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/cache.h> 12 #include <linux/errno.h> 13 #include <linux/time.h> 14 #include <linux/proc_fs.h> 15 #include <linux/stat.h> 16 #include <linux/mm.h> 17 #include <linux/module.h> 18 #include <linux/slab.h> 19 #include <linux/printk.h> 20 #include <linux/mount.h> 21 #include <linux/init.h> 22 #include <linux/idr.h> 23 #include <linux/bitops.h> 24 #include <linux/spinlock.h> 25 #include <linux/completion.h> 26 #include <linux/uaccess.h> 27 28 #include "internal.h" 29 30 static DEFINE_RWLOCK(proc_subdir_lock); 31 32 struct kmem_cache *proc_dir_entry_cache __ro_after_init; 33 34 void pde_free(struct proc_dir_entry *pde) 35 { 36 if (S_ISLNK(pde->mode)) 37 kfree(pde->data); 38 if (pde->name != pde->inline_name) 39 kfree(pde->name); 40 kmem_cache_free(proc_dir_entry_cache, pde); 41 } 42 43 static int proc_match(const char *name, struct proc_dir_entry *de, unsigned int len) 44 { 45 if (len < de->namelen) 46 return -1; 47 if (len > de->namelen) 48 return 1; 49 50 return memcmp(name, de->name, len); 51 } 52 53 static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir) 54 { 55 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry, 56 subdir_node); 57 } 58 59 static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir) 60 { 61 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry, 62 subdir_node); 63 } 64 65 static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir, 66 const char *name, 67 unsigned int len) 68 { 69 struct rb_node *node = dir->subdir.rb_node; 70 71 while (node) { 72 struct proc_dir_entry *de = rb_entry(node, 73 struct proc_dir_entry, 74 subdir_node); 75 int result = proc_match(name, de, len); 76 77 if (result < 0) 78 node = node->rb_left; 79 else if (result > 0) 80 node = node->rb_right; 81 else 82 return de; 83 } 84 return NULL; 85 } 86 87 static bool pde_subdir_insert(struct proc_dir_entry *dir, 88 struct proc_dir_entry *de) 89 { 90 struct rb_root *root = &dir->subdir; 91 struct rb_node **new = &root->rb_node, *parent = NULL; 92 93 /* Figure out where to put new node */ 94 while (*new) { 95 struct proc_dir_entry *this = rb_entry(*new, 96 struct proc_dir_entry, 97 subdir_node); 98 int result = proc_match(de->name, this, de->namelen); 99 100 parent = *new; 101 if (result < 0) 102 new = &(*new)->rb_left; 103 else if (result > 0) 104 new = &(*new)->rb_right; 105 else 106 return false; 107 } 108 109 /* Add new node and rebalance tree. */ 110 rb_link_node(&de->subdir_node, parent, new); 111 rb_insert_color(&de->subdir_node, root); 112 return true; 113 } 114 115 static int proc_notify_change(struct dentry *dentry, struct iattr *iattr) 116 { 117 struct inode *inode = d_inode(dentry); 118 struct proc_dir_entry *de = PDE(inode); 119 int error; 120 121 error = setattr_prepare(dentry, iattr); 122 if (error) 123 return error; 124 125 setattr_copy(inode, iattr); 126 mark_inode_dirty(inode); 127 128 proc_set_user(de, inode->i_uid, inode->i_gid); 129 de->mode = inode->i_mode; 130 return 0; 131 } 132 133 static int proc_getattr(const struct path *path, struct kstat *stat, 134 u32 request_mask, unsigned int query_flags) 135 { 136 struct inode *inode = d_inode(path->dentry); 137 struct proc_dir_entry *de = PDE(inode); 138 if (de && de->nlink) 139 set_nlink(inode, de->nlink); 140 141 generic_fillattr(inode, stat); 142 return 0; 143 } 144 145 static const struct inode_operations proc_file_inode_operations = { 146 .setattr = proc_notify_change, 147 }; 148 149 /* 150 * This function parses a name such as "tty/driver/serial", and 151 * returns the struct proc_dir_entry for "/proc/tty/driver", and 152 * returns "serial" in residual. 