| 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 | */ |
| 10 | |
| 11 | #include <linux/sched.h> |
| 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 | |
| 21 | DEFINE_MUTEX(kernfs_mutex); |
| 22 | static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */ |
| 23 | static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */ |
| 24 | |
| 25 | #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb) |
| 26 | |
| 27 | static bool kernfs_active(struct kernfs_node *kn) |
| 28 | { |
| 29 | lockdep_assert_held(&kernfs_mutex); |
| 30 | return atomic_read(&kn->active) >= 0; |
| 31 | } |
| 32 | |
| 33 | static bool kernfs_lockdep(struct kernfs_node *kn) |
| 34 | { |
| 35 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 36 | return kn->flags & KERNFS_LOCKDEP; |
| 37 | #else |
| 38 | return false; |
| 39 | #endif |
| 40 | } |
| 41 | |
| 42 | static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen) |
| 43 | { |
| 44 | return strlcpy(buf, kn->parent ? kn->name : "/", buflen); |
| 45 | } |
| 46 | |
| 47 | static char * __must_check kernfs_path_locked(struct kernfs_node *kn, char *buf, |
| 48 | size_t buflen) |
| 49 | { |
| 50 | char *p = buf + buflen; |
| 51 | int len; |
| 52 | |
| 53 | *--p = '\0'; |
| 54 | |
| 55 | do { |
| 56 | len = strlen(kn->name); |
| 57 | if (p - buf < len + 1) { |
| 58 | buf[0] = '\0'; |
| 59 | p = NULL; |
| 60 | break; |
| 61 | } |
| 62 | p -= len; |
| 63 | memcpy(p, kn->name, len); |
| 64 | *--p = '/'; |
| 65 | kn = kn->parent; |
| 66 | } while (kn && kn->parent); |
| 67 | |
| 68 | return p; |
| 69 | } |
| 70 | |
| 71 | /** |
| 72 | * kernfs_name - obtain the name of a given node |
| 73 | * @kn: kernfs_node of interest |
| 74 | * @buf: buffer to copy @kn's name into |
| 75 | * @buflen: size of @buf |
| 76 | * |
| 77 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is |
| 78 | * similar to strlcpy(). It returns the length of @kn's name and if @buf |
| 79 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. |
| 80 | * |
| 81 | * This function can be called from any context. |
| 82 | */ |
| 83 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) |
| 84 | { |
| 85 | unsigned long flags; |
| 86 | int ret; |
| 87 | |
| 88 | spin_lock_irqsave(&kernfs_rename_lock, flags); |
| 89 | ret = kernfs_name_locked(kn, buf, buflen); |
| 90 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
| 91 | return ret; |
| 92 | } |
| 93 | |
| 94 | /** |
| 95 | * kernfs_path_len - determine the length of the full path of a given node |
| 96 | * @kn: kernfs_node of interest |
| 97 | * |
| 98 | * The returned length doesn't include the space for the terminating '\0'. |
| 99 | */ |
| 100 | size_t kernfs_path_len(struct kernfs_node *kn) |
| 101 | { |
| 102 | size_t len = 0; |
| 103 | unsigned long flags; |
| 104 | |
| 105 | spin_lock_irqsave(&kernfs_rename_lock, flags); |
| 106 | |
| 107 | do { |
| 108 | len += strlen(kn->name) + 1; |
| 109 | kn = kn->parent; |
| 110 | } while (kn && kn->parent); |
| 111 | |
| 112 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
| 113 | |
| 114 | return len; |
| 115 | } |
| 116 | |
| 117 | /** |
| 118 | * kernfs_path - build full path of a given node |
| 119 | * @kn: kernfs_node of interest |
| 120 | * @buf: buffer to copy @kn's name into |
| 121 | * @buflen: size of @buf |
| 122 | * |
| 123 | * Builds and returns the full path of @kn in @buf of @buflen bytes. The |
| 124 | * path is built from the end of @buf so the returned pointer usually |
| 125 | * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated |
| 126 | * and %NULL is returned. |
| 127 | */ |
| 128 | char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen) |
| 129 | { |
| 130 | unsigned long flags; |
| 131 | char *p; |
| 132 | |
| 133 | spin_lock_irqsave(&kernfs_rename_lock, flags); |
| 134 | p = kernfs_path_locked(kn, buf, buflen); |
| 135 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
| 136 | return p; |
| 137 | } |
| 138 | EXPORT_SYMBOL_GPL(kernfs_path); |
| 139 | |
| 140 | /** |
| 141 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node |
| 142 | * @kn: kernfs_node of interest |
| 143 | * |
| 144 | * This function can be called from any context. |
| 145 | */ |
| 146 | void pr_cont_kernfs_name(struct kernfs_node *kn) |
| 147 | { |
| 148 | unsigned long flags; |
| 149 | |
| 150 | spin_lock_irqsave(&kernfs_rename_lock, flags); |
| 151 | |
| 152 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); |
| 153 | pr_cont("%s", kernfs_pr_cont_buf); |
| 154 | |
| 155 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
| 156 | } |
| 157 | |
| 158 | /** |
| 159 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node |
| 160 | * @kn: kernfs_node of interest |
| 161 | * |
| 162 | * This function can be called from any context. |
| 163 | */ |
| 164 | void pr_cont_kernfs_path(struct kernfs_node *kn) |
| 165 | { |
| 166 | unsigned long flags; |
| 167 | char *p; |
| 168 | |
| 169 | spin_lock_irqsave(&kernfs_rename_lock, flags); |
| 170 | |
| 171 | p = kernfs_path_locked(kn, kernfs_pr_cont_buf, |
| 172 | sizeof(kernfs_pr_cont_buf)); |
| 173 | if (p) |
| 174 | pr_cont("%s", p); |
| 175 | else |
| 176 | pr_cont("<name too long>"); |
| 177 | |
| 178 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
| 179 | } |
| 180 | |
| 181 | /** |
| 182 | * kernfs_get_parent - determine the parent node and pin it |
| 183 | * @kn: kernfs_node of interest |
| 184 | * |
| 185 | * Determines @kn's parent, pins and returns it. This function can be |
| 186 | * called from any context. |
| 187 | */ |
| 188 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) |
| 189 | { |
| 190 | struct kernfs_node *parent; |
| 191 | unsigned long flags; |
| 192 | |
| 193 | spin_lock_irqsave(&kernfs_rename_lock, flags); |
| 194 | parent = kn->parent; |
| 195 | kernfs_get(parent); |
| 196 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
| 197 | |
| 198 | return parent; |
| 199 | } |
| 200 | |
| 201 | /** |
| 202 | * kernfs_name_hash |
| 203 | * @name: Null terminated string to hash |
| 204 | * @ns: Namespace tag to hash |
| 205 | * |
| 206 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) |
| 207 | */ |
| 208 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
| 209 | { |
| 210 | unsigned long hash = init_name_hash(); |
| 211 | unsigned int len = strlen(name); |
| 212 | while (len--) |
| 213 | hash = partial_name_hash(*name++, hash); |
