1 /* 2 * /proc/sys support 3 */ 4 #include <linux/init.h> 5 #include <linux/sysctl.h> 6 #include <linux/poll.h> 7 #include <linux/proc_fs.h> 8 #include <linux/security.h> 9 #include <linux/namei.h> 10 #include "internal.h" 11 12 static const struct dentry_operations proc_sys_dentry_operations; 13 static const struct file_operations proc_sys_file_operations; 14 static const struct inode_operations proc_sys_inode_operations; 15 static const struct file_operations proc_sys_dir_file_operations; 16 static const struct inode_operations proc_sys_dir_operations; 17 18 void proc_sys_poll_notify(struct ctl_table_poll *poll) 19 { 20 if (!poll) 21 return; 22 23 atomic_inc(&poll->event); 24 wake_up_interruptible(&poll->wait); 25 } 26 27 static struct inode *proc_sys_make_inode(struct super_block *sb, 28 struct ctl_table_header *head, struct ctl_table *table) 29 { 30 struct inode *inode; 31 struct proc_inode *ei; 32 33 inode = new_inode(sb); 34 if (!inode) 35 goto out; 36 37 inode->i_ino = get_next_ino(); 38 39 sysctl_head_get(head); 40 ei = PROC_I(inode); 41 ei->sysctl = head; 42 ei->sysctl_entry = table; 43 44 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 45 inode->i_mode = table->mode; 46 if (!table->child) { 47 inode->i_mode |= S_IFREG; 48 inode->i_op = &proc_sys_inode_operations; 49 inode->i_fop = &proc_sys_file_operations; 50 } else { 51 inode->i_mode |= S_IFDIR; 52 clear_nlink(inode); 53 inode->i_op = &proc_sys_dir_operations; 54 inode->i_fop = &proc_sys_dir_file_operations; 55 } 56 out: 57 return inode; 58 } 59 60 static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name) 61 { 62 int len; 63 for ( ; p->procname; p++) { 64 65 if (!p->procname) 66 continue; 67 68 len = strlen(p->procname); 69 if (len != name->len) 70 continue; 71 72 if (memcmp(p->procname, name->name, len) != 0) 73 continue; 74 75 /* I have a match */ 76 return p; 77 } 78 return NULL; 79 } 80 81 static struct ctl_table_header *grab_header(struct inode *inode) 82 { 83 if (PROC_I(inode)->sysctl) 84 return sysctl_head_grab(PROC_I(inode)->sysctl); 85 else 86 return sysctl_head_next(NULL); 87 } 88 89 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry, 90 struct nameidata *nd) 91 { 92 struct ctl_table_header *head = grab_header(dir); 93 struct ctl_table *table = PROC_I(dir)->sysctl_entry; 94 struct ctl_table_header *h = NULL; 95 struct qstr *name = &dentry->d_name; 96 struct ctl_table *p; 97 struct inode *inode; 98 struct dentry *err = ERR_PTR(-ENOENT); 99 100 if (IS_ERR(head)) 101 return ERR_CAST(head); 102 103 if (table && !table->child) { 104 WARN_ON(1); 105 goto out; 106 } 107 108 table = table ? table->child : head->ctl_table; 109 110 p = find_in_table(table, name); 111 if (!p) { 112 for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) { 113 if (h->attached_to != table) 114 continue; 115 p = find_in_table(h->attached_by, name); 116 if (p) 117 break; 118 } 119 } 120 121 if (!p) 122 goto out; 123 124 err = ERR_PTR(-ENOMEM); 125 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p); 126 if (h) 127 sysctl_head_finish(h); 128 129 if (!inode) 130 goto out; 131 132 err = NULL; 133 d_set_d_op(dentry, &proc_sys_dentry_operations); 134 d_add(dentry, inode); 135 136 out: 137 sysctl_head_finish(head); 138 return err; 139 } 140 141 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf, 142 size_t count, loff_t *ppos, int write) 143 { 144 struct inode *inode = filp->f_path.dentry->d_inode; 145 struct ctl_table_header *head = grab_header(inode); 146 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 147 ssize_t error; 148 size_t res; 149 150 if (IS_ERR(head)) 151 return PTR_ERR(head); 152 153 /* 154 * At this point we know that the sysctl was not unregistered 155 * and won't be until we finish. 