 | */ |
81c173cb | 199 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 200 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 201 | { |
abd54f02 | 202 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 203 | |
35beab06 | 204 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 205 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 206 | |
35beab06 | 207 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 208 | |
182fd64b | 209 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
210 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
211 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
212 | lock_contended(&kn->dep_map, _RET_IP_); | |
213 | } | |
abd54f02 | 214 | |
35beab06 | 215 | /* but everyone should wait for draining */ |
abd54f02 TH |
216 | wait_event(root->deactivate_waitq, |
217 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 218 | |
182fd64b | 219 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
220 | lock_acquired(&kn->dep_map, _RET_IP_); |
221 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
222 | } | |
35beab06 | 223 | |
ccf02aaf TH |
224 | kernfs_unmap_bin_file(kn); |
225 | ||
35beab06 | 226 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
227 | } |
228 | ||
fd7b9f7b | 229 | /** |
324a56e1 TH |
230 | * kernfs_get - get a reference count on a kernfs_node |
231 | * @kn: the target kernfs_node | |
fd7b9f7b | 232 | */ |
324a56e1 | 233 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 234 | { |
324a56e1 | 235 | if (kn) { |
adc5e8b5 TH |
236 | WARN_ON(!atomic_read(&kn->count)); |
237 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
238 | } |
239 | } | |
240 | EXPORT_SYMBOL_GPL(kernfs_get); | |
241 | ||
242 | /** | |
324a56e1 TH |
243 | * kernfs_put - put a reference count on a kernfs_node |
244 | * @kn: the target kernfs_node | |
fd7b9f7b | 245 | * |
324a56e1 | 246 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 247 | */ |
324a56e1 | 248 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 249 | { |
324a56e1 | 250 | struct kernfs_node *parent; |
ba7443bc | 251 | struct kernfs_root *root; |
fd7b9f7b | 252 | |
adc5e8b5 | 253 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 254 | return; |
324a56e1 | 255 | root = kernfs_root(kn); |
fd7b9f7b | 256 | repeat: |
81c173cb TH |
257 | /* |
258 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 259 | * kn->parent won't change beneath us. |
fd7b9f7b | 260 | */ |
adc5e8b5 | 261 | parent = kn->parent; |
fd7b9f7b | 262 | |
81c173cb TH |
263 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
264 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
265 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 266 | |
df23fc39 | 267 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 268 | kernfs_put(kn->symlink.target_kn); |
2063d608 | 269 | if (!(kn->flags & KERNFS_STATIC_NAME)) |
adc5e8b5 TH |
270 | kfree(kn->name); |
271 | if (kn->iattr) { | |
272 | if (kn->iattr->ia_secdata) | |
273 | security_release_secctx(kn->iattr->ia_secdata, | |
274 | kn->iattr->ia_secdata_len); | |
275 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 276 | } |
adc5e8b5 TH |
277 | kfree(kn->iattr); |
278 | ida_simple_remove(&root->ino_ida, kn->ino); | |
a797bfc3 | 279 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 280 | |
324a56e1 TH |
281 | kn = parent; |
282 | if (kn) { | |
adc5e8b5 | 283 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
284 | goto repeat; |
285 | } else { | |
324a56e1 | 286 | /* just released the root kn, free @root too */ |
bc755553 | 287 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
288 | kfree(root); |
289 | } | |
fd7b9f7b TH |
290 | } |
291 | EXPORT_SYMBOL_GPL(kernfs_put); | |
292 | ||
c637b8ac | 293 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 294 | { |
324a56e1 | 295 | struct kernfs_node *kn; |
fd7b9f7b TH |
296 | |
297 | if (flags & LOOKUP_RCU) | |
298 | return -ECHILD; | |
299 | ||
19bbb926 TH |
300 | /* Always perform fresh lookup for negatives */ |
301 | if (!dentry->d_inode) | |
302 | goto out_bad_unlocked; | |
303 | ||
324a56e1 | 304 | kn = dentry->d_fsdata; |
a797bfc3 | 305 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 306 | |
81c173cb TH |
307 | /* The kernfs node has been deactivated */ |
308 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
309 | goto out_bad; |
310 | ||
c637b8ac | 311 | /* The kernfs node has been moved? */ |
adc5e8b5 | 312 | if (dentry->d_parent->d_fsdata != kn->parent) |
fd7b9f7b TH |
313 | goto out_bad; |
314 | ||
c637b8ac | 315 | /* The kernfs node has been renamed */ |
adc5e8b5 | 316 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
317 | goto out_bad; |
318 | ||
c637b8ac | 319 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 320 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 321 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
322 | goto out_bad; |
323 | ||
a797bfc3 | 324 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
