ipc/mqueue.c

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