1 /* 2 * proc/fs/generic.c --- generic routines for the proc-fs 3 * 4 * This file contains generic proc-fs routines for handling 5 * directories and files. 6 * 7 * Copyright (C) 1991, 1992 Linus Torvalds. 8 * Copyright (C) 1997 Theodore Ts'o 9 */ 10 11 #include <linux/errno.h> 12 #include <linux/time.h> 13 #include <linux/proc_fs.h> 14 #include <linux/stat.h> 15 #include <linux/mm.h> 16 #include <linux/module.h> 17 #include <linux/slab.h> 18 #include <linux/printk.h> 19 #include <linux/mount.h> 20 #include <linux/init.h> 21 #include <linux/idr.h> 22 #include <linux/bitops.h> 23 #include <linux/spinlock.h> 24 #include <linux/completion.h> 25 #include <linux/uaccess.h> 26 27 #include "internal.h" 28 29 static DEFINE_RWLOCK(proc_subdir_lock); 30 31 static int proc_match(unsigned int len, const char *name, struct proc_dir_entry *de) 32 { 33 if (len < de->namelen) 34 return -1; 35 if (len > de->namelen) 36 return 1; 37 38 return memcmp(name, de->name, len); 39 } 40 41 static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir) 42 { 43 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry, 44 subdir_node); 45 } 46 47 static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir) 48 { 49 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry, 50 subdir_node); 51 } 52 53 static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir, 54 const char *name, 55 unsigned int len) 56 { 57 struct rb_node *node = dir->subdir.rb_node; 58 59 while (node) { 60 struct proc_dir_entry *de = rb_entry(node, 61 struct proc_dir_entry, 62 subdir_node); 63 int result = proc_match(len, name, de); 64 65 if (result < 0) 66 node = node->rb_left; 67 else if (result > 0) 68 node = node->rb_right; 69 else 70 return de; 71 } 72 return NULL; 73 } 74 75 static bool pde_subdir_insert(struct proc_dir_entry *dir, 76 struct proc_dir_entry *de) 77 { 78 struct rb_root *root = &dir->subdir; 79 struct rb_node **new = &root->rb_node, *parent = NULL; 80 81 /* Figure out where to put new node */ 82 while (*new) { 83 struct proc_dir_entry *this = rb_entry(*new, 84 struct proc_dir_entry, 85 subdir_node); 86 int result = proc_match(de->namelen, de->name, this); 87 88 parent = *new; 89 if (result < 0) 90 new = &(*new)->rb_left; 91 else if (result > 0) 92 new = &(*new)->rb_right; 93 else 94 return false; 95 } 96 97 /* Add new node and rebalance tree. */ 98 rb_link_node(&de->subdir_node, parent, new); 99 rb_insert_color(&de->subdir_node, root); 100 return true; 101 } 102 103 static int proc_notify_change(struct dentry *dentry, struct iattr *iattr) 104 { 105 struct inode *inode = d_inode(dentry); 106 struct proc_dir_entry *de = PDE(inode); 107 int error; 108 109 error = setattr_prepare(dentry, iattr); 110 if (error) 111 return error; 112 113 setattr_copy(inode, iattr); 114 mark_inode_dirty(inode); 115 116 proc_set_user(de, inode->i_uid, inode->i_gid); 117 de->mode = inode->i_mode; 118 return 0; 119 } 120 121 static int proc_getattr(const struct path *path, struct kstat *stat, 122 u32 request_mask, unsigned int query_flags) 123 { 124 struct inode *inode = d_inode(path->dentry); 125 struct proc_dir_entry *de = PDE(inode); 126 if (de && de->nlink) 127 set_nlink(inode, de->nlink); 128 129 generic_fillattr(inode, stat); 130 return 0; 131 } 132 133 static const struct inode_operations proc_file_inode_operations = { 134 .setattr = proc_notify_change, 135 }; 136 137 /* 138 * This function parses a name such as "tty/driver/serial", and 139 * returns the struct proc_dir_entry for "/proc/tty/driver", and 140 * returns "serial" in residual. 