fs/proc/proc_sysctl.c

   1 /*
   2  * /proc/sys support
   3  */
   4 #include <linux/init.h>
   5 #include <linux/sysctl.h>
   6 #include <linux/proc_fs.h>
   7 #include <linux/security.h>
   8 #include <linux/namei.h>
   9 #include "internal.h"
  10
  11 static const struct dentry_operations proc_sys_dentry_operations;
  12 static const struct file_operations proc_sys_file_operations;
  13 static const struct inode_operations proc_sys_inode_operations;
  14 static const struct file_operations proc_sys_dir_file_operations;
  15 static const struct inode_operations proc_sys_dir_operations;
  16
  17 static struct inode *proc_sys_make_inode(struct super_block *sb,
  18                 struct ctl_table_header *head, struct ctl_table *table)
  19 {
  20         struct inode *inode;
  21         struct proc_inode *ei;
  22
  23         inode = new_inode(sb);
  24         if (!inode)
  25                 goto out;
  26
  27         inode->i_ino = get_next_ino();
  28
  29         sysctl_head_get(head);
  30         ei = PROC_I(inode);
  31         ei->sysctl = head;
  32         ei->sysctl_entry = table;
  33
  34         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  35         inode->i_mode = table->mode;
  36         if (!table->child) {
  37                 inode->i_mode |= S_IFREG;
  38                 inode->i_op = &proc_sys_inode_operations;
  39                 inode->i_fop = &proc_sys_file_operations;
  40         } else {
  41                 inode->i_mode |= S_IFDIR;
  42                 inode->i_nlink = 0;
  43                 inode->i_op = &proc_sys_dir_operations;
  44                 inode->i_fop = &proc_sys_dir_file_operations;
  45         }
  46 out:
  47         return inode;
  48 }
  49
  50 static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name)
  51 {
  52         for ( ; p->procname; p++) {
  53                 if (strlen(p->procname) != name->len)
  54                         continue;
  55
  56                 if (memcmp(p->procname, name->name, name->len) != 0)
  57                         continue;
  58
  59                 /* I have a match */
  60                 return p;
  61         }
  62         return NULL;
  63 }
  64
  65 static struct ctl_table_header *grab_header(struct inode *inode)
  66 {
  67         if (PROC_I(inode)->sysctl)
  68                 return sysctl_head_grab(PROC_I(inode)->sysctl);
  69         else
  70                 return sysctl_head_next(NULL);
  71 }
  72
  73 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
  74                                         struct nameidata *nd)
  75 {
  76         struct ctl_table_header *head = grab_header(dir);
  77         struct ctl_table *table = PROC_I(dir)->sysctl_entry;
  78         struct ctl_table_header *h = NULL;
  79         struct qstr *name = &dentry->d_name;
  80         struct ctl_table *p;
  81         struct inode *inode;
  82         struct dentry *err = ERR_PTR(-ENOENT);
  83
  84         if (IS_ERR(head))
  85                 return ERR_CAST(head);
  86
  87         if (table && !table->child) {
  88                 WARN_ON(1);
  89                 goto out;
  90         }
  91
  92         table = table ? table->child : head->ctl_table;
  93
  94         p = find_in_table(table, name);
  95         if (!p) {
  96                 for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
  97                         if (h->attached_to != table)
  98                                 continue;
  99                         p = find_in_table(h->attached_by, name);
 100                         if (p)
 101                                 break;
 102                 }
 103         }
 104
 105         if (!p)
 106                 goto out;
 107
 108         err = ERR_PTR(-ENOMEM);
 109         inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
 110         if (h)
 111                 sysctl_head_finish(h);
 112
 113         if (!inode)
 114                 goto out;
 115
 116         err = NULL;
 117         d_set_d_op(dentry, &proc_sys_dentry_operations);
 118         d_add(dentry, inode);
 119
 120 out:
 121         sysctl_head_finish(head);
 122         return err;
 123 }
 124
 125 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
 126                 size_t count, loff_t *ppos, int write)
 127 {
 128         struct inode *inode = filp->f_path.dentry->d_inode;
 129         struct ctl_table_header *head = grab_header(inode);
 130         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 131         ssize_t error;
 132         size_t res;
 133
 134         if (IS_ERR(head))
 135                 return PTR_ERR(head);
 136
 137         /*
 138          * At this point we know that the sysctl was not unregistered
 139          * and won't be until we finish.
