1 /* 2 * POSIX message queues filesystem for Linux. 3 * 4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl) 5 * Michal Wronski (michal.wronski@gmail.com) 6 * 7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com) 8 * Lockless receive & send, fd based notify: 9 * Manfred Spraul (manfred@colorfullife.com) 10 * 11 * Audit: George Wilson (ltcgcw@us.ibm.com) 12 * 13 * This file is released under the GPL. 14 */ 15 16 #include <linux/capability.h> 17 #include <linux/init.h> 18 #include <linux/pagemap.h> 19 #include <linux/file.h> 20 #include <linux/mount.h> 21 #include <linux/namei.h> 22 #include <linux/sysctl.h> 23 #include <linux/poll.h> 24 #include <linux/mqueue.h> 25 #include <linux/msg.h> 26 #include <linux/skbuff.h> 27 #include <linux/vmalloc.h> 28 #include <linux/netlink.h> 29 #include <linux/syscalls.h> 30 #include <linux/audit.h> 31 #include <linux/signal.h> 32 #include <linux/mutex.h> 33 #include <linux/nsproxy.h> 34 #include <linux/pid.h> 35 #include <linux/ipc_namespace.h> 36 #include <linux/user_namespace.h> 37 #include <linux/slab.h> 38 #include <linux/sched/wake_q.h> 39 #include <linux/sched/signal.h> 40 #include <linux/sched/user.h> 41 42 #include <net/sock.h> 43 #include "util.h" 44 45 #define MQUEUE_MAGIC 0x19800202 46 #define DIRENT_SIZE 20 47 #define FILENT_SIZE 80 48 49 #define SEND 0 50 #define RECV 1 51 52 #define STATE_NONE 0 53 #define STATE_READY 1 54 55 struct posix_msg_tree_node { 56 struct rb_node rb_node; 57 struct list_head msg_list; 58 int priority; 59 }; 60 61 struct ext_wait_queue { /* queue of sleeping tasks */ 62 struct task_struct *task; 63 struct list_head list; 64 struct msg_msg *msg; /* ptr of loaded message */ 65 int state; /* one of STATE_* values */ 66 }; 67 68 struct mqueue_inode_info { 69 spinlock_t lock; 70 struct inode vfs_inode; 71 wait_queue_head_t wait_q; 72 73 struct rb_root msg_tree; 74 struct posix_msg_tree_node *node_cache; 75 struct mq_attr attr; 76 77 struct sigevent notify; 78 struct pid *notify_owner; 79 struct user_namespace *notify_user_ns; 80 struct user_struct *user; /* user who created, for accounting */ 81 struct sock *notify_sock; 82 struct sk_buff *notify_cookie; 83 84 /* for tasks waiting for free space and messages, respectively */ 85 struct ext_wait_queue e_wait_q[2]; 86 87 unsigned long qsize; /* size of queue in memory (sum of all msgs) */ 88 }; 89 90 static const struct inode_operations mqueue_dir_inode_operations; 91 static const struct file_operations mqueue_file_operations; 92 static const struct super_operations mqueue_super_ops; 93 static void remove_notification(struct mqueue_inode_info *info); 94 95 static struct kmem_cache *mqueue_inode_cachep; 96 97 static struct ctl_table_header *mq_sysctl_table; 98 99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode) 100 { 101 return container_of(inode, struct mqueue_inode_info, vfs_inode); 102 } 103 104 /* 105 * This routine should be called with the mq_lock held. 106 */ 107 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode) 108 { 109 return get_ipc_ns(inode->i_sb->s_fs_info); 110 } 111 112 static struct ipc_namespace *get_ns_from_inode(struct inode *inode) 113 { 114 struct ipc_namespace *ns; 115 116 spin_lock(&mq_lock); 117 ns = __get_ns_from_inode(inode); 118 spin_unlock(&mq_lock); 119 return ns; 120 } 121 122 /* Auxiliary functions to manipulate messages' list */ 123 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info) 124 { 125 struct rb_node **p, *parent = NULL; 126 struct posix_msg_tree_node *leaf; 127 128 p = &info->msg_tree.rb_node; 129 while (*p) { 130 parent = *p; 131 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); 132 133 if (likely(leaf->priority == msg->m_type)) 134 goto insert_msg; 135 else if (msg->m_type < leaf->priority) 136 p = &(*p)->rb_left; 137 else 138 p = &(*p)->rb_right; 139 } 140 if (info->node_cache) { 141 leaf = info->node_cache; 142 info->node_cache = NULL; 143 } else { 144 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC); 145 if (!leaf) 146 return -ENOMEM; 147 INIT_LIST_HEAD(&leaf->msg_list); 148 } 149 leaf->priority = msg->m_type; 150 rb_link_node(&leaf->rb_node, parent, p); 151 rb_insert_color(&leaf->rb_node, &info->msg_tree); 152 insert_msg: 153 info->attr.mq_curmsgs++; 154 info->qsize += msg->m_ts; 155 list_add_tail(&msg->m_list, &leaf->msg_list); 156 return 0; 157 } 158 159 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info) 160 { 161 struct rb_node **p, *parent = NULL; 162 struct posix_msg_tree_node *leaf; 163 struct msg_msg *msg; 164 165 try_again: 166 p = &info->msg_tree.rb_node; 167 while (*p) { 168 parent = *p; 169 /* 170 * During insert, low priorities go to the left and high to the 171 * right. On receive, we want the highest priorities first, so 172 * walk all the way to the right. 173 */ 174 p = &(*p)->rb_right; 175 } 176 if (!parent) { 177 if (info->attr.mq_curmsgs) { 178 pr_warn_once("Inconsistency in POSIX message queue, " 179 "no tree element, but supposedly messages " 180 "should exist!