325 | out_valid: |
326 | return 1; | |
327 | out_bad: | |
a797bfc3 | 328 | mutex_unlock(&kernfs_mutex); |
19bbb926 TH |
329 | out_bad_unlocked: |
330 | /* | |
331 | * @dentry doesn't match the underlying kernfs node, drop the | |
332 | * dentry and force lookup. If we have submounts we must allow the | |
333 | * vfs caches to lie about the state of the filesystem to prevent | |
334 | * leaks and other nasty things, so use check_submounts_and_drop() | |
335 | * instead of d_drop(). | |
fd7b9f7b TH |
336 | */ |
337 | if (check_submounts_and_drop(dentry) != 0) | |
338 | goto out_valid; | |
339 | ||
340 | return 0; | |
341 | } | |
342 | ||
c637b8ac | 343 | static void kernfs_dop_release(struct dentry *dentry) |
fd7b9f7b TH |
344 | { |
345 | kernfs_put(dentry->d_fsdata); | |
346 | } | |
347 | ||
a797bfc3 | 348 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 349 | .d_revalidate = kernfs_dop_revalidate, |
c637b8ac | 350 | .d_release = kernfs_dop_release, |
fd7b9f7b TH |
351 | }; |
352 | ||
db4aad20 TH |
353 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
354 | const char *name, umode_t mode, | |
355 | unsigned flags) | |
fd7b9f7b TH |
356 | { |
357 | char *dup_name = NULL; | |
324a56e1 | 358 | struct kernfs_node *kn; |
bc755553 | 359 | int ret; |
fd7b9f7b | 360 | |
2063d608 | 361 | if (!(flags & KERNFS_STATIC_NAME)) { |
fd7b9f7b TH |
362 | name = dup_name = kstrdup(name, GFP_KERNEL); |
363 | if (!name) | |
364 | return NULL; | |
365 | } | |
366 | ||
a797bfc3 | 367 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 368 | if (!kn) |
fd7b9f7b TH |
369 | goto err_out1; |
370 | ||
bc755553 TH |
371 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); |
372 | if (ret < 0) | |
fd7b9f7b | 373 | goto err_out2; |
adc5e8b5 | 374 | kn->ino = ret; |
fd7b9f7b | 375 | |
adc5e8b5 | 376 | atomic_set(&kn->count, 1); |
81c173cb | 377 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 378 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 379 | |
adc5e8b5 TH |
380 | kn->name = name; |
381 | kn->mode = mode; | |
81c173cb | 382 | kn->flags = flags; |
fd7b9f7b | 383 | |
324a56e1 | 384 | return kn; |
fd7b9f7b TH |
385 | |
386 | err_out2: | |
a797bfc3 | 387 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b TH |
388 | err_out1: |
389 | kfree(dup_name); | |
390 | return NULL; | |
391 | } | |
392 | ||
db4aad20 TH |
393 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
394 | const char *name, umode_t mode, | |
395 | unsigned flags) | |
396 | { | |
397 | struct kernfs_node *kn; | |
398 | ||
399 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | |
400 | if (kn) { | |
401 | kernfs_get(parent); | |
402 | kn->parent = parent; | |
403 | } | |
404 | return kn; | |
405 | } | |
406 | ||
fd7b9f7b | 407 | /** |
c637b8ac | 408 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 409 | * @kn: kernfs_node to be added |
fd7b9f7b | 410 | * |
db4aad20 TH |
411 | * The caller must already have initialized @kn->parent. This |
412 | * function increments nlink of the parent's inode if @kn is a | |
413 | * directory and link into the children list of the parent. | |
fd7b9f7b | 414 | * |
fd7b9f7b TH |
415 | * RETURNS: |
416 | * 0 on success, -EEXIST if entry with the given name already | |
417 | * exists. | |
418 | */ | |
988cd7af | 419 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 420 | { |
db4aad20 | 421 | struct kernfs_node *parent = kn->parent; |
c525aadd | 422 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 423 | bool has_ns; |
fd7b9f7b TH |
424 | int ret; |
425 | ||
988cd7af TH |
426 | mutex_lock(&kernfs_mutex); |
427 | ||
428 | ret = -EINVAL; | |
429 | has_ns = kernfs_ns_enabled(parent); | |
430 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
431 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
432 | goto out_unlock; | |
fd7b9f7b | 433 | |
df23fc39 | 434 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 435 | goto out_unlock; |
fd7b9f7b | 436 | |
988cd7af | 437 | ret = -ENOENT; |
d35258ef | 438 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 439 | goto out_unlock; |
798c75a0 | 440 | |
c637b8ac | 441 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 442 | |
c637b8ac | 443 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 444 | if (ret) |
988cd7af | 445 | goto out_unlock; |
fd7b9f7b TH |
446 | |
447 | /* Update timestamps on the parent */ | |
adc5e8b5 | 448 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
449 | if (ps_iattr) { |
450 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
451 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; | |
452 | } | |
453 | ||
d35258ef TH |
454 | mutex_unlock(&kernfs_mutex); |
455 | ||
456 | /* | |
457 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
458 | * If not activated here, the kernfs user is responsible for | |
459 | * activating the node with kernfs_activate(). A node which hasn't | |
460 | * been activated is not visible to userland and its removal won't | |
