fs/proc/proc_sysctl.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
is_empty_dir
set_empty_dir
clear_empty_dir
proc_sys_poll_notify
sysctl_print_dir
namecmp
find_entry
insert_entry
erase_entry
init_header
erase_header
insert_header
use_table
unuse_table
proc_sys_prune_dcache
start_unregistering
sysctl_head_grab
sysctl_head_finish
lookup_header_set
lookup_entry
first_usable_entry
first_entry
next_entry
test_perm
sysctl_perm
proc_sys_make_inode
proc_sys_evict_inode
grab_header
proc_sys_lookup
proc_sys_call_handler
proc_sys_read
proc_sys_write
proc_sys_open
proc_sys_poll
proc_sys_fill_cache
proc_sys_link_fill_cache
scan
proc_sys_readdir
proc_sys_permission
proc_sys_setattr
proc_sys_getattr
proc_sys_revalidate
proc_sys_delete
sysctl_is_seen
proc_sys_compare
find_subdir
new_dir
get_subdir
xlate_dir
sysctl_follow_link
sysctl_err
sysctl_check_table_array
sysctl_check_table
new_links
get_links
insert_links
__register_sysctl_table
register_sysctl
append_path
count_subheaders
register_leaf_sysctl_tables
__register_sysctl_paths
register_sysctl_paths
register_sysctl_table
put_links
drop_sysctl_table
unregister_sysctl_table
setup_sysctl_set
retire_sysctl_set
proc_sys_init
   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/module.h>
  16 #include <linux/bpf-cgroup.h>
  17 #include "internal.h"
  18 
  19 static const struct dentry_operations proc_sys_dentry_operations;
  20 static const struct file_operations proc_sys_file_operations;
  21 static const struct inode_operations proc_sys_inode_operations;
  22 static const struct file_operations proc_sys_dir_file_operations;
  23 static const struct inode_operations proc_sys_dir_operations;
  24 
  25 /* shared constants to be used in various sysctls */
  26 const int sysctl_vals[] = { 0, 1, INT_MAX };
  27 EXPORT_SYMBOL(sysctl_vals);
  28 
  29 /* Support for permanently empty directories */
  30 
  31 struct ctl_table sysctl_mount_point[] = {
  32         { }
  33 };
  34 
  35 static bool is_empty_dir(struct ctl_table_header *head)
  36 {
  37         return head->ctl_table[0].child == sysctl_mount_point;
  38 }
  39 
  40 static void set_empty_dir(struct ctl_dir *dir)
  41 {
  42         dir->header.ctl_table[0].child = sysctl_mount_point;
  43 }
  44 
  45 static void clear_empty_dir(struct ctl_dir *dir)
  46 
  47 {
  48         dir->header.ctl_table[0].child = NULL;
  49 }
  50 
  51 void proc_sys_poll_notify(struct ctl_table_poll *poll)
  52 {
  53         if (!poll)
  54                 return;
  55 
  56         atomic_inc(&poll->event);
  57         wake_up_interruptible(&poll->wait);
  58 }
  59 
  60 static struct ctl_table root_table[] = {
  61         {
  62                 .procname = "",
  63                 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
  64         },
  65         { }
  66 };
  67 static struct ctl_table_root sysctl_table_root = {
  68         .default_set.dir.header = {
  69                 {{.count = 1,
  70                   .nreg = 1,
  71                   .ctl_table = root_table }},
  72                 .ctl_table_arg = root_table,
  73                 .root = &sysctl_table_root,
  74                 .set = &sysctl_table_root.default_set,
  75         },
  76 };
  77 
  78 static DEFINE_SPINLOCK(sysctl_lock);
  79 
  80 static void drop_sysctl_table(struct ctl_table_header *header);
  81 static int sysctl_follow_link(struct ctl_table_header **phead,
  82         struct ctl_table **pentry);
  83 static int insert_links(struct ctl_table_header *head);
  84 static void put_links(struct ctl_table_header *header);
  85 
  86 static void sysctl_print_dir(struct ctl_dir *dir)
  87 {
  88         if (dir->header.parent)
  89                 sysctl_print_dir(dir->header.parent);
  90         pr_cont("%s/", dir->header.ctl_table[0].procname);
  91 }
  92 
  93 static int namecmp(const char *name1, int len1, const char *name2, int len2)
  94 {
  95         int minlen;
  96         int cmp;
  97 
  98         minlen = len1;
  99         if (minlen > len2)
 100                 minlen = len2;
 101 
 102         cmp = memcmp(name1, name2, minlen);
 103         if (cmp == 0)
 104                 cmp = len1 - len2;
 105         return cmp;
 106 }
 107 
 108 /* Called under sysctl_lock */
 109 static struct ctl_table *find_entry(struct ctl_table_header **phead,
 110         struct ctl_dir *dir, const char *name, int namelen)
 111 {
 112         struct ctl_table_header *head;
 113         struct ctl_table *entry;
 114         struct rb_node *node = dir->root.rb_node;
 115 
 116         while (node)
 117         {
 118                 struct ctl_node *ctl_node;
 119                 const char *procname;
 120                 int cmp;
 121 
 122                 ctl_node = rb_entry(node, struct ctl_node, node);
 123                 head = ctl_node->header;
 124                 entry = &head->ctl_table[ctl_node - head->node];
 125                 procname = entry->procname;
 126 
 127                 cmp = namecmp(name, namelen, procname, strlen(procname));
 128                 if (cmp < 0)
 129                         node = node->rb_left;
 130                 else if (cmp > 0)
 131                         node = node->rb_right;
 132                 else {
 133                         *phead = head;
 134                         return entry;
 135                 }
 136         }
 137         return NULL;
 138 }
 139 
 140 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
 141 {
 142         struct rb_node *node = &head->node[entry - head->ctl_table].node;
 143         struct rb_node **p = &head->parent->root.rb_node;
 144         struct rb_node *parent = NULL;
 145         const char *name = entry->procname;
 146         int namelen = strlen(name);
 147 
 148         while (*p) {
 149                 struct ctl_table_header *parent_head;
 150                 struct ctl_table *parent_entry;
 151                 struct ctl_node *parent_node;
 152                 const char *parent_name;
 153                 int cmp;
 154 
 155                 parent = *p;
 156                 parent_node = rb_entry(parent, struct ctl_node, node);
 157                 parent_head = parent_node->header;
 158                 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
 159                 parent_name = parent_entry->procname;
 160 
 161                 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
 162                 if (cmp < 0)
 163                         p = &(*p)->rb_left;
 164                 else if (cmp > 0)
 165                         p = &(*p)->rb_right;
 166                 else {
 167                         pr_err("sysctl duplicate entry: ");
 168                         sysctl_print_dir(head->parent);
 169                         pr_cont("/%s\n", entry->procname);
 170                         return -EEXIST;
 171                 }
 172         }
 173 
 174         rb_link_node(node, parent, p);
 175         rb_insert_color(node, &head->parent->root);
 176         return 0;
 177 }
 178 
 179 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
 180 {
 181         struct rb_node *node = &head->node[entry - head->ctl_table].node;
 182 
 183         rb_erase(node, &head->parent->root);
 184 }
 185 
 186 static void init_header(struct ctl_table_header *head,
 187         struct ctl_table_root *root, struct ctl_table_set *set,
 188         struct ctl_node *node, struct ctl_table *table)
 189 {
 190         head->ctl_table = table;
 191         head->ctl_table_arg = table;
 192         head->used = 0;
 193         head->count = 1;
 194         head->nreg = 1;
 195         head->unregistering = NULL;
 196         head->root = root;
 197         head->set = set;
 198         head->parent = NULL;
 199         head->node = node;
 200         INIT_HLIST_HEAD(&head->inodes);
 201         if (node) {
 202                 struct ctl_table *entry;
 203                 for (entry = table; entry->procname; entry++, node++)
 204                         node->header = head;