\n"); 181 info->attr.mq_curmsgs = 0; 182 } 183 return NULL; 184 } 185 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); 186 if (unlikely(list_empty(&leaf->msg_list))) { 187 pr_warn_once("Inconsistency in POSIX message queue, " 188 "empty leaf node but we haven't implemented " 189 "lazy leaf delete!\n"); 190 rb_erase(&leaf->rb_node, &info->msg_tree); 191 if (info->node_cache) { 192 kfree(leaf); 193 } else { 194 info->node_cache = leaf; 195 } 196 goto try_again; 197 } else { 198 msg = list_first_entry(&leaf->msg_list, 199 struct msg_msg, m_list); 200 list_del(&msg->m_list); 201 if (list_empty(&leaf->msg_list)) { 202 rb_erase(&leaf->rb_node, &info->msg_tree); 203 if (info->node_cache) { 204 kfree(leaf); 205 } else { 206 info->node_cache = leaf; 207 } 208 } 209 } 210 info->attr.mq_curmsgs--; 211 info->qsize -= msg->m_ts; 212 return msg; 213 } 214 215 static struct inode *mqueue_get_inode(struct super_block *sb, 216 struct ipc_namespace *ipc_ns, umode_t mode, 217 struct mq_attr *attr) 218 { 219 struct user_struct *u = current_user(); 220 struct inode *inode; 221 int ret = -ENOMEM; 222 223 inode = new_inode(sb); 224 if (!inode) 225 goto err; 226 227 inode->i_ino = get_next_ino(); 228 inode->i_mode = mode; 229 inode->i_uid = current_fsuid(); 230 inode->i_gid = current_fsgid(); 231 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode); 232 233 if (S_ISREG(mode)) { 234 struct mqueue_inode_info *info; 235 unsigned long mq_bytes, mq_treesize; 236 237 inode->i_fop = &mqueue_file_operations; 238 inode->i_size = FILENT_SIZE; 239 /* mqueue specific info */ 240 info = MQUEUE_I(inode); 241 spin_lock_init(&info->lock); 242 init_waitqueue_head(&info->wait_q); 243 INIT_LIST_HEAD(&info->e_wait_q[0].list); 244 INIT_LIST_HEAD(&info->e_wait_q[1].list); 245 info->notify_owner = NULL; 246 info->notify_user_ns = NULL; 247 info->qsize = 0; 248 info->user = NULL; /* set when all is ok */ 249 info->msg_tree = RB_ROOT; 250 info->node_cache = NULL; 251 memset(&info->attr, 0, sizeof(info->attr)); 252 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max, 253 ipc_ns->mq_msg_default); 254 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max, 255 ipc_ns->mq_msgsize_default); 256 if (attr) { 257 info->attr.mq_maxmsg = attr->mq_maxmsg; 258 info->attr.mq_msgsize = attr->mq_msgsize; 259 } 260 /* 261 * We used to allocate a static array of pointers and account 262 * the size of that array as well as one msg_msg struct per 263 * possible message into the queue size. That's no longer 264 * accurate as the queue is now an rbtree and will grow and 265 * shrink depending on usage patterns. We can, however, still 266 * account one msg_msg struct per message, but the nodes are 267 * allocated depending on priority usage, and most programs 268 * only use one, or a handful, of priorities. However, since 269 * this is pinned memory, we need to assume worst case, so 270 * that means the min(mq_maxmsg, max_priorities) * struct 271 * posix_msg_tree_node. 272 */ 273 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + 274 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * 275 sizeof(struct posix_msg_tree_node); 276 277 mq_bytes = mq_treesize + (info->attr.mq_maxmsg * 278 info->attr.mq_msgsize); 279 280 spin_lock(&mq_lock); 281 if (u->mq_bytes + mq_bytes < u->mq_bytes || 282 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) { 283 spin_unlock(&mq_lock); 284 /* mqueue_evict_inode() releases info->messages */ 285 ret = -EMFILE; 286 goto out_inode; 287 } 288 u->mq_bytes += mq_bytes; 289 spin_unlock(&mq_lock); 290 291 /* all is ok */ 292 info->user = get_uid(u); 293 } else if (S_ISDIR(mode)) { 294 inc_nlink(inode); 295 /* Some things misbehave if size == 0 on a directory */ 296 inode->i_size = 2 * DIRENT_SIZE; 297 inode->i_op = &mqueue_dir_inode_operations; 298 inode->i_fop = &simple_dir_operations; 299 } 300 301 return inode; 302 out_inode: 303 iput(inode); 304 err: 305 return ERR_PTR(ret); 306 } 307 308 static int mqueue_fill_super(struct super_block *sb, void *data, int silent) 309 { 310 struct inode *inode; 311 struct ipc_namespace *ns = sb->s_fs_info; 312 313 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV; 314 sb->s_blocksize = PAGE_SIZE; 315 sb->s_blocksize_bits = PAGE_SHIFT; 316 sb->s_magic = MQUEUE_MAGIC; 317 sb->s_op = &mqueue_super_ops; 318 319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL); 320 if (IS_ERR(inode)) 321 return PTR_ERR(inode); 322 323 sb->s_root = d_make_root(inode); 324 if (!sb->s_root) 325 return -ENOMEM; 326 return 0; 327 } 328 329 static struct dentry *mqueue_mount(struct file_system_type *fs_type, 330 int flags, const char *dev_name, 331 void *data) 332 { 333 struct ipc_namespace *ns; 334 if (flags & MS_KERNMOUNT) { 335 ns = data; 336 data = NULL; 337 } else { 338 ns = current->nsproxy->ipc_ns; 339 } 340 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super); 341 } 342 343 static void init_once(void *foo) 344 { 345 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo; 346 347 inode_init_once(&p->vfs_inode); 348 } 349 350 static struct inode *mqueue_alloc_inode(struct super_block *sb) 351 { 