| 214 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); |
| 215 | hash &= 0x7fffffffU; |
| 216 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ |
| 217 | if (hash < 2) |
| 218 | hash += 2; |
| 219 | if (hash >= INT_MAX) |
| 220 | hash = INT_MAX - 1; |
| 221 | return hash; |
| 222 | } |
| 223 | |
| 224 | static int kernfs_name_compare(unsigned int hash, const char *name, |
| 225 | const void *ns, const struct kernfs_node *kn) |
| 226 | { |
| 227 | if (hash < kn->hash) |
| 228 | return -1; |
| 229 | if (hash > kn->hash) |
| 230 | return 1; |
| 231 | if (ns < kn->ns) |
| 232 | return -1; |
| 233 | if (ns > kn->ns) |
| 234 | return 1; |
| 235 | return strcmp(name, kn->name); |
| 236 | } |
| 237 | |
| 238 | static int kernfs_sd_compare(const struct kernfs_node *left, |
| 239 | const struct kernfs_node *right) |
| 240 | { |
| 241 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
| 242 | } |
| 243 | |
| 244 | /** |
| 245 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
| 246 | * @kn: kernfs_node of interest |
| 247 | * |
| 248 | * Link @kn into its sibling rbtree which starts from |
| 249 | * @kn->parent->dir.children. |
| 250 | * |
| 251 | * Locking: |
| 252 | * mutex_lock(kernfs_mutex) |
| 253 | * |
| 254 | * RETURNS: |
| 255 | * 0 on susccess -EEXIST on failure. |
| 256 | */ |
| 257 | static int kernfs_link_sibling(struct kernfs_node *kn) |
| 258 | { |
| 259 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
| 260 | struct rb_node *parent = NULL; |
| 261 | |
| 262 | while (*node) { |
| 263 | struct kernfs_node *pos; |
| 264 | int result; |
| 265 | |
| 266 | pos = rb_to_kn(*node); |
| 267 | parent = *node; |
| 268 | result = kernfs_sd_compare(kn, pos); |
| 269 | if (result < 0) |
| 270 | node = &pos->rb.rb_left; |
| 271 | else if (result > 0) |
| 272 | node = &pos->rb.rb_right; |
| 273 | else |
| 274 | return -EEXIST; |
| 275 | } |
| 276 | |
| 277 | /* add new node and rebalance the tree */ |
| 278 | rb_link_node(&kn->rb, parent, node); |
| 279 | rb_insert_color(&kn->rb, &kn->parent->dir.children); |
| 280 | |
| 281 | /* successfully added, account subdir number */ |
| 282 | if (kernfs_type(kn) == KERNFS_DIR) |
| 283 | kn->parent->dir.subdirs++; |
| 284 | |
| 285 | return 0; |
| 286 | } |
| 287 | |
| 288 | /** |
| 289 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
| 290 | * @kn: kernfs_node of interest |
| 291 | * |
| 292 | * Try to unlink @kn from its sibling rbtree which starts from |
| 293 | * kn->parent->dir.children. Returns %true if @kn was actually |
| 294 | * removed, %false if @kn wasn't on the rbtree. |
| 295 | * |
| 296 | * Locking: |
| 297 | * mutex_lock(kernfs_mutex) |
| 298 | */ |
| 299 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
| 300 | { |
| 301 | if (RB_EMPTY_NODE(&kn->rb)) |
| 302 | return false; |
| 303 | |
| 304 | if (kernfs_type(kn) == KERNFS_DIR) |
| 305 | kn->parent->dir.subdirs--; |
| 306 | |
| 307 | rb_erase(&kn->rb, &kn->parent->dir.children); |
| 308 | RB_CLEAR_NODE(&kn->rb); |
| 309 | return true; |
| 310 | } |
| 311 | |
| 312 | /** |
| 313 | * kernfs_get_active - get an active reference to kernfs_node |
| 314 | * @kn: kernfs_node to get an active reference to |
| 315 | * |
| 316 | * Get an active reference of @kn. This function is noop if @kn |
| 317 | * is NULL. |
| 318 | * |
| 319 | * RETURNS: |
| 320 | * Pointer to @kn on success, NULL on failure. |
| 321 | */ |
| 322 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
| 323 | { |
| 324 | if (unlikely(!kn)) |
| 325 | return NULL; |
| 326 | |
| 327 | if (!atomic_inc_unless_negative(&kn->active)) |
| 328 | return NULL; |
| 329 | |
| 330 | if (kernfs_lockdep(kn)) |
| 331 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
| 332 | return kn; |
| 333 | } |
| 334 | |
| 335 | /** |
| 336 | * kernfs_put_active - put an active reference to kernfs_node |
| 337 | * @kn: kernfs_node to put an active reference to |
| 338 | * |
| 339 | * Put an active reference to @kn. This function is noop if @kn |
| 340 | * is NULL. |
| 341 | */ |
| 342 | void kernfs_put_active(struct kernfs_node *kn) |
| 343 | { |
| 344 | struct kernfs_root *root = kernfs_root(kn); |
| 345 | int v; |
| 346 | |
| 347 | if (unlikely(!kn)) |
| 348 | return; |
| 349 | |
| 350 | if (kernfs_lockdep(kn)) |
| 351 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
| 352 | v = atomic_dec_return(&kn->active); |
| 353 | if (likely(v != KN_DEACTIVATED_BIAS)) |
| 354 | return; |
| 355 | |
| 356 | wake_up_all(&root->deactivate_waitq); |
| 357 | } |
| 358 | |
| 359 | /** |
| 360 | * kernfs_drain - drain kernfs_node |
| 361 | * @kn: kernfs_node to drain |
| 362 | * |
| 363 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
| 364 | * removers may invoke this function concurrently on @kn and all will |
| 365 | * return after draining is complete. |
| 366 | */ |
| 367 | static void kernfs_drain(struct kernfs_node *kn) |
| 368 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
| 369 | { |
| 370 | struct kernfs_root *root = kernfs_root(kn); |
| 371 | |
| 372 | lockdep_assert_held(&kernfs_mutex); |
| 373 | WARN_ON_ONCE(kernfs_active(kn)); |
| 374 | |
| 375 | mutex_unlock(&kernfs_mutex); |
| 376 | |
| 377 | if (kernfs_lockdep(kn)) { |
| 378 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
| 379 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) |
| 380 | lock_contended(&kn->dep_map, _RET_IP_); |
| 381 | } |
| 382 | |
| 383 | /* but everyone should wait for draining */ |
| 384 | wait_event(root->deactivate_waitq, |
| 385 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); |
| 386 | |
| 387 | if (kernfs_lockdep(kn)) { |
| 388 | lock_acquired(&kn->dep_map, _RET_IP_); |
| 389 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
| 390 | } |
| 391 | |
| 392 | kernfs_unmap_bin_file(kn); |
| 393 | |
| 394 | mutex_lock(&kernfs_mutex); |
| 395 | } |
| 396 | |
| 397 | /** |
| 398 | * kernfs_get - get a reference count on a kernfs_node |
| 399 | * @kn: the target kernfs_node |
| 400 | */ |
| 401 | void kernfs_get(struct kernfs_node *kn) |
| 402 | { |
| 403 | if (kn) { |
| 404 | WARN_ON(!atomic_read(&kn->count)); |
| 405 | atomic_inc(&kn->count); |
| 406 | } |
| 407 | } |
| 408 | EXPORT_SYMBOL_GPL(kernfs_get); |
| 409 | |
| 410 | /** |
| 411 | * kernfs_put - put a reference count on a kernfs_node |
| 412 | * @kn: the target kernfs_node |
| 413 | * |
| 414 | * Put a reference count of @kn and destroy it if it reached zero. |
