Linux Audio

Check our new training course

Loading...
v3.5.6
   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
  53struct posix_msg_tree_node {
  54	struct rb_node		rb_node;
  55	struct list_head	msg_list;
  56	int			priority;
  57};
  58
  59struct 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
  66struct 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
  88static const struct inode_operations mqueue_dir_inode_operations;
  89static const struct file_operations mqueue_file_operations;
  90static const struct super_operations mqueue_super_ops;
  91static void remove_notification(struct mqueue_inode_info *info);
  92
  93static struct kmem_cache *mqueue_inode_cachep;
  94
  95static struct ctl_table_header * mq_sysctl_table;
  96
  97static 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 */
 105static 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
 110static 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 */
 121static 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);
 152insert_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
 159static 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
 165try_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
 217static 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;
 304out_inode:
 305	iput(inode);
 306err:
 307	return ERR_PTR(ret);
 308}
 309
 310static 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
 330static 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
 339static 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
 346static 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
 356static 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
 362static void mqueue_destroy_inode(struct inode *inode)
 363{
 364	call_rcu(&inode->i_rcu, mqueue_i_callback);
 365}
 366
 367static 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
 415static int mqueue_create(struct inode *dir, struct dentry *dentry,
 416				umode_t mode, struct nameidata *nd)
 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;
 453out_unlock:
 454	spin_unlock(&mq_lock);
 455	if (ipc_ns)
 456		put_ipc_ns(ipc_ns);
 457	return error;
 458}
 459
 460static 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*/
 478static 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
 506static 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
 518static 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 */
 537static 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 */
 558static 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);
 595out_unlock:
 596	spin_unlock(&info->lock);
 597out:
 598	return retval;
 599}
 600
 601/*
 602 * Returns waiting task that should be serviced first or NULL if none exists
 603 */
 604static 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
 616static 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 */
 624static 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
 668static 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
 680static 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
 693static 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 */
 724static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 725			struct dentry *dentry, int oflag, umode_t mode,
 726			struct mq_attr *attr)
 727{
 728	const struct cred *cred = current_cred();
 729	struct file *result;
 730	int ret;
 731
 732	if (attr) {
 733		ret = mq_attr_ok(ipc_ns, attr);
 734		if (ret)
 735			goto out;
 736		/* store for use during create */
 737		dentry->d_fsdata = attr;
 738	} else {
 739		struct mq_attr def_attr;
 740
 741		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 742					 ipc_ns->mq_msg_default);
 743		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 744					  ipc_ns->mq_msgsize_default);
 745		ret = mq_attr_ok(ipc_ns, &def_attr);
 746		if (ret)
 747			goto out;
 748	}
 749
 750	mode &= ~current_umask();
 751	ret = mnt_want_write(ipc_ns->mq_mnt);
 752	if (ret)
 753		goto out;
 754	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 755	dentry->d_fsdata = NULL;
 756	if (ret)
 757		goto out_drop_write;
 758
 759	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 760	/*
 761	 * dentry_open() took a persistent mnt_want_write(),
 762	 * so we can now drop this one.
 763	 */
 764	mnt_drop_write(ipc_ns->mq_mnt);
 765	return result;
 766
 767out_drop_write:
 768	mnt_drop_write(ipc_ns->mq_mnt);
 769out:
 770	dput(dentry);
 771	mntput(ipc_ns->mq_mnt);
 772	return ERR_PTR(ret);
 773}
 774
 775/* Opens existing queue */
 776static struct file *do_open(struct ipc_namespace *ipc_ns,
 777				struct dentry *dentry, int oflag)
 778{
 779	int ret;
 780	const struct cred *cred = current_cred();
 781
 782	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 783						  MAY_READ | MAY_WRITE };
 784
 785	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 786		ret = -EINVAL;
 787		goto err;
 788	}
 789
 790	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 791		ret = -EACCES;
 792		goto err;
 793	}
 794
 795	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 796
 797err:
 798	dput(dentry);
 799	mntput(ipc_ns->mq_mnt);
 800	return ERR_PTR(ret);
 801}
 802
 803SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 804		struct mq_attr __user *, u_attr)
 805{
 806	struct dentry *dentry;
 807	struct file *filp;
 808	char *name;
 809	struct mq_attr attr;
 810	int fd, error;
 811	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 
 