461 | * trigger deactivation. | |
462 | */ | |
463 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
464 | kernfs_activate(kn); | |
465 | return 0; | |
466 | ||
988cd7af | 467 | out_unlock: |
a797bfc3 | 468 | mutex_unlock(&kernfs_mutex); |
988cd7af | 469 | return ret; |
fd7b9f7b TH |
470 | } |
471 | ||
472 | /** | |
324a56e1 TH |
473 | * kernfs_find_ns - find kernfs_node with the given name |
474 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
475 | * @name: name to look for |
476 | * @ns: the namespace tag to use | |
477 | * | |
324a56e1 TH |
478 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
479 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 480 | */ |
324a56e1 TH |
481 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
482 | const unsigned char *name, | |
483 | const void *ns) | |
fd7b9f7b | 484 | { |
adc5e8b5 | 485 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 486 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
487 | unsigned int hash; |
488 | ||
a797bfc3 | 489 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
490 | |
491 | if (has_ns != (bool)ns) { | |
c637b8ac | 492 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 493 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
494 | return NULL; |
495 | } | |
496 | ||
c637b8ac | 497 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 498 | while (node) { |
324a56e1 | 499 | struct kernfs_node *kn; |
fd7b9f7b TH |
500 | int result; |
501 | ||
324a56e1 | 502 | kn = rb_to_kn(node); |
c637b8ac | 503 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
504 | if (result < 0) |
505 | node = node->rb_left; | |
506 | else if (result > 0) | |
507 | node = node->rb_right; | |
508 | else | |
324a56e1 | 509 | return kn; |
fd7b9f7b TH |
510 | } |
511 | return NULL; | |
512 | } | |
513 | ||
514 | /** | |
324a56e1 TH |
515 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
516 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
517 | * @name: name to look for |
518 | * @ns: the namespace tag to use | |
519 | * | |
324a56e1 | 520 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 521 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 522 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 523 | */ |
324a56e1 TH |
524 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
525 | const char *name, const void *ns) | |
fd7b9f7b | 526 | { |
324a56e1 | 527 | struct kernfs_node *kn; |
fd7b9f7b | 528 | |
a797bfc3 | 529 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
530 | kn = kernfs_find_ns(parent, name, ns); |
531 | kernfs_get(kn); | |
a797bfc3 | 532 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 533 | |
324a56e1 | 534 | return kn; |
fd7b9f7b TH |
535 | } |
536 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
537 | ||
ba7443bc TH |
538 | /** |
539 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 540 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 541 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
542 | * @priv: opaque data associated with the new directory |
543 | * | |
544 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
545 | * failure. | |
546 | */ | |
90c07c89 | 547 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 548 | unsigned int flags, void *priv) |
ba7443bc TH |
549 | { |
550 | struct kernfs_root *root; | |
324a56e1 | 551 | struct kernfs_node *kn; |
ba7443bc TH |
552 | |
553 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
554 | if (!root) | |
555 | return ERR_PTR(-ENOMEM); | |
556 | ||
bc755553 TH |
557 | ida_init(&root->ino_ida); |
558 | ||
db4aad20 TH |
559 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
560 | KERNFS_DIR); | |
324a56e1 | 561 | if (!kn) { |
bc755553 | 562 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
563 | kfree(root); |
564 | return ERR_PTR(-ENOMEM); | |
565 | } | |
566 | ||
324a56e1 | 567 | kn->priv = priv; |
adc5e8b5 | 568 | kn->dir.root = root; |
ba7443bc | 569 | |
90c07c89 | 570 | root->syscall_ops = scops; |
d35258ef | 571 | root->flags = flags; |
324a56e1 | 572 | root->kn = kn; |
abd54f02 | 573 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 574 | |
d35258ef TH |
575 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
576 | kernfs_activate(kn); | |
577 | ||
ba7443bc TH |
578 | return root; |
579 | } | |
580 | ||
581 | /** | |
582 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
583 | * @root: root of the hierarchy to destroy | |
584 | * | |
585 | * Destroy the hierarchy anchored at @root by removing all existing | |
586 | * directories and destroying @root. | |
587 | */ | |
588 | void kernfs_destroy_root(struct kernfs_root *root) | |
589 | { | |
324a56e1 | 590 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
591 | } |
592 | ||
fd7b9f7b TH |
593 | /** |
594 | * kernfs_create_dir_ns - create a directory | |
595 | * @parent: parent in which to create a new directory | |
596 | * @name: name of the new directory | |