| 415 | */ |
| 416 | void kernfs_put(struct kernfs_node *kn) |
| 417 | { |
| 418 | struct kernfs_node *parent; |
| 419 | struct kernfs_root *root; |
| 420 | |
| 421 | if (!kn || !atomic_dec_and_test(&kn->count)) |
| 422 | return; |
| 423 | root = kernfs_root(kn); |
| 424 | repeat: |
| 425 | /* |
| 426 | * Moving/renaming is always done while holding reference. |
| 427 | * kn->parent won't change beneath us. |
| 428 | */ |
| 429 | parent = kn->parent; |
| 430 | |
| 431 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
| 432 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", |
| 433 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); |
| 434 | |
| 435 | if (kernfs_type(kn) == KERNFS_LINK) |
| 436 | kernfs_put(kn->symlink.target_kn); |
| 437 | |
| 438 | kfree_const(kn->name); |
| 439 | |
| 440 | if (kn->iattr) { |
| 441 | if (kn->iattr->ia_secdata) |
| 442 | security_release_secctx(kn->iattr->ia_secdata, |
| 443 | kn->iattr->ia_secdata_len); |
| 444 | simple_xattrs_free(&kn->iattr->xattrs); |
| 445 | } |
| 446 | kfree(kn->iattr); |
| 447 | ida_simple_remove(&root->ino_ida, kn->ino); |
| 448 | kmem_cache_free(kernfs_node_cache, kn); |
| 449 | |
| 450 | kn = parent; |
| 451 | if (kn) { |
| 452 | if (atomic_dec_and_test(&kn->count)) |
| 453 | goto repeat; |
| 454 | } else { |
| 455 | /* just released the root kn, free @root too */ |
| 456 | ida_destroy(&root->ino_ida); |
| 457 | kfree(root); |
| 458 | } |
| 459 | } |
| 460 | EXPORT_SYMBOL_GPL(kernfs_put); |
| 461 | |
| 462 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
| 463 | { |
| 464 | struct kernfs_node *kn; |
| 465 | |
| 466 | if (flags & LOOKUP_RCU) |
| 467 | return -ECHILD; |
| 468 | |
| 469 | /* Always perform fresh lookup for negatives */ |
| 470 | if (d_really_is_negative(dentry)) |
| 471 | goto out_bad_unlocked; |
| 472 | |
| 473 | kn = dentry->d_fsdata; |
| 474 | mutex_lock(&kernfs_mutex); |
| 475 | |
| 476 | /* The kernfs node has been deactivated */ |
| 477 | if (!kernfs_active(kn)) |
| 478 | goto out_bad; |
| 479 | |
| 480 | /* The kernfs node has been moved? */ |
| 481 | if (dentry->d_parent->d_fsdata != kn->parent) |
| 482 | goto out_bad; |
| 483 | |
| 484 | /* The kernfs node has been renamed */ |
| 485 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
| 486 | goto out_bad; |
| 487 | |
| 488 | /* The kernfs node has been moved to a different namespace */ |
| 489 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
| 490 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
| 491 | goto out_bad; |
| 492 | |
| 493 | mutex_unlock(&kernfs_mutex); |
| 494 | return 1; |
| 495 | out_bad: |
| 496 | mutex_unlock(&kernfs_mutex); |
| 497 | out_bad_unlocked: |
| 498 | return 0; |
| 499 | } |
| 500 | |
| 501 | static void kernfs_dop_release(struct dentry *dentry) |
| 502 | { |
| 503 | kernfs_put(dentry->d_fsdata); |
| 504 | } |
| 505 | |
| 506 | const struct dentry_operations kernfs_dops = { |
| 507 | .d_revalidate = kernfs_dop_revalidate, |
| 508 | .d_release = kernfs_dop_release, |
| 509 | }; |
| 510 | |
| 511 | /** |
| 512 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry |
| 513 | * @dentry: the dentry in question |
| 514 | * |
| 515 | * Return the kernfs_node associated with @dentry. If @dentry is not a |
| 516 | * kernfs one, %NULL is returned. |
| 517 | * |
| 518 | * While the returned kernfs_node will stay accessible as long as @dentry |
| 519 | * is accessible, the returned node can be in any state and the caller is |
| 520 | * fully responsible for determining what's accessible. |
| 521 | */ |
| 522 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) |
| 523 | { |
| 524 | if (dentry->d_sb->s_op == &kernfs_sops) |
| 525 | return dentry->d_fsdata; |
| 526 | return NULL; |
| 527 | } |
| 528 | |
| 529 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
| 530 | const char *name, umode_t mode, |
| 531 | unsigned flags) |
| 532 | { |
| 533 | struct kernfs_node *kn; |
| 534 | int ret; |
| 535 | |
| 536 | name = kstrdup_const(name, GFP_KERNEL); |
| 537 | if (!name) |
| 538 | return NULL; |
| 539 | |
| 540 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
| 541 | if (!kn) |
| 542 | goto err_out1; |
| 543 | |
| 544 | /* |
| 545 | * If the ino of the sysfs entry created for a kmem cache gets |
| 546 | * allocated from an ida layer, which is accounted to the memcg that |
| 547 | * owns the cache, the memcg will get pinned forever. So do not account |
| 548 | * ino ida allocations. |
| 549 | */ |
| 550 | ret = ida_simple_get(&root->ino_ida, 1, 0, |
| 551 | GFP_KERNEL | __GFP_NOACCOUNT); |
| 552 | if (ret < 0) |
| 553 | goto err_out2; |
| 554 | kn->ino = ret; |
| 555 | |
| 556 | atomic_set(&kn->count, 1); |
| 557 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
| 558 | RB_CLEAR_NODE(&kn->rb); |
| 559 | |
| 560 | kn->name = name; |
| 561 | kn->mode = mode; |
| 562 | kn->flags = flags; |
| 563 | |
| 564 | return kn; |
| 565 | |
| 566 | err_out2: |
| 567 | kmem_cache_free(kernfs_node_cache, kn); |
| 568 | err_out1: |
| 569 | kfree_const(name); |
| 570 | return NULL; |
| 571 | } |
| 572 | |
| 573 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
| 574 | const char *name, umode_t mode, |
| 575 | unsigned flags) |
| 576 | { |
| 577 | struct kernfs_node *kn; |
| 578 | |
| 579 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); |
| 580 | if (kn) { |
| 581 | kernfs_get(parent); |
| 582 | kn->parent = parent; |
| 583 | } |
| 584 | return kn; |
| 585 | } |
| 586 | |
| 587 | /** |
| 588 | * kernfs_add_one - add kernfs_node to parent without warning |
| 589 | * @kn: kernfs_node to be added |
| 590 | * |
| 591 | * The caller must already have initialized @kn->parent. This |
| 592 | * function increments nlink of the parent's inode if @kn is a |
| 593 | * directory and link into the children list of the parent. |
| 594 | * |
| 595 | * RETURNS: |
| 596 | * 0 on success, -EEXIST if entry with the given name already |
| 597 | * exists. |
| 598 | */ |
| 599 | int kernfs_add_one(struct kernfs_node *kn) |
| 600 | { |
| 601 | struct kernfs_node *parent = kn->parent; |
| 602 | struct kernfs_iattrs *ps_iattr; |
| 603 | bool has_ns; |
| 604 | int ret; |
| 605 | |
| 606 | mutex_lock(&kernfs_mutex); |
| 607 | |
| 608 | ret = -EINVAL; |
| 609 | has_ns = kernfs_ns_enabled(parent); |
| 610 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
| 611 | has_ns ? "required" : "invalid", parent->name, kn->name)) |