156 */ 157 error = -EPERM; 158 if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ)) 159 goto out; 160 161 /* if that can happen at all, it should be -EINVAL, not -EISDIR */ 162 error = -EINVAL; 163 if (!table->proc_handler) 164 goto out; 165 166 /* careful: calling conventions are nasty here */ 167 res = count; 168 error = table->proc_handler(table, write, buf, &res, ppos); 169 if (!error) 170 error = res; 171 out: 172 sysctl_head_finish(head); 173 174 return error; 175 } 176 177 static ssize_t proc_sys_read(struct file *filp, char __user *buf, 178 size_t count, loff_t *ppos) 179 { 180 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0); 181 } 182 183 static ssize_t proc_sys_write(struct file *filp, const char __user *buf, 184 size_t count, loff_t *ppos) 185 { 186 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1); 187 } 188 189 static int proc_sys_open(struct inode *inode, struct file *filp) 190 { 191 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 192 193 if (table->poll) 194 filp->private_data = proc_sys_poll_event(table->poll); 195 196 return 0; 197 } 198 199 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait) 200 { 201 struct inode *inode = filp->f_path.dentry->d_inode; 202 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 203 unsigned long event = (unsigned long)filp->private_data; 204 unsigned int ret = DEFAULT_POLLMASK; 205 206 if (!table->proc_handler) 207 goto out; 208 209 if (!table->poll) 210 goto out; 211 212 poll_wait(filp, &table->poll->wait, wait); 213 214 if (event != atomic_read(&table->poll->event)) { 215 filp->private_data = proc_sys_poll_event(table->poll); 216 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI; 217 } 218 219 out: 220 return ret; 221 } 222 223 static int proc_sys_fill_cache(struct file *filp, void *dirent, 224 filldir_t filldir, 225 struct ctl_table_header *head, 226 struct ctl_table *table) 227 { 228 struct dentry *child, *dir = filp->f_path.dentry; 229 struct inode *inode; 230 struct qstr qname; 231 ino_t ino = 0; 232 unsigned type = DT_UNKNOWN; 233 234 qname.name = table->procname; 235 qname.len = strlen(table->procname); 236 qname.hash = full_name_hash(qname.name, qname.len); 237 238 child = d_lookup(dir, &qname); 239 if (!child) { 240 child = d_alloc(dir, &qname); 241 if (child) { 242 inode = proc_sys_make_inode(dir->d_sb, head, table); 243 if (!inode) { 244 dput(child); 245 return -ENOMEM; 246 } else { 247 d_set_d_op(child, &proc_sys_dentry_operations); 248 d_add(child, inode); 249 } 250 } else { 251 return -ENOMEM; 252 } 253 } 254 inode = child->d_inode; 255 ino = inode->i_ino; 256 type = inode->i_mode >> 12; 257 dput(child); 258 return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type); 259 } 260 261 static int scan(struct ctl_table_header *head, ctl_table *table, 262 unsigned long *pos, struct file *file, 263 void *dirent, filldir_t filldir) 264 { 265 266 for (; table->procname; table++, (*pos)++) { 267 int res; 268 269 /* Can't do anything without a proc name */ 270 if (!table->procname) 271 continue; 272 273 if (*pos < file->f_pos) 274 continue; 275 276 res = proc_sys_fill_cache(file, dirent, filldir, head, table); 277 if (res) 278 return res; 279 280 file->f_pos = *pos + 1; 281 } 282 return 0; 283 } 284 285 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir) 286 { 287 struct dentry *dentry = filp->f_path.dentry; 288 struct inode *inode = dentry->d_inode; 289 struct ctl_table_header *head = grab_header(inode); 290 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 291 struct ctl_table_header *h = NULL; 292 unsigned long pos; 293 int ret = -EINVAL; 294 295 if (IS_ERR(head)) 296 return PTR_ERR(head); 297 298 if (table && !table->child) { 299 WARN_ON(1); 300 goto out; 301 } 302 303 table = table ? table->child : head->ctl_table; 304 305 ret = 0; 306 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */ 307 if (filp->f_pos == 0) { 308 if (filldir(dirent, ".", 1, filp->f_pos, 309 inode->i_ino, DT_DIR) < 0) 310 goto out; 311 filp->f_pos++; 312 } 313 if (filp->f_pos == 1) { 314 if (filldir(dirent, "..", 2, filp->f_pos, 315 parent_ino(dentry), DT_DIR) < 0) 316 goto out; 317 filp->f_pos++; 318 } 319 pos = 2; 320 321 ret = scan(head, table, &pos, filp, dirent, filldir); 322 if (ret) 323 goto out; 324 325 for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) { 326 if (h->attached_to != table) 327 continue; 328 ret = scan(h, h->attached_by, &pos, filp, dirent, filldir); 329 if (ret) { 330 sysctl_head_finish(h); 331 break; 332 } 333 } 334 