bb8b9d09 | 597 | * @mode: mode of the new directory |
fd7b9f7b TH |
598 | * @priv: opaque data associated with the new directory |
599 | * @ns: optional namespace tag of the directory | |
600 | * | |
601 | * Returns the created node on success, ERR_PTR() value on failure. | |
602 | */ | |
324a56e1 | 603 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 TH |
604 | const char *name, umode_t mode, |
605 | void *priv, const void *ns) | |
fd7b9f7b | 606 | { |
324a56e1 | 607 | struct kernfs_node *kn; |
fd7b9f7b TH |
608 | int rc; |
609 | ||
610 | /* allocate */ | |
db4aad20 | 611 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
324a56e1 | 612 | if (!kn) |
fd7b9f7b TH |
613 | return ERR_PTR(-ENOMEM); |
614 | ||
adc5e8b5 TH |
615 | kn->dir.root = parent->dir.root; |
616 | kn->ns = ns; | |
324a56e1 | 617 | kn->priv = priv; |
fd7b9f7b TH |
618 | |
619 | /* link in */ | |
988cd7af | 620 | rc = kernfs_add_one(kn); |
fd7b9f7b | 621 | if (!rc) |
324a56e1 | 622 | return kn; |
fd7b9f7b | 623 | |
324a56e1 | 624 | kernfs_put(kn); |
fd7b9f7b TH |
625 | return ERR_PTR(rc); |
626 | } | |
627 | ||
c637b8ac TH |
628 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
629 | struct dentry *dentry, | |
630 | unsigned int flags) | |
fd7b9f7b | 631 | { |
19bbb926 | 632 | struct dentry *ret; |
324a56e1 TH |
633 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
634 | struct kernfs_node *kn; | |
fd7b9f7b TH |
635 | struct inode *inode; |
636 | const void *ns = NULL; | |
637 | ||
a797bfc3 | 638 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 639 | |
324a56e1 | 640 | if (kernfs_ns_enabled(parent)) |
c525aadd | 641 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 642 | |
324a56e1 | 643 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
644 | |
645 | /* no such entry */ | |
b9c9dad0 | 646 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 647 | ret = NULL; |
fd7b9f7b TH |
648 | goto out_unlock; |
649 | } | |
324a56e1 TH |
650 | kernfs_get(kn); |
651 | dentry->d_fsdata = kn; | |
fd7b9f7b TH |
652 | |
653 | /* attach dentry and inode */ | |
c637b8ac | 654 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
655 | if (!inode) { |
656 | ret = ERR_PTR(-ENOMEM); | |
657 | goto out_unlock; | |
658 | } | |
659 | ||
660 | /* instantiate and hash dentry */ | |
661 | ret = d_materialise_unique(dentry, inode); | |
662 | out_unlock: | |
a797bfc3 | 663 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
664 | return ret; |
665 | } | |
666 | ||
80b9bbef TH |
667 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
668 | umode_t mode) | |
669 | { | |
670 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 671 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 672 | int ret; |
80b9bbef | 673 | |
90c07c89 | 674 | if (!scops || !scops->mkdir) |
80b9bbef TH |
675 | return -EPERM; |
676 | ||
07c7530d TH |
677 | if (!kernfs_get_active(parent)) |
678 | return -ENODEV; | |
679 | ||
90c07c89 | 680 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
681 | |
682 | kernfs_put_active(parent); | |
683 | return ret; | |
80b9bbef TH |
684 | } |
685 | ||
686 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
687 | { | |
688 | struct kernfs_node *kn = dentry->d_fsdata; | |
90c07c89 | 689 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 690 | int ret; |
80b9bbef | 691 | |
90c07c89 | 692 | if (!scops || !scops->rmdir) |
80b9bbef TH |
693 | return -EPERM; |
694 | ||
07c7530d TH |
695 | if (!kernfs_get_active(kn)) |
696 | return -ENODEV; | |
697 | ||
90c07c89 | 698 | ret = scops->rmdir(kn); |
07c7530d TH |
699 | |
700 | kernfs_put_active(kn); | |
701 | return ret; | |
80b9bbef TH |
702 | } |
703 | ||
704 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
705 | struct inode *new_dir, struct dentry *new_dentry) | |
706 | { | |
707 | struct kernfs_node *kn = old_dentry->d_fsdata; | |
708 | struct kernfs_node *new_parent = new_dir->i_private; | |
90c07c89 | 709 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 710 | int ret; |
80b9bbef | 711 | |
90c07c89 | 712 | if (!scops || !scops->rename) |
80b9bbef TH |
713 | return -EPERM; |
714 | ||
07c7530d TH |
715 | if (!kernfs_get_active(kn)) |
716 | return -ENODEV; | |
717 | ||
718 | if (!kernfs_get_active(new_parent)) { | |
719 | kernfs_put_active(kn); | |
720 | return -ENODEV; | |
721 | } | |
722 | ||
90c07c89 | 723 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
724 | |
725 | kernfs_put_active(new_parent); | |
726 | kernfs_put_active(kn); | |
727 | return ret; | |
80b9bbef TH |
728 | } |
729 | ||
a797bfc3 | 730 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
731 | .lookup = kernfs_iop_lookup, |
732 | .permission = kernfs_iop_permission, | |
733 | .setattr = kernfs_iop_setattr, | |
734 | .getattr = kernfs_iop_getattr, | |
735 | .setxattr = kernfs_iop_setxattr, | |
736 | .removexattr = kernfs_iop_removexattr, | |
737 | .getxattr = kernfs_iop_getxattr, | |
738 | .listxattr = kernfs_iop_listxattr, | |
80b9bbef TH |
739 | |