| 612 | goto out_unlock; |
| 613 | |
| 614 | if (kernfs_type(parent) != KERNFS_DIR) |
| 615 | goto out_unlock; |
| 616 | |
| 617 | ret = -ENOENT; |
| 618 | if (parent->flags & KERNFS_EMPTY_DIR) |
| 619 | goto out_unlock; |
| 620 | |
| 621 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
| 622 | goto out_unlock; |
| 623 | |
| 624 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
| 625 | |
| 626 | ret = kernfs_link_sibling(kn); |
| 627 | if (ret) |
| 628 | goto out_unlock; |
| 629 | |
| 630 | /* Update timestamps on the parent */ |
| 631 | ps_iattr = parent->iattr; |
| 632 | if (ps_iattr) { |
| 633 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; |
| 634 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; |
| 635 | } |
| 636 | |
| 637 | mutex_unlock(&kernfs_mutex); |
| 638 | |
| 639 | /* |
| 640 | * Activate the new node unless CREATE_DEACTIVATED is requested. |
| 641 | * If not activated here, the kernfs user is responsible for |
| 642 | * activating the node with kernfs_activate(). A node which hasn't |
| 643 | * been activated is not visible to userland and its removal won't |
| 644 | * trigger deactivation. |
| 645 | */ |
| 646 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
| 647 | kernfs_activate(kn); |
| 648 | return 0; |
| 649 | |
| 650 | out_unlock: |
| 651 | mutex_unlock(&kernfs_mutex); |
| 652 | return ret; |
| 653 | } |
| 654 | |
| 655 | /** |
| 656 | * kernfs_find_ns - find kernfs_node with the given name |
| 657 | * @parent: kernfs_node to search under |
| 658 | * @name: name to look for |
| 659 | * @ns: the namespace tag to use |
| 660 | * |
| 661 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
| 662 | * the found kernfs_node on success, %NULL on failure. |
| 663 | */ |
| 664 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
| 665 | const unsigned char *name, |
| 666 | const void *ns) |
| 667 | { |
| 668 | struct rb_node *node = parent->dir.children.rb_node; |
| 669 | bool has_ns = kernfs_ns_enabled(parent); |
| 670 | unsigned int hash; |
| 671 | |
| 672 | lockdep_assert_held(&kernfs_mutex); |
| 673 | |
| 674 | if (has_ns != (bool)ns) { |
| 675 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
| 676 | has_ns ? "required" : "invalid", parent->name, name); |
| 677 | return NULL; |
| 678 | } |
| 679 | |
| 680 | hash = kernfs_name_hash(name, ns); |
| 681 | while (node) { |
| 682 | struct kernfs_node *kn; |
| 683 | int result; |
| 684 | |
| 685 | kn = rb_to_kn(node); |
| 686 | result = kernfs_name_compare(hash, name, ns, kn); |
| 687 | if (result < 0) |
| 688 | node = node->rb_left; |
| 689 | else if (result > 0) |
| 690 | node = node->rb_right; |
| 691 | else |
| 692 | return kn; |
| 693 | } |
| 694 | return NULL; |
| 695 | } |
| 696 | |
| 697 | static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent, |
| 698 | const unsigned char *path, |
| 699 | const void *ns) |
| 700 | { |
| 701 | static char path_buf[PATH_MAX]; /* protected by kernfs_mutex */ |
| 702 | size_t len = strlcpy(path_buf, path, PATH_MAX); |
| 703 | char *p = path_buf; |
| 704 | char *name; |
| 705 | |
| 706 | lockdep_assert_held(&kernfs_mutex); |
| 707 | |
| 708 | if (len >= PATH_MAX) |
| 709 | return NULL; |
| 710 | |
| 711 | while ((name = strsep(&p, "/")) && parent) { |
| 712 | if (*name == '\0') |
| 713 | continue; |
| 714 | parent = kernfs_find_ns(parent, name, ns); |
| 715 | } |
| 716 | |
| 717 | return parent; |
| 718 | } |
| 719 | |
| 720 | /** |
| 721 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
| 722 | * @parent: kernfs_node to search under |
| 723 | * @name: name to look for |
| 724 | * @ns: the namespace tag to use |
| 725 | * |
| 726 | * Look for kernfs_node with name @name under @parent and get a reference |
| 727 | * if found. This function may sleep and returns pointer to the found |
| 728 | * kernfs_node on success, %NULL on failure. |
| 729 | */ |
| 730 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
| 731 | const char *name, const void *ns) |
| 732 | { |
| 733 | struct kernfs_node *kn; |
| 734 | |
| 735 | mutex_lock(&kernfs_mutex); |
| 736 | kn = kernfs_find_ns(parent, name, ns); |
| 737 | kernfs_get(kn); |
| 738 | mutex_unlock(&kernfs_mutex); |
| 739 | |
| 740 | return kn; |
| 741 | } |
| 742 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); |
| 743 | |
| 744 | /** |
| 745 | * kernfs_walk_and_get_ns - find and get kernfs_node with the given path |
| 746 | * @parent: kernfs_node to search under |
| 747 | * @path: path to look for |
| 748 | * @ns: the namespace tag to use |
| 749 | * |
| 750 | * Look for kernfs_node with path @path under @parent and get a reference |
| 751 | * if found. This function may sleep and returns pointer to the found |
| 752 | * kernfs_node on success, %NULL on failure. |
| 753 | */ |
| 754 | struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent, |
| 755 | const char *path, const void *ns) |
| 756 | { |
| 757 | struct kernfs_node *kn; |
| 758 | |
| 759 | mutex_lock(&kernfs_mutex); |
| 760 | kn = kernfs_walk_ns(parent, path, ns); |
| 761 | kernfs_get(kn); |
| 762 | mutex_unlock(&kernfs_mutex); |
| 763 | |
| 764 | return kn; |
| 765 | } |
| 766 | |
| 767 | /** |
| 768 | * kernfs_create_root - create a new kernfs hierarchy |
| 769 | * @scops: optional syscall operations for the hierarchy |
| 770 | * @flags: KERNFS_ROOT_* flags |
| 771 | * @priv: opaque data associated with the new directory |
| 772 | * |
| 773 | * Returns the root of the new hierarchy on success, ERR_PTR() value on |
| 774 | * failure. |
| 775 | */ |
| 776 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
| 777 | unsigned int flags, void *priv) |
| 778 | { |
| 779 | struct kernfs_root *root; |
| 780 | struct kernfs_node *kn; |
| 781 | |
| 782 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
| 783 | if (!root) |
| 784 | return ERR_PTR(-ENOMEM); |
| 785 | |
| 786 | ida_init(&root->ino_ida); |
| 787 | INIT_LIST_HEAD(&root->supers); |
| 788 | |
| 789 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
| 790 | KERNFS_DIR); |
| 791 | if (!kn) { |
| 792 | ida_destroy(&root->ino_ida); |
| 793 | kfree(root); |
| 794 | return ERR_PTR(-ENOMEM); |
| 795 | } |
| 796 | |
| 797 | kn->priv = priv; |
| 798 | kn->dir.root = root; |
| 799 | |
| 800 | root->syscall_ops = scops; |
| 801 | root->flags = flags; |
| 802 | root->kn = kn; |
| 803 | init_waitqueue_head(&root->deactivate_waitq); |