 
 812
 813	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 814		return -EFAULT;
 815
 816	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 817
 818	if (IS_ERR(name = getname(u_name)))
 819		return PTR_ERR(name);
 820
 821	fd = get_unused_fd_flags(O_CLOEXEC);
 822	if (fd < 0)
 823		goto out_putname;
 824
 825	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 826	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 827	if (IS_ERR(dentry)) {
 828		error = PTR_ERR(dentry);
 
 
 829		goto out_putfd;
 830	}
 831	mntget(ipc_ns->mq_mnt);
 832
 833	if (oflag & O_CREAT) {
 834		if (dentry->d_inode) {	/* entry already exists */
 835			audit_inode(name, dentry);
 836			if (oflag & O_EXCL) {
 837				error = -EEXIST;
 838				goto out;
 839			}
 840			filp = do_open(ipc_ns, dentry, oflag);
 841		} else {
 842			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 843						dentry, oflag, mode,
 
 
 
 
 
 844						u_attr ? &attr : NULL);
 845		}
 846	} else {
 847		if (!dentry->d_inode) {
 848			error = -ENOENT;
 849			goto out;
 850		}
 851		audit_inode(name, dentry);
 852		filp = do_open(ipc_ns, dentry, oflag);
 853	}
 854
 855	if (IS_ERR(filp)) {
 
 
 856		error = PTR_ERR(filp);
 857		goto out_putfd;
 858	}
 859
 860	fd_install(fd, filp);
 861	goto out_upsem;
 862
 863out:
 864	dput(dentry);
 865	mntput(ipc_ns->mq_mnt);
 866out_putfd:
 867	put_unused_fd(fd);
 868	fd = error;
 869out_upsem:
 870	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 
 
 
 871out_putname:
 872	putname(name);
 873	return fd;
 874}
 875
 876SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 877{
 878	int err;
 879	char *name;
 880	struct dentry *dentry;
 881	struct inode *inode = NULL;
 882	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 
 883
 884	name = getname(u_name);
 885	if (IS_ERR(name))
 886		return PTR_ERR(name);
 887
 888	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 889			I_MUTEX_PARENT);
 890	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 
 
 
 
 891	if (IS_ERR(dentry)) {
 892		err = PTR_ERR(dentry);
 893		goto out_unlock;
 894	}
 895
 896	if (!dentry->d_inode) {
 
 897		err = -ENOENT;
 898		goto out_err;
 899	}
 900
 901	inode = dentry->d_inode;
 902	if (inode)
 903		ihold(inode);
 904	err = mnt_want_write(ipc_ns->mq_mnt);
 905	if (err)
 906		goto out_err;
 907	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 908	mnt_drop_write(ipc_ns->mq_mnt);
 909out_err:
 910	dput(dentry);
 911
 912out_unlock:
 913	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 914	putname(name);
 915	if (inode)
 916		iput(inode);
 
 
 
 917
 918	return err;
 919}
 920
 921/* Pipelined send and receive functions.
 922 *
 923 * If a receiver finds no waiting message, then it registers itself in the
 924 * list of waiting receivers. A sender checks that list before adding the new
 925 * message into the message array. If there is a waiting receiver, then it
 926 * bypasses the message array and directly hands the message over to the
 927 * receiver.
 928 * The receiver accepts the message and returns without grabbing the queue
 929 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 930 * are necessary. The same algorithm is used for sysv semaphores, see
 931 * ipc/sem.c for more details.
 