740 | .mkdir = kernfs_iop_mkdir, | |
741 | .rmdir = kernfs_iop_rmdir, | |
742 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
743 | }; |
744 | ||
c637b8ac | 745 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 746 | { |
324a56e1 | 747 | struct kernfs_node *last; |
fd7b9f7b TH |
748 | |
749 | while (true) { | |
750 | struct rb_node *rbn; | |
751 | ||
752 | last = pos; | |
753 | ||
df23fc39 | 754 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
755 | break; |
756 | ||
adc5e8b5 | 757 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
758 | if (!rbn) |
759 | break; | |
760 | ||
324a56e1 | 761 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
762 | } |
763 | ||
764 | return last; | |
765 | } | |
766 | ||
767 | /** | |
c637b8ac | 768 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 769 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 770 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
771 | * |
772 | * Find the next descendant to visit for post-order traversal of @root's | |
773 | * descendants. @root is included in the iteration and the last node to be | |
774 | * visited. | |
775 | */ | |
c637b8ac TH |
776 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
777 | struct kernfs_node *root) | |
fd7b9f7b TH |
778 | { |
779 | struct rb_node *rbn; | |
780 | ||
a797bfc3 | 781 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
782 | |
783 | /* if first iteration, visit leftmost descendant which may be root */ | |
784 | if (!pos) | |
c637b8ac | 785 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
786 | |
787 | /* if we visited @root, we're done */ | |
788 | if (pos == root) | |
789 | return NULL; | |
790 | ||
791 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 792 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 793 | if (rbn) |
c637b8ac | 794 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
795 | |
796 | /* no sibling left, visit parent */ | |
adc5e8b5 | 797 | return pos->parent; |
fd7b9f7b TH |
798 | } |
799 | ||
d35258ef TH |
800 | /** |
801 | * kernfs_activate - activate a node which started deactivated | |
802 | * @kn: kernfs_node whose subtree is to be activated | |
803 | * | |
804 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
805 | * needs to be explicitly activated. A node which hasn't been activated | |
806 | * isn't visible to userland and deactivation is skipped during its | |
807 | * removal. This is useful to construct atomic init sequences where | |
808 | * creation of multiple nodes should either succeed or fail atomically. | |
809 | * | |
810 | * The caller is responsible for ensuring that this function is not called | |
811 | * after kernfs_remove*() is invoked on @kn. | |
812 | */ | |
813 | void kernfs_activate(struct kernfs_node *kn) | |
814 | { | |
815 | struct kernfs_node *pos; | |
816 | ||
817 | mutex_lock(&kernfs_mutex); | |
818 | ||
819 | pos = NULL; | |
820 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
821 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
822 | continue; | |
823 | ||
824 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
825 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
826 | ||
827 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
828 | pos->flags |= KERNFS_ACTIVATED; | |
829 | } | |
830 | ||
831 | mutex_unlock(&kernfs_mutex); | |
832 | } | |
833 | ||
988cd7af | 834 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 835 | { |
35beab06 TH |
836 | struct kernfs_node *pos; |
837 | ||
838 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 839 | |
6b0afc2a TH |
840 | /* |
841 | * Short-circuit if non-root @kn has already finished removal. | |
842 | * This is for kernfs_remove_self() which plays with active ref | |
843 | * after removal. | |
844 | */ | |
845 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
846 | return; |
847 | ||
c637b8ac | 848 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 849 | |
81c173cb | 850 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
851 | pos = NULL; |
852 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
853 | if (kernfs_active(pos)) |
854 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
855 | |
856 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 857 | do { |
35beab06 TH |
858 | pos = kernfs_leftmost_descendant(kn); |
859 | ||
860 | /* | |
81c173cb TH |
861 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
862 | * base ref could have been put by someone else by the time | |
863 | * the function returns. Make sure it doesn't go away | |
864 | * underneath us. | |
35beab06 TH |
865 | */ |
866 | kernfs_get(pos); | |
867 | ||
d35258ef TH |
868 | /* |
869 | * Drain iff @kn was activated. This avoids draining and | |
870 | * its lockdep annotations for nodes which have never been | |
871 | * activated and allows embedding kernfs_remove() in create | |
872 | * error paths without worrying about draining. | |
873 | */ | |
874 | if (kn->flags & KERNFS_ACTIVATED) | |
