1// SPDX-License-Identifier: GPL-2.0 2/* 3 * /proc/sys support 4 */ 5#include <linux/init.h> 6#include <linux/sysctl.h> 7#include <linux/poll.h> 8#include <linux/proc_fs.h> 9#include <linux/printk.h> 10#include <linux/security.h> 11#include <linux/sched.h> 12#include <linux/cred.h> 13#include <linux/namei.h> 14#include <linux/mm.h> 15#include <linux/uio.h> 16#include <linux/module.h> 17#include <linux/bpf-cgroup.h> 18#include <linux/mount.h> 19#include <linux/kmemleak.h> 20#include "internal.h" 21 22#define list_for_each_table_entry(entry, header) \ 23 entry = header->ctl_table; \ 24 for (size_t i = 0 ; i < header->ctl_table_size && entry->procname; ++i, entry++) 25 26static const struct dentry_operations proc_sys_dentry_operations; 27static const struct file_operations proc_sys_file_operations; 28static const struct inode_operations proc_sys_inode_operations; 29static const struct file_operations proc_sys_dir_file_operations; 30static const struct inode_operations proc_sys_dir_operations; 31 32/* Support for permanently empty directories */ 33static struct ctl_table sysctl_mount_point[] = { 34 {.type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY } 35}; 36 37/** 38 * register_sysctl_mount_point() - registers a sysctl mount point 39 * @path: path for the mount point 40 * 41 * Used to create a permanently empty directory to serve as mount point. 42 * There are some subtle but important permission checks this allows in the 43 * case of unprivileged mounts. 44 */ 45struct ctl_table_header *register_sysctl_mount_point(const char *path) 46{ 47 return register_sysctl(path, sysctl_mount_point); 48} 49EXPORT_SYMBOL(register_sysctl_mount_point); 50 51#define sysctl_is_perm_empty_ctl_table(tptr) \ 52 (tptr[0].type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY) 53#define sysctl_is_perm_empty_ctl_header(hptr) \ 54 (sysctl_is_perm_empty_ctl_table(hptr->ctl_table)) 55#define sysctl_set_perm_empty_ctl_header(hptr) \ 56 (hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY) 57#define sysctl_clear_perm_empty_ctl_header(hptr) \ 58 (hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_DEFAULT) 59 60void proc_sys_poll_notify(struct ctl_table_poll *poll) 61{ 62 if (!poll) 63 return; 64 65 atomic_inc(&poll->event); 66 wake_up_interruptible(&poll->wait); 67} 68 69static struct ctl_table root_table[] = { 70 { 71 .procname = "", 72 .mode = S_IFDIR|S_IRUGO|S_IXUGO, 73 }, 74 { } 75}; 76static struct ctl_table_root sysctl_table_root = { 77 .default_set.dir.header = { 78 {{.count = 1, 79 .nreg = 1, 80 .ctl_table = root_table }}, 81 .ctl_table_arg = root_table, 82 .root = &sysctl_table_root, 83 .set = &sysctl_table_root.default_set, 84 }, 85}; 86 87static DEFINE_SPINLOCK(sysctl_lock); 88 89static void drop_sysctl_table(struct ctl_table_header *header); 90static int sysctl_follow_link(struct ctl_table_header **phead, 91 struct ctl_table **pentry); 92static int insert_links(struct ctl_table_header *head); 93static void put_links(struct ctl_table_header *header); 94 95static void sysctl_print_dir(struct ctl_dir *dir) 96{ 97 if (dir->header.parent) 98 sysctl_print_dir(dir->header.parent); 99 pr_cont("%s/", dir->header.ctl_table[0].procname); 100} 101 102static int namecmp(const char *name1, int len1, const char *name2, int len2) 103{ 104 int cmp; 105 106 cmp = memcmp(name1, name2, min(len1, len2)); 107 if (cmp == 0) 108 cmp = len1 - len2; 109 return cmp; 110} 111 112/* Called under sysctl_lock */ 113static struct ctl_table *find_entry(struct ctl_table_header **phead, 114 struct ctl_dir *dir, const char *name, int namelen) 115{ 116 struct ctl_table_header *head; 117 struct ctl_table *entry; 118 struct rb_node *node = dir->root.rb_node; 119 120 while (node) 121 { 122 struct ctl_node *ctl_node; 123 const char *procname; 124 int cmp; 125 126 ctl_node = rb_entry(node, struct ctl_node, node); 127 head = ctl_node->header; 128 entry = &head->ctl_table[ctl_node - head->node]; 129 procname = entry->procname; 130 131 cmp = namecmp(name, namelen, procname, strlen(procname)); 132 if (cmp < 0) 133 node = node->rb_left; 134 else if (cmp > 0) 135 node = node->rb_right; 136 else { 137 *phead = head; 138 return entry; 139 } 140 } 141 return NULL; 142} 143 144static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry) 145{ 146 struct rb_node *node = &head->node[entry - head->ctl_table].node; 147 struct rb_node **p = &head->parent->root.rb_node; 148 struct rb_node *parent = NULL; 149 const char *name = entry->procname; 150 int namelen = strlen(name); 151 152 while (*p) { 153 struct ctl_table_header *parent_head; 154 struct ctl_table *parent_entry; 155 struct ctl_node *parent_node; 156 const char *parent_name; 157 int cmp; 158 159 parent = *p; 160 parent_node = rb_entry(parent, struct ctl_node, node); 161 parent_head = parent_node->header; 162 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node]; 163 parent_name = parent_entry->procname; 164 165 cmp = namecmp(name, namelen, parent_name, strlen(parent_name)); 166 if (cmp < 0) 167 p = &(*p)->rb_left; 168 else if (cmp > 0) 169 p = &(*p)->rb_right; 170 else { 171 pr_err("sysctl duplicate entry: "); 172 sysctl_print_dir(head->parent); 173 pr_cont("%s\n", entry->procname); 174 return -EEXIST; 175 } 176 } 177 178 rb_link_node(node, parent, p); 179 rb_insert_color(node, &head->parent->root); 180 return 0; 181} 182 183static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry) 184{ 185 struct rb_node *node = &head->node[entry - head->ctl_table].node; 186 187 rb_erase(node, &head->parent->root); 188} 189 190static void init_header(struct ctl_table_header *head, 191 struct ctl_table_root *root, struct ctl_table_set *set, 192 struct ctl_node *node, struct ctl_table *table, size_t table_size) 193{ 194 head->ctl_table = table; 195 head->ctl_table_size = table_size; 196 head->ctl_table_arg = table; 197 head->used = 0; 198 head->count = 1; 199 head->nreg = 1; 200 head->unregistering = NULL; 201 head->root = root; 202 head->set = set; 203 head->parent = NULL; 204 head->node = node; 205 INIT_HLIST_HEAD(&head->inodes); 206 if (node) { 207 struct ctl_table *entry; 208 209 list_for_each_table_entry(entry, head) { 210 node->header = head; 