 
 
 
 932 *
 933 * The same algorithm is used for senders.
 934 */
 935
 936/* pipelined_send() - send a message directly to the task waiting in
 937 * sys_mq_timedreceive() (without inserting message into a queue).
 938 */
 939static inline void pipelined_send(struct mqueue_inode_info *info,
 
 940				  struct msg_msg *message,
 941				  struct ext_wait_queue *receiver)
 942{
 943	receiver->msg = message;
 944	list_del(&receiver->list);
 945	receiver->state = STATE_PENDING;
 946	wake_up_process(receiver->task);
 947	smp_wmb();
 
 
 
 
 
 
 948	receiver->state = STATE_READY;
 949}
 950
 951/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 952 * gets its message and put to the queue (we have one free place for sure). */
 953static inline void pipelined_receive(struct mqueue_inode_info *info)
 
 954{
 955	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 956
 957	if (!sender) {
 958		/* for poll */
 959		wake_up_interruptible(&info->wait_q);
 960		return;
 961	}
 962	if (msg_insert(sender->msg, info))
 963		return;
 
 964	list_del(&sender->list);
 965	sender->state = STATE_PENDING;
 966	wake_up_process(sender->task);
 967	smp_wmb();
 968	sender->state = STATE_READY;
 969}
 970
 971SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 972		size_t, msg_len, unsigned int, msg_prio,
 973		const struct timespec __user *, u_abs_timeout)
 974{
 975	struct file *filp;
 976	struct inode *inode;
 977	struct ext_wait_queue wait;
 978	struct ext_wait_queue *receiver;
 979	struct msg_msg *msg_ptr;
 980	struct mqueue_inode_info *info;
 981	ktime_t expires, *timeout = NULL;
 982	struct timespec ts;
 983	struct posix_msg_tree_node *new_leaf = NULL;
 984	int ret = 0;
 
 985
 986	if (u_abs_timeout) {
 987		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 988		if (res)
 989			return res;
 990		timeout = &expires;
 991	}
 992
 993	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 994		return -EINVAL;
 995
 996	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 997
 998	filp = fget(mqdes);
 999	if (unlikely(!filp)) {
1000		ret = -EBADF;
1001		goto out;
1002	}
1003
1004	inode = filp->f_path.dentry->d_inode;
1005	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1006		ret = -EBADF;
1007		goto out_fput;
1008	}
1009	info = MQUEUE_I(inode);
1010	audit_inode(NULL, filp->f_path.dentry);
1011
1012	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
1013		ret = -EBADF;
1014		goto out_fput;
1015	}
1016
1017	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1018		ret = -EMSGSIZE;
1019		goto out_fput;
1020	}
1021
1022	/* First try to allocate memory, before doing anything with
1023	 * existing queues. */
1024	msg_ptr = load_msg(u_msg_ptr, msg_len);
1025	if (IS_ERR(msg_ptr)) {
1026		ret = PTR_ERR(msg_ptr);
1027		goto out_fput;
1028	}
1029	msg_ptr->m_ts = msg_len;
1030	msg_ptr->m_type = msg_prio;
1031
1032	/*
1033	 * msg_insert really wants us to have a valid, spare node struct so
1034	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1035	 * fall back to that if necessary.
1036	 */
1037	if (!info->node_cache)
1038		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1039
1040	spin_lock(&info->lock);
1041
1042	if (!info->node_cache && new_leaf) {
1043		/* Save our speculative allocation into the cache */
1044		rb_init_node(&new_leaf->rb_node);
1045		INIT_LIST_HEAD(&new_leaf->msg_list);
1046		info->node_cache = new_leaf;
1047		info->qsize += sizeof(*new_leaf);
1048		new_leaf = NULL;
1049	} else {
1050		kfree(new_leaf);
1051	}
1052
1053	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1054		if (filp->f_flags & O_NONBLOCK) {
1055			ret = -EAGAIN;
1056		} else {
1057			wait.task = current;
1058			wait.msg = (void *) msg_ptr;
1059			wait.state = STATE_NONE;
1060			ret = wq_sleep(info, SEND, timeout, &wait);
1061			/*
1062			 * wq_sleep must be called with info->lock held, and
1063			 * returns with the lock released
1064			 */
1065			goto out_free;
1066		}
1067	} else {
1068		receiver = wq_get_first_waiter(info, RECV);
1069		if (receiver) {
1070			pipelined_send(info, msg_ptr, receiver);
1071		} else {
1072			/* adds message to the queue */
1073			ret = msg_insert(msg_ptr, info);
1074			if (ret)
1075				goto out_unlock;
1076			__do_notify(info);
1077		}
1078		inode->i_atime = inode->i_mtime = inode->i_ctime =
1079				CURRENT_TIME;
1080	}
1081out_unlock:
1082	spin_unlock(&info->lock);
 