 205         }
 206 }
 207 
 208 static void erase_header(struct ctl_table_header *head)
 209 {
 210         struct ctl_table *entry;
 211         for (entry = head->ctl_table; entry->procname; entry++)
 212                 erase_entry(head, entry);
 213 }
 214 
 215 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
 216 {
 217         struct ctl_table *entry;
 218         int err;
 219 
 220         /* Is this a permanently empty directory? */
 221         if (is_empty_dir(&dir->header))
 222                 return -EROFS;
 223 
 224         /* Am I creating a permanently empty directory? */
 225         if (header->ctl_table == sysctl_mount_point) {
 226                 if (!RB_EMPTY_ROOT(&dir->root))
 227                         return -EINVAL;
 228                 set_empty_dir(dir);
 229         }
 230 
 231         dir->header.nreg++;
 232         header->parent = dir;
 233         err = insert_links(header);
 234         if (err)
 235                 goto fail_links;
 236         for (entry = header->ctl_table; entry->procname; entry++) {
 237                 err = insert_entry(header, entry);
 238                 if (err)
 239                         goto fail;
 240         }
 241         return 0;
 242 fail:
 243         erase_header(header);
 244         put_links(header);
 245 fail_links:
 246         if (header->ctl_table == sysctl_mount_point)
 247                 clear_empty_dir(dir);
 248         header->parent = NULL;
 249         drop_sysctl_table(&dir->header);
 250         return err;
 251 }
 252 
 253 /* called under sysctl_lock */
 254 static int use_table(struct ctl_table_header *p)
 255 {
 256         if (unlikely(p->unregistering))
 257                 return 0;
 258         p->used++;
 259         return 1;
 260 }
 261 
 262 /* called under sysctl_lock */
 263 static void unuse_table(struct ctl_table_header *p)
 264 {
 265         if (!--p->used)
 266                 if (unlikely(p->unregistering))
 267                         complete(p->unregistering);
 268 }
 269 
 270 static void proc_sys_prune_dcache(struct ctl_table_header *head)
 271 {
 272         struct inode *inode;
 273         struct proc_inode *ei;
 274         struct hlist_node *node;
 275         struct super_block *sb;
 276 
 277         rcu_read_lock();
 278         for (;;) {
 279                 node = hlist_first_rcu(&head->inodes);
 280                 if (!node)
 281                         break;
 282                 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
 283                 spin_lock(&sysctl_lock);
 284                 hlist_del_init_rcu(&ei->sysctl_inodes);
 285                 spin_unlock(&sysctl_lock);
 286 
 287                 inode = &ei->vfs_inode;
 288                 sb = inode->i_sb;
 289                 if (!atomic_inc_not_zero(&sb->s_active))
 290                         continue;
 291                 inode = igrab(inode);
 292                 rcu_read_unlock();
 293                 if (unlikely(!inode)) {
 294                         deactivate_super(sb);
 295                         rcu_read_lock();
 296                         continue;
 297                 }
 298 
 299                 d_prune_aliases(inode);
 300                 iput(inode);
 301                 deactivate_super(sb);
 302 
 303                 rcu_read_lock();
 304         }
 305         rcu_read_unlock();
 306 }
 307 
 308 /* called under sysctl_lock, will reacquire if has to wait */
 309 static void start_unregistering(struct ctl_table_header *p)
 310 {
 311         /*
 312          * if p->used is 0, nobody will ever touch that entry again;
 313          * we'll eliminate all paths to it before dropping sysctl_lock
 314          */
 315         if (unlikely(p->used)) {
 316                 struct completion wait;
 317                 init_completion(&wait);
 318                 p->unregistering = &wait;
 319                 spin_unlock(&sysctl_lock);
 320                 wait_for_completion(&wait);
 321         } else {
 322                 /* anything non-NULL; we'll never dereference it */
 323                 p->unregistering = ERR_PTR(-EINVAL);
 324                 spin_unlock(&sysctl_lock);
 325         }
 326         /*
 327          * Prune dentries for unregistered sysctls: namespaced sysctls
 328          * can have duplicate names and contaminate dcache very badly.
 329          */
 330         proc_sys_prune_dcache(p);
 331         /*
 332          * do not remove from the list until nobody holds it; walking the
 333          * list in do_sysctl() relies on that.
 334          */
 335         spin_lock(&sysctl_lock);
 336         erase_header(p);
 337 }
 338 
 339 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
 340 {
 341         BUG_ON(!head);
 342         spin_lock(&sysctl_lock);
 343         if (!use_table(head))
 344                 head = ERR_PTR(-ENOENT);
 345         spin_unlock(&sysctl_lock);
 346         return head;
 347 }
 348 
 349 static void sysctl_head_finish(struct ctl_table_header *head)
 350 {
 351         if (!head)
 352                 return;
 353         spin_lock(&sysctl_lock);
 354         unuse_table(head);
 355         spin_unlock(&sysctl_lock);
 356 }
 357 
 358 static struct ctl_table_set *
 359 lookup_header_set(struct ctl_table_root *root)
 360 {
 361         struct ctl_table_set *set = &root->default_set;
 362         if (root->lookup)
 363                 set = root->lookup(root);
 364         return set;
 365 }
 366 
 367 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
 368                                       struct ctl_dir *dir,
 369                                       const char *name, int namelen)
 370 {
 371         struct ctl_table_header *head;
 372         struct ctl_table *entry;
 373 
 374         spin_lock(&sysctl_lock);
 375         entry = find_entry(&head, dir, name, namelen);
 376         if (entry && use_table(head))
 377                 *phead = head;
 378         else
 379                 entry = NULL;
 380         spin_unlock(&sysctl_lock);
 381         return entry;
 382 }
 383 
 384 static struct ctl_node *first_usable_entry(struct rb_node *node)
 385 {
 386         struct ctl_node *ctl_node;
 387 
 388         for (;node; node = rb_next(node)) {
 389                 ctl_node = rb_entry(node, struct ctl_node, node);
 390                 if (use_table(ctl_node->header))
 391                         return ctl_node;
 392         }
 393         return NULL;
 394 }
 395 
 396 static void first_entry(struct ctl_dir *dir,
 397         struct ctl_table_header **phead, struct ctl_table **pentry)
 398 {
 399         struct ctl_table_header *head = NULL;
 400         struct ctl_table *entry = NULL;
 401         struct ctl_node *ctl_node;
 402 
 403         spin_lock(&sysctl_lock);
 404         ctl_node = first_usable_entry(rb_first(&dir->root));
 405         spin_unlock(&sysctl_lock);
 406         if (ctl_node) {
 407                 head = ctl_node->header;
 408                 entry = &head->ctl_table[ctl_node - head->node];
 409         }
 410         *phead = head;
 411         *pentry = entry;
 412 }
 413 
 414 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
 415 {
 416         struct ctl_table_header *head = *phead;
 417         struct ctl_table *entry = *pentry;
 418         struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
 419 
 420         spin_lock(&sysctl_lock);
 421         unuse_table(head);
 422 
 423         ctl_node = first_usable_entry(rb_next(&ctl_node->node));
 424         spin_unlock(&sysctl_lock);
 425         head = NULL;
 426         if (ctl_node) {
 427                 head = ctl_node->header;
 428                 entry = &head->ctl_table[ctl_node - head->node];
 429         }
 430         *phead = head;
 431         *pentry = entry;
 432 }
 433 
 434 /*
 435  * sysctl_perm does NOT grant the superuser all rights automatically, because
 436  * some sysctl variables are readonly even to root.
