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                 INIT_LIST_HEAD(&leaf->msg_list);
 146                 info->qsize += sizeof(*leaf);
 147         }
 148         leaf->priority = msg->m_type;
 149         rb_link_node(&leaf->rb_node, parent, p);
 150         rb_insert_color(&leaf->rb_node, &info->msg_tree);
 151 insert_msg:
 152         info->attr.mq_curmsgs++;
 153         info->qsize += msg->m_ts;
 154         list_add_tail(&msg->m_list, &leaf->msg_list);
 155         return 0;
 156 }
 157
 158 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 159 {
 160         struct rb_node **p, *parent = NULL;
 161         struct posix_msg_tree_node *leaf;
 162         struct msg_msg *msg;
 163
 164 try_again:
 165         p = &info->msg_tree.rb_node;
 166         while (*p) {
 167                 parent = *p;
 168                 /*
 169                  * During insert, low priorities go to the left and high to the
 170                  * right.  On receive, we want the highest priorities first, so
 171                  * walk all the way to the right.
 172                  */
 173                 p = &(*p)->rb_right;
 174         }
 175         if (!parent) {
 176                 if (info->attr.mq_curmsgs) {
 177                         pr_warn_once("Inconsistency in POSIX message queue, "
 178                                      "no tree element, but supposedly messages "
 179                                      "should exist!\n");
 180                         info->attr.mq_curmsgs = 0;
 181                 }
 182                 return NULL;
 183         }
 184         leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 185         if (unlikely(list_empty(&leaf->msg_list))) {
 186                 pr_warn_once("Inconsistency in POSIX message queue, "
 187                              "empty leaf node but we haven't implemented "
 188                              "lazy leaf delete!\n");
 189                 rb_erase(&leaf->rb_node, &info->msg_tree);
 190                 if (info->node_cache) {
 191                         info->qsize -= sizeof(*leaf);
 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                                 info->qsize -= sizeof(*leaf);
 205                                 kfree(leaf);
 206                         } else {
 207                                 info->node_cache = leaf;
 208                         }
 209                 }
 210         }
 211         info->attr.mq_curmsgs--;
 212         info->qsize -= msg->m_ts;
 213         return msg;
 214 }
 215
 216 static struct inode *mqueue_get_inode(struct super_block *sb,
 217                 struct ipc_namespace *ipc_ns, umode_t mode,
 218                 struct mq_attr *attr)
 219 {
 220         struct user_struct *u = current_user();
 221         struct inode *inode;
 222         int ret = -ENOMEM;
 223
 224         inode = new_inode(sb);
 225         if (!inode)
 226                 goto err;
 227
 228         inode->i_ino = get_next_ino();
 229         inode->i_mode = mode;
 230         inode->i_uid = current_fsuid();
 231         inode->i_gid = current_fsgid();
 232         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
 233
 234         if (S_ISREG(mode)) {
 235                 struct mqueue_inode_info *info;
 236                 unsigned long mq_bytes, mq_treesize;
 237
 238                 inode->i_fop = &mqueue_file_operations;
 239                 inode->i_size = FILENT_SIZE;
 240                 /* mqueue specific info */
 241                 info = MQUEUE_I(inode);
 242                 spin_lock_init(&info->lock);
 243                 init_waitqueue_head(&info->wait_q);
 244                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
 245                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
 246                 info->notify_owner = NULL;
 247                 info->notify_user_ns = NULL;
 248                 info->qsize = 0;
 249                 info->user = NULL;      /* set when all is ok */
 250                 info->msg_tree = RB_ROOT;
 251                 info->node_cache = NULL;
 252                 memset(&info->attr, 0, sizeof(info->attr));
 253                 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 254                                            ipc_ns->mq_msg_default);
 255                 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 256                                             ipc_ns->mq_msgsize_default);
 257                 if (attr) {
 258                         info->attr.mq_maxmsg = attr->mq_maxmsg;
 259                         info->attr.mq_msgsize = attr->mq_msgsize;
 260                 }
 261                 /*
 262                  * We used to allocate a static array of pointers and account
 263                  * the size of that array as well as one msg_msg struct per
 264                  * possible message into the queue size. That's no longer
 265                  * accurate as the queue is now an rbtree and will grow and
 266                  * shrink depending on usage patterns.  We can, however, still
 267                  * account one msg_msg struct per message, but the nodes are
 268                  * allocated depending on priority usage, and most programs
 269                  * only use one, or a handful, of priorities.  However, since
 270                  * this is pinned memory, we need to assume worst case, so
 271                  * that means the min(mq_maxmsg, max_priorities) * struct
 272                  * posix_msg_tree_node.