1083out_free:
1084	if (ret)
1085		free_msg(msg_ptr);
1086out_fput:
1087	fput(filp);
1088out:
1089	return ret;
1090}
1091
1092SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1093		size_t, msg_len, unsigned int __user *, u_msg_prio,
1094		const struct timespec __user *, u_abs_timeout)
1095{
1096	ssize_t ret;
1097	struct msg_msg *msg_ptr;
1098	struct file *filp;
1099	struct inode *inode;
1100	struct mqueue_inode_info *info;
1101	struct ext_wait_queue wait;
1102	ktime_t expires, *timeout = NULL;
1103	struct timespec ts;
1104	struct posix_msg_tree_node *new_leaf = NULL;
1105
1106	if (u_abs_timeout) {
1107		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1108		if (res)
1109			return res;
1110		timeout = &expires;
1111	}
1112
1113	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1114
1115	filp = fget(mqdes);
1116	if (unlikely(!filp)) {
1117		ret = -EBADF;
1118		goto out;
1119	}
1120
1121	inode = filp->f_path.dentry->d_inode;
1122	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1123		ret = -EBADF;
1124		goto out_fput;
1125	}
1126	info = MQUEUE_I(inode);
1127	audit_inode(NULL, filp->f_path.dentry);
1128
1129	if (unlikely(!(filp->f_mode & FMODE_READ))) {
1130		ret = -EBADF;
1131		goto out_fput;
1132	}
1133
1134	/* checks if buffer is big enough */
1135	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1136		ret = -EMSGSIZE;
1137		goto out_fput;
1138	}
1139
1140	/*
1141	 * msg_insert really wants us to have a valid, spare node struct so
1142	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1143	 * fall back to that if necessary.
1144	 */
1145	if (!info->node_cache)
1146		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1147
1148	spin_lock(&info->lock);
1149
1150	if (!info->node_cache && new_leaf) {
1151		/* Save our speculative allocation into the cache */
1152		rb_init_node(&new_leaf->rb_node);
1153		INIT_LIST_HEAD(&new_leaf->msg_list);
1154		info->node_cache = new_leaf;
1155		info->qsize += sizeof(*new_leaf);
1156	} else {
1157		kfree(new_leaf);
1158	}
1159
1160	if (info->attr.mq_curmsgs == 0) {
1161		if (filp->f_flags & O_NONBLOCK) {
1162			spin_unlock(&info->lock);
1163			ret = -EAGAIN;
1164		} else {
1165			wait.task = current;
1166			wait.state = STATE_NONE;
1167			ret = wq_sleep(info, RECV, timeout, &wait);
1168			msg_ptr = wait.msg;
1169		}
1170	} else {
 
 
1171		msg_ptr = msg_get(info);
1172
1173		inode->i_atime = inode->i_mtime = inode->i_ctime =
1174				CURRENT_TIME;
1175
1176		/* There is now free space in queue. */
1177		pipelined_receive(info);
1178		spin_unlock(&info->lock);
 