 437  */
 438 
 439 static int test_perm(int mode, int op)
 440 {
 441         if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
 442                 mode >>= 6;
 443         else if (in_egroup_p(GLOBAL_ROOT_GID))
 444                 mode >>= 3;
 445         if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
 446                 return 0;
 447         return -EACCES;
 448 }
 449 
 450 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
 451 {
 452         struct ctl_table_root *root = head->root;
 453         int mode;
 454 
 455         if (root->permissions)
 456                 mode = root->permissions(head, table);
 457         else
 458                 mode = table->mode;
 459 
 460         return test_perm(mode, op);
 461 }
 462 
 463 static struct inode *proc_sys_make_inode(struct super_block *sb,
 464                 struct ctl_table_header *head, struct ctl_table *table)
 465 {
 466         struct ctl_table_root *root = head->root;
 467         struct inode *inode;
 468         struct proc_inode *ei;
 469 
 470         inode = new_inode(sb);
 471         if (!inode)
 472                 return ERR_PTR(-ENOMEM);
 473 
 474         inode->i_ino = get_next_ino();
 475 
 476         ei = PROC_I(inode);
 477 
 478         spin_lock(&sysctl_lock);
 479         if (unlikely(head->unregistering)) {
 480                 spin_unlock(&sysctl_lock);
 481                 iput(inode);
 482                 return ERR_PTR(-ENOENT);
 483         }
 484         ei->sysctl = head;
 485         ei->sysctl_entry = table;
 486         hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
 487         head->count++;
 488         spin_unlock(&sysctl_lock);
 489 
 490         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 491         inode->i_mode = table->mode;
 492         if (!S_ISDIR(table->mode)) {
 493                 inode->i_mode |= S_IFREG;
 494                 inode->i_op = &proc_sys_inode_operations;
 495                 inode->i_fop = &proc_sys_file_operations;
 496         } else {
 497                 inode->i_mode |= S_IFDIR;
 498                 inode->i_op = &proc_sys_dir_operations;
 499                 inode->i_fop = &proc_sys_dir_file_operations;
 500                 if (is_empty_dir(head))
 501                         make_empty_dir_inode(inode);
 502         }
 503 
 504         if (root->set_ownership)
 505                 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
 506         else {
 507                 inode->i_uid = GLOBAL_ROOT_UID;
 508                 inode->i_gid = GLOBAL_ROOT_GID;
 509         }
 510 
 511         return inode;
 512 }
 513 
 514 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
 515 {
 516         spin_lock(&sysctl_lock);
 517         hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
 518         if (!--head->count)
 519                 kfree_rcu(head, rcu);
 520         spin_unlock(&sysctl_lock);
 521 }
 522 
 523 static struct ctl_table_header *grab_header(struct inode *inode)
 524 {
 525         struct ctl_table_header *head = PROC_I(inode)->sysctl;
 526         if (!head)
 527                 head = &sysctl_table_root.default_set.dir.header;
 528         return sysctl_head_grab(head);
 529 }
 530 
 531 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
 532                                         unsigned int flags)
 533 {
 534         struct ctl_table_header *head = grab_header(dir);
 535         struct ctl_table_header *h = NULL;
 536         const struct qstr *name = &dentry->d_name;
 537         struct ctl_table *p;
 538         struct inode *inode;
 539         struct dentry *err = ERR_PTR(-ENOENT);
 540         struct ctl_dir *ctl_dir;
 541         int ret;
 542 
 543         if (IS_ERR(head))
 544                 return ERR_CAST(head);
 545 
 546         ctl_dir = container_of(head, struct ctl_dir, header);
 547 
 548         p = lookup_entry(&h, ctl_dir, name->name, name->len);
 549         if (!p)
 550                 goto out;
 551 
 552         if (S_ISLNK(p->mode)) {
 553                 ret = sysctl_follow_link(&h, &p);
 554                 err = ERR_PTR(ret);
 555                 if (ret)
 556                         goto out;
 557         }
 558 
 559         inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
 560         if (IS_ERR(inode)) {
 561                 err = ERR_CAST(inode);
 562                 goto out;
 563         }
 564 
 565         d_set_d_op(dentry, &proc_sys_dentry_operations);
 566         err = d_splice_alias(inode, dentry);
 567 
 568 out:
 569         if (h)
 570                 sysctl_head_finish(h);
 571         sysctl_head_finish(head);
 572         return err;
 573 }
 574 
 575 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
 576                 size_t count, loff_t *ppos, int write)
 577 {
 578         struct inode *inode = file_inode(filp);
 579         struct ctl_table_header *head = grab_header(inode);
 580         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 581         void *new_buf = NULL;
 582         ssize_t error;
 583 
 584         if (IS_ERR(head))
 585                 return PTR_ERR(head);
 586 
 587         /*
 588          * At this point we know that the sysctl was not unregistered
 589          * and won't be until we finish.