875 | kernfs_drain(pos); | |
876 | else | |
877 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
878 | |
879 | /* | |
880 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
881 | * to decide who's responsible for cleanups. | |
882 | */ | |
883 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
884 | struct kernfs_iattrs *ps_iattr = | |
885 | pos->parent ? pos->parent->iattr : NULL; | |
886 | ||
887 | /* update timestamps on the parent */ | |
888 | if (ps_iattr) { | |
889 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; | |
890 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; | |
891 | } | |
892 | ||
988cd7af | 893 | kernfs_put(pos); |
35beab06 TH |
894 | } |
895 | ||
896 | kernfs_put(pos); | |
897 | } while (pos != kn); | |
fd7b9f7b TH |
898 | } |
899 | ||
900 | /** | |
324a56e1 TH |
901 | * kernfs_remove - remove a kernfs_node recursively |
902 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 903 | * |
324a56e1 | 904 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 905 | */ |
324a56e1 | 906 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 907 | { |
988cd7af TH |
908 | mutex_lock(&kernfs_mutex); |
909 | __kernfs_remove(kn); | |
910 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
911 | } |
912 | ||
6b0afc2a TH |
913 | /** |
914 | * kernfs_break_active_protection - break out of active protection | |
915 | * @kn: the self kernfs_node | |
916 | * | |
917 | * The caller must be running off of a kernfs operation which is invoked | |
918 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
919 | * this function must also be matched with an invocation of | |
920 | * kernfs_unbreak_active_protection(). | |
921 | * | |
922 | * This function releases the active reference of @kn the caller is | |
923 | * holding. Once this function is called, @kn may be removed at any point | |
924 | * and the caller is solely responsible for ensuring that the objects it | |
925 | * dereferences are accessible. | |
926 | */ | |
927 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
928 | { | |
929 | /* | |
930 | * Take out ourself out of the active ref dependency chain. If | |
931 | * we're called without an active ref, lockdep will complain. | |
932 | */ | |
933 | kernfs_put_active(kn); | |
934 | } | |
935 | ||
936 | /** | |
937 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
938 | * @kn: the self kernfs_node | |
939 | * | |
940 | * If kernfs_break_active_protection() was called, this function must be | |
941 | * invoked before finishing the kernfs operation. Note that while this | |
942 | * function restores the active reference, it doesn't and can't actually | |
943 | * restore the active protection - @kn may already or be in the process of | |
944 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
945 | * protection is irreversibly gone for the kernfs operation instance. | |
946 | * | |
947 | * While this function may be called at any point after | |
948 | * kernfs_break_active_protection() is invoked, its most useful location | |
949 | * would be right before the enclosing kernfs operation returns. | |
950 | */ | |
951 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
952 | { | |
953 | /* | |
954 | * @kn->active could be in any state; however, the increment we do | |
955 | * here will be undone as soon as the enclosing kernfs operation | |
956 | * finishes and this temporary bump can't break anything. If @kn | |
957 | * is alive, nothing changes. If @kn is being deactivated, the | |
958 | * soon-to-follow put will either finish deactivation or restore | |
959 | * deactivated state. If @kn is already removed, the temporary | |
960 | * bump is guaranteed to be gone before @kn is released. | |
961 | */ | |
962 | atomic_inc(&kn->active); | |
963 | if (kernfs_lockdep(kn)) | |
964 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
965 | } | |
966 | ||
967 | /** | |
968 | * kernfs_remove_self - remove a kernfs_node from its own method | |
969 | * @kn: the self kernfs_node to remove | |
970 | * | |
971 | * The caller must be running off of a kernfs operation which is invoked | |
972 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
973 | * implement a file operation which deletes itself. | |
974 | * | |
975 | * For example, the "delete" file for a sysfs device directory can be | |
976 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
977 | * itself. This function breaks the circular dependency of trying to | |
978 | * deactivate self while holding an active ref itself. It isn't necessary | |
979 | * to modify the usual removal path to use kernfs_remove_self(). The | |
980 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
981 | * before proceeding with the usual removal path. kernfs will ignore later | |
982 | * kernfs_remove() on self. | |
983 | * | |
984 | * kernfs_remove_self() can be called multiple times concurrently on the | |
985 | * same kernfs_node. Only the first one actually performs removal and | |