352 struct mqueue_inode_info *ei; 353 354 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL); 355 if (!ei) 356 return NULL; 357 return &ei->vfs_inode; 358 } 359 360 static void mqueue_i_callback(struct rcu_head *head) 361 { 362 struct inode *inode = container_of(head, struct inode, i_rcu); 363 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode)); 364 } 365 366 static void mqueue_destroy_inode(struct inode *inode) 367 { 368 call_rcu(&inode->i_rcu, mqueue_i_callback); 369 } 370 371 static void mqueue_evict_inode(struct inode *inode) 372 { 373 struct mqueue_inode_info *info; 374 struct user_struct *user; 375 unsigned long mq_bytes, mq_treesize; 376 struct ipc_namespace *ipc_ns; 377 struct msg_msg *msg; 378 379 clear_inode(inode); 380 381 if (S_ISDIR(inode->i_mode)) 382 return; 383 384 ipc_ns = get_ns_from_inode(inode); 385 info = MQUEUE_I(inode); 386 spin_lock(&info->lock); 387 while ((msg = msg_get(info)) != NULL) 388 free_msg(msg); 389 kfree(info->node_cache); 390 spin_unlock(&info->lock); 391 392 /* Total amount of bytes accounted for the mqueue */ 393 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + 394 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * 395 sizeof(struct posix_msg_tree_node); 396 397 mq_bytes = mq_treesize + (info->attr.mq_maxmsg * 398 info->attr.mq_msgsize); 399 400 user = info->user; 401 if (user) { 402 spin_lock(&mq_lock); 403 user->mq_bytes -= mq_bytes; 404 /* 405 * get_ns_from_inode() ensures that the 406 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns 407 * to which we now hold a reference, or it is NULL. 408 * We can't put it here under mq_lock, though. 409 */ 410 if (ipc_ns) 411 ipc_ns->mq_queues_count--; 412 spin_unlock(&mq_lock); 413 free_uid(user); 414 } 415 if (ipc_ns) 416 put_ipc_ns(ipc_ns); 417 } 418 419 static int mqueue_create(struct inode *dir, struct dentry *dentry, 420 umode_t mode, bool excl) 421 { 422 struct inode *inode; 423 struct mq_attr *attr = dentry->d_fsdata; 424 int error; 425 struct ipc_namespace *ipc_ns; 426 427 spin_lock(&mq_lock); 428 ipc_ns = __get_ns_from_inode(dir); 429 if (!ipc_ns) { 430 error = -EACCES; 431 goto out_unlock; 432 } 433 434 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max && 435 !capable(CAP_SYS_RESOURCE)) { 436 error = -ENOSPC; 437 goto out_unlock; 438 } 439 ipc_ns->mq_queues_count++; 440 spin_unlock(&mq_lock); 441 442 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr); 443 if (IS_ERR(inode)) { 444 error = PTR_ERR(inode); 445 spin_lock(&mq_lock); 446 ipc_ns->mq_queues_count--; 447 goto out_unlock; 448 } 449 450 put_ipc_ns(ipc_ns); 451 dir->i_size += DIRENT_SIZE; 452 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir); 453 454 d_instantiate(dentry, inode); 455 dget(dentry); 456 return 0; 457 out_unlock: 458 spin_unlock(&mq_lock); 459 if (ipc_ns) 460 put_ipc_ns(ipc_ns); 461 return error; 462 } 463 464 static int mqueue_unlink(struct inode *dir, struct dentry *dentry) 465 { 466 struct inode *inode = d_inode(dentry); 467 468 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir); 469 dir->i_size -= DIRENT_SIZE; 470 drop_nlink(inode); 471 dput(dentry); 472 return 0; 473 } 474 475 /* 476 * This is routine for system read from queue file. 477 * To avoid mess with doing here some sort of mq_receive we allow 478 * to read only queue size & notification info (the only values 479 * that are interesting from user point of view and aren't accessible 480 * through std routines) 481 */ 482 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data, 483 size_t count, loff_t *off) 484 { 485 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); 486 char buffer[FILENT_SIZE]; 487 ssize_t ret; 488 489 spin_lock(&info->lock); 490 snprintf(buffer, sizeof(buffer), 491 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n", 492 info->qsize, 493 info->notify_owner ? info->notify.sigev_notify : 0, 494 (info->notify_owner && 495 info->notify.sigev_notify == SIGEV_SIGNAL) ? 496 info->notify.sigev_signo : 0, 497 pid_vnr(info->notify_owner)); 498 spin_unlock(&info->lock); 499 buffer[sizeof(buffer)-1] = '\0'; 500 501 ret = simple_read_from_buffer(u_data, count, off, buffer, 502 strlen(buffer)); 503 if (ret <= 0) 504 return ret; 505 506 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp)); 507 return ret; 508 } 509 510 static int mqueue_flush_file(struct file *filp, fl_owner_t id) 511 { 512 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); 513 514 spin_lock(&info->lock); 515 if (task_tgid(current) == info->notify_owner) 516 remove_notification(info); 517 518 spin_unlock(&info->lock); 519 return 0; 520 } 521 522 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab) 523 { 524 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); 525 int retval = 0; 526 527 poll_wait(filp, &info->wait_q, poll_tab); 528 529 spin_lock(&info->lock); 530 if (info->attr.mq_curmsgs) 531 retval = POLLIN | POLLRDNORM; 532 533 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg) 534 retval |= POLLOUT | POLLWRNORM; 535 spin_unlock(&info->lock); 536 537 return retval; 538 } 539 