 590          */
 591         error = -EPERM;
 592         if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
 593                 goto out;
 594 
 595         /* if that can happen at all, it should be -EINVAL, not -EISDIR */
 596         error = -EINVAL;
 597         if (!table->proc_handler)
 598                 goto out;
 599 
 600         error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, &count,
 601                                            ppos, &new_buf);
 602         if (error)
 603                 goto out;
 604 
 605         /* careful: calling conventions are nasty here */
 606         if (new_buf) {
 607                 mm_segment_t old_fs;
 608 
 609                 old_fs = get_fs();
 610                 set_fs(KERNEL_DS);
 611                 error = table->proc_handler(table, write, (void __user *)new_buf,
 612                                             &count, ppos);
 613                 set_fs(old_fs);
 614                 kfree(new_buf);
 615         } else {
 616                 error = table->proc_handler(table, write, buf, &count, ppos);
 617         }
 618 
 619         if (!error)
 620                 error = count;
 621 out:
 622         sysctl_head_finish(head);
 623 
 624         return error;
 625 }
 626 
 627 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
 628                                 size_t count, loff_t *ppos)
 629 {
 630         return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
 631 }
 632 
 633 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
 634                                 size_t count, loff_t *ppos)
 635 {
 636         return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
 637 }
 638 
 639 static int proc_sys_open(struct inode *inode, struct file *filp)
 640 {
 641         struct ctl_table_header *head = grab_header(inode);
 642         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 643 
 644         /* sysctl was unregistered */
 645         if (IS_ERR(head))
 646                 return PTR_ERR(head);
 647 
 648         if (table->poll)
 649                 filp->private_data = proc_sys_poll_event(table->poll);
 650 
 651         sysctl_head_finish(head);
 652 
 653         return 0;
 654 }
 655 
 656 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
 657 {
 658         struct inode *inode = file_inode(filp);
 659         struct ctl_table_header *head = grab_header(inode);
 660         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 661         __poll_t ret = DEFAULT_POLLMASK;
 662         unsigned long event;
 663 
 664         /* sysctl was unregistered */
 665         if (IS_ERR(head))
 666                 return EPOLLERR | EPOLLHUP;
 667 
 668         if (!table->proc_handler)
 669                 goto out;
 670 
 671         if (!table->poll)
 672                 goto out;
 673 
 674         event = (unsigned long)filp->private_data;
 675         poll_wait(filp, &table->poll->wait, wait);
 676 
 677         if (event != atomic_read(&table->poll->event)) {
 678                 filp->private_data = proc_sys_poll_event(table->poll);
 679                 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
 680         }
 681 
 682 out:
 683         sysctl_head_finish(head);
 684 
 685         return ret;
 686 }
 687 
 688 static bool proc_sys_fill_cache(struct file *file,
 689                                 struct dir_context *ctx,
 690                                 struct ctl_table_header *head,
 691                                 struct ctl_table *table)
 692 {
 693         struct dentry *child, *dir = file->f_path.dentry;
 694         struct inode *inode;
 695         struct qstr qname;
 696         ino_t ino = 0;
 697         unsigned type = DT_UNKNOWN;
 698 
 699         qname.name = table->procname;
 700         qname.len  = strlen(table->procname);
 701         qname.hash = full_name_hash(dir, qname.name, qname.len);
 702 
 703         child = d_lookup(dir, &qname);
 704         if (!child) {
 705                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 706                 child = d_alloc_parallel(dir, &qname, &wq);
 707                 if (IS_ERR(child))
 708                         return false;
 709                 if (d_in_lookup(child)) {
 710                         struct dentry *res;
 711                         inode = proc_sys_make_inode(dir->d_sb, head, table);
 712                         if (IS_ERR(inode)) {
 713                                 d_lookup_done(child);
 714                                 dput(child);
 715                                 return false;
 716                         }
 717                         d_set_d_op(child, &proc_sys_dentry_operations);
 718                         res = d_splice_alias(inode, child);
 719                         d_lookup_done(child);
 720                         if (unlikely(res)) {
 721                                 if (IS_ERR(res)) {
 722                                         dput(child);
 723                                         return false;
 724                                 }
 725                                 dput(child);
 726                                 child = res;
 727                         }
 728                 }
 729         }
 730         inode = d_inode(child);
 731         ino  = inode->i_ino;
 732         type = inode->i_mode >> 12;
 733         dput(child);
 734         return dir_emit(ctx, qname.name, qname.len, ino, type);
 735 }
 736 
 737 static bool proc_sys_link_fill_cache(struct file *file,
 738                                     struct dir_context *ctx,
 739                                     struct ctl_table_header *head,
 740                                     struct ctl_table *table)
 741 {
 742         bool ret = true;
 743 
 744         head = sysctl_head_grab(head);
 745         if (IS_ERR(head))
 746                 return false;
 747 
 748         /* It is not an error if we can not follow the link ignore it */
 749         if (sysctl_follow_link(&head, &table))
 750                 goto out;
 751 
 752         ret = proc_sys_fill_cache(file, ctx, head, table);
 753 out:
 754         sysctl_head_finish(head);
 755         return ret;
 756 }
 757 
 758 static int scan(struct ctl_table_header *head, struct ctl_table *table,
 759                 unsigned long *pos, struct file *file,
 760                 struct dir_context *ctx)
 761 {
 762         bool res;
 763 
 764         if ((*pos)++ < ctx->pos)
 765                 return true;
 766 
 767         if (unlikely(S_ISLNK(table->mode)))
 768                 res = proc_sys_link_fill_cache(file, ctx, head, table);
 769         else
 770                 res = proc_sys_fill_cache(file, ctx, head, table);
 771 
 772         if (res)
 773                 ctx->pos = *pos;
 774 
 775         return res;
 776 }
 777 
 778 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
 779 {
 780         struct ctl_table_header *head = grab_header(file_inode(file));
 781         struct ctl_table_header *h = NULL;
 782         struct ctl_table *entry;
 783         struct ctl_dir *ctl_dir;
 784         unsigned long pos;
 785 
 786         if (IS_ERR(head))
 787                 return PTR_ERR(head);
 788 
 789         ctl_dir = container_of(head, struct ctl_dir, header);
 790 
 791         if (!dir_emit_dots(file, ctx))
 792                 goto out;
 793 
 794         pos = 2;
 795 
 796         for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
 797                 if (!scan(h, entry, &pos, file, ctx)) {
 798                         sysctl_head_finish(h);
 799                         break;
 800                 }
 801         }
 802 out:
 803         sysctl_head_finish(head);
 804         return 0;
 805 }
 806 
 807 static int proc_sys_permission(struct inode *inode, int mask)
 808 {
 809         /*
 810          * sysctl entries that are not writeable,
 811          * are _NOT_ writeable, capabilities or not.