211 node++; 212 } 213 } 214} 215 216static void erase_header(struct ctl_table_header *head) 217{ 218 struct ctl_table *entry; 219 220 list_for_each_table_entry(entry, head) 221 erase_entry(head, entry); 222} 223 224static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header) 225{ 226 struct ctl_table *entry; 227 struct ctl_table_header *dir_h = &dir->header; 228 int err; 229 230 231 /* Is this a permanently empty directory? */ 232 if (sysctl_is_perm_empty_ctl_header(dir_h)) 233 return -EROFS; 234 235 /* Am I creating a permanently empty directory? */ 236 if (header->ctl_table_size > 0 && 237 sysctl_is_perm_empty_ctl_table(header->ctl_table)) { 238 if (!RB_EMPTY_ROOT(&dir->root)) 239 return -EINVAL; 240 sysctl_set_perm_empty_ctl_header(dir_h); 241 } 242 243 dir_h->nreg++; 244 header->parent = dir; 245 err = insert_links(header); 246 if (err) 247 goto fail_links; 248 list_for_each_table_entry(entry, header) { 249 err = insert_entry(header, entry); 250 if (err) 251 goto fail; 252 } 253 return 0; 254fail: 255 erase_header(header); 256 put_links(header); 257fail_links: 258 if (header->ctl_table == sysctl_mount_point) 259 sysctl_clear_perm_empty_ctl_header(dir_h); 260 header->parent = NULL; 261 drop_sysctl_table(dir_h); 262 return err; 263} 264 265/* called under sysctl_lock */ 266static int use_table(struct ctl_table_header *p) 267{ 268 if (unlikely(p->unregistering)) 269 return 0; 270 p->used++; 271 return 1; 272} 273 274/* called under sysctl_lock */ 275static void unuse_table(struct ctl_table_header *p) 276{ 277 if (!--p->used) 278 if (unlikely(p->unregistering)) 279 complete(p->unregistering); 280} 281 282static void proc_sys_invalidate_dcache(struct ctl_table_header *head) 283{ 284 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock); 285} 286 287/* called under sysctl_lock, will reacquire if has to wait */ 288static void start_unregistering(struct ctl_table_header *p) 289{ 290 /* 291 * if p->used is 0, nobody will ever touch that entry again; 292 * we'll eliminate all paths to it before dropping sysctl_lock 293 */ 294 if (unlikely(p->used)) { 295 struct completion wait; 296 init_completion(&wait); 297 p->unregistering = &wait; 298 spin_unlock(&sysctl_lock); 299 wait_for_completion(&wait); 300 } else { 301 /* anything non-NULL; we'll never dereference it */ 302 p->unregistering = ERR_PTR(-EINVAL); 303 spin_unlock(&sysctl_lock); 304 } 305 /* 306 * Invalidate dentries for unregistered sysctls: namespaced sysctls 307 * can have duplicate names and contaminate dcache very badly. 308 */ 309 proc_sys_invalidate_dcache(p); 310 /* 311 * do not remove from the list until nobody holds it; walking the 312 * list in do_sysctl() relies on that. 313 */ 314 spin_lock(&sysctl_lock); 315 erase_header(p); 316} 317 318static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head) 319{ 320 BUG_ON(!head); 321 spin_lock(&sysctl_lock); 322 if (!use_table(head)) 323 head = ERR_PTR(-ENOENT); 324 spin_unlock(&sysctl_lock); 325 return head; 326} 327 328static void sysctl_head_finish(struct ctl_table_header *head) 329{ 330 if (!head) 331 return; 332 spin_lock(&sysctl_lock); 333 unuse_table(head); 334 spin_unlock(&sysctl_lock); 335} 336 337static struct ctl_table_set * 338lookup_header_set(struct ctl_table_root *root) 339{ 340 struct ctl_table_set *set = &root->default_set; 341 if (root->lookup) 342 set = root->lookup(root); 343 return set; 344} 345 346static struct ctl_table *lookup_entry(struct ctl_table_header **phead, 347 struct ctl_dir *dir, 348 const char *name, int namelen) 349{ 350 struct ctl_table_header *head; 351 struct ctl_table *entry; 352 353 spin_lock(&sysctl_lock); 354 entry = find_entry(&head, dir, name, namelen); 355 if (entry && use_table(head)) 356 *phead = head; 357 else 358 entry = NULL; 359 spin_unlock(&sysctl_lock); 360 return entry; 361} 362 363static struct ctl_node *first_usable_entry(struct rb_node *node) 364{ 365 struct ctl_node *ctl_node; 366 367 for (;node; node = rb_next(node)) { 368 ctl_node = rb_entry(node, struct ctl_node, node); 369 if (use_table(ctl_node->header)) 370 return ctl_node; 371 } 372 return NULL; 373} 374 375static void first_entry(struct ctl_dir *dir, 376 struct ctl_table_header **phead, struct ctl_table **pentry) 377{ 378 struct ctl_table_header *head = NULL; 379 struct ctl_table *entry = NULL; 380 struct ctl_node *ctl_node; 381 382 spin_lock(&sysctl_lock); 383 ctl_node = first_usable_entry(rb_first(&dir->root)); 384 spin_unlock(&sysctl_lock); 385 if (ctl_node) { 386 head = ctl_node->header; 387 entry = &head->ctl_table[ctl_node - head->node]; 388 } 389 *phead = head; 390 *pentry = entry; 391} 392 393static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry) 394{ 395 struct ctl_table_header *head = *phead; 396 struct ctl_table *entry = *pentry; 397 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table]; 398 399 spin_lock(&sysctl_lock); 400 unuse_table(head); 401 402 ctl_node = first_usable_entry(rb_next(&ctl_node->node)); 403 spin_unlock(&sysctl_lock); 404 head = NULL; 405 if (ctl_node) { 406 head = ctl_node->header; 407 entry = &head->ctl_table[ctl_node - head->node]; 408 } 409 *phead = head; 410 *pentry = entry; 411} 412 413/* 414 * sysctl_perm does NOT grant the superuser all rights automatically, because 415 * some sysctl variables are readonly even to root. 416 */ 417 418static int test_perm(int mode, int op) 419{ 420 if (uid_eq(current_euid(), GLOBAL_ROOT_UID)) 421 mode >>= 6; 422 else if (in_egroup_p(GLOBAL_ROOT_GID)) 423 mode >>= 3; 424 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0) 425 return 0; 426 return -EACCES; 427} 428 429static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op) 430{ 431 struct ctl_table_root *root = head->root; 432 int mode; 433 434 if (root->permissions) 435 mode = root->permissions(head, table); 436 else 437 mode = table->mode; 438 439 return test_perm(mode, op); 440} 441 442static struct inode *proc_sys_make_inode(struct super_block *sb, 443 struct ctl_table_header *head, struct ctl_table *table) 444{ 445 struct ctl_table_root *root = head->root; 446 struct inode *inode; 447 struct proc_inode *ei; 448 449 inode = new_inode(sb); 450 if (!inode) 451 return