540 /* Adds current to info->e_wait_q[sr] before element with smaller prio */ 541 static void wq_add(struct mqueue_inode_info *info, int sr, 542 struct ext_wait_queue *ewp) 543 { 544 struct ext_wait_queue *walk; 545 546 ewp->task = current; 547 548 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) { 549 if (walk->task->static_prio <= current->static_prio) { 550 list_add_tail(&ewp->list, &walk->list); 551 return; 552 } 553 } 554 list_add_tail(&ewp->list, &info->e_wait_q[sr].list); 555 } 556 557 /* 558 * Puts current task to sleep. Caller must hold queue lock. After return 559 * lock isn't held. 560 * sr: SEND or RECV 561 */ 562 static int wq_sleep(struct mqueue_inode_info *info, int sr, 563 ktime_t *timeout, struct ext_wait_queue *ewp) 564 __releases(&info->lock) 565 { 566 int retval; 567 signed long time; 568 569 wq_add(info, sr, ewp); 570 571 for (;;) { 572 __set_current_state(TASK_INTERRUPTIBLE); 573 574 spin_unlock(&info->lock); 575 time = schedule_hrtimeout_range_clock(timeout, 0, 576 HRTIMER_MODE_ABS, CLOCK_REALTIME); 577 578 if (ewp->state == STATE_READY) { 579 retval = 0; 580 goto out; 581 } 582 spin_lock(&info->lock); 583 if (ewp->state == STATE_READY) { 584 retval = 0; 585 goto out_unlock; 586 } 587 if (signal_pending(current)) { 588 retval = -ERESTARTSYS; 589 break; 590 } 591 if (time == 0) { 592 retval = -ETIMEDOUT; 593 break; 594 } 595 } 596 list_del(&ewp->list); 597 out_unlock: 598 spin_unlock(&info->lock); 599 out: 600 return retval; 601 } 602 603 /* 604 * Returns waiting task that should be serviced first or NULL if none exists 605 */ 606 static struct ext_wait_queue *wq_get_first_waiter( 607 struct mqueue_inode_info *info, int sr) 608 { 609 struct list_head *ptr; 610 611 ptr = info->e_wait_q[sr].list.prev; 612 if (ptr == &info->e_wait_q[sr].list) 613 return NULL; 614 return list_entry(ptr, struct ext_wait_queue, list); 615 } 616 617 618 static inline void set_cookie(struct sk_buff *skb, char code) 619 { 620 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code; 621 } 622 623 /* 624 * The next function is only to split too long sys_mq_timedsend 625 */ 626 static void __do_notify(struct mqueue_inode_info *info) 627 { 628 /* notification 629 * invoked when there is registered process and there isn't process 630 * waiting synchronously for message AND state of queue changed from 631 * empty to not empty. Here we are sure that no one is waiting 632 * synchronously. */ 633 if (info->notify_owner && 634 info->attr.mq_curmsgs == 1) { 635 struct siginfo sig_i; 636 switch (info->notify.sigev_notify) { 637 case SIGEV_NONE: 638 break; 639 case SIGEV_SIGNAL: 640 /* sends signal */ 641 642 sig_i.si_signo = info->notify.sigev_signo; 643 sig_i.si_errno = 0; 644 sig_i.si_code = SI_MESGQ; 645 sig_i.si_value = info->notify.sigev_value; 646 /* map current pid/uid into info->owner's namespaces */ 647 rcu_read_lock(); 648 sig_i.si_pid = task_tgid_nr_ns(current, 649 ns_of_pid(info->notify_owner)); 650 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid()); 651 rcu_read_unlock(); 652 653 kill_pid_info(info->notify.sigev_signo, 654 &sig_i, info->notify_owner); 655 break; 656 case SIGEV_THREAD: 657 set_cookie(info->notify_cookie, NOTIFY_WOKENUP); 658 netlink_sendskb(info->notify_sock, info->notify_cookie); 659 break; 660 } 661 /* after notification unregisters process */ 662 put_pid(info->notify_owner); 663 put_user_ns(info->notify_user_ns); 664 info->notify_owner = NULL; 665 info->notify_user_ns = NULL; 666 } 667 wake_up(&info->wait_q); 668 } 669 670 static int prepare_timeout(const struct timespec __user *u_abs_timeout, 671 ktime_t *expires, struct timespec *ts) 672 { 673 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec))) 674 return -EFAULT; 675 if (!timespec_valid(ts)) 676 return -EINVAL; 677 678 *expires = timespec_to_ktime(*ts); 679 return 0; 680 } 681 682 static void remove_notification(struct mqueue_inode_info *info) 683 { 684 if (info->notify_owner != NULL && 685 info->notify.sigev_notify == SIGEV_THREAD) { 686 set_cookie(info->notify_cookie, NOTIFY_REMOVED); 687 netlink_sendskb(info->notify_sock, info->notify_cookie); 688 } 689 put_pid(info->notify_owner); 690 put_user_ns(info->notify_user_ns); 691 info->notify_owner = NULL; 692 info->notify_user_ns = NULL; 693 } 694 695 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr) 696 { 697 int mq_treesize; 698 unsigned long total_size; 699 700 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0) 701 return -EINVAL; 702 if (capable(CAP_SYS_RESOURCE)) { 703 if (attr->mq_maxmsg > HARD_MSGMAX || 704 attr->mq_msgsize > HARD_MSGSIZEMAX) 705 return -EINVAL; 706 } else { 707 if (attr->mq_maxmsg > ipc_ns->mq_msg_max || 708 attr->mq_msgsize > ipc_ns->mq_msgsize_max) 709 return -EINVAL; 710 } 711 /* check for overflow */ 712 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg) 713 return -EOVERFLOW; 714 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) + 715 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) * 716 sizeof(struct posix_msg_tree_node); 717 total_size = attr->mq_maxmsg * attr->mq_msgsize; 718 if (total_size + mq_treesize < total_size) 719 return -EOVERFLOW; 720 return 0; 721 } 722 723 /* 724 * Invoked when creating a new queue via sys_mq_open 725 */ 726 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir, 727 struct path *path, int oflag, umode_t mode, 728 struct mq_attr *attr) 729 { 730 const struct cred *cred = current_cred(); 731 int ret; 732 733 if (attr) { 734 ret = mq_attr_ok(ipc_ns, attr); 735 if (ret) 736 return ERR_PTR(ret); 737 /* store for use during create */ 738 path->dentry->d_fsdata = attr; 739 } else { 740 struct mq_attr def_attr; 741 742 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max, 743 ipc_ns->mq_msg_default); 744 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max, 745 ipc_ns->mq_msgsize_default); 746 ret = mq_attr_ok(ipc_ns, &def_attr); 747 if (ret) 748 return ERR_PTR(ret); 749 } 750 751 mode &= ~current_umask(); 752 ret = vfs_create(dir, path->dentry, mode, true); 753 path->dentry->d_fsdata = NULL; 754 if (ret) 755 return ERR_PTR(ret); 756 return dentry_open(path, oflag, cred); 757 } 758 759 /* Opens existing queue */ 760 static struct file *do_open(struct path *path, int oflag) 761 { 762 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE, 763 MAY_READ | MAY_WRITE }; 764 int acc; 765 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) 766 return ERR_PTR(-EINVAL); 767 acc = oflag2acc[oflag & O_ACCMODE]; 768 if (inode_permission(d_inode(path->dentry), acc)) 769 return ERR_PTR(-EACCES); 770 return dentry_open(path, oflag, current_cred()); 771 } 772 773 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode, 774 struct mq_attr __user *, u_attr) 775 { 776 struct path path; 777 struct file *filp; 778 struct filename *name; 779 struct mq_attr attr; 780 int fd, error; 781 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns; 782 struct vfsmount *mnt = ipc_ns->mq_mnt; 783 struct dentry *root = mnt->mnt_root; 784 int ro; 785 786 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr))) 787 return -EFAULT; 788 789 audit_mq_open(oflag, mode, u_attr ? &attr : NULL); 790 791 if (IS_ERR(name = getname(u_name))) 792 return PTR_ERR(name); 793 794 fd = get_unused_fd_flags(O_CLOEXEC); 795 if (fd < 0) 796 goto out_putname; 797 798 ro = mnt_want_write(mnt); /* we'll drop it in any case */ 799 error = 0; 800 inode_lock(d_inode(root)); 801 path.dentry = lookup_one_len(name->name, root, strlen(name->name)); 802 if (IS_ERR(path.dentry)) { 803 error = PTR_ERR(path.dentry); 804 goto out_putfd; 805 } 806 path.mnt = mntget(mnt); 807 808 if (oflag & O_CREAT) { 809 if (d_really_is_positive(path.dentry)) { /* entry already exists */ 810 audit_inode(name, path.dentry, 0); 811 if (oflag & O_EXCL) { 812 error = -EEXIST; 813 goto out; 814 } 815 filp = do_open(&path, oflag); 816 } else { 817 if (ro) { 818 error = ro; 819 goto out; 820 } 821 audit_inode_parent_hidden(name, root); 822 filp = do_create(ipc_ns, d_inode(root), 823 &path, oflag, mode, 824 u_attr ? &attr : NULL); 825 } 826 } else { 827 if (d_really_is_negative(path.dentry)) { 828 error = -ENOENT; 829 goto out; 830 } 831 audit_inode(name, path.dentry, 0); 832 filp = do_open(&path, oflag); 833 } 834 835 if (!IS_ERR(filp)) 836 fd_install(fd, filp); 837 else 838 error = PTR_ERR(filp); 839 out: 840 path_put(&path); 841 out_putfd: 842 if (error) { 843 put_unused_fd(fd); 844 fd = error; 845 } 846 inode_unlock(d_inode(root)); 847 if (!ro) 848 mnt_drop_write(mnt); 849 out_putname: 850 putname(name); 851 return fd; 852 } 853 854 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name) 855 { 856 int err; 857 struct filename *name; 858 struct dentry *dentry; 859 struct inode *inode = NULL; 860 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns; 861 struct vfsmount *mnt = ipc_ns->mq_mnt; 862 863 name = getname(u_name); 864 if (IS_ERR(name)) 865 return PTR_ERR(name); 866 867 audit_inode_parent_hidden(name, mnt->mnt_root); 868 err = mnt_want_write(mnt); 869 if (err) 870 goto out_name; 871 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT); 872 dentry = lookup_one_len(name->name, mnt->mnt_root, 873 strlen(name->name)); 874 if (IS_ERR(dentry)) { 875 err = PTR_ERR(dentry); 876 goto out_unlock; 877 } 878 879 inode = d_inode(dentry); 880 if (!inode) { 881 err = -ENOENT; 882 } else { 883 ihold(inode); 884 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL); 885 } 886 dput(dentry); 887 888 out_unlock: 889 inode_unlock(d_inode(mnt->mnt_root)); 890 if (inode) 891 iput(inode); 892 mnt_drop_write(mnt); 893 out_name: 894 putname(name); 895 896 return err; 897 } 898 899 /* Pipelined send and receive functions. 900 * 901 * If a receiver finds no waiting message, then it registers itself in the 902 * list of waiting receivers. A sender checks that list before adding the new 903 * message into the message array. If there is a waiting receiver, then it 904 * bypasses the message array and directly hands the message over to the 905 * receiver. The receiver accepts the message and returns without grabbing the 906 * queue spinlock: 907 * 908 * - Set pointer to message. 