 812          */
 813         struct ctl_table_header *head;
 814         struct ctl_table *table;
 815         int error;
 816 
 817         /* Executable files are not allowed under /proc/sys/ */
 818         if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
 819                 return -EACCES;
 820 
 821         head = grab_header(inode);
 822         if (IS_ERR(head))
 823                 return PTR_ERR(head);
 824 
 825         table = PROC_I(inode)->sysctl_entry;
 826         if (!table) /* global root - r-xr-xr-x */
 827                 error = mask & MAY_WRITE ? -EACCES : 0;
 828         else /* Use the permissions on the sysctl table entry */
 829                 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
 830 
 831         sysctl_head_finish(head);
 832         return error;
 833 }
 834 
 835 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
 836 {
 837         struct inode *inode = d_inode(dentry);
 838         int error;
 839 
 840         if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
 841                 return -EPERM;
 842 
 843         error = setattr_prepare(dentry, attr);
 844         if (error)
 845                 return error;
 846 
 847         setattr_copy(inode, attr);
 848         mark_inode_dirty(inode);
 849         return 0;
 850 }
 851 
 852 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
 853                             u32 request_mask, unsigned int query_flags)
 854 {
 855         struct inode *inode = d_inode(path->dentry);
 856         struct ctl_table_header *head = grab_header(inode);
 857         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 858 
 859         if (IS_ERR(head))
 860                 return PTR_ERR(head);
 861 
 862         generic_fillattr(inode, stat);
 863         if (table)
 864                 stat->mode = (stat->mode & S_IFMT) | table->mode;
 865 
 866         sysctl_head_finish(head);
 867         return 0;
 868 }
 869 
 870 static const struct file_operations proc_sys_file_operations = {
 871         .open           = proc_sys_open,
 872         .poll           = proc_sys_poll,
 873         .read           = proc_sys_read,
 874         .write          = proc_sys_write,
 875         .llseek         = default_llseek,
 876 };
 877 
 878 static const struct file_operations proc_sys_dir_file_operations = {
 879         .read           = generic_read_dir,
 880         .iterate_shared = proc_sys_readdir,
 881         .llseek         = generic_file_llseek,
 882 };
 883 
 884 static const struct inode_operations proc_sys_inode_operations = {
 885         .permission     = proc_sys_permission,
 886         .setattr        = proc_sys_setattr,
 887         .getattr        = proc_sys_getattr,
 888 };
 889 
 890 static const struct inode_operations proc_sys_dir_operations = {
 891         .lookup         = proc_sys_lookup,
 892         .permission     = proc_sys_permission,
 893         .setattr        = proc_sys_setattr,
 894         .getattr        = proc_sys_getattr,
 895 };
 896 
 897 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
 898 {
 899         if (flags & LOOKUP_RCU)
 900                 return -ECHILD;
 901         return !PROC_I(d_inode(dentry))->sysctl->unregistering;
 902 }
 903 
 904 static int proc_sys_delete(const struct dentry *dentry)
 905 {
 906         return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
 907 }
 908 
 909 static int sysctl_is_seen(struct ctl_table_header *p)
 910 {
 911         struct ctl_table_set *set = p->set;
 912         int res;
 913         spin_lock(&sysctl_lock);
 914         if (p->unregistering)
 915                 res = 0;
 916         else if (!set->is_seen)
 917                 res = 1;
 918         else
 919                 res = set->is_seen(set);
 920         spin_unlock(&sysctl_lock);
 921         return res;
 922 }
 923 
 924 static int proc_sys_compare(const struct dentry *dentry,
 925                 unsigned int len, const char *str, const struct qstr *name)
 926 {
 927         struct ctl_table_header *head;
 928         struct inode *inode;
 929 
 930         /* Although proc doesn't have negative dentries, rcu-walk means
 931          * that inode here can be NULL */
 932         /* AV: can it, indeed? */
 933         inode = d_inode_rcu(dentry);
 934         if (!inode)
 935                 return 1;
 936         if (name->len != len)
 937                 return 1;
 938         if (memcmp(name->name, str, len))
 939                 return 1;
 940         head = rcu_dereference(PROC_I(inode)->sysctl);
 941         return !head || !sysctl_is_seen(head);
 942 }
 943 
 944 static const struct dentry_operations proc_sys_dentry_operations = {
 945         .d_revalidate   = proc_sys_revalidate,
 946         .d_delete       = proc_sys_delete,
 947         .d_compare      = proc_sys_compare,
 948 };
 949 
 950 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
 951                                    const char *name, int namelen)
 952 {
 953         struct ctl_table_header *head;
 954         struct ctl_table *entry;
 955 
 956         entry = find_entry(&head, dir, name, namelen);
 957         if (!entry)
 958                 return ERR_PTR(-ENOENT);
 959         if (!S_ISDIR(entry->mode))
 960                 return ERR_PTR(-ENOTDIR);
 961         return container_of(head, struct ctl_dir, header);
 962 }
 963 
 964 static struct ctl_dir *new_dir(struct ctl_table_set *set,
 965                                const char *name, int namelen)
 966 {
 967         struct ctl_table *table;
 968         struct ctl_dir *new;
 969         struct ctl_node *node;
 970         char *new_name;
 971 
 972         new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
 973                       sizeof(struct ctl_table)*2 +  namelen + 1,
 974                       GFP_KERNEL);
 975         if (!new)
 976                 return NULL;
 977 
 978         node = (struct ctl_node *)(new + 1);
 979         table = (struct ctl_table *)(node + 1);
 980         new_name = (char *)(table + 2);
 981         memcpy(new_name, name, namelen);
 982         new_name[namelen] = '\0';
 983         table[0].procname = new_name;
 984         table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
 985         init_header(&new->header, set->dir.header.root, set, node, table);
 986 
 987         return new;
 988 }
 989 
 990 /**
 991  * get_subdir - find or create a subdir with the specified name.
 992  * @dir:  Directory to create the subdirectory in
 993  * @name: The name of the subdirectory to find or create
 994  * @namelen: The length of name
 995  *
 996  * Takes a directory with an elevated reference count so we know that
 997  * if we drop the lock the directory will not go away.  Upon success
 998  * the reference is moved from @dir to the returned subdirectory.
 999  * Upon error an error code is returned and the reference on @dir is
1000  * simply dropped.
1001  */
1002 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
1003                                   const char *name, int namelen)
1004 {
1005         struct ctl_table_set *set = dir->header.set;
1006         struct ctl_dir *subdir, *new = NULL;
1007         int err;
1008 
1009         spin_lock(&sysctl_lock);
1010         subdir = find_subdir(dir, name, namelen);
1011         if (!IS_ERR(subdir))
1012                 goto found;
1013         if (PTR_ERR(subdir) != -ENOENT)
1014                 goto failed;
1015 
1016         spin_unlock(&sysctl_lock);
1017         new = new_dir(set, name, namelen);
1018         spin_lock(&sysctl_lock);
1019         subdir = ERR_PTR(-ENOMEM);
1020         if (!new)
1021                 goto failed;
1022 
1023         /* Was the subdir added while we dropped the lock? */
1024         subdir = find_subdir(dir, name, namelen);
1025         if (!IS_ERR(subdir))
1026                 goto found;
1027         if (PTR_ERR(subdir) != -ENOENT)
1028                 goto failed;
1029 
1030         /* Nope.  Use the our freshly made directory entry. */
1031         err = insert_header(dir, &new->header);
1032         subdir = ERR_PTR(err);
1033         if (err)
1034                 goto failed;
1035         subdir = new;
1036 found:
1037         subdir->header.nreg++;
1038 failed:
1039         if (IS_ERR(subdir)) {
1040                 pr_err("sysctl could not get directory: ");
1041                 sysctl_print_dir(dir);
1042                 pr_cont("/%*.*s %ld\n",
1043                         namelen, namelen, name, PTR_ERR(subdir));
1044         }
1045         drop_sysctl_table(&dir->header);
1046         if (new)
1047                 drop_sysctl_table(&new->header);
1048         spin_unlock(&sysctl_lock);
1049         return subdir;
1050 }
1051 
1052 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1053 {
1054         struct ctl_dir *parent;
1055         const char *procname;
1056         if (!dir->header.parent)
1057                 return &set->dir;
1058         parent = xlate_dir(set, dir->header.parent);