 273                  */
 274                 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 275                         min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 276                         sizeof(struct posix_msg_tree_node);
 277
 278                 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 279                                           info->attr.mq_msgsize);
 280
 281                 spin_lock(&mq_lock);
 282                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 283                     u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 284                         spin_unlock(&mq_lock);
 285                         /* mqueue_evict_inode() releases info->messages */
 286                         ret = -EMFILE;
 287                         goto out_inode;
 288                 }
 289                 u->mq_bytes += mq_bytes;
 290                 spin_unlock(&mq_lock);
 291
 292                 /* all is ok */
 293                 info->user = get_uid(u);
 294         } else if (S_ISDIR(mode)) {
 295                 inc_nlink(inode);
 296                 /* Some things misbehave if size == 0 on a directory */
 297                 inode->i_size = 2 * DIRENT_SIZE;
 298                 inode->i_op = &mqueue_dir_inode_operations;
 299                 inode->i_fop = &simple_dir_operations;
 300         }
 301
 302         return inode;
 303 out_inode:
 304         iput(inode);
 305 err:
 306         return ERR_PTR(ret);
 307 }
 308
 309 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 310 {
 311         struct inode *inode;
 312         struct ipc_namespace *ns = data;
 313
 314         sb->s_blocksize = PAGE_CACHE_SIZE;
 315         sb->s_blocksize_bits = PAGE_CACHE_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         if (!(flags & MS_KERNMOUNT))
 334                 data = current->nsproxy->ipc_ns;
 335         return mount_ns(fs_type, flags, data, mqueue_fill_super);
 336 }
 337
 338 static void init_once(void *foo)
 339 {
 340         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 341
 342         inode_init_once(&p->vfs_inode);
 343 }
 344
 345 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 346 {
 347         struct mqueue_inode_info *ei;
 348
 349         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 350         if (!ei)
 351                 return NULL;
 352         return &ei->vfs_inode;
 353 }
 354
 355 static void mqueue_i_callback(struct rcu_head *head)
 356 {
 357         struct inode *inode = container_of(head, struct inode, i_rcu);
 358         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 359 }
 360
 361 static void mqueue_destroy_inode(struct inode *inode)
 362 {
 363         call_rcu(&inode->i_rcu, mqueue_i_callback);
 364 }
 365
 366 static void mqueue_evict_inode(struct inode *inode)
 367 {
 368         struct mqueue_inode_info *info;
 369         struct user_struct *user;
 370         unsigned long mq_bytes, mq_treesize;
 371         struct ipc_namespace *ipc_ns;
 372         struct msg_msg *msg;
 373
 374         clear_inode(inode);
 375
 376         if (S_ISDIR(inode->i_mode))
 377                 return;
 378
 379         ipc_ns = get_ns_from_inode(inode);
 380         info = MQUEUE_I(inode);
 381         spin_lock(&info->lock);
 382         while ((msg = msg_get(info)) != NULL)
 383                 free_msg(msg);
 384         kfree(info->node_cache);
 385         spin_unlock(&info->lock);
 386
 387         /* Total amount of bytes accounted for the mqueue */
 388         mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 389                 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 390                 sizeof(struct posix_msg_tree_node);
 391
 392         mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 393                                   info->attr.mq_msgsize);
 394
 395         user = info->user;
 396         if (user) {
 397                 spin_lock(&mq_lock);
 398                 user->mq_bytes -= mq_bytes;
 399                 /*
 400                  * get_ns_from_inode() ensures that the
 401                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 402                  * to which we now hold a reference, or it is NULL.
 403                  * We can't put it here under mq_lock, though.