909 * - Queue the receiver task for later wakeup (without the info->lock). 910 * - Update its state to STATE_READY. Now the receiver can continue. 911 * - Wake up the process after the lock is dropped. Should the process wake up 912 * before this wakeup (due to a timeout or a signal) it will either see 913 * STATE_READY and continue or acquire the lock to check the state again. 914 * 915 * The same algorithm is used for senders. 916 */ 917 918 /* pipelined_send() - send a message directly to the task waiting in 919 * sys_mq_timedreceive() (without inserting message into a queue). 920 */ 921 static inline void pipelined_send(struct wake_q_head *wake_q, 922 struct mqueue_inode_info *info, 923 struct msg_msg *message, 924 struct ext_wait_queue *receiver) 925 { 926 receiver->msg = message; 927 list_del(&receiver->list); 928 wake_q_add(wake_q, receiver->task); 929 /* 930 * Rely on the implicit cmpxchg barrier from wake_q_add such 931 * that we can ensure that updating receiver->state is the last 932 * write operation: As once set, the receiver can continue, 933 * and if we don't have the reference count from the wake_q, 934 * yet, at that point we can later have a use-after-free 935 * condition and bogus wakeup. 936 */ 937 receiver->state = STATE_READY; 938 } 939 940 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend() 941 * gets its message and put to the queue (we have one free place for sure). */ 942 static inline void pipelined_receive(struct wake_q_head *wake_q, 943 struct mqueue_inode_info *info) 944 { 945 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND); 946 947 if (!sender) { 948 /* for poll */ 949 wake_up_interruptible(&info->wait_q); 950 return; 951 } 952 if (msg_insert(sender->msg, info)) 953 return; 954 955 list_del(&sender->list); 956 wake_q_add(wake_q, sender->task); 957 sender->state = STATE_READY; 958 } 959 960 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr, 961 size_t, msg_len, unsigned int, msg_prio, 962 const struct timespec __user *, u_abs_timeout) 963 { 964 struct fd f; 965 struct inode *inode; 966 struct ext_wait_queue wait; 967 struct ext_wait_queue *receiver; 968 struct msg_msg *msg_ptr; 969 struct mqueue_inode_info *info; 970 ktime_t expires, *timeout = NULL; 971 struct timespec ts; 972 struct posix_msg_tree_node *new_leaf = NULL; 973 int ret = 0; 974 DEFINE_WAKE_Q(wake_q); 975 976 if (u_abs_timeout) { 977 int res = prepare_timeout(u_abs_timeout, &expires, &ts); 978 if (res) 979 return res; 980 timeout = &expires; 981 } 982 983 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX)) 984 return -EINVAL; 985 986 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL); 987 988 f = fdget(mqdes); 989 if (unlikely(!f.file)) { 990 ret = -EBADF; 991 goto out; 992 } 993 994 inode = file_inode(f.file); 995 if (unlikely(f.file->f_op != &mqueue_file_operations)) { 996 ret = -EBADF; 997 goto out_fput; 998 } 999 info = MQUEUE_I(inode); 1000 audit_file(f.file); 1001 1002 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) { 1003 ret = -EBADF; 1004 goto out_fput; 1005 } 1006 1007 if (unlikely(msg_len > info->attr.mq_msgsize)) { 1008 ret = -EMSGSIZE; 1009 goto out_fput; 1010 } 1011 1012 /* First try to allocate memory, before doing anything with 1013 * existing queues. */ 1014 msg_ptr = load_msg(u_msg_ptr, msg_len); 1015 if (IS_ERR(msg_ptr)) { 1016 ret = PTR_ERR(msg_ptr); 1017 goto out_fput; 1018 } 1019 msg_ptr->m_ts = msg_len; 1020 msg_ptr->m_type = msg_prio; 1021 1022 /* 1023 * msg_insert really wants us to have a valid, spare node struct so 1024 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will 1025 * fall back to that if necessary. 1026 */ 1027 if (!info->node_cache) 1028 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL); 1029 1030 spin_lock(&info->lock); 1031 1032 if (!info->node_cache && new_leaf) { 1033 /* Save our speculative allocation into the cache */ 1034 INIT_LIST_HEAD(&new_leaf->msg_list); 1035 info->node_cache = new_leaf; 1036 new_leaf = NULL; 1037 } else { 1038 kfree(new_leaf); 1039 } 1040 1041 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) { 1042 if (f.file->f_flags & O_NONBLOCK) { 1043 ret = -EAGAIN; 1044 } else { 1045 wait.task = current; 1046 wait.msg = (void *) msg_ptr; 1047 wait.state = STATE_NONE; 1048 ret = wq_sleep(info, SEND, timeout, &wait); 1049 /* 1050 * wq_sleep must be called with info->lock held, and 1051 * returns with the lock released 1052 */ 1053 goto out_free; 1054 } 1055 } else { 1056 receiver = wq_get_first_waiter(info, RECV); 1057 if (receiver) { 1058 pipelined_send(&wake_q, info, msg_ptr, receiver); 1059 } else { 1060 /* adds message to the queue */ 1061 ret = msg_insert(msg_ptr, info); 1062 if (ret) 1063 goto out_unlock; 1064 __do_notify(info); 1065 } 1066 inode->i_atime = inode->i_mtime = inode->i_ctime = 1067 current_time(inode); 1068 } 1069 out_unlock: 1070 spin_unlock(&info->lock); 1071 wake_up_q(&wake_q); 1072 out_free: 1073 if (ret) 1074 free_msg(msg_ptr); 1075 out_fput: 1076 fdput(f); 1077 out: 1078 return ret; 1079 } 1080 1081 