1179		ret = 0;
1180	}
1181	if (ret == 0) {
1182		ret = msg_ptr->m_ts;
1183
1184		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1185			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1186			ret = -EFAULT;
1187		}
1188		free_msg(msg_ptr);
1189	}
1190out_fput:
1191	fput(filp);
1192out:
1193	return ret;
1194}
1195
1196/*
1197 * Notes: the case when user wants us to deregister (with NULL as pointer)
1198 * and he isn't currently owner of notification, will be silently discarded.
1199 * It isn't explicitly defined in the POSIX.
1200 */
1201SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1202		const struct sigevent __user *, u_notification)
1203{
1204	int ret;
1205	struct file *filp;
1206	struct sock *sock;
1207	struct inode *inode;
1208	struct sigevent notification;
1209	struct mqueue_inode_info *info;
1210	struct sk_buff *nc;
1211
1212	if (u_notification) {
1213		if (copy_from_user(&notification, u_notification,
1214					sizeof(struct sigevent)))
1215			return -EFAULT;
1216	}
1217
1218	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1219
1220	nc = NULL;
1221	sock = NULL;
1222	if (u_notification != NULL) {
1223		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1224			     notification.sigev_notify != SIGEV_SIGNAL &&
1225			     notification.sigev_notify != SIGEV_THREAD))
1226			return -EINVAL;
1227		if (notification.sigev_notify == SIGEV_SIGNAL &&
1228			!valid_signal(notification.sigev_signo)) {
1229			return -EINVAL;
1230		}
1231		if (notification.sigev_notify == SIGEV_THREAD) {
1232			long timeo;
1233
1234			/* create the notify skb */
1235			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1236			if (!nc) {
1237				ret = -ENOMEM;
1238				goto out;
1239			}
1240			if (copy_from_user(nc->data,
1241					notification.sigev_value.sival_ptr,
1242					NOTIFY_COOKIE_LEN)) {
1243				ret = -EFAULT;
1244				goto out;
1245			}
1246
1247			/* TODO: add a header? */
1248			skb_put(nc, NOTIFY_COOKIE_LEN);
1249			/* and attach it to the socket */
1250retry:
1251			filp = fget(notification.sigev_signo);
1252			if (!filp) {
1253				ret = -EBADF;
1254				goto out;
1255			}
1256			sock = netlink_getsockbyfilp(filp);
1257			fput(filp);
1258			if (IS_ERR(sock)) {
1259				ret = PTR_ERR(sock);
1260				sock = NULL;
1261				goto out;
1262			}
1263
1264			timeo = MAX_SCHEDULE_TIMEOUT;
1265			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1266			if (ret == 1)
1267				goto retry;
1268			if (ret) {
1269				sock = NULL;
1270				nc = NULL;
1271				goto out;
1272			}
1273		}
1274	}
1275
1276	filp = fget(mqdes);
1277	if (!filp) {
1278		ret = -EBADF;
1279		goto out;
1280	}
1281
1282	inode = filp->f_path.dentry->d_inode;
1283	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1284		ret = -EBADF;
1285		goto out_fput;
1286	}
1287	info = MQUEUE_I(inode);
1288
1289	ret = 0;
1290	spin_lock(&info->lock);
1291	if (u_notification == NULL) {
1292		if (info->notify_owner == task_tgid(current)) {
1293			remove_notification(info);
1294			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1295		}
1296	} else if (info->notify_owner != NULL) {
1297		ret = -EBUSY;
1298	} else {
1299		switch (notification.sigev_notify) {
1300		case SIGEV_NONE:
1301			info->notify.sigev_notify = SIGEV_NONE;
1302			break;
1303		case SIGEV_THREAD:
1304			info->notify_sock = sock;
1305			info->notify_cookie = nc;
1306			sock = NULL;
1307			nc = NULL;
1308			info->notify.sigev_notify = SIGEV_THREAD;
1309			break;
1310		case SIGEV_SIGNAL:
1311			info->notify.sigev_signo = notification.sigev_signo;
1312			info->notify.sigev_value = notification.sigev_value;
1313			info->notify.sigev_notify = SIGEV_SIGNAL;
1314			break;
1315		}
1316
1317		info->notify_owner = get_pid(task_tgid(current));
1318		info->notify_user_ns = get_user_ns(current_user_ns());
1319		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1320	}
1321	spin_unlock(&info->lock);
1322out_fput:
1323	fput(filp);
1324out:
1325	if (sock) {
1326		netlink_detachskb(sock, nc);
1327	} else if (nc) {
1328		dev_kfree_skb(nc);
1329	}
1330	return ret;
1331}
1332
1333SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1334		const struct mq_attr __user *, u_mqstat,
1335		struct mq_attr __user *, u_omqstat)
1336{
1337	int ret;
1338	struct mq_attr mqstat, omqstat;
1339	struct file *filp;
1340	struct inode *inode;
1341	struct mqueue_inode_info *info;
1342
1343	if (u_mqstat != NULL) {
1344		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1345			return -EFAULT;
1346		if (mqstat.mq_flags & (~O_NONBLOCK))
1347			return -EINVAL;
1348	}
1349
1350	filp = fget(mqdes);
1351	if (!filp) {
1352		ret = -EBADF;
1353		goto out;
1354	}
1355
1356	inode = filp->f_path.dentry->d_inode;
1357	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1358		ret = -EBADF;
1359		goto out_fput;
1360	}
1361	info = MQUEUE_I(inode);
1362
1363	spin_lock(&info->lock);
1364
1365	omqstat = info->attr;
1366	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1367	if (u_mqstat) {
1368		audit_mq_getsetattr(mqdes, &mqstat);
1369		spin_lock(&filp->f_lock);
1370		if (mqstat.mq_flags & O_NONBLOCK)
1371			filp->f_flags |= O_NONBLOCK;
1372		else
1373			filp->f_flags &= ~O_NONBLOCK;
1374		spin_unlock(&filp->f_lock);
1375
1376		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1377	}
1378
1379	spin_unlock(&info->lock);
1380
1381	ret = 0;
1382	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1383						sizeof(struct mq_attr)))
1384		ret = -EFAULT;
1385
1386out_fput:
1387	fput(filp);
1388out:
1389	return ret;
1390}
1391
1392static const struct inode_operations mqueue_dir_inode_operations = {
1393	.lookup = simple_lookup,
1394	.create = mqueue_create,
1395	.unlink = mqueue_unlink,
1396};
1397
1398static const struct file_operations mqueue_file_operations = {
1399	.flush = mqueue_flush_file,
1400	.poll = mqueue_poll_file,
1401	.read = mqueue_read_file,
1402	.llseek = default_llseek,
1403};
1404
1405static const struct super_operations mqueue_super_ops = {
1406	.alloc_inode = mqueue_alloc_inode,
1407	.destroy_inode = mqueue_destroy_inode,
1408	.evict_inode = mqueue_evict_inode,
1409	.statfs = simple_statfs,
1410};
1411
1412static struct file_system_type mqueue_fs_type = {
1413	.name = "mqueue",
1414	.mount = mqueue_mount,
1415	.kill_sb = kill_litter_super,
 