ERR_PTR(-ENOMEM); 452 453 inode->i_ino = get_next_ino(); 454 455 ei = PROC_I(inode); 456 457 spin_lock(&sysctl_lock); 458 if (unlikely(head->unregistering)) { 459 spin_unlock(&sysctl_lock); 460 iput(inode); 461 return ERR_PTR(-ENOENT); 462 } 463 ei->sysctl = head; 464 ei->sysctl_entry = table; 465 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes); 466 head->count++; 467 spin_unlock(&sysctl_lock); 468 469 inode->i_mtime = inode->i_atime = inode_set_ctime_current(inode); 470 inode->i_mode = table->mode; 471 if (!S_ISDIR(table->mode)) { 472 inode->i_mode |= S_IFREG; 473 inode->i_op = &proc_sys_inode_operations; 474 inode->i_fop = &proc_sys_file_operations; 475 } else { 476 inode->i_mode |= S_IFDIR; 477 inode->i_op = &proc_sys_dir_operations; 478 inode->i_fop = &proc_sys_dir_file_operations; 479 if (sysctl_is_perm_empty_ctl_header(head)) 480 make_empty_dir_inode(inode); 481 } 482 483 if (root->set_ownership) 484 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid); 485 else { 486 inode->i_uid = GLOBAL_ROOT_UID; 487 inode->i_gid = GLOBAL_ROOT_GID; 488 } 489 490 return inode; 491} 492 493void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head) 494{ 495 spin_lock(&sysctl_lock); 496 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes); 497 if (!--head->count) 498 kfree_rcu(head, rcu); 499 spin_unlock(&sysctl_lock); 500} 501 502static struct ctl_table_header *grab_header(struct inode *inode) 503{ 504 struct ctl_table_header *head = PROC_I(inode)->sysctl; 505 if (!head) 506 head = &sysctl_table_root.default_set.dir.header; 507 return sysctl_head_grab(head); 508} 509 510static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry, 511 unsigned int flags) 512{ 513 struct ctl_table_header *head = grab_header(dir); 514 struct ctl_table_header *h = NULL; 515 const struct qstr *name = &dentry->d_name; 516 struct ctl_table *p; 517 struct inode *inode; 518 struct dentry *err = ERR_PTR(-ENOENT); 519 struct ctl_dir *ctl_dir; 520 int ret; 521 522 if (IS_ERR(head)) 523 return ERR_CAST(head); 524 525 ctl_dir = container_of(head, struct ctl_dir, header); 526 527 p = lookup_entry(&h, ctl_dir, name->name, name->len); 528 if (!p) 529 goto out; 530 531 if (S_ISLNK(p->mode)) { 532 ret = sysctl_follow_link(&h, &p); 533 err = ERR_PTR(ret); 534 if (ret) 535 goto out; 536 } 537 538 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p); 539 if (IS_ERR(inode)) { 540 err = ERR_CAST(inode); 541 goto out; 542 } 543 544 d_set_d_op(dentry, &proc_sys_dentry_operations); 545 err = d_splice_alias(inode, dentry); 546 547out: 548 if (h) 549 sysctl_head_finish(h); 550 sysctl_head_finish(head); 551 return err; 552} 553 554static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter, 555 int write) 556{ 557 struct inode *inode = file_inode(iocb->ki_filp); 558 struct ctl_table_header *head = grab_header(inode); 559 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 560 size_t count = iov_iter_count(iter); 561 char *kbuf; 562 ssize_t error; 563 564 if (IS_ERR(head)) 565 return PTR_ERR(head); 566 567 /* 568 * At this point we know that the sysctl was not unregistered 569 * and won't be until we finish. 570 */ 571 error = -EPERM; 572 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ)) 573 goto out; 574 575 /* if that can happen at all, it should be -EINVAL, not -EISDIR */ 576 error = -EINVAL; 577 if (!table->proc_handler) 578 goto out; 579 580 /* don't even try if the size is too large */ 581 error = -ENOMEM; 582 if (count >= KMALLOC_MAX_SIZE) 583 goto out; 584 kbuf = kvzalloc(count + 1, GFP_KERNEL); 585 if (!kbuf) 586 goto out; 587 588 if (write) { 589 error = -EFAULT; 590 if (!copy_from_iter_full(kbuf, count, iter)) 591 goto out_free_buf; 592 kbuf[count] = '\0'; 593 } 594 595 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count, 596 &iocb->ki_pos); 597 if (error) 598 goto out_free_buf; 599 600 /* careful: calling conventions are nasty here */ 601 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos); 602 if (error) 603 goto out_free_buf; 604 605 if (!write) { 606 error = -EFAULT; 607 if (copy_to_iter(kbuf, count, iter) < count) 608 goto out_free_buf; 609 } 610 611 error = count; 612out_free_buf: 613 kvfree(kbuf); 614out: 615 sysctl_head_finish(head); 616 617 return error; 618} 619 620static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter) 621{ 622 return proc_sys_call_handler(iocb, iter, 0); 623} 624 625static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter) 626{ 627 return proc_sys_call_handler(iocb, iter, 1); 628} 629 630static int proc_sys_open(struct inode *inode, struct file *filp) 631{ 632 struct ctl_table_header *head = grab_header(inode); 633 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 634 635 /* sysctl was unregistered */ 636 if (IS_ERR(head)) 637 return PTR_ERR(head); 638 639 if (table->poll) 640 filp->private_data = proc_sys_poll_event(table->poll); 641 642 sysctl_head_finish(head); 643 644 return 0; 645} 646 647static __poll_t proc_sys_poll(struct file *filp, poll_table *wait) 648{ 649 struct inode *inode = file_inode(filp); 650 struct ctl_table_header *head = grab_header(inode); 651 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 652 __poll_t ret = DEFAULT_POLLMASK; 653 unsigned long event; 654 655 /* sysctl was unregistered */ 656 if (IS_ERR(head)) 657 return EPOLLERR | EPOLLHUP; 658 659 if (!table->proc_handler) 660 goto out; 661 662 if (!table->poll) 663 goto out; 664 665 event = (unsigned long)filp->private_data; 666 poll_wait(filp, &table->poll->wait, wait); 667 668 if (event != atomic_read(&table->poll->event)) { 669 filp->private_data = proc_sys_poll_event(table->poll); 670 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; 671 } 672 673out: 674 sysctl_head_finish(head); 675 676 return ret; 677} 678 679static bool proc_sys_fill_cache(struct file *file, 680 struct dir_context *ctx, 681 struct ctl_table_header *head, 682 struct ctl_table *table) 683{ 684 struct dentry *child, *dir = file->f_path.dentry; 685 struct inode *inode; 686 struct qstr qname; 687 ino_t ino = 0; 688 unsigned type = DT_UNKNOWN; 689 690 qname.name = table->procname; 691 qname.len = strlen(table->procname); 692 qname.hash = full_name_hash(dir, qname.name, qname.len); 693 694 child = d_lookup(dir, &qname); 695 if (!child) { 696 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 697 child = d_alloc_parallel(dir, &qname, &wq); 698 if (IS_ERR(child)) 699 return false; 700 if (d_in_lookup(child)) { 701 struct dentry *res; 702 inode = proc_sys_make_inode(dir->d_sb, head, table); 703 if (IS_ERR(inode)) { 704 d_lookup_done(child); 705 dput(child); 706 return false; 707 } 708 d_set_d_op(child, &proc_sys_dentry_operations); 709 res = d_splice_alias(inode, child); 710 d_lookup_done(child); 711 if (unlikely(res)) { 712 if (IS_ERR(res)) { 713 dput(child); 714 return false; 715 } 716 dput(child); 717 child = res; 718 } 719 } 720 } 721 inode = d_inode(child); 722 ino = inode->i_ino; 723 type = inode->i_mode >> 12; 724 dput(child); 725 return dir_emit(ctx, qname.name, qname.len, ino, type); 726} 727 728static bool proc_sys_link_fill_cache(struct file *file, 729 struct dir_context *ctx, 730 struct ctl_table_header *head, 731 struct ctl_table *table) 732{ 733 bool ret = true; 734 735 head = sysctl_head_grab(head); 736 if (IS_ERR(head)) 737 return false; 738 739 /* It is not an error if we can not follow the link ignore it */ 740 if (sysctl_follow_link(&head, &table)) 741 goto out; 742 743 ret = proc_sys_fill_cache(file, ctx, head, table); 744out: 745 sysctl_head_finish(head); 746 return ret; 747} 748 749static int scan(struct ctl_table_header *head, struct ctl_table *table, 750 unsigned long *pos, struct file *file, 751 struct dir_context *ctx) 752{ 753 bool res; 754 755 if ((*pos)++ < ctx->pos) 756 return true; 757 758 if (unlikely(S_ISLNK(table->mode))) 759 res = proc_sys_link_fill_cache(file, ctx, head, table); 760 else 761 res = proc_sys_fill_cache(file, ctx, head, table); 762 763 if (res) 764 ctx->pos = *pos; 765 766 return res; 767} 768 769static int proc_sys_readdir(struct file *file, struct dir_context *ctx) 770{ 771 struct ctl_table_header *head = grab_header(file_inode(file)); 772 struct ctl_table_header *h = NULL; 773 struct ctl_table *entry; 774 struct ctl_dir *ctl_dir; 775 unsigned long pos; 776 777 if (IS_ERR(head)) 778 return PTR_ERR(head); 779 780 ctl_dir = container_of(head, struct ctl_dir, header); 781 782 if (!dir_emit_dots(file, ctx)) 783 goto out; 784 785 pos = 2; 786 787 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) { 788 if (!scan(h, entry, &pos, file, ctx)) { 789 sysctl_head_finish(h); 790 break; 791 } 792 } 793out: 794 sysctl_head_finish(head); 795 return 0; 796} 797 798static int proc_sys_permission(struct mnt_idmap *idmap, 799 struct inode *inode, int mask) 800{ 801 /* 802 * sysctl entries that are not writeable, 803 * are _NOT_ writeable, capabilities or not. 804 */ 805 struct ctl_table_header *head; 806 struct ctl_table *table; 807 int error; 808 809 /* Executable files are not allowed under /proc/sys/ */ 810 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) 811 return -EACCES; 812 813 head = grab_header(inode); 814 if (IS_ERR(head)) 815 return PTR_ERR(head); 816 817 table = PROC_I(inode)->sysctl_entry; 818 if (!table) /* global root - r-xr-xr-x */ 819 error = mask & MAY_WRITE ? -EACCES : 0; 820 else /* Use the permissions on the sysctl table entry */ 821 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK); 822 823 sysctl_head_finish(head); 824 return error; 825} 826 827static int proc_sys_setattr(struct mnt_idmap *idmap, 828 struct dentry *dentry, struct iattr *attr) 829{ 830 struct inode *inode = d_inode(dentry); 831 int error; 832 833 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID)) 834 return -EPERM; 835 836 error = setattr_prepare(&nop_mnt_idmap, dentry, attr); 837 if (error) 838 return error; 839 840 setattr_copy(&nop_mnt_idmap, inode, attr); 841 return 0; 842} 843 844static int proc_sys_getattr(struct mnt_idmap *idmap, 845 const struct path *path, struct kstat *stat, 846 u32 request_mask, unsigned int query_flags) 847{ 848 struct inode *inode = d_inode(path->dentry); 849 struct ctl_table_header *head = grab_header(inode); 850 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 851 852 if (IS_ERR(head)) 853 return PTR_ERR(head); 854 855 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 856 if (table) 857 stat->mode = (stat->mode & S_IFMT) | table->mode; 858 859 sysctl_head_finish(head); 860 return 0; 861} 862 863static const struct file_operations proc_sys_file_operations = { 864 .open = proc_sys_open, 865 .poll = proc_sys_poll, 866 .read_iter = proc_sys_read, 867 .write_iter = proc_sys_write, 868 .splice_read = copy_splice_read, 869 .splice_write = iter_file_splice_write, 870 .llseek = default_llseek, 871}; 872 873static const struct file_operations proc_sys_dir_file_operations = { 874 .read = generic_read_dir, 875 .iterate_shared = proc_sys_readdir, 876 .llseek = generic_file_llseek, 877}; 878 879static const struct inode_operations proc_sys_inode_operations = { 880 .permission = proc_sys_permission, 881 .setattr = proc_sys_setattr, 882 .getattr = proc_sys_getattr, 883}; 884 885static const struct inode_operations proc_sys_dir_operations = { 886 .lookup = proc_sys_lookup, 887 .permission = proc_sys_permission, 888 .setattr = proc_sys_setattr, 889 .getattr = proc_sys_getattr, 890}; 891 892static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags) 893{ 894 if (flags & LOOKUP_RCU) 895 return -ECHILD; 896 return !PROC_I(d_inode(dentry))->sysctl->unregistering; 897} 898 899static int proc_sys_delete(const struct dentry *dentry) 900{ 901 return !!PROC_I(d_inode(dentry))->sysctl->unregistering; 902} 903 904static int sysctl_is_seen(struct ctl_table_header *p) 905{ 906 struct ctl_table_set *set = p->set; 907 int res; 908 spin_lock(&sysctl_lock); 909 if (p->unregistering) 910 res = 0; 911 else if (!set->is_seen) 912 res = 1; 913 else 914 res = set->is_seen(set); 915 spin_unlock(&sysctl_lock); 916 return res; 917} 918 919static int proc_sys_compare(const struct dentry *dentry, 920 unsigned int len, const char *str, const struct qstr *name) 921{ 922 struct ctl_table_header *head; 923 struct inode *inode; 924 925 /* Although proc doesn't have negative dentries, rcu-walk means 926 * that inode here can be NULL */ 927 /* AV: can