141 */ 142 static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret, 143 const char **residual) 144 { 145 const char *cp = name, *next; 146 struct proc_dir_entry *de; 147 unsigned int len; 148 149 de = *ret; 150 if (!de) 151 de = &proc_root; 152 153 while (1) { 154 next = strchr(cp, '/'); 155 if (!next) 156 break; 157 158 len = next - cp; 159 de = pde_subdir_find(de, cp, len); 160 if (!de) { 161 WARN(1, "name '%s'\n", name); 162 return -ENOENT; 163 } 164 cp += len + 1; 165 } 166 *residual = cp; 167 *ret = de; 168 return 0; 169 } 170 171 static int xlate_proc_name(const char *name, struct proc_dir_entry **ret, 172 const char **residual) 173 { 174 int rv; 175 176 read_lock(&proc_subdir_lock); 177 rv = __xlate_proc_name(name, ret, residual); 178 read_unlock(&proc_subdir_lock); 179 return rv; 180 } 181 182 static DEFINE_IDA(proc_inum_ida); 183 static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */ 184 185 #define PROC_DYNAMIC_FIRST 0xF0000000U 186 187 /* 188 * Return an inode number between PROC_DYNAMIC_FIRST and 189 * 0xffffffff, or zero on failure. 190 */ 191 int proc_alloc_inum(unsigned int *inum) 192 { 193 unsigned int i; 194 int error; 195 196 retry: 197 if (!ida_pre_get(&proc_inum_ida, GFP_KERNEL)) 198 return -ENOMEM; 199 200 spin_lock_irq(&proc_inum_lock); 201 error = ida_get_new(&proc_inum_ida, &i); 202 spin_unlock_irq(&proc_inum_lock); 203 if (error == -EAGAIN) 204 goto retry; 205 else if (error) 206 return error; 207 208 if (i > UINT_MAX - PROC_DYNAMIC_FIRST) { 209 spin_lock_irq(&proc_inum_lock); 210 ida_remove(&proc_inum_ida, i); 211 spin_unlock_irq(&proc_inum_lock); 212 return -ENOSPC; 213 } 214 *inum = PROC_DYNAMIC_FIRST + i; 215 return 0; 216 } 217 218 void proc_free_inum(unsigned int inum) 219 { 220 unsigned long flags; 221 spin_lock_irqsave(&proc_inum_lock, flags); 222 ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST); 223 spin_unlock_irqrestore(&proc_inum_lock, flags); 224 } 225 226 /* 227 * Don't create negative dentries here, return -ENOENT by hand 228 * instead. 229 */ 230 struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir, 231 struct dentry *dentry) 232 { 233 struct inode *inode; 234 235 read_lock(&proc_subdir_lock); 236 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len); 237 if (de) { 238 pde_get(de); 239 read_unlock(&proc_subdir_lock); 240 inode = proc_get_inode(dir->i_sb, de); 241 if (!inode) 242 return ERR_PTR(-ENOMEM); 243 d_set_d_op(dentry, &simple_dentry_operations); 244 d_add(dentry, inode); 245 return NULL; 246 } 247 read_unlock(&proc_subdir_lock); 248 return ERR_PTR(-ENOENT); 249 } 250 251 struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry, 252 unsigned int flags) 253 { 254 return proc_lookup_de(PDE(dir), dir, dentry); 255 } 256 257 /* 258 * This returns non-zero if at EOF, so that the /proc 259 * root directory can use this and check if it should 260 * continue with the <pid> entries.. 