| 804 | |
| 805 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
| 806 | kernfs_activate(kn); |
| 807 | |
| 808 | return root; |
| 809 | } |
| 810 | |
| 811 | /** |
| 812 | * kernfs_destroy_root - destroy a kernfs hierarchy |
| 813 | * @root: root of the hierarchy to destroy |
| 814 | * |
| 815 | * Destroy the hierarchy anchored at @root by removing all existing |
| 816 | * directories and destroying @root. |
| 817 | */ |
| 818 | void kernfs_destroy_root(struct kernfs_root *root) |
| 819 | { |
| 820 | kernfs_remove(root->kn); /* will also free @root */ |
| 821 | } |
| 822 | |
| 823 | /** |
| 824 | * kernfs_create_dir_ns - create a directory |
| 825 | * @parent: parent in which to create a new directory |
| 826 | * @name: name of the new directory |
| 827 | * @mode: mode of the new directory |
| 828 | * @priv: opaque data associated with the new directory |
| 829 | * @ns: optional namespace tag of the directory |
| 830 | * |
| 831 | * Returns the created node on success, ERR_PTR() value on failure. |
| 832 | */ |
| 833 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
| 834 | const char *name, umode_t mode, |
| 835 | void *priv, const void *ns) |
| 836 | { |
| 837 | struct kernfs_node *kn; |
| 838 | int rc; |
| 839 | |
| 840 | /* allocate */ |
| 841 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
| 842 | if (!kn) |
| 843 | return ERR_PTR(-ENOMEM); |
| 844 | |
| 845 | kn->dir.root = parent->dir.root; |
| 846 | kn->ns = ns; |
| 847 | kn->priv = priv; |
| 848 | |
| 849 | /* link in */ |
| 850 | rc = kernfs_add_one(kn); |
| 851 | if (!rc) |
| 852 | return kn; |
| 853 | |
| 854 | kernfs_put(kn); |
| 855 | return ERR_PTR(rc); |
| 856 | } |
| 857 | |
| 858 | /** |
| 859 | * kernfs_create_empty_dir - create an always empty directory |
| 860 | * @parent: parent in which to create a new directory |
| 861 | * @name: name of the new directory |
| 862 | * |
| 863 | * Returns the created node on success, ERR_PTR() value on failure. |
| 864 | */ |
| 865 | struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent, |
| 866 | const char *name) |
| 867 | { |
| 868 | struct kernfs_node *kn; |
| 869 | int rc; |
| 870 | |
| 871 | /* allocate */ |
| 872 | kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR); |
| 873 | if (!kn) |
| 874 | return ERR_PTR(-ENOMEM); |
| 875 | |
| 876 | kn->flags |= KERNFS_EMPTY_DIR; |
| 877 | kn->dir.root = parent->dir.root; |
| 878 | kn->ns = NULL; |
| 879 | kn->priv = NULL; |
| 880 | |
| 881 | /* link in */ |
| 882 | rc = kernfs_add_one(kn); |
| 883 | if (!rc) |
| 884 | return kn; |
| 885 | |
| 886 | kernfs_put(kn); |
| 887 | return ERR_PTR(rc); |
| 888 | } |
| 889 | |
| 890 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
| 891 | struct dentry *dentry, |
| 892 | unsigned int flags) |
| 893 | { |
| 894 | struct dentry *ret; |
| 895 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
| 896 | struct kernfs_node *kn; |
| 897 | struct inode *inode; |
| 898 | const void *ns = NULL; |
| 899 | |
| 900 | mutex_lock(&kernfs_mutex); |
| 901 | |
| 902 | if (kernfs_ns_enabled(parent)) |
| 903 | ns = kernfs_info(dir->i_sb)->ns; |
| 904 | |
| 905 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
| 906 | |
| 907 | /* no such entry */ |
| 908 | if (!kn || !kernfs_active(kn)) { |
| 909 | ret = NULL; |
| 910 | goto out_unlock; |
| 911 | } |
| 912 | kernfs_get(kn); |
| 913 | dentry->d_fsdata = kn; |
| 914 | |
| 915 | /* attach dentry and inode */ |
| 916 | inode = kernfs_get_inode(dir->i_sb, kn); |
| 917 | if (!inode) { |
| 918 | ret = ERR_PTR(-ENOMEM); |
| 919 | goto out_unlock; |
| 920 | } |
| 921 | |
| 922 | /* instantiate and hash dentry */ |
| 923 | ret = d_splice_alias(inode, dentry); |
| 924 | out_unlock: |
| 925 | mutex_unlock(&kernfs_mutex); |
| 926 | return ret; |
| 927 | } |
| 928 | |
| 929 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
| 930 | umode_t mode) |
| 931 | { |
| 932 | struct kernfs_node *parent = dir->i_private; |
| 933 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
| 934 | int ret; |
| 935 | |
| 936 | if (!scops || !scops->mkdir) |
| 937 | return -EPERM; |
| 938 | |
| 939 | if (!kernfs_get_active(parent)) |
| 940 | return -ENODEV; |
| 941 | |
| 942 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
| 943 | |
| 944 | kernfs_put_active(parent); |
| 945 | return ret; |
| 946 | } |
| 947 | |
| 948 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) |
| 949 | { |
| 950 | struct kernfs_node *kn = dentry->d_fsdata; |
| 951 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
| 952 | int ret; |
| 953 | |
| 954 | if (!scops || !scops->rmdir) |
| 955 | return -EPERM; |
| 956 | |
| 957 | if (!kernfs_get_active(kn)) |
| 958 | return -ENODEV; |
| 959 | |
| 960 | ret = scops->rmdir(kn); |
| 961 | |
| 962 | kernfs_put_active(kn); |
| 963 | return ret; |
| 964 | } |
| 965 | |
| 966 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, |
| 967 | struct inode *new_dir, struct dentry *new_dentry) |
| 968 | { |
| 969 | struct kernfs_node *kn = old_dentry->d_fsdata; |
| 970 | struct kernfs_node *new_parent = new_dir->i_private; |
| 971 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
| 972 | int ret; |
| 973 | |
| 974 | if (!scops || !scops->rename) |
| 975 | return -EPERM; |
| 976 | |
| 977 | if (!kernfs_get_active(kn)) |
| 978 | return -ENODEV; |
| 979 | |
| 980 | if (!kernfs_get_active(new_parent)) { |
| 981 | kernfs_put_active(kn); |
| 982 | return -ENODEV; |
| 983 | } |
| 984 | |
| 985 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
| 986 | |
| 987 | kernfs_put_active(new_parent); |
| 988 | kernfs_put_active(kn); |
| 989 | return ret; |
| 990 | } |
| 991 | |
| 992 | const struct inode_operations kernfs_dir_iops = { |
| 993 | .lookup = kernfs_iop_lookup, |
| 994 | .permission = kernfs_iop_permission, |
| 995 | .setattr = kernfs_iop_setattr, |
| 996 | .getattr = kernfs_iop_getattr, |
| 997 | .setxattr = kernfs_iop_setxattr, |
| 998 | .removexattr = kernfs_iop_removexattr, |
| 999 | .getxattr = kernfs_iop_getxattr, |
| 1000 | .listxattr = kernfs_iop_listxattr, |
| 1001 | |
| 1002 | .mkdir = kernfs_iop_mkdir, |
| 1003 | .rmdir = kernfs_iop_rmdir, |
| 1004 | .rename = kernfs_iop_rename, |
| 1005 | }; |
| 1006 | |
| 1007 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
| 1008 | { |
| 1009 | struct kernfs_node *last; |
| 1010 | |
| 1011 | while (true) { |
| 1012 | struct rb_node *rbn; |
| 1013 | |