 404                  */
 405                 if (ipc_ns)
 406                         ipc_ns->mq_queues_count--;
 407                 spin_unlock(&mq_lock);
 408                 free_uid(user);
 409         }
 410         if (ipc_ns)
 411                 put_ipc_ns(ipc_ns);
 412 }
 413
 414 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 415                                 umode_t mode, bool excl)
 416 {
 417         struct inode *inode;
 418         struct mq_attr *attr = dentry->d_fsdata;
 419         int error;
 420         struct ipc_namespace *ipc_ns;
 421
 422         spin_lock(&mq_lock);
 423         ipc_ns = __get_ns_from_inode(dir);
 424         if (!ipc_ns) {
 425                 error = -EACCES;
 426                 goto out_unlock;
 427         }
 428         if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
 429             (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 430              !capable(CAP_SYS_RESOURCE))) {
 431                 error = -ENOSPC;
 432                 goto out_unlock;
 433         }
 434         ipc_ns->mq_queues_count++;
 435         spin_unlock(&mq_lock);
 436
 437         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 438         if (IS_ERR(inode)) {
 439                 error = PTR_ERR(inode);
 440                 spin_lock(&mq_lock);
 441                 ipc_ns->mq_queues_count--;
 442                 goto out_unlock;
 443         }
 444
 445         put_ipc_ns(ipc_ns);
 446         dir->i_size += DIRENT_SIZE;
 447         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 448
 449         d_instantiate(dentry, inode);
 450         dget(dentry);
 451         return 0;
 452 out_unlock:
 453         spin_unlock(&mq_lock);
 454         if (ipc_ns)
 455                 put_ipc_ns(ipc_ns);
 456         return error;
 457 }
 458
 459 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 460 {
 461         struct inode *inode = dentry->d_inode;
 462
 463         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 464         dir->i_size -= DIRENT_SIZE;
 465         drop_nlink(inode);
 466         dput(dentry);
 467         return 0;
 468 }
 469
 470 /*
 471 *       This is routine for system read from queue file.
 472 *       To avoid mess with doing here some sort of mq_receive we allow
 473 *       to read only queue size & notification info (the only values
 474 *       that are interesting from user point of view and aren't accessible
 475 *       through std routines)
 476 */
 477 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 478                                 size_t count, loff_t *off)
 479 {
 480         struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 481         char buffer[FILENT_SIZE];
 482         ssize_t ret;
 483
 484         spin_lock(&info->lock);
 485         snprintf(buffer, sizeof(buffer),
 486                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 487                         info->qsize,
 488                         info->notify_owner ? info->notify.sigev_notify : 0,
 489                         (info->notify_owner &&
 490                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 491                                 info->notify.sigev_signo : 0,
 492                         pid_vnr(info->notify_owner));
 493         spin_unlock(&info->lock);
 494         buffer[sizeof(buffer)-1] = '\0';
 495
 496         ret = simple_read_from_buffer(u_data, count, off, buffer,
 497                                 strlen(buffer));
 498         if (ret <= 0)
 499                 return ret;
 500
 501         file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
 502         return ret;
 503 }
 504
 505 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 506 {
 507         struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 508
 509         spin_lock(&info->lock);
 510         if (task_tgid(current) == info->notify_owner)
 511                 remove_notification(info);
 512
 513         spin_unlock(&info->lock);
 514         return 0;
 515 }
 516
 517 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 518 {
 519         struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 520         int retval = 0;
 521
 522         poll_wait(filp, &info->wait_q, poll_tab);
 523
 524         spin_lock(&info->lock);
 525         if (info->attr.mq_curmsgs)
 526                 retval = POLLIN | POLLRDNORM;
 527
 528         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 529                 retval |= POLLOUT | POLLWRNORM;
 530         spin_unlock(&info->lock);
 531
 532         return retval;
 533 }
 534
 535 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 536 static void wq_add(struct mqueue_inode_info *info, int sr,
 537                         struct ext_wait_queue *ewp)
 538 {
 539         struct ext_wait_queue *walk;
 540
 541         ewp->task = current;
 542
 543         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 544                 if (walk->task->static_prio <= current->static_prio) {
 545                         list_add_tail(&ewp->list, &walk->list);
 546                         return;
 547                 }
 548         }
 549         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 550 }
 551
 552 /*
 553  * Puts current task to sleep. Caller must hold queue lock. After return
 554  * lock isn't held.