986 | * returns %true. All others will wait until the kernfs operation which | |
987 | * won self-removal finishes and return %false. Note that the losers wait | |
988 | * for the completion of not only the winning kernfs_remove_self() but also | |
989 | * the whole kernfs_ops which won the arbitration. This can be used to | |
990 | * guarantee, for example, all concurrent writes to a "delete" file to | |
991 | * finish only after the whole operation is complete. | |
992 | */ | |
993 | bool kernfs_remove_self(struct kernfs_node *kn) | |
994 | { | |
995 | bool ret; | |
996 | ||
997 | mutex_lock(&kernfs_mutex); | |
998 | kernfs_break_active_protection(kn); | |
999 | ||
1000 | /* | |
1001 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1002 | * the first one will actually perform removal. When the removal | |
1003 | * is complete, SUICIDED is set and the active ref is restored | |
1004 | * while holding kernfs_mutex. The ones which lost arbitration | |
1005 | * waits for SUICDED && drained which can happen only after the | |
1006 | * enclosing kernfs operation which executed the winning instance | |
1007 | * of kernfs_remove_self() finished. | |
1008 | */ | |
1009 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1010 | kn->flags |= KERNFS_SUICIDAL; | |
1011 | __kernfs_remove(kn); | |
1012 | kn->flags |= KERNFS_SUICIDED; | |
1013 | ret = true; | |
1014 | } else { | |
1015 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1016 | DEFINE_WAIT(wait); | |
1017 | ||
1018 | while (true) { | |
1019 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1020 | ||
1021 | if ((kn->flags & KERNFS_SUICIDED) && | |
1022 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1023 | break; | |
1024 | ||
1025 | mutex_unlock(&kernfs_mutex); | |
1026 | schedule(); | |
1027 | mutex_lock(&kernfs_mutex); | |
1028 | } | |
1029 | finish_wait(waitq, &wait); | |
1030 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1031 | ret = false; | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1036 | * for SUICIDED && deactivated could finish prematurely. | |
1037 | */ | |
1038 | kernfs_unbreak_active_protection(kn); | |
1039 | ||
1040 | mutex_unlock(&kernfs_mutex); | |
1041 | return ret; | |
1042 | } | |
1043 | ||
fd7b9f7b | 1044 | /** |
324a56e1 TH |
1045 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1046 | * @parent: parent of the target | |
1047 | * @name: name of the kernfs_node to remove | |
1048 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1049 | * |
324a56e1 TH |
1050 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1051 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1052 | */ |
324a56e1 | 1053 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1054 | const void *ns) |
1055 | { | |
324a56e1 | 1056 | struct kernfs_node *kn; |
fd7b9f7b | 1057 | |
324a56e1 | 1058 | if (!parent) { |
c637b8ac | 1059 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1060 | name); |
1061 | return -ENOENT; | |
1062 | } | |
1063 | ||
988cd7af | 1064 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1065 | |
324a56e1 TH |
1066 | kn = kernfs_find_ns(parent, name, ns); |
1067 | if (kn) | |
988cd7af | 1068 | __kernfs_remove(kn); |
fd7b9f7b | 1069 | |
988cd7af | 1070 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1071 | |
324a56e1 | 1072 | if (kn) |
fd7b9f7b TH |
1073 | return 0; |
1074 | else | |
1075 | return -ENOENT; | |
1076 | } | |
1077 | ||
1078 | /** | |
1079 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1080 | * @kn: target node |
fd7b9f7b TH |
1081 | * @new_parent: new parent to put @sd under |
1082 | * @new_name: new name | |
1083 | * @new_ns: new namespace tag | |
1084 | */ | |
324a56e1 | 1085 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1086 | const char *new_name, const void *new_ns) |
1087 | { | |
1088 | int error; | |
1089 | ||
798c75a0 GKH |
1090 | mutex_lock(&kernfs_mutex); |
1091 | ||
d0ae3d43 | 1092 | error = -ENOENT; |
81c173cb | 1093 | if (!kernfs_active(kn) || !kernfs_active(new_parent)) |
d0ae3d43 TH |
1094 | goto out; |
1095 | ||
fd7b9f7b | 1096 | error = 0; |
adc5e8b5 TH |
1097 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1098 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1099 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1100 | |
1101 | error = -EEXIST; | |
1102 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1103 | goto out; |
fd7b9f7b | 1104 | |
324a56e1 | 1105 | /* rename kernfs_node */ |
adc5e8b5 | 1106 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b TH |
1107 | error = -ENOMEM; |
1108 | new_name = kstrdup(new_name, GFP_KERNEL); | |
1109 | if (!new_name) | |
798c75a0 | 1110 | goto out; |
fd7b9f7b | 1111 | |
47a52e91 TH |
1112 | if (kn->flags & KERNFS_STATIC_NAME) |
1113 | kn->flags &= ~KERNFS_STATIC_NAME; | |
1114 | else | |
1115 | kfree(kn->name); | |
1116 | ||
adc5e8b5 | 1117 | kn->name = new_name; |
fd7b9f7b TH |
1118 | } |
1119 | ||
1120 | /* | |