| 1014 | last = pos; |
| 1015 | |
| 1016 | if (kernfs_type(pos) != KERNFS_DIR) |
| 1017 | break; |
| 1018 | |
| 1019 | rbn = rb_first(&pos->dir.children); |
| 1020 | if (!rbn) |
| 1021 | break; |
| 1022 | |
| 1023 | pos = rb_to_kn(rbn); |
| 1024 | } |
| 1025 | |
| 1026 | return last; |
| 1027 | } |
| 1028 | |
| 1029 | /** |
| 1030 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
| 1031 | * @pos: the current position (%NULL to initiate traversal) |
| 1032 | * @root: kernfs_node whose descendants to walk |
| 1033 | * |
| 1034 | * Find the next descendant to visit for post-order traversal of @root's |
| 1035 | * descendants. @root is included in the iteration and the last node to be |
| 1036 | * visited. |
| 1037 | */ |
| 1038 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
| 1039 | struct kernfs_node *root) |
| 1040 | { |
| 1041 | struct rb_node *rbn; |
| 1042 | |
| 1043 | lockdep_assert_held(&kernfs_mutex); |
| 1044 | |
| 1045 | /* if first iteration, visit leftmost descendant which may be root */ |
| 1046 | if (!pos) |
| 1047 | return kernfs_leftmost_descendant(root); |
| 1048 | |
| 1049 | /* if we visited @root, we're done */ |
| 1050 | if (pos == root) |
| 1051 | return NULL; |
| 1052 | |
| 1053 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
| 1054 | rbn = rb_next(&pos->rb); |
| 1055 | if (rbn) |
| 1056 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
| 1057 | |
| 1058 | /* no sibling left, visit parent */ |
| 1059 | return pos->parent; |
| 1060 | } |
| 1061 | |
| 1062 | /** |
| 1063 | * kernfs_activate - activate a node which started deactivated |
| 1064 | * @kn: kernfs_node whose subtree is to be activated |
| 1065 | * |
| 1066 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node |
| 1067 | * needs to be explicitly activated. A node which hasn't been activated |
| 1068 | * isn't visible to userland and deactivation is skipped during its |
| 1069 | * removal. This is useful to construct atomic init sequences where |
| 1070 | * creation of multiple nodes should either succeed or fail atomically. |
| 1071 | * |
| 1072 | * The caller is responsible for ensuring that this function is not called |
| 1073 | * after kernfs_remove*() is invoked on @kn. |
| 1074 | */ |
| 1075 | void kernfs_activate(struct kernfs_node *kn) |
| 1076 | { |
| 1077 | struct kernfs_node *pos; |
| 1078 | |
| 1079 | mutex_lock(&kernfs_mutex); |
| 1080 | |
| 1081 | pos = NULL; |
| 1082 | while ((pos = kernfs_next_descendant_post(pos, kn))) { |
| 1083 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) |
| 1084 | continue; |
| 1085 | |
| 1086 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); |
| 1087 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); |
| 1088 | |
| 1089 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); |
| 1090 | pos->flags |= KERNFS_ACTIVATED; |
| 1091 | } |
| 1092 | |
| 1093 | mutex_unlock(&kernfs_mutex); |
| 1094 | } |
| 1095 | |
| 1096 | static void __kernfs_remove(struct kernfs_node *kn) |
| 1097 | { |
| 1098 | struct kernfs_node *pos; |
| 1099 | |
| 1100 | lockdep_assert_held(&kernfs_mutex); |
| 1101 | |
| 1102 | /* |
| 1103 | * Short-circuit if non-root @kn has already finished removal. |
| 1104 | * This is for kernfs_remove_self() which plays with active ref |
| 1105 | * after removal. |
| 1106 | */ |
| 1107 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) |
| 1108 | return; |
| 1109 | |
| 1110 | pr_debug("kernfs %s: removing\n", kn->name); |
| 1111 | |
| 1112 | /* prevent any new usage under @kn by deactivating all nodes */ |
| 1113 | pos = NULL; |
| 1114 | while ((pos = kernfs_next_descendant_post(pos, kn))) |
| 1115 | if (kernfs_active(pos)) |
| 1116 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); |
| 1117 | |
| 1118 | /* deactivate and unlink the subtree node-by-node */ |
| 1119 | do { |
| 1120 | pos = kernfs_leftmost_descendant(kn); |
| 1121 | |
| 1122 | /* |
| 1123 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
| 1124 | * base ref could have been put by someone else by the time |
| 1125 | * the function returns. Make sure it doesn't go away |
| 1126 | * underneath us. |
| 1127 | */ |
| 1128 | kernfs_get(pos); |
| 1129 | |
| 1130 | /* |
| 1131 | * Drain iff @kn was activated. This avoids draining and |
| 1132 | * its lockdep annotations for nodes which have never been |
| 1133 | * activated and allows embedding kernfs_remove() in create |
| 1134 | * error paths without worrying about draining. |
| 1135 | */ |
| 1136 | if (kn->flags & KERNFS_ACTIVATED) |
| 1137 | kernfs_drain(pos); |
| 1138 | else |
| 1139 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); |
| 1140 | |
| 1141 | /* |
| 1142 | * kernfs_unlink_sibling() succeeds once per node. Use it |
| 1143 | * to decide who's responsible for cleanups. |
| 1144 | */ |
| 1145 | if (!pos->parent || kernfs_unlink_sibling(pos)) { |
| 1146 | struct kernfs_iattrs *ps_iattr = |
| 1147 | pos->parent ? pos->parent->iattr : NULL; |
| 1148 | |
| 1149 | /* update timestamps on the parent */ |
| 1150 | if (ps_iattr) { |
| 1151 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; |
| 1152 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; |
| 1153 | } |
| 1154 | |
| 1155 | kernfs_put(pos); |
| 1156 | } |
| 1157 | |
| 1158 | kernfs_put(pos); |
| 1159 | } while (pos != kn); |
| 1160 | } |
| 1161 | |
| 1162 | /** |
| 1163 | * kernfs_remove - remove a kernfs_node recursively |
| 1164 | * @kn: the kernfs_node to remove |
| 1165 | * |
| 1166 | * Remove @kn along with all its subdirectories and files. |
| 1167 | */ |
| 1168 | void kernfs_remove(struct kernfs_node *kn) |
| 1169 | { |
| 1170 | mutex_lock(&kernfs_mutex); |
| 1171 | __kernfs_remove(kn); |
| 1172 | mutex_unlock(&kernfs_mutex); |
| 1173 | } |
| 1174 | |
| 1175 | /** |
| 1176 | * kernfs_break_active_protection - break out of active protection |
| 1177 | * @kn: the self kernfs_node |
| 1178 | * |
| 1179 | * The caller must be running off of a kernfs operation which is invoked |
| 1180 | * with an active reference - e.g. one of kernfs_ops. Each invocation of |
| 1181 | * this function must also be matched with an invocation of |
| 1182 | * kernfs_unbreak_active_protection(). |
| 1183 | * |
| 1184 | * This function releases the active reference of @kn the caller is |
| 1185 | * holding. Once this function is called, @kn may be removed at any point |
| 1186 | * and the caller is solely responsible for ensuring that the objects it |