 555  * sr: SEND or RECV
 556  */
 557 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 558                     ktime_t *timeout, struct ext_wait_queue *ewp)
 559 {
 560         int retval;
 561         signed long time;
 562
 563         wq_add(info, sr, ewp);
 564
 565         for (;;) {
 566                 set_current_state(TASK_INTERRUPTIBLE);
 567
 568                 spin_unlock(&info->lock);
 569                 time = schedule_hrtimeout_range_clock(timeout, 0,
 570                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
 571
 572                 while (ewp->state == STATE_PENDING)
 573                         cpu_relax();
 574
 575                 if (ewp->state == STATE_READY) {
 576                         retval = 0;
 577                         goto out;
 578                 }
 579                 spin_lock(&info->lock);
 580                 if (ewp->state == STATE_READY) {
 581                         retval = 0;
 582                         goto out_unlock;
 583                 }
 584                 if (signal_pending(current)) {
 585                         retval = -ERESTARTSYS;
 586                         break;
 587                 }
 588                 if (time == 0) {
 589                         retval = -ETIMEDOUT;
 590                         break;
 591                 }
 592         }
 593         list_del(&ewp->list);
 594 out_unlock:
 595         spin_unlock(&info->lock);
 596 out:
 597         return retval;
 598 }
 599
 600 /*
 601  * Returns waiting task that should be serviced first or NULL if none exists
 602  */
 603 static struct ext_wait_queue *wq_get_first_waiter(
 604                 struct mqueue_inode_info *info, int sr)
 605 {
 606         struct list_head *ptr;
 607
 608         ptr = info->e_wait_q[sr].list.prev;
 609         if (ptr == &info->e_wait_q[sr].list)
 610                 return NULL;
 611         return list_entry(ptr, struct ext_wait_queue, list);
 612 }
 613
 614
 615 static inline void set_cookie(struct sk_buff *skb, char code)
 616 {
 617         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 618 }
 619
 620 /*
 621  * The next function is only to split too long sys_mq_timedsend
 622  */
 623 static void __do_notify(struct mqueue_inode_info *info)
 624 {
 625         /* notification
 626          * invoked when there is registered process and there isn't process
 627          * waiting synchronously for message AND state of queue changed from
 628          * empty to not empty. Here we are sure that no one is waiting
 629          * synchronously. */
 630         if (info->notify_owner &&
 631             info->attr.mq_curmsgs == 1) {
 632                 struct siginfo sig_i;
 633                 switch (info->notify.sigev_notify) {
 634                 case SIGEV_NONE:
 635                         break;
 636                 case SIGEV_SIGNAL:
 637                         /* sends signal */
 638
 639                         sig_i.si_signo = info->notify.sigev_signo;
 640                         sig_i.si_errno = 0;
 641                         sig_i.si_code = SI_MESGQ;
 642                         sig_i.si_value = info->notify.sigev_value;
 643                         /* map current pid/uid into info->owner's namespaces */
 644                         rcu_read_lock();
 645                         sig_i.si_pid = task_tgid_nr_ns(current,
 646                                                 ns_of_pid(info->notify_owner));
 647                         sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
 648                         rcu_read_unlock();
 649
 650                         kill_pid_info(info->notify.sigev_signo,
 651                                       &sig_i, info->notify_owner);
 652                         break;
 653                 case SIGEV_THREAD:
 654                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 655                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 656                         break;
 657                 }
 658                 /* after notification unregisters process */
 659                 put_pid(info->notify_owner);
 660                 put_user_ns(info->notify_user_ns);
 661                 info->notify_owner = NULL;
 662                 info->notify_user_ns = NULL;
 663         }
 664         wake_up(&info->wait_q);
 665 }
 666
 667 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 668                            ktime_t *expires, struct timespec *ts)
 669 {
 670         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 671                 return -EFAULT;
 672         if (!timespec_valid(ts))
 673                 return -EINVAL;
 674
 675         *expires = timespec_to_ktime(*ts);
 676         return 0;
 677 }
 678
 679 static void remove_notification(struct mqueue_inode_info *info)
 680 {
 681         if (info->notify_owner != NULL &&
 682             info->notify.sigev_notify == SIGEV_THREAD) {
 683                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 684                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 685         }
 686         put_pid(info->notify_owner);
 687         put_user_ns(info->notify_user_ns);
 688         info->notify_owner = NULL;
 689         info->notify_user_ns = NULL;
 690 }
 691
 692 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 693 {
 694         int mq_treesize;
 695         unsigned long total_size;
 696
 697         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 698                 return -EINVAL;
 699         if (capable(CAP_SYS_RESOURCE)) {
 700                 if (attr->mq_maxmsg > HARD_MSGMAX ||
 701                     attr->mq_msgsize > HARD_MSGSIZEMAX)
 702                         return -EINVAL;
 703         } else {
 704                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 705                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 706                         return -EINVAL;
 707         }
 708         /* check for overflow */
 709         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 710                 return -EOVERFLOW;
 711         mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