1059         if (IS_ERR(parent))
1060                 return parent;
1061         procname = dir->header.ctl_table[0].procname;
1062         return find_subdir(parent, procname, strlen(procname));
1063 }
1064 
1065 static int sysctl_follow_link(struct ctl_table_header **phead,
1066         struct ctl_table **pentry)
1067 {
1068         struct ctl_table_header *head;
1069         struct ctl_table_root *root;
1070         struct ctl_table_set *set;
1071         struct ctl_table *entry;
1072         struct ctl_dir *dir;
1073         int ret;
1074 
1075         ret = 0;
1076         spin_lock(&sysctl_lock);
1077         root = (*pentry)->data;
1078         set = lookup_header_set(root);
1079         dir = xlate_dir(set, (*phead)->parent);
1080         if (IS_ERR(dir))
1081                 ret = PTR_ERR(dir);
1082         else {
1083                 const char *procname = (*pentry)->procname;
1084                 head = NULL;
1085                 entry = find_entry(&head, dir, procname, strlen(procname));
1086                 ret = -ENOENT;
1087                 if (entry && use_table(head)) {
1088                         unuse_table(*phead);
1089                         *phead = head;
1090                         *pentry = entry;
1091                         ret = 0;
1092                 }
1093         }
1094 
1095         spin_unlock(&sysctl_lock);
1096         return ret;
1097 }
1098 
1099 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1100 {
1101         struct va_format vaf;
1102         va_list args;
1103 
1104         va_start(args, fmt);
1105         vaf.fmt = fmt;
1106         vaf.va = &args;
1107 
1108         pr_err("sysctl table check failed: %s/%s %pV\n",
1109                path, table->procname, &vaf);
1110 
1111         va_end(args);
1112         return -EINVAL;
1113 }
1114 
1115 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1116 {
1117         int err = 0;
1118 
1119         if ((table->proc_handler == proc_douintvec) ||
1120             (table->proc_handler == proc_douintvec_minmax)) {
1121                 if (table->maxlen != sizeof(unsigned int))
1122                         err |= sysctl_err(path, table, "array not allowed");
1123         }
1124 
1125         return err;
1126 }
1127 
1128 static int sysctl_check_table(const char *path, struct ctl_table *table)
1129 {
1130         int err = 0;
1131         for (; table->procname; table++) {
1132                 if (table->child)
1133                         err |= sysctl_err(path, table, "Not a file");
1134 
1135                 if ((table->proc_handler == proc_dostring) ||
1136                     (table->proc_handler == proc_dointvec) ||
1137                     (table->proc_handler == proc_douintvec) ||
1138                     (table->proc_handler == proc_douintvec_minmax) ||
1139                     (table->proc_handler == proc_dointvec_minmax) ||
1140                     (table->proc_handler == proc_dointvec_jiffies) ||
1141                     (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1142                     (table->proc_handler == proc_dointvec_ms_jiffies) ||
1143                     (table->proc_handler == proc_doulongvec_minmax) ||
1144                     (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1145                         if (!table->data)
1146                                 err |= sysctl_err(path, table, "No data");
1147                         if (!table->maxlen)
1148                                 err |= sysctl_err(path, table, "No maxlen");
1149                         else
1150                                 err |= sysctl_check_table_array(path, table);
1151                 }
1152                 if (!table->proc_handler)
1153                         err |= sysctl_err(path, table, "No proc_handler");
1154 
1155                 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1156                         err |= sysctl_err(path, table, "bogus .mode 0%o",
1157                                 table->mode);
1158         }
1159         return err;
1160 }
1161 
1162 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1163         struct ctl_table_root *link_root)
1164 {
1165         struct ctl_table *link_table, *entry, *link;
1166         struct ctl_table_header *links;
1167         struct ctl_node *node;
1168         char *link_name;
1169         int nr_entries, name_bytes;
1170 
1171         name_bytes = 0;
1172         nr_entries = 0;
1173         for (entry = table; entry->procname; entry++) {
1174                 nr_entries++;
1175                 name_bytes += strlen(entry->procname) + 1;
1176         }
1177 
1178         links = kzalloc(sizeof(struct ctl_table_header) +
1179                         sizeof(struct ctl_node)*nr_entries +
1180                         sizeof(struct ctl_table)*(nr_entries + 1) +
1181                         name_bytes,
1182                         GFP_KERNEL);
1183 
1184         if (!links)
1185                 return NULL;
1186 
1187         node = (struct ctl_node *)(links + 1);
1188         link_table = (struct ctl_table *)(node + nr_entries);
1189         link_name = (char *)&link_table[nr_entries + 1];
1190 
1191         for (link = link_table, entry = table; entry->procname; link++, entry++) {
1192                 int len = strlen(entry->procname) + 1;
1193                 memcpy(link_name, entry->procname, len);
1194                 link->procname = link_name;
1195                 link->mode = S_IFLNK|S_IRWXUGO;
1196                 link->data = link_root;
1197                 link_name += len;
1198         }
1199         init_header(links, dir->header.root, dir->header.set, node, link_table);
1200         links->nreg = nr_entries;
1201 
1202         return links;
1203 }
1204 
1205 static bool get_links(struct ctl_dir *dir,
1206         struct ctl_table *table, struct ctl_table_root *link_root)
1207 {
1208         struct ctl_table_header *head;
1209         struct ctl_table *entry, *link;
1210 
1211         /* Are there links available for every entry in table? */
1212         for (entry = table; entry->procname; entry++) {
1213                 const char *procname = entry->procname;
1214                 link = find_entry(&head, dir, procname, strlen(procname));
1215                 if (!link)
1216                         return false;
1217                 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1218                         continue;
1219                 if (S_ISLNK(link->mode) && (link->data == link_root))
1220                         continue;
1221                 return false;
1222         }
1223 
1224         /* The checks passed.  Increase the registration count on the links */
1225         for (entry = table; entry->procname; entry++) {
1226                 const char *procname = entry->procname;
1227                 link = find_entry(&head, dir, procname, strlen(procname));
1228                 head->nreg++;
1229         }
1230         return true;
1231 }
1232 
1233 static int insert_links(struct ctl_table_header *head)
1234 {
1235         struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1236         struct ctl_dir *core_parent = NULL;
1237         struct ctl_table_header *links;
1238         int err;
1239 
1240         if (head->set == root_set)
1241                 return 0;
1242 
1243         core_parent = xlate_dir(root_set, head->parent);
1244         if (IS_ERR(core_parent))
1245                 return 0;
1246 
1247         if (get_links(core_parent, head->ctl_table, head->root))
1248                 return 0;
1249 
1250         core_parent->header.nreg++;
1251         spin_unlock(&sysctl_lock);
1252 
1253         links = new_links(core_parent, head->ctl_table, head->root);
1254 
1255         spin_lock(&sysctl_lock);
1256         err = -ENOMEM;
1257         if (!links)
1258                 goto out;
1259 
1260         err = 0;
1261         if (get_links(core_parent, head->ctl_table, head->root)) {
1262                 kfree(links);
1263                 goto out;
1264         }
1265 
1266         err = insert_header(core_parent, links);
1267         if (err)
1268                 kfree(links);
1269 out:
1270         drop_sysctl_table(&core_parent->header);
1271         return err;
1272 }
1273 
1274 /**
1275  * __register_sysctl_table - register a leaf sysctl table
1276  * @set: Sysctl tree to register on
1277  * @path: The path to the directory the sysctl table is in.
1278  * @table: the top-level table structure
1279  *
1280  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1281  * array. A completely 0 filled entry terminates the table.
1282  *
1283  * The members of the &struct ctl_table structure are used as follows:
1284  *
1285  * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1286  *            enter a sysctl file
1287  *
1288  * data - a pointer to data for use by proc_handler
1289  *
1290  * maxlen - the maximum size in bytes of the data
1291  *
1292  * mode - the file permissions for the /proc/sys file
1293  *
1294  * child - must be %NULL.