1416};
1417
1418int mq_init_ns(struct ipc_namespace *ns)
1419{
1420	ns->mq_queues_count  = 0;
1421	ns->mq_queues_max    = DFLT_QUEUESMAX;
1422	ns->mq_msg_max       = DFLT_MSGMAX;
1423	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1424	ns->mq_msg_default   = DFLT_MSG;
1425	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1426
1427	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1428	if (IS_ERR(ns->mq_mnt)) {
1429		int err = PTR_ERR(ns->mq_mnt);
1430		ns->mq_mnt = NULL;
1431		return err;
1432	}
1433	return 0;
1434}
1435
1436void mq_clear_sbinfo(struct ipc_namespace *ns)
1437{
1438	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1439}
1440
1441void mq_put_mnt(struct ipc_namespace *ns)
1442{
1443	kern_unmount(ns->mq_mnt);
1444}
1445
1446static int __init init_mqueue_fs(void)
1447{
1448	int error;
1449
1450	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1451				sizeof(struct mqueue_inode_info), 0,
1452				SLAB_HWCACHE_ALIGN, init_once);
1453	if (mqueue_inode_cachep == NULL)
1454		return -ENOMEM;
1455
1456	/* ignore failures - they are not fatal */
1457	mq_sysctl_table = mq_register_sysctl_table();
1458
1459	error = register_filesystem(&mqueue_fs_type);
1460	if (error)
1461		goto out_sysctl;
1462
1463	spin_lock_init(&mq_lock);
1464
1465	error = mq_init_ns(&init_ipc_ns);
1466	if (error)
1467		goto out_filesystem;
1468
1469	return 0;
1470
1471out_filesystem:
1472	unregister_filesystem(&mqueue_fs_type);
1473out_sysctl:
1474	if (mq_sysctl_table)
1475		unregister_sysctl_table(mq_sysctl_table);
1476	kmem_cache_destroy(mqueue_inode_cachep);
1477	return error;
1478}
1479
1480__initcall(init_mqueue_fs);
v4.6
   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_READY	1
 