 140          */
 141         error = -EPERM;
 142         if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
 143                 goto out;
 144
 145         /* if that can happen at all, it should be -EINVAL, not -EISDIR */
 146         error = -EINVAL;
 147         if (!table->proc_handler)
 148                 goto out;
 149
 150         /* careful: calling conventions are nasty here */
 151         res = count;
 152         error = table->proc_handler(table, write, buf, &res, ppos);
 153         if (!error)
 154                 error = res;
 155 out:
 156         sysctl_head_finish(head);
 157
 158         return error;
 159 }
 160
 161 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
 162                                 size_t count, loff_t *ppos)
 163 {
 164         return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
 165 }
 166
 167 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
 168                                 size_t count, loff_t *ppos)
 169 {
 170         return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
 171 }
 172
 173
 174 static int proc_sys_fill_cache(struct file *filp, void *dirent,
 175                                 filldir_t filldir,
 176                                 struct ctl_table_header *head,
 177                                 struct ctl_table *table)
 178 {
 179         struct dentry *child, *dir = filp->f_path.dentry;
 180         struct inode *inode;
 181         struct qstr qname;
 182         ino_t ino = 0;
 183         unsigned type = DT_UNKNOWN;
 184
 185         qname.name = table->procname;
 186         qname.len  = strlen(table->procname);
 187         qname.hash = full_name_hash(qname.name, qname.len);
 188
 189         child = d_lookup(dir, &qname);
 190         if (!child) {
 191                 child = d_alloc(dir, &qname);
 192                 if (child) {
 193                         inode = proc_sys_make_inode(dir->d_sb, head, table);
 194                         if (!inode) {
 195                                 dput(child);
 196                                 return -ENOMEM;
 197                         } else {
 198                                 d_set_d_op(child, &proc_sys_dentry_operations);
 199                                 d_add(child, inode);
 200                         }
 201                 } else {
 202                         return -ENOMEM;
 203                 }
 204         }
 205         inode = child->d_inode;
 206         ino  = inode->i_ino;
 207         type = inode->i_mode >> 12;
 208         dput(child);
 209         return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
 210 }
 211
 212 static int scan(struct ctl_table_header *head, ctl_table *table,
 213                 unsigned long *pos, struct file *file,
 214                 void *dirent, filldir_t filldir)
 215 {
 216
 217         for (; table->procname; table++, (*pos)++) {
 218                 int res;
 219
 220                 if (*pos < file->f_pos)
 221                         continue;
 222
 223                 res = proc_sys_fill_cache(file, dirent, filldir, head, table);
 224                 if (res)
 225                         return res;
 226
 227                 file->f_pos = *pos + 1;
 228         }
 229         return 0;
 230 }
 231
 232 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
 233 {
 234         struct dentry *dentry = filp->f_path.dentry;
 235         struct inode *inode = dentry->d_inode;
 236         struct ctl_table_header *head = grab_header(inode);
 237         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 238         struct ctl_table_header *h = NULL;
 239         unsigned long pos;
 240         int ret = -EINVAL;
 241
 242         if (IS_ERR(head))
 243                 return PTR_ERR(head);
 244
 245         if (table && !table->child) {
 246                 WARN_ON(1);
 247                 goto out;
 248         }
 249
 250         table = table ? table->child : head->ctl_table;
 251
 252         ret = 0;
 253         /* Avoid a switch here: arm builds fail with missing __cmpdi2 */
 254         if (filp->f_pos == 0) {
 255                 if (filldir(dirent, ".", 1, filp->f_pos,
 256                                 inode->i_ino, DT_DIR) < 0)
 257                         goto out;
 258                 filp->f_pos++;
 259         }
 260         if (filp->f_pos == 1) {
 261                 if (filldir(dirent, "..", 2, filp->f_pos,
 262                                 parent_ino(dentry), DT_DIR) < 0)
 263                         goto out;
 264                 filp->f_pos++;
 265         }
 266         pos = 2;
 267
 268         ret = scan(head, table, &pos, filp, dirent, filldir);
 269         if (ret)
 270                 goto out;
 271
 272         for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
 273                 if (h->attached_to != table)
 274                         continue;
 275                 ret = scan(h, h->attached_by, &pos, filp, dirent, filldir);
 276                 if (ret) {
 277                         sysctl_head_finish(h);
 278                         break;
 279                 }
 280         }
 281         ret = 1;
 282 out:
 283         sysctl_head_finish(head);
 284         return ret;
 285 }
 286
 287 static int proc_sys_permission(struct inode *inode, int mask)
 288 {
 289         /*
 290          * sysctl entries that are not writeable,
 291          * are _NOT_ writeable, capabilities or not.