153 */ 154 static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret, 155 const char **residual) 156 { 157 const char *cp = name, *next; 158 struct proc_dir_entry *de; 159 unsigned int len; 160 161 de = *ret; 162 if (!de) 163 de = &proc_root; 164 165 while (1) { 166 next = strchr(cp, '/'); 167 if (!next) 168 break; 169 170 len = next - cp; 171 de = pde_subdir_find(de, cp, len); 172 if (!de) { 173 WARN(1, "name '%s'\n", name); 174 return -ENOENT; 175 } 176 cp += len + 1; 177 } 178 *residual = cp; 179 *ret = de; 180 return 0; 181 } 182 183 static int xlate_proc_name(const char *name, struct proc_dir_entry **ret, 184 const char **residual) 185 { 186 int rv; 187 188 read_lock(&proc_subdir_lock); 189 rv = __xlate_proc_name(name, ret, residual); 190 read_unlock(&proc_subdir_lock); 191 return rv; 192 } 193 194 static DEFINE_IDA(proc_inum_ida); 195 196 #define PROC_DYNAMIC_FIRST 0xF0000000U 197 198 /* 199 * Return an inode number between PROC_DYNAMIC_FIRST and 200 * 0xffffffff, or zero on failure. 201 */ 202 int proc_alloc_inum(unsigned int *inum) 203 { 204 int i; 205 206 i = ida_simple_get(&proc_inum_ida, 0, UINT_MAX - PROC_DYNAMIC_FIRST + 1, 207 GFP_KERNEL); 208 if (i < 0) 209 return i; 210 211 *inum = PROC_DYNAMIC_FIRST + (unsigned int)i; 212 return 0; 213 } 214 215 void proc_free_inum(unsigned int inum) 216 { 217 ida_simple_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST); 218 } 219 220 /* 221 * Don't create negative dentries here, return -ENOENT by hand 222 * instead. 223 */ 224 struct dentry *proc_lookup_de(struct inode *dir, struct dentry *dentry, 225 struct proc_dir_entry *de) 226 { 227 struct inode *inode; 228 229 read_lock(&proc_subdir_lock); 230 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len); 231 if (de) { 232 pde_get(de); 233 read_unlock(&proc_subdir_lock); 234 inode = proc_get_inode(dir->i_sb, de); 235 if (!inode) 236 return ERR_PTR(-ENOMEM); 237 d_set_d_op(dentry, &simple_dentry_operations); 238 d_add(dentry, inode); 239 return NULL; 240 } 241 read_unlock(&proc_subdir_lock); 242 return ERR_PTR(-ENOENT); 243 } 244 245 struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry, 246 unsigned int flags) 247 { 248 return proc_lookup_de(dir, dentry, PDE(dir)); 249 } 250 251 /* 252 * This returns non-zero if at EOF, so that the /proc 253 * root directory can use this and check if it should 254 * continue with the <pid> entries.. 255 * 256 * Note that the VFS-layer doesn't care about the return 257 * value of the readdir() call, as long as it's non-negative 258 * for success.. 