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr, 1082 size_t, msg_len, unsigned int __user *, u_msg_prio, 1083 const struct timespec __user *, u_abs_timeout) 1084 { 1085 ssize_t ret; 1086 struct msg_msg *msg_ptr; 1087 struct fd f; 1088 struct inode *inode; 1089 struct mqueue_inode_info *info; 1090 struct ext_wait_queue wait; 1091 ktime_t expires, *timeout = NULL; 1092 struct timespec ts; 1093 struct posix_msg_tree_node *new_leaf = NULL; 1094 1095 if (u_abs_timeout) { 1096 int res = prepare_timeout(u_abs_timeout, &expires, &ts); 1097 if (res) 1098 return res; 1099 timeout = &expires; 1100 } 1101 1102 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL); 1103 1104 f = fdget(mqdes); 1105 if (unlikely(!f.file)) { 1106 ret = -EBADF; 1107 goto out; 1108 } 1109 1110 inode = file_inode(f.file); 1111 if (unlikely(f.file->f_op != &mqueue_file_operations)) { 1112 ret = -EBADF; 1113 goto out_fput; 1114 } 1115 info = MQUEUE_I(inode); 1116 audit_file(f.file); 1117 1118 if (unlikely(!(f.file->f_mode & FMODE_READ))) { 1119 ret = -EBADF; 1120 goto out_fput; 1121 } 1122 1123 /* checks if buffer is big enough */ 1124 if (unlikely(msg_len < info->attr.mq_msgsize)) { 1125 ret = -EMSGSIZE; 1126 goto out_fput; 1127 } 1128 1129 /* 1130 * msg_insert really wants us to have a valid, spare node struct so 1131 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will 1132 * fall back to that if necessary. 1133 */ 1134 if (!info->node_cache) 1135 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL); 1136 1137 spin_lock(&info->lock); 1138 1139 if (!info->node_cache && new_leaf) { 1140 /* Save our speculative allocation into the cache */ 1141 INIT_LIST_HEAD(&new_leaf->msg_list); 1142 info->node_cache = new_leaf; 1143 } else { 1144 kfree(new_leaf); 1145 } 1146 1147 if (info->attr.mq_curmsgs == 0) { 1148 if (f.file->f_flags & O_NONBLOCK) { 1149 spin_unlock(&info->lock); 1150 ret = -EAGAIN; 1151 } else { 1152 wait.task = current; 1153 wait.state = STATE_NONE; 1154 ret = wq_sleep(info, RECV, timeout, &wait); 1155 msg_ptr = wait.msg; 1156 } 1157 } else { 1158 DEFINE_WAKE_Q(wake_q); 1159 1160 msg_ptr = msg_get(info); 1161 1162 inode->i_atime = inode->i_mtime = inode->i_ctime = 1163 current_time(inode); 1164 1165 /* There is now free space in queue. */ 1166 pipelined_receive(&wake_q, info); 1167 spin_unlock(&info->lock); 1168 wake_up_q(&wake_q); 1169 ret = 0; 1170 } 1171 if (ret == 0) { 1172 ret = msg_ptr->m_ts; 1173 1174 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) || 1175 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) { 1176 ret = -EFAULT; 1177 } 1178 free_msg(msg_ptr); 1179 } 1180 out_fput: 1181 fdput(f); 1182 out: 1183 return ret; 1184 } 1185 1186 /* 1187 * Notes: the case when user wants us to deregister (with NULL as pointer) 1188 * and he isn't currently owner of notification, will be silently discarded. 1189 * It isn't explicitly defined in the POSIX. 1190 */ 1191 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes, 1192 const struct sigevent __user *, u_notification) 1193 { 1194 int ret; 1195 struct fd f; 1196 struct sock *sock; 1197 struct inode *inode; 1198 struct sigevent notification; 1199 struct mqueue_inode_info *info; 1200 struct sk_buff *nc; 1201 1202 if (u_notification) { 1203 if (copy_from_user(¬ification, u_notification, 1204 sizeof(struct sigevent))) 1205 return -EFAULT; 1206 } 1207 1208 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL); 1209 1210 nc = NULL; 1211 sock = NULL; 1212 if (u_notification != NULL) { 1213 if (unlikely(notification.sigev_notify != SIGEV_NONE && 1214 notification.sigev_notify != SIGEV_SIGNAL && 1215 notification.sigev_notify != SIGEV_THREAD)) 1216 return -EINVAL; 1217 if (notification.sigev_notify == SIGEV_SIGNAL && 1218 !valid_signal(notification.sigev_signo)) { 1219 return -EINVAL; 1220 } 1221 if (notification.sigev_notify == SIGEV_THREAD) { 1222 long timeo; 1223 1224 /* create the notify skb */ 1225 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL); 1226 if (!nc) { 1227 ret = -ENOMEM; 1228 goto out; 1229 } 1230 if (copy_from_user(nc->data, 1231 notification.sigev_value.sival_ptr, 1232 NOTIFY_COOKIE_LEN)) { 1233 ret = -EFAULT; 1234 goto out; 1235 } 1236 1237 /* TODO: add a header? */ 1238 skb_put(nc, NOTIFY_COOKIE_LEN); 1239 /* and attach it to the socket */ 1240 retry: 1241 f = fdget(notification.sigev_signo); 1242 if (!f.file) { 1243 ret = -EBADF; 1244 goto out; 1245 } 1246 sock = netlink_getsockbyfilp(f.file); 1247 fdput(f); 1248 if (IS_ERR(sock)) { 1249 ret = PTR_ERR(sock); 1250 sock = NULL; 1251 goto out; 1252 } 1253 1254 timeo = MAX_SCHEDULE_TIMEOUT; 1255 ret = netlink_attachskb(sock, nc, &timeo, NULL); 1256 if (ret == 1) 1257 goto retry; 1258 if (ret) { 1259 sock = NULL; 1260 nc = NULL; 1261 goto out; 1262 } 1263 } 1264 } 1265 1266 f = fdget(mqdes); 1267 if (!f.file) { 1268 ret = -EBADF; 1269 goto out; 1270 } 1271 1272 inode = file_inode(f.file); 1273 if (unlikely(f.file->f_op != &mqueue_file_operations)) { 1274 ret = -EBADF; 1275 goto out_fput; 1276 } 1277 info = MQUEUE_I(inode); 1278 1279 ret = 0; 