 292          */
 293         struct ctl_table_header *head;
 294         struct ctl_table *table;
 295         int error;
 296
 297         /* Executable files are not allowed under /proc/sys/ */
 298         if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
 299                 return -EACCES;
 300
 301         head = grab_header(inode);
 302         if (IS_ERR(head))
 303                 return PTR_ERR(head);
 304
 305         table = PROC_I(inode)->sysctl_entry;
 306         if (!table) /* global root - r-xr-xr-x */
 307                 error = mask & MAY_WRITE ? -EACCES : 0;
 308         else /* Use the permissions on the sysctl table entry */
 309                 error = sysctl_perm(head->root, table, mask & ~MAY_NOT_BLOCK);
 310
 311         sysctl_head_finish(head);
 312         return error;
 313 }
 314
 315 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
 316 {
 317         struct inode *inode = dentry->d_inode;
 318         int error;
 319
 320         if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
 321                 return -EPERM;
 322
 323         error = inode_change_ok(inode, attr);
 324         if (error)
 325                 return error;
 326
 327         if ((attr->ia_valid & ATTR_SIZE) &&
 328             attr->ia_size != i_size_read(inode)) {
 329                 error = vmtruncate(inode, attr->ia_size);
 330                 if (error)
 331                         return error;
 332         }
 333
 334         setattr_copy(inode, attr);
 335         mark_inode_dirty(inode);
 336         return 0;
 337 }
 338
 339 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 340 {
 341         struct inode *inode = dentry->d_inode;
 342         struct ctl_table_header *head = grab_header(inode);
 343         struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 344
 345         if (IS_ERR(head))
 346                 return PTR_ERR(head);
 347
 348         generic_fillattr(inode, stat);
 349         if (table)
 350                 stat->mode = (stat->mode & S_IFMT) | table->mode;
 351
 352         sysctl_head_finish(head);
 353         return 0;
 354 }
 355
 356 static const struct file_operations proc_sys_file_operations = {
 357         .read           = proc_sys_read,
 358         .write          = proc_sys_write,
 359         .llseek         = default_llseek,
 360 };
 361
 362 static const struct file_operations proc_sys_dir_file_operations = {
 363         .read           = generic_read_dir,
 364         .readdir        = proc_sys_readdir,
 365         .llseek         = generic_file_llseek,
 366 };
 367
 368 static const struct inode_operations proc_sys_inode_operations = {
 369         .permission     = proc_sys_permission,
 370         .setattr        = proc_sys_setattr,
 371         .getattr        = proc_sys_getattr,
 372 };
 373
 374 static const struct inode_operations proc_sys_dir_operations = {
 375         .lookup         = proc_sys_lookup,
 376         .permission     = proc_sys_permission,
 377         .setattr        = proc_sys_setattr,
 378         .getattr        = proc_sys_getattr,
 379 };
 380
 381 static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
 382 {
 383         if (nd->flags & LOOKUP_RCU)
 384                 return -ECHILD;
 385         return !PROC_I(dentry->d_inode)->sysctl->unregistering;
 386 }
 387
 388 static int proc_sys_delete(const struct dentry *dentry)
 389 {
 390         return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
 391 }
 392
 393 static int proc_sys_compare(const struct dentry *parent,
 394                 const struct inode *pinode,
 395                 const struct dentry *dentry, const struct inode *inode,
 396                 unsigned int len, const char *str, const struct qstr *name)
 397 {
 398         struct ctl_table_header *head;
 399         /* Although proc doesn't have negative dentries, rcu-walk means
 400          * that inode here can be NULL */
 401         /* AV: can it, indeed? */
 402         if (!inode)
 403                 return 1;
 404         if (name->len != len)
 405                 return 1;
 406         if (memcmp(name->name, str, len))
 407                 return 1;
 408         head = rcu_dereference(PROC_I(inode)->sysctl);
 409         return !head || !sysctl_is_seen(head);
 410 }
 411
 412 static const struct dentry_operations proc_sys_dentry_operations = {
 413         .d_revalidate   = proc_sys_revalidate,
 414         .d_delete       = proc_sys_delete,
 415         .d_compare      = proc_sys_compare,
 416 };
 417
 418 int __init proc_sys_init(void)
 419 {
 420         struct proc_dir_entry *proc_sys_root;
 421
 422         proc_sys_root = proc_mkdir("sys", NULL);
 423         proc_sys_root->proc_iops = &proc_sys_dir_operations;
 424         proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
 425         proc_sys_root->nlink = 0;
 426         return 0;
 427 }