  51
  52struct posix_msg_tree_node {
  53	struct rb_node		rb_node;
  54	struct list_head	msg_list;
  55	int			priority;
  56};
  57
  58struct ext_wait_queue {		/* queue of sleeping tasks */
  59	struct task_struct *task;
  60	struct list_head list;
  61	struct msg_msg *msg;	/* ptr of loaded message */
  62	int state;		/* one of STATE_* values */
  63};
  64
  65struct mqueue_inode_info {
  66	spinlock_t lock;
  67	struct inode vfs_inode;
  68	wait_queue_head_t wait_q;
  69
  70	struct rb_root msg_tree;
  71	struct posix_msg_tree_node *node_cache;
  72	struct mq_attr attr;
  73
  74	struct sigevent notify;
  75	struct pid *notify_owner;
  76	struct user_namespace *notify_user_ns;
  77	struct user_struct *user;	/* user who created, for accounting */
  78	struct sock *notify_sock;
  79	struct sk_buff *notify_cookie;
  80
  81	/* for tasks waiting for free space and messages, respectively */
  82	struct ext_wait_queue e_wait_q[2];
  83
  84	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  85};
  86
  87static const struct inode_operations mqueue_dir_inode_operations;
  88static const struct file_operations mqueue_file_operations;
  89static const struct super_operations mqueue_super_ops;
  90static void remove_notification(struct mqueue_inode_info *info);
  91
  92static struct kmem_cache *mqueue_inode_cachep;
  93
  94static struct ctl_table_header *mq_sysctl_table;
  95
  96static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  97{
  98	return container_of(inode, struct mqueue_inode_info, vfs_inode);
  99}
 100
 101/*
 102 * This routine should be called with the mq_lock held.
 103 */
 104static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
 105{
 106	return get_ipc_ns(inode->i_sb->s_fs_info);
 107}
 108
 109static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 110{
 111	struct ipc_namespace *ns;
 112
 113	spin_lock(&mq_lock);
 114	ns = __get_ns_from_inode(inode);
 115	spin_unlock(&mq_lock);
 116	return ns;
 117}
 118
 119/* Auxiliary functions to manipulate messages' list */
 120static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
 121{
 122	struct rb_node **p, *parent = NULL;
 123	struct posix_msg_tree_node *leaf;
 124
 125	p = &info->msg_tree.rb_node;
 126	while (*p) {
 127		parent = *p;
 128		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 129
 130		if (likely(leaf->priority == msg->m_type))
 131			goto insert_msg;
 132		else if (msg->m_type < leaf->priority)
 133			p = &(*p)->rb_left;
 134		else
 135			p = &(*p)->rb_right;
 136	}
 137	if (info->node_cache) {
 138		leaf = info->node_cache;
 139		info->node_cache = NULL;
 140	} else {
 141		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
 142		if (!leaf)
 143			return -ENOMEM;
 