 712                 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
 713                 sizeof(struct posix_msg_tree_node);
 714         total_size = attr->mq_maxmsg * attr->mq_msgsize;
 715         if (total_size + mq_treesize < total_size)
 716                 return -EOVERFLOW;
 717         return 0;
 718 }
 719
 720 /*
 721  * Invoked when creating a new queue via sys_mq_open
 722  */
 723 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
 724                         struct path *path, int oflag, umode_t mode,
 725                         struct mq_attr *attr)
 726 {
 727         const struct cred *cred = current_cred();
 728         int ret;
 729
 730         if (attr) {
 731                 ret = mq_attr_ok(ipc_ns, attr);
 732                 if (ret)
 733                         return ERR_PTR(ret);
 734                 /* store for use during create */
 735                 path->dentry->d_fsdata = attr;
 736         } else {
 737                 struct mq_attr def_attr;
 738
 739                 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 740                                          ipc_ns->mq_msg_default);
 741                 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 742                                           ipc_ns->mq_msgsize_default);
 743                 ret = mq_attr_ok(ipc_ns, &def_attr);
 744                 if (ret)
 745                         return ERR_PTR(ret);
 746         }
 747
 748         mode &= ~current_umask();
 749         ret = vfs_create(dir, path->dentry, mode, true);
 750         path->dentry->d_fsdata = NULL;
 751         if (ret)
 752                 return ERR_PTR(ret);
 753         return dentry_open(path, oflag, cred);
 754 }
 755
 756 /* Opens existing queue */
 757 static struct file *do_open(struct path *path, int oflag)
 758 {
 759         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 760                                                   MAY_READ | MAY_WRITE };
 761         int acc;
 762         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 763                 return ERR_PTR(-EINVAL);
 764         acc = oflag2acc[oflag & O_ACCMODE];
 765         if (inode_permission(path->dentry->d_inode, acc))
 766                 return ERR_PTR(-EACCES);
 767         return dentry_open(path, oflag, current_cred());
 768 }
 769
 770 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 771                 struct mq_attr __user *, u_attr)
 772 {
 773         struct path path;
 774         struct file *filp;
 775         struct filename *name;
 776         struct mq_attr attr;
 777         int fd, error;
 778         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 779         struct vfsmount *mnt = ipc_ns->mq_mnt;
 780         struct dentry *root = mnt->mnt_root;
 781         int ro;
 782
 783         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 784                 return -EFAULT;
 785
 786         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 787
 788         if (IS_ERR(name = getname(u_name)))
 789                 return PTR_ERR(name);
 790
 791         fd = get_unused_fd_flags(O_CLOEXEC);
 792         if (fd < 0)
 793                 goto out_putname;
 794
 795         ro = mnt_want_write(mnt);       /* we'll drop it in any case */
 796         error = 0;
 797         mutex_lock(&root->d_inode->i_mutex);
 798         path.dentry = lookup_one_len(name->name, root, strlen(name->name));
 799         if (IS_ERR(path.dentry)) {
 800                 error = PTR_ERR(path.dentry);
 801                 goto out_putfd;
 802         }
 803         path.mnt = mntget(mnt);
 804
 805         if (oflag & O_CREAT) {
 806                 if (path.dentry->d_inode) {     /* entry already exists */
 807                         audit_inode(name, path.dentry, 0);
 808                         if (oflag & O_EXCL) {
 809                                 error = -EEXIST;
 810                                 goto out;
 811                         }
 812                         filp = do_open(&path, oflag);
 813                 } else {
 814                         if (ro) {
 815                                 error = ro;
 816                                 goto out;
 817                         }
 818                         filp = do_create(ipc_ns, root->d_inode,
 819                                                 &path, oflag, mode,
 820                                                 u_attr ? &attr : NULL);
 821                 }
 822         } else {
 823                 if (!path.dentry->d_inode) {
 824                         error = -ENOENT;
 825                         goto out;
 826                 }
 827                 audit_inode(name, path.dentry, 0);
 828                 filp = do_open(&path, oflag);
 829         }
 830
 831         if (!IS_ERR(filp))
 832                 fd_install(fd, filp);
 833         else
 834                 error = PTR_ERR(filp);
 835 out:
 836         path_put(&path);
 837 out_putfd:
 838         if (error) {
 839                 put_unused_fd(fd);
 840                 fd = error;
 841         }
 842         mutex_unlock(&root->d_inode->i_mutex);
 843         if (!ro)
 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         struct filename *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->name, mnt->mnt_root,
 868                                 strlen(name->name));
 869         if (IS_ERR(dentry)) {
 870                 err = PTR_ERR(dentry);
 871                 goto out_unlock;
 872         }
 873
 874         inode = dentry->d_inode;
 875         if (!inode) {
 876                 err = -ENOENT;
 877         } else {
 878                 ihold(inode);
 879                 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 880         }
 881         dput(dentry);
 882
 883 out_unlock:
 884         mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
 885         if (inode)
 886                 iput(inode);
 887         mnt_drop_write(mnt);
 888 out_name:
 889         putname(name);
 890
 891         return err;
 892 }
 893
 894 /* Pipelined send and receive functions.