| 1187 | * dereferences are accessible. |
| 1188 | */ |
| 1189 | void kernfs_break_active_protection(struct kernfs_node *kn) |
| 1190 | { |
| 1191 | /* |
| 1192 | * Take out ourself out of the active ref dependency chain. If |
| 1193 | * we're called without an active ref, lockdep will complain. |
| 1194 | */ |
| 1195 | kernfs_put_active(kn); |
| 1196 | } |
| 1197 | |
| 1198 | /** |
| 1199 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() |
| 1200 | * @kn: the self kernfs_node |
| 1201 | * |
| 1202 | * If kernfs_break_active_protection() was called, this function must be |
| 1203 | * invoked before finishing the kernfs operation. Note that while this |
| 1204 | * function restores the active reference, it doesn't and can't actually |
| 1205 | * restore the active protection - @kn may already or be in the process of |
| 1206 | * being removed. Once kernfs_break_active_protection() is invoked, that |
| 1207 | * protection is irreversibly gone for the kernfs operation instance. |
| 1208 | * |
| 1209 | * While this function may be called at any point after |
| 1210 | * kernfs_break_active_protection() is invoked, its most useful location |
| 1211 | * would be right before the enclosing kernfs operation returns. |
| 1212 | */ |
| 1213 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) |
| 1214 | { |
| 1215 | /* |
| 1216 | * @kn->active could be in any state; however, the increment we do |
| 1217 | * here will be undone as soon as the enclosing kernfs operation |
| 1218 | * finishes and this temporary bump can't break anything. If @kn |
| 1219 | * is alive, nothing changes. If @kn is being deactivated, the |
| 1220 | * soon-to-follow put will either finish deactivation or restore |
| 1221 | * deactivated state. If @kn is already removed, the temporary |
| 1222 | * bump is guaranteed to be gone before @kn is released. |
| 1223 | */ |
| 1224 | atomic_inc(&kn->active); |
| 1225 | if (kernfs_lockdep(kn)) |
| 1226 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); |
| 1227 | } |
| 1228 | |
| 1229 | /** |
| 1230 | * kernfs_remove_self - remove a kernfs_node from its own method |
| 1231 | * @kn: the self kernfs_node to remove |
| 1232 | * |
| 1233 | * The caller must be running off of a kernfs operation which is invoked |
| 1234 | * with an active reference - e.g. one of kernfs_ops. This can be used to |
| 1235 | * implement a file operation which deletes itself. |
| 1236 | * |
| 1237 | * For example, the "delete" file for a sysfs device directory can be |
| 1238 | * implemented by invoking kernfs_remove_self() on the "delete" file |
| 1239 | * itself. This function breaks the circular dependency of trying to |
| 1240 | * deactivate self while holding an active ref itself. It isn't necessary |
| 1241 | * to modify the usual removal path to use kernfs_remove_self(). The |
| 1242 | * "delete" implementation can simply invoke kernfs_remove_self() on self |
| 1243 | * before proceeding with the usual removal path. kernfs will ignore later |
| 1244 | * kernfs_remove() on self. |
| 1245 | * |
| 1246 | * kernfs_remove_self() can be called multiple times concurrently on the |
| 1247 | * same kernfs_node. Only the first one actually performs removal and |
| 1248 | * returns %true. All others will wait until the kernfs operation which |
| 1249 | * won self-removal finishes and return %false. Note that the losers wait |
| 1250 | * for the completion of not only the winning kernfs_remove_self() but also |
| 1251 | * the whole kernfs_ops which won the arbitration. This can be used to |
| 1252 | * guarantee, for example, all concurrent writes to a "delete" file to |
| 1253 | * finish only after the whole operation is complete. |
| 1254 | */ |
| 1255 | bool kernfs_remove_self(struct kernfs_node *kn) |
| 1256 | { |
| 1257 | bool ret; |
| 1258 | |
| 1259 | mutex_lock(&kernfs_mutex); |
| 1260 | kernfs_break_active_protection(kn); |
| 1261 | |
| 1262 | /* |
| 1263 | * SUICIDAL is used to arbitrate among competing invocations. Only |
| 1264 | * the first one will actually perform removal. When the removal |
| 1265 | * is complete, SUICIDED is set and the active ref is restored |
| 1266 | * while holding kernfs_mutex. The ones which lost arbitration |
| 1267 | * waits for SUICDED && drained which can happen only after the |
| 1268 | * enclosing kernfs operation which executed the winning instance |
| 1269 | * of kernfs_remove_self() finished. |
| 1270 | */ |
| 1271 | if (!(kn->flags & KERNFS_SUICIDAL)) { |
| 1272 | kn->flags |= KERNFS_SUICIDAL; |
| 1273 | __kernfs_remove(kn); |
| 1274 | kn->flags |= KERNFS_SUICIDED; |
| 1275 | ret = true; |
| 1276 | } else { |
| 1277 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; |
| 1278 | DEFINE_WAIT(wait); |
| 1279 | |
| 1280 | while (true) { |
| 1281 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); |
| 1282 | |
| 1283 | if ((kn->flags & KERNFS_SUICIDED) && |
| 1284 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) |
| 1285 | break; |
| 1286 | |
| 1287 | mutex_unlock(&kernfs_mutex); |
| 1288 | schedule(); |
| 1289 | mutex_lock(&kernfs_mutex); |
| 1290 | } |
| 1291 | finish_wait(waitq, &wait); |
| 1292 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); |
| 1293 | ret = false; |
| 1294 | } |
| 1295 | |
| 1296 | /* |
| 1297 | * This must be done while holding kernfs_mutex; otherwise, waiting |
| 1298 | * for SUICIDED && deactivated could finish prematurely. |
| 1299 | */ |
| 1300 | kernfs_unbreak_active_protection(kn); |
| 1301 | |
| 1302 | mutex_unlock(&kernfs_mutex); |
| 1303 | return ret; |
| 1304 | } |
| 1305 | |
| 1306 | /** |
| 1307 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
| 1308 | * @parent: parent of the target |
| 1309 | * @name: name of the kernfs_node to remove |
| 1310 | * @ns: namespace tag of the kernfs_node to remove |
| 1311 | * |
| 1312 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
| 1313 | * Returns 0 on success, -ENOENT if such entry doesn't exist. |
| 1314 | */ |
| 1315 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
| 1316 | const void *ns) |
| 1317 | { |
| 1318 | struct kernfs_node *kn; |
| 1319 | |
| 1320 | if (!parent) { |
| 1321 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
| 1322 | name); |
| 1323 | return -ENOENT; |
| 1324 | } |
| 1325 | |
| 1326 | mutex_lock(&kernfs_mutex); |
| 1327 | |
| 1328 | kn = kernfs_find_ns(parent, name, ns); |