1295  *
1296  * proc_handler - the text handler routine (described below)
1297  *
1298  * extra1, extra2 - extra pointers usable by the proc handler routines
1299  *
1300  * Leaf nodes in the sysctl tree will be represented by a single file
1301  * under /proc; non-leaf nodes will be represented by directories.
1302  *
1303  * There must be a proc_handler routine for any terminal nodes.
1304  * Several default handlers are available to cover common cases -
1305  *
1306  * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1307  * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1308  * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1309  *
1310  * It is the handler's job to read the input buffer from user memory
1311  * and process it. The handler should return 0 on success.
1312  *
1313  * This routine returns %NULL on a failure to register, and a pointer
1314  * to the table header on success.
1315  */
1316 struct ctl_table_header *__register_sysctl_table(
1317         struct ctl_table_set *set,
1318         const char *path, struct ctl_table *table)
1319 {
1320         struct ctl_table_root *root = set->dir.header.root;
1321         struct ctl_table_header *header;
1322         const char *name, *nextname;
1323         struct ctl_dir *dir;
1324         struct ctl_table *entry;
1325         struct ctl_node *node;
1326         int nr_entries = 0;
1327 
1328         for (entry = table; entry->procname; entry++)
1329                 nr_entries++;
1330 
1331         header = kzalloc(sizeof(struct ctl_table_header) +
1332                          sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1333         if (!header)
1334                 return NULL;
1335 
1336         node = (struct ctl_node *)(header + 1);
1337         init_header(header, root, set, node, table);
1338         if (sysctl_check_table(path, table))
1339                 goto fail;
1340 
1341         spin_lock(&sysctl_lock);
1342         dir = &set->dir;
1343         /* Reference moved down the diretory tree get_subdir */
1344         dir->header.nreg++;
1345         spin_unlock(&sysctl_lock);
1346 
1347         /* Find the directory for the ctl_table */
1348         for (name = path; name; name = nextname) {
1349                 int namelen;
1350                 nextname = strchr(name, '/');
1351                 if (nextname) {
1352                         namelen = nextname - name;
1353                         nextname++;
1354                 } else {
1355                         namelen = strlen(name);
1356                 }
1357                 if (namelen == 0)
1358                         continue;
1359 
1360                 dir = get_subdir(dir, name, namelen);
1361                 if (IS_ERR(dir))
1362                         goto fail;
1363         }
1364 
1365         spin_lock(&sysctl_lock);
1366         if (insert_header(dir, header))
1367                 goto fail_put_dir_locked;
1368 
1369         drop_sysctl_table(&dir->header);
1370         spin_unlock(&sysctl_lock);
1371 
1372         return header;
1373 
1374 fail_put_dir_locked:
1375         drop_sysctl_table(&dir->header);
1376         spin_unlock(&sysctl_lock);
1377 fail:
1378         kfree(header);
1379         dump_stack();
1380         return NULL;
1381 }
1382 
1383 /**
1384  * register_sysctl - register a sysctl table
1385  * @path: The path to the directory the sysctl table is in.
1386  * @table: the table structure
1387  *
1388  * Register a sysctl table. @table should be a filled in ctl_table
1389  * array. A completely 0 filled entry terminates the table.
1390  *
1391  * See __register_sysctl_table for more details.
1392  */
1393 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1394 {
1395         return __register_sysctl_table(&sysctl_table_root.default_set,
1396                                         path, table);
1397 }
1398 EXPORT_SYMBOL(register_sysctl);
1399 
1400 static char *append_path(const char *path, char *pos, const char *name)
1401 {
1402         int namelen;
1403         namelen = strlen(name);
1404         if (((pos - path) + namelen + 2) >= PATH_MAX)
1405                 return NULL;
1406         memcpy(pos, name, namelen);
1407         pos[namelen] = '/';
1408         pos[namelen + 1] = '\0';
1409         pos += namelen + 1;
1410         return pos;
1411 }
1412 
1413 static int count_subheaders(struct ctl_table *table)
1414 {
1415         int has_files = 0;
1416         int nr_subheaders = 0;
1417         struct ctl_table *entry;
1418 
1419         /* special case: no directory and empty directory */
1420         if (!table || !table->procname)
1421                 return 1;
1422 
1423         for (entry = table; entry->procname; entry++) {
1424                 if (entry->child)
1425                         nr_subheaders += count_subheaders(entry->child);
1426                 else
1427                         has_files = 1;
1428         }
1429         return nr_subheaders + has_files;
1430 }
1431 
1432 static int register_leaf_sysctl_tables(const char *path, char *pos,
1433         struct ctl_table_header ***subheader, struct ctl_table_set *set,
1434         struct ctl_table *table)
1435 {
1436         struct ctl_table *ctl_table_arg = NULL;
1437         struct ctl_table *entry, *files;
1438         int nr_files = 0;
1439         int nr_dirs = 0;
1440         int err = -ENOMEM;
1441 
1442         for (entry = table; entry->procname; entry++) {
1443                 if (entry->child)
1444                         nr_dirs++;
1445                 else
1446                         nr_files++;
1447         }
1448 
1449         files = table;
1450         /* If there are mixed files and directories we need a new table */
1451         if (nr_dirs && nr_files) {
1452                 struct ctl_table *new;
1453                 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1454                                 GFP_KERNEL);
1455                 if (!files)
1456                         goto out;
1457 
1458                 ctl_table_arg = files;
1459                 for (new = files, entry = table; entry->procname; entry++) {
1460                         if (entry->child)
1461                                 continue;
1462                         *new = *entry;
1463                         new++;
1464                 }
1465         }
1466 
1467         /* Register everything except a directory full of subdirectories */
1468         if (nr_files || !nr_dirs) {
1469                 struct ctl_table_header *header;
1470                 header = __register_sysctl_table(set, path, files);
1471                 if (!header) {
1472                         kfree(ctl_table_arg);
1473                         goto out;
1474                 }
1475 
1476                 /* Remember if we need to free the file table */
1477                 header->ctl_table_arg = ctl_table_arg;
1478                 **subheader = header;
1479                 (*subheader)++;
1480         }
1481 
1482         /* Recurse into the subdirectories. */
1483         for (entry = table; entry->procname; entry++) {
1484                 char *child_pos;
1485 
1486                 if (!entry->child)
1487                         continue;
1488 
1489                 err = -ENAMETOOLONG;
1490                 child_pos = append_path(path, pos, entry->procname);
1491                 if (!child_pos)
1492                         goto out;
1493 
1494                 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1495                                                   set, entry->child);
1496                 pos[0] = '\0';
1497                 if (err)
1498                         goto out;
1499         }
1500         err = 0;
1501 out:
1502         /* On failure our caller will unregister all registered subheaders */
1503         return err;
1504 }
1505 
1506 /**
1507  * __register_sysctl_paths - register a sysctl table hierarchy
1508  * @set: Sysctl tree to register on
1509  * @path: The path to the directory the sysctl table is in.
1510  * @table: the top-level table structure
1511  *
1512  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1513  * array. A completely 0 filled entry terminates the table.