259 */ 260 int proc_readdir_de(struct file *file, struct dir_context *ctx, 261 struct proc_dir_entry *de) 262 { 263 int i; 264 265 if (!dir_emit_dots(file, ctx)) 266 return 0; 267 268 read_lock(&proc_subdir_lock); 269 de = pde_subdir_first(de); 270 i = ctx->pos - 2; 271 for (;;) { 272 if (!de) { 273 read_unlock(&proc_subdir_lock); 274 return 0; 275 } 276 if (!i) 277 break; 278 de = pde_subdir_next(de); 279 i--; 280 } 281 282 do { 283 struct proc_dir_entry *next; 284 pde_get(de); 285 read_unlock(&proc_subdir_lock); 286 if (!dir_emit(ctx, de->name, de->namelen, 287 de->low_ino, de->mode >> 12)) { 288 pde_put(de); 289 return 0; 290 } 291 read_lock(&proc_subdir_lock); 292 ctx->pos++; 293 next = pde_subdir_next(de); 294 pde_put(de); 295 de = next; 296 } while (de); 297 read_unlock(&proc_subdir_lock); 298 return 1; 299 } 300 301 int proc_readdir(struct file *file, struct dir_context *ctx) 302 { 303 struct inode *inode = file_inode(file); 304 305 return proc_readdir_de(file, ctx, PDE(inode)); 306 } 307 308 /* 309 * These are the generic /proc directory operations. They 310 * use the in-memory "struct proc_dir_entry" tree to parse 311 * the /proc directory. 312 */ 313 static const struct file_operations proc_dir_operations = { 314 .llseek = generic_file_llseek, 315 .read = generic_read_dir, 316 .iterate_shared = proc_readdir, 317 }; 318 319 /* 320 * proc directories can do almost nothing.. 321 */ 322 static const struct inode_operations proc_dir_inode_operations = { 323 .lookup = proc_lookup, 324 .getattr = proc_getattr, 325 .setattr = proc_notify_change, 326 }; 327 328 static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp) 329 { 330 int ret; 331 332 ret = proc_alloc_inum(&dp->low_ino); 333 if (ret) 334 return ret; 335 336 write_lock(&proc_subdir_lock); 337 dp->parent = dir; 338 if (pde_subdir_insert(dir, dp) == false) { 339 WARN(1, "proc_dir_entry '%s/%s' already registered\n", 340 dir->name, dp->name); 341 write_unlock(&proc_subdir_lock); 342 proc_free_inum(dp->low_ino); 343 return -EEXIST; 344 } 345 write_unlock(&proc_subdir_lock); 346 347 return 0; 348 } 349 350 static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent, 351 const char *name, 352 umode_t mode, 353 nlink_t nlink) 354 { 355 struct proc_dir_entry *ent = NULL; 356 const char *fn; 357 struct qstr qstr; 358 359 if (xlate_proc_name(name, parent, &fn) != 0) 360 goto out; 361 qstr.name = fn; 362 qstr.len = strlen(fn); 363 if (qstr.len == 0 || qstr.len >= 256) { 364 WARN(1, "name len %u\n", qstr.len); 365 return NULL; 366 } 367 if (qstr.len == 1 && fn[0] == '.') { 368 WARN(1, "name '.'\n"); 369 return NULL; 370 } 371 if (qstr.len == 2 && fn[0] == '.' && fn[1] == '.') { 372 WARN(1, "name '..'\n"); 373 return NULL; 374 } 375 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) { 376 WARN(1, "create '/proc/%s' by hand\n", qstr.name); 377 return NULL; 378 } 379 if (is_empty_pde(*parent)) { 380 WARN(1, "attempt to add to permanently empty directory"); 381 return NULL; 382 } 383 384 ent = kmem_cache_zalloc(proc_dir_entry_cache, GFP_KERNEL); 385 if (!ent) 386 goto out; 387 388 if (qstr.len + 1 <= sizeof(ent->inline_name)) { 389 ent->name = ent->inline_name; 390 } else { 391 ent->name = kmalloc(qstr.len + 1, GFP_KERNEL); 392 if (!ent->name) { 393 pde_free(ent); 394 return NULL; 395 } 396 } 397 398 memcpy(ent->name, fn, qstr.len + 1); 399 ent->namelen = qstr.len; 400 ent->mode = mode; 401 ent->nlink = nlink; 