| 1329 | if (kn) |
| 1330 | __kernfs_remove(kn); |
| 1331 | |
| 1332 | mutex_unlock(&kernfs_mutex); |
| 1333 | |
| 1334 | if (kn) |
| 1335 | return 0; |
| 1336 | else |
| 1337 | return -ENOENT; |
| 1338 | } |
| 1339 | |
| 1340 | /** |
| 1341 | * kernfs_rename_ns - move and rename a kernfs_node |
| 1342 | * @kn: target node |
| 1343 | * @new_parent: new parent to put @sd under |
| 1344 | * @new_name: new name |
| 1345 | * @new_ns: new namespace tag |
| 1346 | */ |
| 1347 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
| 1348 | const char *new_name, const void *new_ns) |
| 1349 | { |
| 1350 | struct kernfs_node *old_parent; |
| 1351 | const char *old_name = NULL; |
| 1352 | int error; |
| 1353 | |
| 1354 | /* can't move or rename root */ |
| 1355 | if (!kn->parent) |
| 1356 | return -EINVAL; |
| 1357 | |
| 1358 | mutex_lock(&kernfs_mutex); |
| 1359 | |
| 1360 | error = -ENOENT; |
| 1361 | if (!kernfs_active(kn) || !kernfs_active(new_parent) || |
| 1362 | (new_parent->flags & KERNFS_EMPTY_DIR)) |
| 1363 | goto out; |
| 1364 | |
| 1365 | error = 0; |
| 1366 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
| 1367 | (strcmp(kn->name, new_name) == 0)) |
| 1368 | goto out; /* nothing to rename */ |
| 1369 | |
| 1370 | error = -EEXIST; |
| 1371 | if (kernfs_find_ns(new_parent, new_name, new_ns)) |
| 1372 | goto out; |
| 1373 | |
| 1374 | /* rename kernfs_node */ |
| 1375 | if (strcmp(kn->name, new_name) != 0) { |
| 1376 | error = -ENOMEM; |
| 1377 | new_name = kstrdup_const(new_name, GFP_KERNEL); |
| 1378 | if (!new_name) |
| 1379 | goto out; |
| 1380 | } else { |
| 1381 | new_name = NULL; |
| 1382 | } |
| 1383 | |
| 1384 | /* |
| 1385 | * Move to the appropriate place in the appropriate directories rbtree. |
| 1386 | */ |
| 1387 | kernfs_unlink_sibling(kn); |
| 1388 | kernfs_get(new_parent); |
| 1389 | |
| 1390 | /* rename_lock protects ->parent and ->name accessors */ |
| 1391 | spin_lock_irq(&kernfs_rename_lock); |
| 1392 | |
| 1393 | old_parent = kn->parent; |
| 1394 | kn->parent = new_parent; |
| 1395 | |
| 1396 | kn->ns = new_ns; |
| 1397 | if (new_name) { |
| 1398 | old_name = kn->name; |
| 1399 | kn->name = new_name; |
| 1400 | } |
| 1401 | |
| 1402 | spin_unlock_irq(&kernfs_rename_lock); |
| 1403 | |
| 1404 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
| 1405 | kernfs_link_sibling(kn); |
| 1406 | |
| 1407 | kernfs_put(old_parent); |
| 1408 | kfree_const(old_name); |
| 1409 | |
| 1410 | error = 0; |
| 1411 | out: |
| 1412 | mutex_unlock(&kernfs_mutex); |
| 1413 | return error; |
| 1414 | } |
| 1415 | |
| 1416 | /* Relationship between s_mode and the DT_xxx types */ |
| 1417 | static inline unsigned char dt_type(struct kernfs_node *kn) |
| 1418 | { |
| 1419 | return (kn->mode >> 12) & 15; |
| 1420 | } |
| 1421 | |
| 1422 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
| 1423 | { |
| 1424 | kernfs_put(filp->private_data); |
| 1425 | return 0; |
| 1426 | } |
| 1427 | |
| 1428 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
| 1429 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
| 1430 | { |
| 1431 | if (pos) { |
| 1432 | int valid = kernfs_active(pos) && |
| 1433 | pos->parent == parent && hash == pos->hash; |
| 1434 | kernfs_put(pos); |
| 1435 | if (!valid) |
| 1436 | pos = NULL; |
| 1437 | } |
| 1438 | if (!pos && (hash > 1) && (hash < INT_MAX)) { |
| 1439 | struct rb_node *node = parent->dir.children.rb_node; |
| 1440 | while (node) { |
| 1441 | pos = rb_to_kn(node); |
| 1442 | |
| 1443 | if (hash < pos->hash) |
| 1444 | node = node->rb_left; |
| 1445 | else if (hash > pos->hash) |
| 1446 | node = node->rb_right; |
| 1447 | else |
| 1448 | break; |
| 1449 | } |
| 1450 | } |
| 1451 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
| 1452 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { |
| 1453 | struct rb_node *node = rb_next(&pos->rb); |
| 1454 | if (!node) |
| 1455 | pos = NULL; |
| 1456 | else |
| 1457 | pos = rb_to_kn(node); |
| 1458 | } |
| 1459 | return pos; |
| 1460 | } |
| 1461 | |
| 1462 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
| 1463 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
| 1464 | { |
| 1465 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
| 1466 | if (pos) { |
| 1467 | do { |
| 1468 | struct rb_node *node = rb_next(&pos->rb); |
| 1469 | if (!node) |
| 1470 | pos = NULL; |
| 1471 | else |
| 1472 | pos = rb_to_kn(node); |
| 1473 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
| 1474 | } |
| 1475 | return pos; |
| 1476 | } |
| 1477 | |
| 1478 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
| 1479 | { |
| 1480 | struct dentry *dentry = file->f_path.dentry; |
| 1481 | struct kernfs_node *parent = dentry->d_fsdata; |
| 1482 | struct kernfs_node *pos = file->private_data; |
| 1483 | const void *ns = NULL; |
| 1484 | |
| 1485 | if (!dir_emit_dots(file, ctx)) |
| 1486 | return 0; |
| 1487 | mutex_lock(&kernfs_mutex); |
| 1488 | |
| 1489 | if (kernfs_ns_enabled(parent)) |
| 1490 | ns = kernfs_info(dentry->d_sb)->ns; |
| 1491 | |
| 1492 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
| 1493 | pos; |
| 1494 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
| 1495 | const char *name = pos->name; |
| 1496 | unsigned int type = dt_type(pos); |
| 1497 | int len = strlen(name); |
| 1498 | ino_t ino = pos->ino; |
| 1499 | |
| 1500 | ctx->pos = pos->hash; |
| 1501 | file->private_data = pos; |
| 1502 | kernfs_get(pos); |
| 1503 | |
| 1504 | mutex_unlock(&kernfs_mutex); |
| 1505 | if (!dir_emit(ctx, name, len, ino, type)) |
| 1506 | return 0; |
| 1507 | mutex_lock(&kernfs_mutex); |
| 1508 | } |
| 1509 | mutex_unlock(&kernfs_mutex); |
| 1510 | file->private_data = NULL; |
| 1511 | ctx->pos = INT_MAX; |
| 1512 | return 0; |
| 1513 | } |
| 1514 | |
| 1515 | static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset, |
| 1516 | int whence) |
| 1517 | { |
| 1518 | struct inode *inode = file_inode(file); |
| 1519 | loff_t ret; |
| 1520 | |
| 1521 | mutex_lock(&inode->i_mutex); |
| 1522 | ret = generic_file_llseek(file, offset, whence); |
| 1523 | mutex_unlock(&inode->i_mutex); |
| 1524 | |
| 1525 | return ret; |
| 1526 | } |
| 1527 | |
| 1528 | const struct file_operations kernfs_dir_fops = { |
| 1529 | .read = generic_read_dir, |
| 1530 | .iterate = kernfs_fop_readdir, |
| 1531 | .release = kernfs_dir_fop_release, |
| 1532 | .llseek = kernfs_dir_fop_llseek, |
| 1533 | }; |