 895  *
 896  * If a receiver finds no waiting message, then it registers itself in the
 897  * list of waiting receivers. A sender checks that list before adding the new
 898  * message into the message array. If there is a waiting receiver, then it
 899  * bypasses the message array and directly hands the message over to the
 900  * receiver.
 901  * The receiver accepts the message and returns without grabbing the queue
 902  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 903  * are necessary. The same algorithm is used for sysv semaphores, see
 904  * ipc/sem.c for more details.
 905  *
 906  * The same algorithm is used for senders.
 907  */
 908
 909 /* pipelined_send() - send a message directly to the task waiting in
 910  * sys_mq_timedreceive() (without inserting message into a queue).
 911  */
 912 static inline void pipelined_send(struct mqueue_inode_info *info,
 913                                   struct msg_msg *message,
 914                                   struct ext_wait_queue *receiver)
 915 {
 916         receiver->msg = message;
 917         list_del(&receiver->list);
 918         receiver->state = STATE_PENDING;
 919         wake_up_process(receiver->task);
 920         smp_wmb();
 921         receiver->state = STATE_READY;
 922 }
 923
 924 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 925  * gets its message and put to the queue (we have one free place for sure). */
 926 static inline void pipelined_receive(struct mqueue_inode_info *info)
 927 {
 928         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 929
 930         if (!sender) {
 931                 /* for poll */
 932                 wake_up_interruptible(&info->wait_q);
 933                 return;
 934         }
 935         if (msg_insert(sender->msg, info))
 936                 return;
 937         list_del(&sender->list);
 938         sender->state = STATE_PENDING;
 939         wake_up_process(sender->task);
 940         smp_wmb();
 941         sender->state = STATE_READY;
 942 }
 943
 944 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 945                 size_t, msg_len, unsigned int, msg_prio,
 946                 const struct timespec __user *, u_abs_timeout)
 947 {
 948         struct fd f;
 949         struct inode *inode;
 950         struct ext_wait_queue wait;
 951         struct ext_wait_queue *receiver;
 952         struct msg_msg *msg_ptr;
 953         struct mqueue_inode_info *info;
 954         ktime_t expires, *timeout = NULL;
 955         struct timespec ts;
 956         struct posix_msg_tree_node *new_leaf = NULL;
 957         int ret = 0;
 958
 959         if (u_abs_timeout) {
 960                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 961                 if (res)
 962                         return res;
 963                 timeout = &expires;
 964         }
 965
 966         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 967                 return -EINVAL;
 968
 969         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 970
 971         f = fdget(mqdes);
 972         if (unlikely(!f.file)) {
 973                 ret = -EBADF;
 974                 goto out;
 975         }
 976
 977         inode = file_inode(f.file);
 978         if (unlikely(f.file->f_op != &mqueue_file_operations)) {
 979                 ret = -EBADF;
 980                 goto out_fput;
 981         }
 982         info = MQUEUE_I(inode);
 983         audit_inode(NULL, f.file->f_path.dentry, 0);
 984
 985         if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
 986                 ret = -EBADF;
 987                 goto out_fput;
 988         }
 989
 990         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 991                 ret = -EMSGSIZE;
 992                 goto out_fput;
 993         }
 994
 995         /* First try to allocate memory, before doing anything with
 996          * existing queues. */
 997         msg_ptr = load_msg(u_msg_ptr, msg_len);
 998         if (IS_ERR(msg_ptr)) {
 999                 ret = PTR_ERR(msg_ptr);
1000                 goto out_fput;
1001         }
1002         msg_ptr->m_ts = msg_len;
1003         msg_ptr->m_type = msg_prio;
1004
1005         /*
1006          * msg_insert really wants us to have a valid, spare node struct so
1007          * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1008          * fall back to that if necessary.