261 * 262 * Note that the VFS-layer doesn't care about the return 263 * value of the readdir() call, as long as it's non-negative 264 * for success.. 265 */ 266 int proc_readdir_de(struct proc_dir_entry *de, struct file *file, 267 struct dir_context *ctx) 268 { 269 int i; 270 271 if (!dir_emit_dots(file, ctx)) 272 return 0; 273 274 read_lock(&proc_subdir_lock); 275 de = pde_subdir_first(de); 276 i = ctx->pos - 2; 277 for (;;) { 278 if (!de) { 279 read_unlock(&proc_subdir_lock); 280 return 0; 281 } 282 if (!i) 283 break; 284 de = pde_subdir_next(de); 285 i--; 286 } 287 288 do { 289 struct proc_dir_entry *next; 290 pde_get(de); 291 read_unlock(&proc_subdir_lock); 292 if (!dir_emit(ctx, de->name, de->namelen, 293 de->low_ino, de->mode >> 12)) { 294 pde_put(de); 295 return 0; 296 } 297 read_lock(&proc_subdir_lock); 298 ctx->pos++; 299 next = pde_subdir_next(de); 300 pde_put(de); 301 de = next; 302 } while (de); 303 read_unlock(&proc_subdir_lock); 304 return 1; 305 } 306 307 int proc_readdir(struct file *file, struct dir_context *ctx) 308 { 309 struct inode *inode = file_inode(file); 310 311 return proc_readdir_de(PDE(inode), file, ctx); 312 } 313 314 /* 315 * These are the generic /proc directory operations. They 316 * use the in-memory "struct proc_dir_entry" tree to parse 317 * the /proc directory. 318 */ 319 static const struct file_operations proc_dir_operations = { 320 .llseek = generic_file_llseek, 321 .read = generic_read_dir, 322 .iterate_shared = proc_readdir, 323 }; 324 325 /* 326 * proc directories can do almost nothing.. 327 */ 328 static const struct inode_operations proc_dir_inode_operations = { 329 .lookup = proc_lookup, 330 .getattr = proc_getattr, 331 .setattr = proc_notify_change, 332 }; 333 334 static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp) 335 { 336 int ret; 337 338 ret = proc_alloc_inum(&dp->low_ino); 339 if (ret) 340 return ret; 341 342 write_lock(&proc_subdir_lock); 343 dp->parent = dir; 344 if (pde_subdir_insert(dir, dp) == false) { 345 WARN(1, "proc_dir_entry '%s/%s' already registered\n", 346 dir->name, dp->name); 347 write_unlock(&proc_subdir_lock); 348 proc_free_inum(dp->low_ino); 349 return -EEXIST; 350 } 351 write_unlock(&proc_subdir_lock); 352 353 return 0; 354 } 355 356 static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent, 357 const char *name, 358 umode_t mode, 359 nlink_t nlink) 360 { 361 struct proc_dir_entry *ent = NULL; 362 const char *fn; 363 struct qstr qstr; 364 365 if (xlate_proc_name(name, parent, &fn) != 0) 366 goto out; 367 qstr.name = fn; 368 qstr.len = strlen(fn); 369 if (qstr.len == 0 || qstr.len >= 256) { 370 WARN(1, "name len %u\n", qstr.len); 371 return NULL; 372 } 373 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) { 374 WARN(1, "create '/proc/%s' by hand\n", qstr.name); 375 return NULL; 376 } 377 if (is_empty_pde(*parent)) { 378 WARN(1, "attempt to add to permanently empty directory"); 379 return NULL; 380 } 381 382 ent = kzalloc(sizeof(struct proc_dir_entry) + qstr.len + 1, GFP_KERNEL); 383 if (!ent) 384 goto out; 385 386 memcpy(ent->name, fn, qstr.len + 1); 