ret = 1; 335 out: 336 sysctl_head_finish(head); 337 return ret; 338 } 339 340 static int proc_sys_permission(struct inode *inode, int mask) 341 { 342 /* 343 * sysctl entries that are not writeable, 344 * are _NOT_ writeable, capabilities or not. 345 */ 346 struct ctl_table_header *head; 347 struct ctl_table *table; 348 int error; 349 350 /* Executable files are not allowed under /proc/sys/ */ 351 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) 352 return -EACCES; 353 354 head = grab_header(inode); 355 if (IS_ERR(head)) 356 return PTR_ERR(head); 357 358 table = PROC_I(inode)->sysctl_entry; 359 if (!table) /* global root - r-xr-xr-x */ 360 error = mask & MAY_WRITE ? -EACCES : 0; 361 else /* Use the permissions on the sysctl table entry */ 362 error = sysctl_perm(head->root, table, mask & ~MAY_NOT_BLOCK); 363 364 sysctl_head_finish(head); 365 return error; 366 } 367 368 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr) 369 { 370 struct inode *inode = dentry->d_inode; 371 int error; 372 373 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID)) 374 return -EPERM; 375 376 error = inode_change_ok(inode, attr); 377 if (error) 378 return error; 379 380 if ((attr->ia_valid & ATTR_SIZE) && 381 attr->ia_size != i_size_read(inode)) { 382 error = vmtruncate(inode, attr->ia_size); 383 if (error) 384 return error; 385 } 386 387 setattr_copy(inode, attr); 388 mark_inode_dirty(inode); 389 return 0; 390 } 391 392 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 393 { 394 struct inode *inode = dentry->d_inode; 395 struct ctl_table_header *head = grab_header(inode); 396 struct ctl_table *table = PROC_I(inode)->sysctl_entry; 397 398 if (IS_ERR(head)) 399 return PTR_ERR(head); 400 401 generic_fillattr(inode, stat); 402 if (table) 403 stat->mode = (stat->mode & S_IFMT) | table->mode; 404 405 sysctl_head_finish(head); 406 return 0; 407 } 408 409 static const struct file_operations proc_sys_file_operations = { 410 .open = proc_sys_open, 411 .poll = proc_sys_poll, 412 .read = proc_sys_read, 413 .write = proc_sys_write, 414 .llseek = default_llseek, 415 }; 416 417 static const struct file_operations proc_sys_dir_file_operations = { 418 .read = generic_read_dir, 419 .readdir = proc_sys_readdir, 420 .llseek = generic_file_llseek, 421 }; 422 423 static const struct inode_operations proc_sys_inode_operations = { 424 .permission = proc_sys_permission, 425 .setattr = proc_sys_setattr, 426 .getattr = proc_sys_getattr, 427 }; 428 429 static const struct inode_operations proc_sys_dir_operations = { 430 .lookup = proc_sys_lookup, 431 .permission = proc_sys_permission, 432 .setattr = proc_sys_setattr, 433 .getattr = proc_sys_getattr, 434 }; 435 436 static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd) 437 { 438 if (nd->flags & LOOKUP_RCU) 439 return -ECHILD; 440 return !PROC_I(dentry->d_inode)->sysctl->unregistering; 441 } 442 443 static int proc_sys_delete(const struct dentry *dentry) 444 { 445 return !!PROC_I(dentry->d_inode)->sysctl->unregistering; 446 } 447 448 static int proc_sys_compare(const struct dentry *parent, 449 const struct inode *pinode, 450 const struct dentry *dentry, const struct inode *inode, 451 unsigned int len, const char *str, const struct qstr *name) 452 { 453 struct ctl_table_header *head; 454 /* Although proc doesn't have negative dentries, rcu-walk means 455 * that inode here can be NULL */ 456 /* AV: can it, indeed? */ 457 if (!inode) 458 return 1; 459 if (name->len != len) 460 return 1; 461 if (memcmp(name->name, str, len)) 462 return 1; 463 head = rcu_dereference(PROC_I(inode)->sysctl); 464 return !head || !sysctl_is_seen(head); 465 } 466 467 static const struct dentry_operations proc_sys_dentry_operations = { 468 .d_revalidate = proc_sys_revalidate, 469 .d_delete = proc_sys_delete, 470 .d_compare = proc_sys_compare, 471 }; 472 473 int __init proc_sys_init(void) 474 { 475 struct proc_dir_entry *proc_sys_root; 476 477 proc_sys_root = proc_mkdir("sys", NULL); 478 proc_sys_root->proc_iops = &proc_sys_dir_operations; 479 proc_sys_root->proc_fops = &proc_sys_dir_file_operations; 480 proc_sys_root->nlink = 0; 481 return 0; 482 } 483