it, indeed? */ 928 inode = d_inode_rcu(dentry); 929 if (!inode) 930 return 1; 931 if (name->len != len) 932 return 1; 933 if (memcmp(name->name, str, len)) 934 return 1; 935 head = rcu_dereference(PROC_I(inode)->sysctl); 936 return !head || !sysctl_is_seen(head); 937} 938 939static const struct dentry_operations proc_sys_dentry_operations = { 940 .d_revalidate = proc_sys_revalidate, 941 .d_delete = proc_sys_delete, 942 .d_compare = proc_sys_compare, 943}; 944 945static struct ctl_dir *find_subdir(struct ctl_dir *dir, 946 const char *name, int namelen) 947{ 948 struct ctl_table_header *head; 949 struct ctl_table *entry; 950 951 entry = find_entry(&head, dir, name, namelen); 952 if (!entry) 953 return ERR_PTR(-ENOENT); 954 if (!S_ISDIR(entry->mode)) 955 return ERR_PTR(-ENOTDIR); 956 return container_of(head, struct ctl_dir, header); 957} 958 959static struct ctl_dir *new_dir(struct ctl_table_set *set, 960 const char *name, int namelen) 961{ 962 struct ctl_table *table; 963 struct ctl_dir *new; 964 struct ctl_node *node; 965 char *new_name; 966 967 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) + 968 sizeof(struct ctl_table)*2 + namelen + 1, 969 GFP_KERNEL); 970 if (!new) 971 return NULL; 972 973 node = (struct ctl_node *)(new + 1); 974 table = (struct ctl_table *)(node + 1); 975 new_name = (char *)(table + 2); 976 memcpy(new_name, name, namelen); 977 table[0].procname = new_name; 978 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO; 979 init_header(&new->header, set->dir.header.root, set, node, table, 1); 980 981 return new; 982} 983 984/** 985 * get_subdir - find or create a subdir with the specified name. 986 * @dir: Directory to create the subdirectory in 987 * @name: The name of the subdirectory to find or create 988 * @namelen: The length of name 989 * 990 * Takes a directory with an elevated reference count so we know that 991 * if we drop the lock the directory will not go away. Upon success 992 * the reference is moved from @dir to the returned subdirectory. 993 * Upon error an error code is returned and the reference on @dir is 994 * simply dropped. 995 */ 996static struct ctl_dir *get_subdir(struct ctl_dir *dir, 997 const char *name, int namelen) 998{ 999 struct ctl_table_set *set = dir->header.set; 1000 struct ctl_dir *subdir, *new = NULL; 1001 int err; 1002 1003 spin_lock(&sysctl_lock); 1004 subdir = find_subdir(dir, name, namelen); 1005 if (!IS_ERR(subdir)) 1006 goto found; 1007 if (PTR_ERR(subdir) != -ENOENT) 1008 goto failed; 1009 1010 spin_unlock(&sysctl_lock); 1011 new = new_dir(set, name, namelen); 1012 spin_lock(&sysctl_lock); 1013 subdir = ERR_PTR(-ENOMEM); 1014 if (!new) 1015 goto failed; 1016 1017 /* Was the subdir added while we dropped the lock? */ 1018 subdir = find_subdir(dir, name, namelen); 1019 if (!IS_ERR(subdir)) 1020 goto found; 1021 if (PTR_ERR(subdir) != -ENOENT) 1022 goto failed; 1023 1024 /* Nope. Use the our freshly made directory entry. */ 1025 err = insert_header(dir, &new->header); 1026 subdir = ERR_PTR(err); 1027 if (err) 1028 goto failed; 1029 subdir = new; 1030found: 1031 subdir->header.nreg++; 1032failed: 1033 if (IS_ERR(subdir)) { 1034 pr_err("sysctl could not get directory: "); 1035 sysctl_print_dir(dir); 1036 pr_cont("%*.*s %ld\n", namelen, namelen, name, 1037 PTR_ERR(subdir)); 1038 } 1039 drop_sysctl_table(&dir->header); 1040 if (new) 1041 drop_sysctl_table(&new->header); 1042 spin_unlock(&sysctl_lock); 1043 return subdir; 1044} 1045 1046static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir) 1047{ 1048 struct ctl_dir *parent; 1049 const char *procname; 1050 if (!dir->header.parent) 1051 return &set->dir; 1052 parent = xlate_dir(set, dir->header.parent); 1053 if (IS_ERR(parent)) 1054 return parent; 1055 procname = dir->header.ctl_table[0].procname; 1056 return find_subdir(parent, procname, strlen(procname)); 1057} 1058 1059static int sysctl_follow_link(struct ctl_table_header **phead, 1060 struct ctl_table **pentry) 1061{ 1062 struct ctl_table_header *head; 1063 struct ctl_table_root *root; 1064 struct ctl_table_set *set; 1065 struct ctl_table *entry; 1066 struct ctl_dir *dir; 1067 int ret; 1068 1069 spin_lock(&sysctl_lock); 1070 root = (*pentry)->data; 1071 set = lookup_header_set(root); 1072 dir = xlate_dir(set, (*phead)->parent); 1073 if (IS_ERR(dir)) 1074 ret = PTR_ERR(dir); 1075 else { 1076 const char *procname = (*pentry)->procname; 1077 head = NULL; 1078 entry = find_entry(&head, dir, procname, strlen(procname)); 1079 ret = -ENOENT; 1080 if (entry && use_table(head)) { 1081 unuse_table(*phead); 1082 *phead = head; 1083 *pentry = entry; 1084 ret = 0; 1085 } 1086 } 1087 1088 spin_unlock(&sysctl_lock); 1089 return ret; 1090} 1091 1092static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...) 1093{ 1094 struct va_format vaf; 1095 va_list args; 1096 1097 va_start(args, fmt); 1098 vaf.fmt = fmt; 1099 vaf.va = &args; 1100 1101 pr_err("sysctl table check failed: %s/%s %pV\n", 1102 path, table->procname, &vaf); 1103 1104 va_end(args); 1105 return -EINVAL; 1106} 1107 1108static int sysctl_check_table_array(const char *path, struct ctl_table *table) 1109{ 1110 int err = 0; 1111 1112 if ((table->proc_handler == proc_douintvec) || 1113 (table->proc_handler == proc_douintvec_minmax)) { 1114 if (table->maxlen != sizeof(unsigned int)) 1115 err |= sysctl_err(path, table, "array not allowed"); 1116 } 1117 1118 if (table->proc_handler == proc_dou8vec_minmax) { 1119 if (table->maxlen != sizeof(u8)) 1120 err |= sysctl_err(path, table, "array not allowed"); 1121 } 1122 1123 if (table->proc_handler == proc_dobool) { 1124 if (table->maxlen != sizeof(bool)) 1125 err |= sysctl_err(path, table, "array not allowed"); 1126 } 1127 1128 return err; 1129} 1130 1131static int sysctl_check_table(const char *path, struct ctl_table_header *header) 1132{ 1133 struct ctl_table *entry; 1134 int err = 0; 1135 list_for_each_table_entry(entry, header) { 1136 if ((entry->proc_handler == proc_dostring) || 1137 (entry->proc_handler == proc_dobool) || 1138 (entry->proc_handler == proc_dointvec) || 1139 (entry->proc_handler == proc_douintvec) || 1140 (entry->proc_handler == proc_douintvec_minmax) || 1141 (entry->proc_handler == proc_dointvec_minmax) || 1142 (entry->proc_handler == proc_dou8vec_minmax) || 1143 (entry->proc_handler == proc_dointvec_jiffies) || 1144 (entry->proc_handler == proc_dointvec_userhz_jiffies) || 1145 (entry->proc_handler == proc_dointvec_ms_jiffies) || 1146 (entry->proc_handler == proc_doulongvec_minmax) || 1147 (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) { 1148 if (!entry->data) 1149 err |= sysctl_err(path, entry, "No data"); 1150 if (!entry->maxlen) 1151 err |= sysctl_err(path, entry, "No maxlen"); 1152 else 1153 err |= sysctl_check_table_array(path, entry); 1154 } 1155 if (!entry->proc_handler) 1156 err |= sysctl_err(path, entry, "No proc_handler"); 1157 1158 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode) 1159 err |= sysctl_err(path, entry, "bogus .mode 0%o", 1160 entry->mode); 1161 } 1162 return err; 1163} 1164 1165static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head) 1166{ 1167 struct ctl_table *link_table, *entry, *link; 1168 struct ctl_table_header *links; 1169 struct ctl_node *node; 1170 char *link_name; 1171 int nr_entries, name_bytes; 1172 1173 name_bytes = 0; 1174 nr_entries = 0; 1175 list_for_each_table_entry(entry, head) { 1176 nr_entries++; 1177 name_bytes += strlen(entry->procname) + 1; 1178 } 1179 1180 links = kzalloc(sizeof(struct ctl_table_header) + 1181 sizeof(struct ctl_node)*nr_entries + 1182 sizeof(struct ctl_table)*(nr_entries + 1) + 1183 name_bytes, 1184 GFP_KERNEL); 1185 1186 if (!links) 1187 return NULL; 1188 1189 node = (struct ctl_node *)(links + 1); 1190 link_table = (struct ctl_table *)(node + nr_entries); 1191 link_name = (char *)&link_table[nr_entries + 1]; 1192 link = link_table; 1193 1194 list_for_each_table_entry(entry, head) { 1195 int len = strlen(entry->procname) + 1; 1196 memcpy(link_name, entry->procname, len); 1197 link->procname = link_name; 1198 link->mode = S_IFLNK|S_IRWXUGO; 1199 link->data = head->root; 1200 link_name += len; 1201 link++; 1202 } 1203 init_header(links, dir->header.root, dir->header.set, node, link_table, 1204 head->ctl_table_size); 1205 links->nreg = nr_entries; 1206 1207 return links; 1208} 1209 1210static bool get_links(struct ctl_dir *dir, 1211 struct ctl_table_header *header, 1212 struct ctl_table_root *link_root) 1213{ 1214 struct ctl_table_header *tmp_head; 1215 struct ctl_table *entry, *link; 1216 1217 if (header->ctl_table_size == 0 || 1218 sysctl_is_perm_empty_ctl_table(header->ctl_table)) 1219 return true; 1220 1221 /* Are there links available for every entry in table? */ 1222 list_for_each_table_entry(entry, header) { 1223 const char *procname = entry->procname; 1224 link = find_entry(&tmp_head, dir, procname, strlen(procname)); 1225 if (!link) 1226 return false; 1227 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode)) 1228 continue; 1229 if (S_ISLNK(link->mode) && (link->data == link_root)) 1230 continue; 1231 return false; 1232 } 1233 1234 /* The checks passed. Increase the registration count on the links */ 1235 list_for_each_table_entry(entry, header) { 1236 const char *procname = entry->procname; 1237 link = find_entry(&tmp_head, dir, procname, strlen(procname)); 1238 tmp_head->nreg++; 1239 } 1240 return true; 1241} 1242 1243static int insert_links(struct ctl_table_header *head) 1244{ 1245 struct ctl_table_set *root_set = &sysctl_table_root.default_set; 1246 struct ctl_dir *core_parent; 1247 struct ctl_table_header *links; 1248 int err; 1249 1250 if (head->set == root_set) 1251 return 0; 1252 1253 core_parent = xlate_dir(root_set, head->parent); 1254 if (IS_ERR(core_parent)) 1255 return 0; 1256 1257 if (get_links(core_parent, head, head->root)) 1258 return 0; 1259 1260 core_parent->header.nreg++; 1261 spin_unlock(&sysctl_lock); 1262 1263 links = new_links(core_parent, head); 1264 1265 spin_lock(&sysctl_lock); 1266 err = -ENOMEM; 1267 if (!links) 1268 goto out; 1269 1270 err = 0; 1271 if (get_links(core_parent, head, head->root)) { 1272 kfree(links); 1273 goto out; 1274 } 1275 1276 err = insert_header(core_parent, links); 1277 if (err) 1278 kfree(links); 1279out: 1280 drop_sysctl_table(&core_parent->header); 1281 return err; 1282} 1283 1284/* Find the directory for the ctl_table. If one is not found create it. */ 1285static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path) 1286{ 1287 const char *name, *nextname; 1288 1289 for (name = path; name; name = nextname) { 1290 int namelen; 1291 nextname = strchr(name, '/'); 1292 if (nextname) { 1293 namelen = nextname - name; 1294 nextname++; 1295 } else { 1296 namelen = strlen(name); 1297 } 1298 if (namelen == 0) 1299 continue; 1300 1301 /* 1302 * namelen ensures if name is "foo/bar/yay" only foo is 1303 * registered first. We traverse as if using mkdir -p and 1304 * return a ctl_dir for the last directory entry. 1305 */ 1306 dir = get_subdir(dir, name, namelen); 1307 if (IS_ERR(dir)) 1308 break; 1309 } 1310 return dir; 1311} 1312 1313/** 1314 * __register_sysctl_table - register a leaf sysctl table 1315 * @set: Sysctl tree to register on 1316 * @path: The path to the directory the sysctl table is in. 1317 * @table: the top-level table structure without any child. This table 1318 * should not be free'd after registration. So it should not be 1319 * used on stack. It can either be a global or dynamically allocated 1320 * by the caller and free'd later after sysctl unregistration. 1321 * @table_size : The number of elements in table 1322 * 1323 * Register a sysctl table hierarchy. @table should be a filled in ctl_table 1324 * array. A completely 0 filled entry terminates the table. 1325 * 1326 * The members of the &struct ctl_table structure are used as follows: 1327 * 1328 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not 1329 * enter a sysctl file 1330 * 1331 * data - a pointer to data for use by proc_handler 1332 * 1333 * maxlen - the maximum size in bytes of the data 1334 * 1335 * mode - the file permissions for the /proc/sys file 1336 * 1337 * child - must be %NULL. 1338 * 1339 * proc_handler - the text handler routine (described below) 1340 * 1341 * extra1, extra2 - extra pointers usable by the proc handler routines 1342 * XXX: we should eventually modify these to use long min / max [0] 1343 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org 1344 * 1345 * Leaf nodes in the sysctl tree will be represented by a single file 1346 * under /proc; non-leaf nodes (where child is not NULL) are not allowed, 1347 * sysctl_check_table() verifies this. 1348 * 1349 * There must be a proc_handler routine for any terminal nodes. 