387 ent->namelen = qstr.len; 388 ent->mode = mode; 389 ent->nlink = nlink; 390 ent->subdir = RB_ROOT; 391 atomic_set(&ent->count, 1); 392 spin_lock_init(&ent->pde_unload_lock); 393 INIT_LIST_HEAD(&ent->pde_openers); 394 proc_set_user(ent, (*parent)->uid, (*parent)->gid); 395 396 out: 397 return ent; 398 } 399 400 struct proc_dir_entry *proc_symlink(const char *name, 401 struct proc_dir_entry *parent, const char *dest) 402 { 403 struct proc_dir_entry *ent; 404 405 ent = __proc_create(&parent, name, 406 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1); 407 408 if (ent) { 409 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL); 410 if (ent->data) { 411 strcpy((char*)ent->data,dest); 412 ent->proc_iops = &proc_link_inode_operations; 413 if (proc_register(parent, ent) < 0) { 414 kfree(ent->data); 415 kfree(ent); 416 ent = NULL; 417 } 418 } else { 419 kfree(ent); 420 ent = NULL; 421 } 422 } 423 return ent; 424 } 425 EXPORT_SYMBOL(proc_symlink); 426 427 struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode, 428 struct proc_dir_entry *parent, void *data) 429 { 430 struct proc_dir_entry *ent; 431 432 if (mode == 0) 433 mode = S_IRUGO | S_IXUGO; 434 435 ent = __proc_create(&parent, name, S_IFDIR | mode, 2); 436 if (ent) { 437 ent->data = data; 438 ent->proc_fops = &proc_dir_operations; 439 ent->proc_iops = &proc_dir_inode_operations; 440 parent->nlink++; 441 if (proc_register(parent, ent) < 0) { 442 kfree(ent); 443 parent->nlink--; 444 ent = NULL; 445 } 446 } 447 return ent; 448 } 449 EXPORT_SYMBOL_GPL(proc_mkdir_data); 450 451 struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode, 452 struct proc_dir_entry *parent) 453 { 454 return proc_mkdir_data(name, mode, parent, NULL); 455 } 456 EXPORT_SYMBOL(proc_mkdir_mode); 457 458 struct proc_dir_entry *proc_mkdir(const char *name, 459 struct proc_dir_entry *parent) 460 { 461 return proc_mkdir_data(name, 0, parent, NULL); 462 } 463 EXPORT_SYMBOL(proc_mkdir); 464 465 struct proc_dir_entry *proc_create_mount_point(const char *name) 466 { 467 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO; 468 struct proc_dir_entry *ent, *parent = NULL; 469 470 ent = __proc_create(&parent, name, mode, 2); 471 if (ent) { 472 ent->data = NULL; 473 ent->proc_fops = NULL; 474 ent->proc_iops = NULL; 475 parent->nlink++; 476 if (proc_register(parent, ent) < 0) { 477 kfree(ent); 478 parent->nlink--; 479 ent = NULL; 480 } 481 } 482 return ent; 483 } 484 EXPORT_SYMBOL(proc_create_mount_point); 485 486 struct proc_dir_entry *proc_create_data(const char *name, umode_t mode, 487 struct proc_dir_entry *parent, 488 const struct file_operations *proc_fops, 489 void *data) 490 { 491 struct proc_dir_entry *pde; 492 if ((mode & S_IFMT) == 0) 493 mode |= S_IFREG; 494 495 if (!S_ISREG(mode)) { 496 WARN_ON(1); /* use proc_mkdir() */ 497 return NULL; 498 } 499 500 BUG_ON(proc_fops == NULL); 501 502 if ((mode & S_IALLUGO) == 0) 503 mode |= S_IRUGO; 504 pde = __proc_create(&parent, name, mode, 1); 505 if (!pde) 506 goto out; 507 pde->proc_fops = proc_fops; 508 pde->data = data; 509 pde->proc_iops = &proc_file_inode_operations; 510 if (proc_register(parent, pde) < 0) 511 goto out_free; 512 return pde; 