1009          */
1010         if (!info->node_cache)
1011                 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1012
1013         spin_lock(&info->lock);
1014
1015         if (!info->node_cache && new_leaf) {
1016                 /* Save our speculative allocation into the cache */
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 (f.file->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         fdput(f);
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 fd f;
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         f = fdget(mqdes);
1088         if (unlikely(!f.file)) {
1089                 ret = -EBADF;
1090                 goto out;
1091         }
1092
1093         inode = file_inode(f.file);
1094         if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1095                 ret = -EBADF;
1096                 goto out_fput;
1097         }
1098         info = MQUEUE_I(inode);
1099         audit_inode(NULL, f.file->f_path.dentry, 0);
1100
1101         if (unlikely(!(f.file->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                 INIT_LIST_HEAD(&new_leaf->msg_list);
1125                 info->node_cache = new_leaf;
1126                 info->qsize += sizeof(*new_leaf);
1127         } else {
1128                 kfree(new_leaf);
1129         }
1130
1131         if (info->attr.mq_curmsgs == 0) {
1132                 if (f.file->f_flags & O_NONBLOCK) {
1133                         spin_unlock(&info->lock);
1134                         ret = -EAGAIN;
1135                 } else {
1136                         wait.task = current;
1137                         wait.state = STATE_NONE;
1138                         ret = wq_sleep(info, RECV, timeout, &wait);
1139                         msg_ptr = wait.msg;
1140                 }
1141         } else {
1142                 msg_ptr = msg_get(info);
1143
1144                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1145                                 CURRENT_TIME;
1146
1147                 /* There is now free space in queue. */
1148                 pipelined_receive(info);
1149                 spin_unlock(&info->lock);
1150                 ret = 0;
1151         }
1152         if (ret == 0) {
1153                 ret = msg_ptr->m_ts;
1154
1155                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1156                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1157                         ret = -EFAULT;
1158                 }
1159                 free_msg(msg_ptr);
1160         }
1161 out_fput:
1162         fdput(f);
1163 out:
1164         return ret;
1165 }
1166
1167 /*
1168  * Notes: the case when user wants us to deregister (with NULL as pointer)
1169  * and he isn't currently owner of notification, will be silently discarded.
1170  * It isn't explicitly defined in the POSIX.
1171  */
1172 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1173                 const struct sigevent __user *, u_notification)
1174 {
1175         int ret;
1176         struct fd f;
1177         struct sock *sock;
1178         struct inode *inode;
1179         struct sigevent notification;
1180         struct mqueue_inode_info *info;
1181         struct sk_buff *nc;
1182
1183         if (u_notification) {
1184                 if (copy_from_user(&notification, u_notification,
1185                                         sizeof(struct sigevent)))
1186                         return -EFAULT;
1187         }
1188
1189         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1190
1191         nc = NULL;
1192         sock = NULL;
1193         if (u_notification != NULL) {
1194                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1195                              notification.sigev_notify != SIGEV_SIGNAL &&
1196                              notification.sigev_notify != SIGEV_THREAD))
1197                         return -EINVAL;
1198                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1199                         !valid_signal(notification.sigev_signo)) {
1200                         return -EINVAL;
1201                 }
1202                 if (notification.sigev_notify == SIGEV_THREAD) {
1203                         long timeo;
1204
1205                         /* create the notify skb */
1206                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1207                         if (!nc) {
1208                                 ret = -ENOMEM;
1209                                 goto out;
1210                         }
1211                         if (copy_from_user(nc->data,
1212                                         notification.sigev_value.sival_ptr,
1213                                         NOTIFY_COOKIE_LEN)) {
1214                                 ret = -EFAULT;
1215                                 goto out;
1216                         }
1217
1218                         /* TODO: add a header? */
1219                         skb_put(nc, NOTIFY_COOKIE_LEN);
1220                         /* and attach it to the socket */
1221 retry:
1222                         f = fdget(notification.sigev_signo);
1223                         if (!f.file) {
1224                                 ret = -EBADF;
1225                                 goto out;
1226                         }
1227                         sock = netlink_getsockbyfilp(f.file);
1228                         fdput(f);
1229                         if (IS_ERR(sock)) {
1230                                 ret = PTR_ERR(sock);
1231                                 sock = NULL;
1232                                 goto out;
1233                         }