1350 * Several default handlers are available to cover common cases - 1351 * 1352 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(), 1353 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(), 1354 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax() 1355 * 1356 * It is the handler's job to read the input buffer from user memory 1357 * and process it. The handler should return 0 on success. 1358 * 1359 * This routine returns %NULL on a failure to register, and a pointer 1360 * to the table header on success. 1361 */ 1362struct ctl_table_header *__register_sysctl_table( 1363 struct ctl_table_set *set, 1364 const char *path, struct ctl_table *table, size_t table_size) 1365{ 1366 struct ctl_table_root *root = set->dir.header.root; 1367 struct ctl_table_header *header; 1368 struct ctl_dir *dir; 1369 struct ctl_node *node; 1370 1371 header = kzalloc(sizeof(struct ctl_table_header) + 1372 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT); 1373 if (!header) 1374 return NULL; 1375 1376 node = (struct ctl_node *)(header + 1); 1377 init_header(header, root, set, node, table, table_size); 1378 if (sysctl_check_table(path, header)) 1379 goto fail; 1380 1381 spin_lock(&sysctl_lock); 1382 dir = &set->dir; 1383 /* Reference moved down the directory tree get_subdir */ 1384 dir->header.nreg++; 1385 spin_unlock(&sysctl_lock); 1386 1387 dir = sysctl_mkdir_p(dir, path); 1388 if (IS_ERR(dir)) 1389 goto fail; 1390 spin_lock(&sysctl_lock); 1391 if (insert_header(dir, header)) 1392 goto fail_put_dir_locked; 1393 1394 drop_sysctl_table(&dir->header); 1395 spin_unlock(&sysctl_lock); 1396 1397 return header; 1398 1399fail_put_dir_locked: 1400 drop_sysctl_table(&dir->header); 1401 spin_unlock(&sysctl_lock); 1402fail: 1403 kfree(header); 1404 return NULL; 1405} 1406 1407/** 1408 * register_sysctl_sz - register a sysctl table 1409 * @path: The path to the directory the sysctl table is in. If the path 1410 * doesn't exist we will create it for you. 1411 * @table: the table structure. The calller must ensure the life of the @table 1412 * will be kept during the lifetime use of the syctl. It must not be freed 1413 * until unregister_sysctl_table() is called with the given returned table 1414 * with this registration. If your code is non modular then you don't need 1415 * to call unregister_sysctl_table() and can instead use something like 1416 * register_sysctl_init() which does not care for the result of the syctl 1417 * registration. 1418 * @table_size: The number of elements in table. 1419 * 1420 * Register a sysctl table. @table should be a filled in ctl_table 1421 * array. A completely 0 filled entry terminates the table. 1422 * 1423 * See __register_sysctl_table for more details. 1424 */ 1425struct ctl_table_header *register_sysctl_sz(const char *path, struct ctl_table *table, 1426 size_t table_size) 1427{ 1428 return __register_sysctl_table(&sysctl_table_root.default_set, 1429 path, table, table_size); 1430} 1431EXPORT_SYMBOL(register_sysctl_sz); 1432 1433/** 1434 * __register_sysctl_init() - register sysctl table to path 1435 * @path: path name for sysctl base. If that path doesn't exist we will create 1436 * it for you. 1437 * @table: This is the sysctl table that needs to be registered to the path. 1438 * The caller must ensure the life of the @table will be kept during the 1439 * lifetime use of the sysctl. 1440 * @table_name: The name of sysctl table, only used for log printing when 1441 * registration fails 1442 * @table_size: The number of elements in table 1443 * 1444 * The sysctl interface is used by userspace to query or modify at runtime 1445 * a predefined value set on a variable. These variables however have default 1446 * values pre-set. Code which depends on these variables will always work even 1447 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the 1448 * ability to query or modify the sysctls dynamically at run time. Chances of 1449 * register_sysctl() failing on init are extremely low, and so for both reasons 1450 * this function does not return any error as it is used by initialization code. 1451 * 1452 * Context: if your base directory does not exist it will be created for you. 1453 */ 1454void __init __register_sysctl_init(const char *path, struct ctl_table *table, 1455 const char *table_name, size_t table_size) 1456{ 1457 struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size); 1458 1459 if (unlikely(!hdr)) { 1460 pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path); 1461 return; 1462 } 1463 kmemleak_not_leak(hdr); 1464} 1465 1466static void put_links(struct ctl_table_header *header) 1467{ 1468 struct ctl_table_set *root_set = &sysctl_table_root.default_set; 1469 struct ctl_table_root *root = header->root; 1470 struct ctl_dir *parent = header->parent; 1471 struct ctl_dir *core_parent; 1472 struct ctl_table *entry; 1473 1474 if (header->set == root_set) 1475 return; 1476 1477 core_parent = xlate_dir(root_set, parent); 1478 if (IS_ERR(core_parent)) 1479 return; 1480 1481 list_for_each_table_entry(entry, header) { 1482 struct ctl_table_header *link_head; 1483 struct ctl_table *link; 1484 const char *name = entry->procname; 1485 1486 link = find_entry(&link_head, core_parent, name, strlen(name)); 1487 if (link && 1488 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) || 1489 (S_ISLNK(link->mode) && (link->data == root)))) { 1490 drop_sysctl_table(link_head); 1491 } 1492 else { 1493 pr_err("sysctl link missing during unregister: "); 1494 sysctl_print_dir(parent); 1495 pr_cont("%s\n", name); 1496 } 1497 } 1498} 1499 1500static void drop_sysctl_table(struct ctl_table_header *header) 1501{ 1502 struct ctl_dir *parent = header->parent; 1503 1504 if (--header->nreg) 1505 return; 1506 1507 if (parent) { 1508 put_links(header); 1509 start_unregistering(header); 1510 } 1511 1512 if (!