513 out_free: 514 kfree(pde); 515 out: 516 return NULL; 517 } 518 EXPORT_SYMBOL(proc_create_data); 519 520 void proc_set_size(struct proc_dir_entry *de, loff_t size) 521 { 522 de->size = size; 523 } 524 EXPORT_SYMBOL(proc_set_size); 525 526 void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid) 527 { 528 de->uid = uid; 529 de->gid = gid; 530 } 531 EXPORT_SYMBOL(proc_set_user); 532 533 static void free_proc_entry(struct proc_dir_entry *de) 534 { 535 proc_free_inum(de->low_ino); 536 537 if (S_ISLNK(de->mode)) 538 kfree(de->data); 539 kfree(de); 540 } 541 542 void pde_put(struct proc_dir_entry *pde) 543 { 544 if (atomic_dec_and_test(&pde->count)) 545 free_proc_entry(pde); 546 } 547 548 /* 549 * Remove a /proc entry and free it if it's not currently in use. 550 */ 551 void remove_proc_entry(const char *name, struct proc_dir_entry *parent) 552 { 553 struct proc_dir_entry *de = NULL; 554 const char *fn = name; 555 unsigned int len; 556 557 write_lock(&proc_subdir_lock); 558 if (__xlate_proc_name(name, &parent, &fn) != 0) { 559 write_unlock(&proc_subdir_lock); 560 return; 561 } 562 len = strlen(fn); 563 564 de = pde_subdir_find(parent, fn, len); 565 if (de) 566 rb_erase(&de->subdir_node, &parent->subdir); 567 write_unlock(&proc_subdir_lock); 568 if (!de) { 569 WARN(1, "name '%s'\n", name); 570 return; 571 } 572 573 proc_entry_rundown(de); 574 575 if (S_ISDIR(de->mode)) 576 parent->nlink--; 577 de->nlink = 0; 578 WARN(pde_subdir_first(de), 579 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n", 580 __func__, de->parent->name, de->name, pde_subdir_first(de)->name); 581 pde_put(de); 582 } 583 EXPORT_SYMBOL(remove_proc_entry); 584 585 int remove_proc_subtree(const char *name, struct proc_dir_entry *parent) 586 { 587 struct proc_dir_entry *root = NULL, *de, *next; 588 const char *fn = name; 589 unsigned int len; 590 591 write_lock(&proc_subdir_lock); 592 if (__xlate_proc_name(name, &parent, &fn) != 0) { 593 write_unlock(&proc_subdir_lock); 594 return -ENOENT; 595 } 596 len = strlen(fn); 597 598 root = pde_subdir_find(parent, fn, len); 599 if (!root) { 600 write_unlock(&proc_subdir_lock); 601 return -ENOENT; 602 } 603 rb_erase(&root->subdir_node, &parent->subdir); 604 605 de = root; 606 while (1) { 607 next = pde_subdir_first(de); 608 if (next) { 609 rb_erase(&next->subdir_node, &de->subdir); 610 de = next; 611 continue; 612 } 613 write_unlock(&proc_subdir_lock); 614 615 proc_entry_rundown(de); 616 next = de->parent; 617 if (S_ISDIR(de->mode)) 618 next->nlink--; 619 de->nlink = 0; 620 if (de == root) 621 break; 622 pde_put(de); 623 624 write_lock(&proc_subdir_lock); 625 de = next; 626 } 627 pde_put(root); 628 return 0; 629 } 630 EXPORT_SYMBOL(remove_proc_subtree); 631 632 void *proc_get_parent_data(const struct inode *inode) 633 { 634 struct proc_dir_entry *de = PDE(inode); 635 return de->parent->data; 636 } 637 EXPORT_SYMBOL_GPL(proc_get_parent_data); 638 639 void proc_remove(struct proc_dir_entry *de) 640 { 641 if (de) 642 remove_proc_subtree(de->name, de->parent); 643 } 644 EXPORT_SYMBOL(proc_remove); 645 646 void *PDE_DATA(const struct inode *inode) 647 { 648 return __PDE_DATA(inode); 649 } 650 EXPORT_SYMBOL(PDE_DATA); 651