1514  *
1515  * See __register_sysctl_table for more details.
1516  */
1517 struct ctl_table_header *__register_sysctl_paths(
1518         struct ctl_table_set *set,
1519         const struct ctl_path *path, struct ctl_table *table)
1520 {
1521         struct ctl_table *ctl_table_arg = table;
1522         int nr_subheaders = count_subheaders(table);
1523         struct ctl_table_header *header = NULL, **subheaders, **subheader;
1524         const struct ctl_path *component;
1525         char *new_path, *pos;
1526 
1527         pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1528         if (!new_path)
1529                 return NULL;
1530 
1531         pos[0] = '\0';
1532         for (component = path; component->procname; component++) {
1533                 pos = append_path(new_path, pos, component->procname);
1534                 if (!pos)
1535                         goto out;
1536         }
1537         while (table->procname && table->child && !table[1].procname) {
1538                 pos = append_path(new_path, pos, table->procname);
1539                 if (!pos)
1540                         goto out;
1541                 table = table->child;
1542         }
1543         if (nr_subheaders == 1) {
1544                 header = __register_sysctl_table(set, new_path, table);
1545                 if (header)
1546                         header->ctl_table_arg = ctl_table_arg;
1547         } else {
1548                 header = kzalloc(sizeof(*header) +
1549                                  sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1550                 if (!header)
1551                         goto out;
1552 
1553                 subheaders = (struct ctl_table_header **) (header + 1);
1554                 subheader = subheaders;
1555                 header->ctl_table_arg = ctl_table_arg;
1556 
1557                 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1558                                                 set, table))
1559                         goto err_register_leaves;
1560         }
1561 
1562 out:
1563         kfree(new_path);
1564         return header;
1565 
1566 err_register_leaves:
1567         while (subheader > subheaders) {
1568                 struct ctl_table_header *subh = *(--subheader);
1569                 struct ctl_table *table = subh->ctl_table_arg;
1570                 unregister_sysctl_table(subh);
1571                 kfree(table);
1572         }
1573         kfree(header);
1574         header = NULL;
1575         goto out;
1576 }
1577 
1578 /**
1579  * register_sysctl_table_path - register a sysctl table hierarchy
1580  * @path: The path to the directory the sysctl table is in.
1581  * @table: the top-level table structure
1582  *
1583  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1584  * array. A completely 0 filled entry terminates the table.
1585  *
1586  * See __register_sysctl_paths for more details.
1587  */
1588 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1589                                                 struct ctl_table *table)
1590 {
1591         return __register_sysctl_paths(&sysctl_table_root.default_set,
1592                                         path, table);
1593 }
1594 EXPORT_SYMBOL(register_sysctl_paths);
1595 
1596 /**
1597  * register_sysctl_table - register a sysctl table hierarchy
1598  * @table: the top-level table structure
1599  *
1600  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1601  * array. A completely 0 filled entry terminates the table.
1602  *
1603  * See register_sysctl_paths for more details.
1604  */
1605 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1606 {
1607         static const struct ctl_path null_path[] = { {} };
1608 
1609         return register_sysctl_paths(null_path, table);
1610 }
1611 EXPORT_SYMBOL(register_sysctl_table);
1612 
1613 static void put_links(struct ctl_table_header *header)
1614 {
1615         struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1616         struct ctl_table_root *root = header->root;
1617         struct ctl_dir *parent = header->parent;
1618         struct ctl_dir *core_parent;
1619         struct ctl_table *entry;
1620 
1621         if (header->set == root_set)
1622                 return;
1623 
1624         core_parent = xlate_dir(root_set, parent);
1625         if (IS_ERR(core_parent))
1626                 return;
1627 
1628         for (entry = header->ctl_table; entry->procname; entry++) {
1629                 struct ctl_table_header *link_head;
1630                 struct ctl_table *link;
1631                 const char *name = entry->procname;
1632 
1633                 link = find_entry(&link_head, core_parent, name, strlen(name));
1634                 if (link &&
1635                     ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1636                      (S_ISLNK(link->mode) && (link->data == root)))) {
1637                         drop_sysctl_table(link_head);
1638                 }
1639                 else {
1640                         pr_err("sysctl link missing during unregister: ");
1641                         sysctl_print_dir(parent);
1642                         pr_cont("/%s\n", name);
1643                 }
1644         }
1645 }
1646 
1647 static void drop_sysctl_table(struct ctl_table_header *header)
1648 {
1649         struct ctl_dir *parent = header->parent;
1650 
1651         if (--header->nreg)
1652                 return;
1653 
1654         if (parent) {
1655                 put_links(header);
1656                 start_unregistering(header);
1657         }
1658 
1659         if (!--header->count)
1660                 kfree_rcu(header, rcu);
1661 
1662         if (parent)
1663                 drop_sysctl_table(&parent->header);
1664 }
1665 
1666 /**
1667  * unregister_sysctl_table - unregister a sysctl table hierarchy
1668  * @header: the header returned from register_sysctl_table
1669  *
1670  * Unregisters the sysctl table and all children. proc entries may not
1671  * actually be removed until they are no longer used by anyone.
1672  */
1673 void unregister_sysctl_table(struct ctl_table_header * header)
1674 {
1675         int nr_subheaders;
1676         might_sleep();
1677 
1678         if (header == NULL)
1679                 return;
1680 
1681         nr_subheaders = count_subheaders(header->ctl_table_arg);
1682         if (unlikely(nr_subheaders > 1)) {
1683                 struct ctl_table_header **subheaders;
1684                 int i;
1685 
1686                 subheaders = (struct ctl_table_header **)(header + 1);
1687                 for (i = nr_subheaders -1; i >= 0; i--) {
1688                         struct ctl_table_header *subh = subheaders[i];
1689                         struct ctl_table *table = subh->ctl_table_arg;
1690                         unregister_sysctl_table(subh);
1691                         kfree(table);
1692                 }
1693                 kfree(header);
1694                 return;
1695         }
1696 
1697         spin_lock(&sysctl_lock);
1698         drop_sysctl_table(header);
1699         spin_unlock(&sysctl_lock);
1700 }
1701 EXPORT_SYMBOL(unregister_sysctl_table);
1702 
1703 void setup_sysctl_set(struct ctl_table_set *set,
1704         struct ctl_table_root *root,
1705         int (*is_seen)(struct ctl_table_set *))
1706 {
1707         memset(set, 0, sizeof(*set));
1708         set->is_seen = is_seen;
1709         init_header(&set->dir.header, root, set, NULL, root_table);
1710 }
1711 
1712 void retire_sysctl_set(struct ctl_table_set *set)
1713 {
1714         WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1715 }
1716 
1717 int __init proc_sys_init(void)
1718 {
1719         struct proc_dir_entry *proc_sys_root;
1720 
1721         proc_sys_root = proc_mkdir("sys", NULL);
1722         proc_sys_root->proc_iops = &proc_sys_dir_operations;
1723         proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1724         proc_sys_root->nlink = 0;
1725 
1726         return sysctl_init();
1727 }
/* [<][>][^][v][top][bottom][index][help] */
root/fs/proc/proc_sysctl.c

DEFINITIONS