--header->count) 1513 kfree_rcu(header, rcu); 1514 1515 if (parent) 1516 drop_sysctl_table(&parent->header); 1517} 1518 1519/** 1520 * unregister_sysctl_table - unregister a sysctl table hierarchy 1521 * @header: the header returned from register_sysctl or __register_sysctl_table 1522 * 1523 * Unregisters the sysctl table and all children. proc entries may not 1524 * actually be removed until they are no longer used by anyone. 1525 */ 1526void unregister_sysctl_table(struct ctl_table_header * header) 1527{ 1528 might_sleep(); 1529 1530 if (header == NULL) 1531 return; 1532 1533 spin_lock(&sysctl_lock); 1534 drop_sysctl_table(header); 1535 spin_unlock(&sysctl_lock); 1536} 1537EXPORT_SYMBOL(unregister_sysctl_table); 1538 1539void setup_sysctl_set(struct ctl_table_set *set, 1540 struct ctl_table_root *root, 1541 int (*is_seen)(struct ctl_table_set *)) 1542{ 1543 memset(set, 0, sizeof(*set)); 1544 set->is_seen = is_seen; 1545 init_header(&set->dir.header, root, set, NULL, root_table, 1); 1546} 1547 1548void retire_sysctl_set(struct ctl_table_set *set) 1549{ 1550 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root)); 1551} 1552 1553int __init proc_sys_init(void) 1554{ 1555 struct proc_dir_entry *proc_sys_root; 1556 1557 proc_sys_root = proc_mkdir("sys", NULL); 1558 proc_sys_root->proc_iops = &proc_sys_dir_operations; 1559 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations; 1560 proc_sys_root->nlink = 0; 1561 1562 return sysctl_init_bases(); 1563} 1564 1565struct sysctl_alias { 1566 const char *kernel_param; 1567 const char *sysctl_param; 1568}; 1569 1570/* 1571 * Historically some settings had both sysctl and a command line parameter. 1572 * With the generic sysctl. parameter support, we can handle them at a single 1573 * place and only keep the historical name for compatibility. This is not meant 1574 * to add brand new aliases. When adding existing aliases, consider whether 1575 * the possibly different moment of changing the value (e.g. from early_param 1576 * to the moment do_sysctl_args() is called) is an issue for the specific 1577 * parameter. 1578 */ 1579static const struct sysctl_alias sysctl_aliases[] = { 1580 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" }, 1581 {"hung_task_panic", "kernel.hung_task_panic" }, 1582 {"numa_zonelist_order", "vm.numa_zonelist_order" }, 1583 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" }, 1584 { } 1585}; 1586 1587static const char *sysctl_find_alias(char *param) 1588{ 1589 const struct sysctl_alias *alias; 1590 1591 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) { 1592 if (strcmp(alias->kernel_param, param) == 0) 1593 return alias->sysctl_param; 1594 } 1595 1596 return NULL; 1597} 1598 1599bool sysctl_is_alias(char *param) 1600{ 1601 const char *alias = sysctl_find_alias(param); 1602 1603 return alias != NULL; 1604} 1605 1606/* Set sysctl value passed on kernel command line. */ 1607static int process_sysctl_arg(char *param, char *val, 1608 const char *unused, void *arg) 1609{ 1610 char *path; 1611 struct vfsmount **proc_mnt = arg; 1612 struct file_system_type *proc_fs_type; 1613 struct file *file; 1614 int len; 1615 int err; 1616 loff_t pos = 0; 1617 ssize_t wret; 1618 1619 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) { 1620 param += sizeof("sysctl") - 1; 1621 1622 if (param[0] != '/' && param[0] != '.') 1623 return 0; 1624 1625 param++; 1626 } else { 1627 param = (char *) sysctl_find_alias(param); 1628 if (!param) 1629 return 0; 1630 } 1631 1632 if (!val) 1633 return -EINVAL; 1634 len = strlen(val); 1635 if (len == 0) 1636 return -EINVAL; 1637 1638 /* 1639 * To set sysctl options, we use a temporary mount of proc, look up the 1640 * respective sys/ file and write to it. To avoid mounting it when no 1641 * options were given, we mount it only when the first sysctl option is 1642 * found. Why not a persistent mount? There are problems with a 1643 * persistent mount of proc in that it forces userspace not to use any 1644 * proc mount options. 1645 */ 1646 if (!*proc_mnt) { 1647 proc_fs_type = get_fs_type("proc"); 1648 if (!proc_fs_type) { 1649 pr_err("Failed to find procfs to set sysctl from command line\n"); 1650 return 0; 1651 } 1652 *proc_mnt = kern_mount(proc_fs_type); 1653 put_filesystem(proc_fs_type); 1654 if (IS_ERR(*proc_mnt)) { 1655 pr_err("Failed to mount procfs to set sysctl from command line\n"); 1656 return 0; 1657 } 1658 } 1659 1660 path = kasprintf(GFP_KERNEL, "sys/%s", param); 1661 if (!path) 1662 panic("%s: Failed to allocate path for %s\n", __func__, param); 1663 strreplace(path, '.', '/'); 1664 1665 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0); 1666 if (IS_ERR(file)) { 1667 err = PTR_ERR(file); 1668 if (err == -ENOENT) 1669 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n", 1670 param, val); 1671 else if (err == -EACCES) 1672 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n", 1673 param, val); 1674 else 1675 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n", 1676 file, param, val); 1677 goto out; 1678 } 1679 wret = kernel_write(file, val, len, &pos); 1680 if (wret < 0) { 1681 err = wret; 1682 if (err == -EINVAL) 1683 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n", 1684 param, val); 1685 else 1686 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n", 1687 ERR_PTR(err), param, val); 1688 } else if (wret != len) { 1689 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n", 1690 wret, len, path, param, val); 1691 } 1692 1693 err = filp_close(file, NULL); 1694 if (err) 1695 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n", 1696 ERR_PTR(err), param, val); 1697out: 1698 kfree(path); 1699 return 0; 1700} 1701 1702void do_sysctl_args(void) 1703{ 1704 char *command_line; 1705 struct vfsmount *proc_mnt = NULL; 1706 1707 command_line = kstrdup(saved_command_line, GFP_KERNEL); 1708 if (!command_line) 1709 panic("%s: Failed to allocate copy of command line\n", __func__); 1710 1711 parse_args("Setting sysctl args", command_line, 1712 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg); 1713 1714 if (proc_mnt) 1715 kern_unmount(proc_mnt); 1716 1717 kfree(command_line); 1718} 1719