1280 spin_lock(&info->lock); 1281 if (u_notification == NULL) { 1282 if (info->notify_owner == task_tgid(current)) { 1283 remove_notification(info); 1284 inode->i_atime = inode->i_ctime = current_time(inode); 1285 } 1286 } else if (info->notify_owner != NULL) { 1287 ret = -EBUSY; 1288 } else { 1289 switch (notification.sigev_notify) { 1290 case SIGEV_NONE: 1291 info->notify.sigev_notify = SIGEV_NONE; 1292 break; 1293 case SIGEV_THREAD: 1294 info->notify_sock = sock; 1295 info->notify_cookie = nc; 1296 sock = NULL; 1297 nc = NULL; 1298 info->notify.sigev_notify = SIGEV_THREAD; 1299 break; 1300 case SIGEV_SIGNAL: 1301 info->notify.sigev_signo = notification.sigev_signo; 1302 info->notify.sigev_value = notification.sigev_value; 1303 info->notify.sigev_notify = SIGEV_SIGNAL; 1304 break; 1305 } 1306 1307 info->notify_owner = get_pid(task_tgid(current)); 1308 info->notify_user_ns = get_user_ns(current_user_ns()); 1309 inode->i_atime = inode->i_ctime = current_time(inode); 1310 } 1311 spin_unlock(&info->lock); 1312 out_fput: 1313 fdput(f); 1314 out: 1315 if (sock) 1316 netlink_detachskb(sock, nc); 1317 else if (nc) 1318 dev_kfree_skb(nc); 1319 1320 return ret; 1321 } 1322 1323 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes, 1324 const struct mq_attr __user *, u_mqstat, 1325 struct mq_attr __user *, u_omqstat) 1326 { 1327 int ret; 1328 struct mq_attr mqstat, omqstat; 1329 struct fd f; 1330 struct inode *inode; 1331 struct mqueue_inode_info *info; 1332 1333 if (u_mqstat != NULL) { 1334 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr))) 1335 return -EFAULT; 1336 if (mqstat.mq_flags & (~O_NONBLOCK)) 1337 return -EINVAL; 1338 } 1339 1340 f = fdget(mqdes); 1341 if (!f.file) { 1342 ret = -EBADF; 1343 goto out; 1344 } 1345 1346 inode = file_inode(f.file); 1347 if (unlikely(f.file->f_op != &mqueue_file_operations)) { 1348 ret = -EBADF; 1349 goto out_fput; 1350 } 1351 info = MQUEUE_I(inode); 1352 1353 spin_lock(&info->lock); 1354 1355 omqstat = info->attr; 1356 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK; 1357 if (u_mqstat) { 1358 audit_mq_getsetattr(mqdes, &mqstat); 1359 spin_lock(&f.file->f_lock); 1360 if (mqstat.mq_flags & O_NONBLOCK) 1361 f.file->f_flags |= O_NONBLOCK; 1362 else 1363 f.file->f_flags &= ~O_NONBLOCK; 1364 spin_unlock(&f.file->f_lock); 1365 1366 inode->i_atime = inode->i_ctime = current_time(inode); 1367 } 1368 1369 spin_unlock(&info->lock); 1370 1371 ret = 0; 1372 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat, 1373 sizeof(struct mq_attr))) 1374 ret = -EFAULT; 1375 1376 out_fput: 1377 fdput(f); 1378 out: 1379 return ret; 1380 } 1381 1382 static const struct inode_operations mqueue_dir_inode_operations = { 1383 .lookup = simple_lookup, 1384 .create = mqueue_create, 1385 .unlink = mqueue_unlink, 1386 }; 1387 1388 static const struct file_operations mqueue_file_operations = { 1389 .flush = mqueue_flush_file, 1390 .poll = mqueue_poll_file, 1391 .read = mqueue_read_file, 1392 .llseek = default_llseek, 1393 }; 1394 1395 static const struct super_operations mqueue_super_ops = { 1396 .alloc_inode = mqueue_alloc_inode, 1397 .destroy_inode = mqueue_destroy_inode, 1398 .evict_inode = mqueue_evict_inode, 1399 .statfs = simple_statfs, 1400 }; 1401 1402 static struct file_system_type mqueue_fs_type = { 1403 .name = "mqueue", 1404 .mount = mqueue_mount, 1405 .kill_sb = kill_litter_super, 1406 .fs_flags = FS_USERNS_MOUNT, 1407 }; 1408 1409 int mq_init_ns(struct ipc_namespace *ns) 1410 { 1411 ns->mq_queues_count = 0; 1412 ns->mq_queues_max = DFLT_QUEUESMAX; 1413 ns->mq_msg_max = DFLT_MSGMAX; 1414 ns->mq_msgsize_max = DFLT_MSGSIZEMAX; 1415 ns->mq_msg_default = DFLT_MSG; 1416 ns->mq_msgsize_default = DFLT_MSGSIZE; 1417 1418 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns); 1419 if (IS_ERR(ns->mq_mnt)) { 1420 int err = PTR_ERR(ns->mq_mnt); 1421 ns->mq_mnt = NULL; 1422 return err; 1423 } 1424 return 0; 1425 } 1426 1427 void mq_clear_sbinfo(struct ipc_namespace *ns) 1428 { 1429 ns->mq_mnt->mnt_sb->s_fs_info = NULL; 1430 } 1431 1432 void mq_put_mnt(struct ipc_namespace *ns) 1433 { 1434 kern_unmount(ns->mq_mnt); 1435 } 1436 1437 static int __init init_mqueue_fs(void) 1438 { 1439 int error; 1440 1441 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache", 1442 sizeof(struct mqueue_inode_info), 0, 1443 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once); 1444 if (mqueue_inode_cachep == NULL) 1445 return -ENOMEM; 1446 1447 /* ignore failures - they are not fatal */ 1448 mq_sysctl_table = mq_register_sysctl_table(); 1449 1450 error = register_filesystem(&mqueue_fs_type); 1451 if (error) 1452 goto out_sysctl; 1453 1454 spin_lock_init(&mq_lock); 1455 1456 error = mq_init_ns(&init_ipc_ns); 1457 if (error) 1458 goto out_filesystem; 1459 1460 return 0; 1461 1462 out_filesystem: 1463 unregister_filesystem(&mqueue_fs_type); 1464 out_sysctl: 1465 if (mq_sysctl_table) 1466 unregister_sysctl_table(mq_sysctl_table); 1467 kmem_cache_destroy(mqueue_inode_cachep); 1468 return error; 1469 } 1470 1471 device_initcall(init_mqueue_fs); 1472