402 ent->subdir = RB_ROOT; 403 refcount_set(&ent->refcnt, 1); 404 spin_lock_init(&ent->pde_unload_lock); 405 INIT_LIST_HEAD(&ent->pde_openers); 406 proc_set_user(ent, (*parent)->uid, (*parent)->gid); 407 408 out: 409 return ent; 410 } 411 412 struct proc_dir_entry *proc_symlink(const char *name, 413 struct proc_dir_entry *parent, const char *dest) 414 { 415 struct proc_dir_entry *ent; 416 417 ent = __proc_create(&parent, name, 418 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1); 419 420 if (ent) { 421 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL); 422 if (ent->data) { 423 strcpy((char*)ent->data,dest); 424 ent->proc_iops = &proc_link_inode_operations; 425 if (proc_register(parent, ent) < 0) { 426 pde_free(ent); 427 ent = NULL; 428 } 429 } else { 430 pde_free(ent); 431 ent = NULL; 432 } 433 } 434 return ent; 435 } 436 EXPORT_SYMBOL(proc_symlink); 437 438 struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode, 439 struct proc_dir_entry *parent, void *data) 440 { 441 struct proc_dir_entry *ent; 442 443 if (mode == 0) 444 mode = S_IRUGO | S_IXUGO; 445 446 ent = __proc_create(&parent, name, S_IFDIR | mode, 2); 447 if (ent) { 448 ent->data = data; 449 ent->proc_fops = &proc_dir_operations; 450 ent->proc_iops = &proc_dir_inode_operations; 451 parent->nlink++; 452 if (proc_register(parent, ent) < 0) { 453 pde_free(ent); 454 parent->nlink--; 455 ent = NULL; 456 } 457 } 458 return ent; 459 } 460 EXPORT_SYMBOL_GPL(proc_mkdir_data); 461 462 struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode, 463 struct proc_dir_entry *parent) 464 { 465 return proc_mkdir_data(name, mode, parent, NULL); 466 } 467 EXPORT_SYMBOL(proc_mkdir_mode); 468 469 struct proc_dir_entry *proc_mkdir(const char *name, 470 struct proc_dir_entry *parent) 471 { 472 return proc_mkdir_data(name, 0, parent, NULL); 473 } 474 EXPORT_SYMBOL(proc_mkdir); 475 476 struct proc_dir_entry *proc_create_mount_point(const char *name) 477 { 478 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO; 479 struct proc_dir_entry *ent, *parent = NULL; 480 481 ent = __proc_create(&parent, name, mode, 2); 482 if (ent) { 483 ent->data = NULL; 484 ent->proc_fops = NULL; 485 ent->proc_iops = NULL; 486 parent->nlink++; 487 if (proc_register(parent, ent) < 0) { 488 pde_free(ent); 489 parent->nlink--; 490 ent = NULL; 491 } 492 } 493 return ent; 494 } 495 EXPORT_SYMBOL(proc_create_mount_point); 496 497 struct proc_dir_entry *proc_create_data(const char *name, umode_t mode, 498 struct proc_dir_entry *parent, 499 const struct file_operations *proc_fops, 500 void *data) 501 { 502 struct proc_dir_entry *pde; 503 if ((mode & S_IFMT) == 0) 504 mode |= S_IFREG; 505 506 if (!S_ISREG(mode)) { 507 WARN_ON(1); /* use proc_mkdir() */ 508 return NULL; 509 } 510 511 BUG_ON(proc_fops == NULL); 512 513 if ((mode & S_IALLUGO) == 0) 514 mode |= S_IRUGO; 515 pde = __proc_create(&parent, name, mode, 1); 516 if (!pde) 517 goto out; 518 pde->proc_fops = proc_fops; 519 pde->data = data; 520 pde->proc_iops = &proc_file_inode_operations; 521 if (proc_register(parent, pde) < 0) 522 goto out_free; 523 return pde; 524 out_free: 525 pde_free(pde); 526 out: 527 return NULL; 528 } 529 EXPORT_SYMBOL(proc_create_data); 530 