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