1121 | * Move to the appropriate place in the appropriate directories rbtree. | |
1122 | */ | |
c637b8ac | 1123 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1124 | kernfs_get(new_parent); |
adc5e8b5 TH |
1125 | kernfs_put(kn->parent); |
1126 | kn->ns = new_ns; | |
c637b8ac | 1127 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
adc5e8b5 | 1128 | kn->parent = new_parent; |
c637b8ac | 1129 | kernfs_link_sibling(kn); |
fd7b9f7b TH |
1130 | |
1131 | error = 0; | |
798c75a0 | 1132 | out: |
a797bfc3 | 1133 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1134 | return error; |
1135 | } | |
1136 | ||
fd7b9f7b | 1137 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1138 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1139 | { |
adc5e8b5 | 1140 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1141 | } |
1142 | ||
c637b8ac | 1143 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1144 | { |
1145 | kernfs_put(filp->private_data); | |
1146 | return 0; | |
1147 | } | |
1148 | ||
c637b8ac | 1149 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1150 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1151 | { |
1152 | if (pos) { | |
81c173cb | 1153 | int valid = kernfs_active(pos) && |
798c75a0 | 1154 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1155 | kernfs_put(pos); |
1156 | if (!valid) | |
1157 | pos = NULL; | |
1158 | } | |
1159 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1160 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1161 | while (node) { |
324a56e1 | 1162 | pos = rb_to_kn(node); |
fd7b9f7b | 1163 | |
adc5e8b5 | 1164 | if (hash < pos->hash) |
fd7b9f7b | 1165 | node = node->rb_left; |
adc5e8b5 | 1166 | else if (hash > pos->hash) |
fd7b9f7b TH |
1167 | node = node->rb_right; |
1168 | else | |
1169 | break; | |
1170 | } | |
1171 | } | |
b9c9dad0 TH |
1172 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1173 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1174 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1175 | if (!node) |
1176 | pos = NULL; | |
1177 | else | |
324a56e1 | 1178 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1179 | } |
1180 | return pos; | |
1181 | } | |
1182 | ||
c637b8ac | 1183 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1184 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1185 | { |
c637b8ac | 1186 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1187 | if (pos) { |
fd7b9f7b | 1188 | do { |
adc5e8b5 | 1189 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1190 | if (!node) |
1191 | pos = NULL; | |
1192 | else | |
324a56e1 | 1193 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1194 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1195 | } | |
fd7b9f7b TH |
1196 | return pos; |
1197 | } | |
1198 | ||
c637b8ac | 1199 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1200 | { |
1201 | struct dentry *dentry = file->f_path.dentry; | |
324a56e1 TH |
1202 | struct kernfs_node *parent = dentry->d_fsdata; |
1203 | struct kernfs_node *pos = file->private_data; | |
fd7b9f7b TH |
1204 | const void *ns = NULL; |
1205 | ||
1206 | if (!dir_emit_dots(file, ctx)) | |
1207 | return 0; | |
a797bfc3 | 1208 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1209 | |
324a56e1 | 1210 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1211 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1212 | |
c637b8ac | 1213 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1214 | pos; |
c637b8ac | 1215 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1216 | const char *name = pos->name; |
fd7b9f7b TH |
1217 | unsigned int type = dt_type(pos); |
1218 | int len = strlen(name); | |
adc5e8b5 | 1219 | ino_t ino = pos->ino; |
fd7b9f7b | 1220 | |
adc5e8b5 | 1221 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1222 | file->private_data = pos; |
1223 | kernfs_get(pos); | |
1224 | ||
a797bfc3 | 1225 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1226 | if (!dir_emit(ctx, name, len, ino, type)) |
1227 | return 0; | |
a797bfc3 | 1228 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1229 | } |
a797bfc3 | 1230 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1231 | file->private_data = NULL; |
1232 | ctx->pos = INT_MAX; | |
1233 | return 0; | |
1234 | } | |
1235 | ||
c637b8ac TH |
1236 | static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset, |
1237 | int whence) | |
fd7b9f7b TH |
1238 | { |
1239 | struct inode *inode = file_inode(file); | |
1240 | loff_t ret; | |
1241 | ||
1242 | mutex_lock(&inode->i_mutex); | |
1243 | ret = generic_file_llseek(file, offset, whence); | |
1244 | mutex_unlock(&inode->i_mutex); | |
1245 | ||
1246 | return ret; | |
1247 | } | |
1248 | ||
a797bfc3 | 1249 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1250 | .read = generic_read_dir, |
c637b8ac TH |
1251 | .iterate = kernfs_fop_readdir, |
1252 | .release = kernfs_dir_fop_release, | |
1253 | .llseek = kernfs_dir_fop_llseek, | |
fd7b9f7b | 1254 | }; |