1234
1235                         timeo = MAX_SCHEDULE_TIMEOUT;
1236                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1237                         if (ret == 1)
1238                                 goto retry;
1239                         if (ret) {
1240                                 sock = NULL;
1241                                 nc = NULL;
1242                                 goto out;
1243                         }
1244                 }
1245         }
1246
1247         f = fdget(mqdes);
1248         if (!f.file) {
1249                 ret = -EBADF;
1250                 goto out;
1251         }
1252
1253         inode = file_inode(f.file);
1254         if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1255                 ret = -EBADF;
1256                 goto out_fput;
1257         }
1258         info = MQUEUE_I(inode);
1259
1260         ret = 0;
1261         spin_lock(&info->lock);
1262         if (u_notification == NULL) {
1263                 if (info->notify_owner == task_tgid(current)) {
1264                         remove_notification(info);
1265                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1266                 }
1267         } else if (info->notify_owner != NULL) {
1268                 ret = -EBUSY;
1269         } else {
1270                 switch (notification.sigev_notify) {
1271                 case SIGEV_NONE:
1272                         info->notify.sigev_notify = SIGEV_NONE;
1273                         break;
1274                 case SIGEV_THREAD:
1275                         info->notify_sock = sock;
1276                         info->notify_cookie = nc;
1277                         sock = NULL;
1278                         nc = NULL;
1279                         info->notify.sigev_notify = SIGEV_THREAD;
1280                         break;
1281                 case SIGEV_SIGNAL:
1282                         info->notify.sigev_signo = notification.sigev_signo;
1283                         info->notify.sigev_value = notification.sigev_value;
1284                         info->notify.sigev_notify = SIGEV_SIGNAL;
1285                         break;
1286                 }
1287
1288                 info->notify_owner = get_pid(task_tgid(current));
1289                 info->notify_user_ns = get_user_ns(current_user_ns());
1290                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1291         }
1292         spin_unlock(&info->lock);
1293 out_fput:
1294         fdput(f);
1295 out:
1296         if (sock) {
1297                 netlink_detachskb(sock, nc);
1298         } else if (nc) {
1299                 dev_kfree_skb(nc);
1300         }
1301         return ret;
1302 }
1303
1304 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1305                 const struct mq_attr __user *, u_mqstat,
1306                 struct mq_attr __user *, u_omqstat)
1307 {
1308         int ret;
1309         struct mq_attr mqstat, omqstat;
1310         struct fd f;
1311         struct inode *inode;
1312         struct mqueue_inode_info *info;
1313
1314         if (u_mqstat != NULL) {
1315                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1316                         return -EFAULT;
1317                 if (mqstat.mq_flags & (~O_NONBLOCK))
1318                         return -EINVAL;
1319         }
1320
1321         f = fdget(mqdes);
1322         if (!f.file) {
1323                 ret = -EBADF;
1324                 goto out;
1325         }
1326
1327         inode = file_inode(f.file);
1328         if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1329                 ret = -EBADF;
1330                 goto out_fput;
1331         }
1332         info = MQUEUE_I(inode);
1333
1334         spin_lock(&info->lock);
1335
1336         omqstat = info->attr;
1337         omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1338         if (u_mqstat) {
1339                 audit_mq_getsetattr(mqdes, &mqstat);
1340                 spin_lock(&f.file->f_lock);
1341                 if (mqstat.mq_flags & O_NONBLOCK)
1342                         f.file->f_flags |= O_NONBLOCK;
1343                 else
1344                         f.file->f_flags &= ~O_NONBLOCK;
1345                 spin_unlock(&f.file->f_lock);
1346
1347                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1348         }
1349
1350         spin_unlock(&info->lock);
1351
1352         ret = 0;
1353         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1354                                                 sizeof(struct mq_attr)))
1355                 ret = -EFAULT;
1356
1357 out_fput:
1358         fdput(f);
1359 out:
1360         return ret;
1361 }
1362
1363 static const struct inode_operations mqueue_dir_inode_operations = {
1364         .lookup = simple_lookup,
1365         .create = mqueue_create,
1366         .unlink = mqueue_unlink,
1367 };
1368
1369 static const struct file_operations mqueue_file_operations = {
1370         .flush = mqueue_flush_file,
1371         .poll = mqueue_poll_file,
1372         .read = mqueue_read_file,
1373         .llseek = default_llseek,
1374 };
1375
1376 static const struct super_operations mqueue_super_ops = {
1377         .alloc_inode = mqueue_alloc_inode,
1378         .destroy_inode = mqueue_destroy_inode,
1379         .evict_inode = mqueue_evict_inode,
1380         .statfs = simple_statfs,
1381 };
1382
1383 static struct file_system_type mqueue_fs_type = {
1384         .name = "mqueue",
1385         .mount = mqueue_mount,
1386         .kill_sb = kill_litter_super,
1387         .fs_flags = FS_USERNS_MOUNT,
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);