531 struct proc_dir_entry *proc_create(const char *name, umode_t mode, 532 struct proc_dir_entry *parent, 533 const struct file_operations *proc_fops) 534 { 535 return proc_create_data(name, mode, parent, proc_fops, NULL); 536 } 537 EXPORT_SYMBOL(proc_create); 538 539 void proc_set_size(struct proc_dir_entry *de, loff_t size) 540 { 541 de->size = size; 542 } 543 EXPORT_SYMBOL(proc_set_size); 544 545 void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid) 546 { 547 de->uid = uid; 548 de->gid = gid; 549 } 550 EXPORT_SYMBOL(proc_set_user); 551 552 void pde_put(struct proc_dir_entry *pde) 553 { 554 if (refcount_dec_and_test(&pde->refcnt)) { 555 proc_free_inum(pde->low_ino); 556 pde_free(pde); 557 } 558 } 559 560 /* 561 * Remove a /proc entry and free it if it's not currently in use. 562 */ 563 void remove_proc_entry(const char *name, struct proc_dir_entry *parent) 564 { 565 struct proc_dir_entry *de = NULL; 566 const char *fn = name; 567 unsigned int len; 568 569 write_lock(&proc_subdir_lock); 570 if (__xlate_proc_name(name, &parent, &fn) != 0) { 571 write_unlock(&proc_subdir_lock); 572 return; 573 } 574 len = strlen(fn); 575 576 de = pde_subdir_find(parent, fn, len); 577 if (de) 578 rb_erase(&de->subdir_node, &parent->subdir); 579 write_unlock(&proc_subdir_lock); 580 if (!de) { 581 WARN(1, "name '%s'\n", name); 582 return; 583 } 584 585 proc_entry_rundown(de); 586 587 if (S_ISDIR(de->mode)) 588 parent->nlink--; 589 de->nlink = 0; 590 WARN(pde_subdir_first(de), 591 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n", 592 __func__, de->parent->name, de->name, pde_subdir_first(de)->name); 593 pde_put(de); 594 } 595 EXPORT_SYMBOL(remove_proc_entry); 596 597 int remove_proc_subtree(const char *name, struct proc_dir_entry *parent) 598 { 599 struct proc_dir_entry *root = NULL, *de, *next; 600 const char *fn = name; 601 unsigned int len; 602 603 write_lock(&proc_subdir_lock); 604 if (__xlate_proc_name(name, &parent, &fn) != 0) { 605 write_unlock(&proc_subdir_lock); 606 return -ENOENT; 607 } 608 len = strlen(fn); 609 610 root = pde_subdir_find(parent, fn, len); 611 if (!root) { 612 write_unlock(&proc_subdir_lock); 613 return -ENOENT; 614 } 615 rb_erase(&root->subdir_node, &parent->subdir); 616 617 de = root; 618 while (1) { 619 next = pde_subdir_first(de); 620 if (next) { 621 rb_erase(&next->subdir_node, &de->subdir); 622 de = next; 623 continue; 624 } 625 write_unlock(&proc_subdir_lock); 626 627 proc_entry_rundown(de); 628 next = de->parent; 629 if (S_ISDIR(de->mode)) 630 next->nlink--; 631 de->nlink = 0; 632 if (de == root) 633 break; 634 pde_put(de); 635 636 write_lock(&proc_subdir_lock); 637 de = next; 638 } 639 pde_put(root); 640 return 0; 641 } 642 EXPORT_SYMBOL(remove_proc_subtree); 643 644 void *proc_get_parent_data(const struct inode *inode) 645 { 646 struct proc_dir_entry *de = PDE(inode); 647 return de->parent->data; 648 } 649 EXPORT_SYMBOL_GPL(proc_get_parent_data); 650 651 void proc_remove(struct proc_dir_entry *de) 652 { 653 if (de) 654 remove_proc_subtree(de->name, de->parent); 655 } 656 EXPORT_SYMBOL(proc_remove); 657 658 void *PDE_DATA(const struct inode *inode) 659 { 660 return __PDE_DATA(inode); 661 } 662 EXPORT_SYMBOL(PDE_DATA); 663