 144		INIT_LIST_HEAD(&leaf->msg_list);
 
 145	}
 146	leaf->priority = msg->m_type;
 147	rb_link_node(&leaf->rb_node, parent, p);
 148	rb_insert_color(&leaf->rb_node, &info->msg_tree);
 149insert_msg:
 150	info->attr.mq_curmsgs++;
 151	info->qsize += msg->m_ts;
 152	list_add_tail(&msg->m_list, &leaf->msg_list);
 153	return 0;
 154}
 155
 156static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 157{
 158	struct rb_node **p, *parent = NULL;
 159	struct posix_msg_tree_node *leaf;
 160	struct msg_msg *msg;
 161
 162try_again:
 163	p = &info->msg_tree.rb_node;
 164	while (*p) {
 165		parent = *p;
 166		/*
 167		 * During insert, low priorities go to the left and high to the
 168		 * right.  On receive, we want the highest priorities first, so
 169		 * walk all the way to the right.
 170		 */
 171		p = &(*p)->rb_right;
 172	}
 173	if (!parent) {
 174		if (info->attr.mq_curmsgs) {
 175			pr_warn_once("Inconsistency in POSIX message queue, "
 176				     "no tree element, but supposedly messages "
 177				     "should exist!\n");
 178			info->attr.mq_curmsgs = 0;
 179		}
 180		return NULL;
 181	}
 182	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 183	if (unlikely(list_empty(&leaf->msg_list))) {
 184		pr_warn_once("Inconsistency in POSIX message queue, "
 185			     "empty leaf node but we haven't implemented "
 186			     "lazy leaf delete!\n");
 187		rb_erase(&leaf->rb_node, &info->msg_tree);
 188		if (info->node_cache) {
 
 189			kfree(leaf);
 190		} else {
 191			info->node_cache = leaf;
 192		}
 193		goto try_again;
 194	} else {
 195		msg = list_first_entry(&leaf->msg_list,
 196				       struct msg_msg, m_list);
 197		list_del(&msg->m_list);
 198		if (list_empty(&leaf->msg_list)) {
 199			rb_erase(&leaf->rb_node, &info->msg_tree);
 200			if (info->node_cache) {
 
 201				kfree(leaf);
 202			} else {
 203				info->node_cache = leaf;
 204			}
 205		}
 206	}
 207	info->attr.mq_curmsgs--;
 208	info->qsize -= msg->m_ts;
 209	return msg;
 210}
 211
 212static struct inode *mqueue_get_inode(struct super_block *sb,
 213		struct ipc_namespace *ipc_ns, umode_t mode,
 214		struct mq_attr *attr)
 215{
 216	struct user_struct *u = current_user();
 217	struct inode *inode;
 218	int ret = -ENOMEM;
 219
 220	inode = new_inode(sb);
 221	if (!inode)
 222		goto err;
 223
 224	inode->i_ino = get_next_ino();
 225	inode->i_mode = mode;
 226	inode->i_uid = current_fsuid();
 227	inode->i_gid = current_fsgid();
 228	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
 229
 230	if (S_ISREG(mode)) {
 231		struct mqueue_inode_info *info;
 232		unsigned long mq_bytes, mq_treesize;
 233
 234		inode->i_fop = &mqueue_file_operations;
 235		inode->i_size = FILENT_SIZE;
 236		/* mqueue specific info */
 237		info = MQUEUE_I(inode);
 238		spin_lock_init(&info->lock);
 239		init_waitqueue_head(&info->wait_q);
 240		INIT_LIST_HEAD(&info->e_wait_q[0].list);
 241		INIT_LIST_HEAD(&info->e_wait_q[1].list);
 242		info->notify_owner = NULL;
 243		info->notify_user_ns = NULL;
 244		info->qsize = 0;
 245		info->user = NULL;	/* set when all is ok */
 246		info->msg_tree = RB_ROOT;
 247		info->node_cache = NULL;
 248		memset(&info->attr, 0, sizeof(info->attr));
 249		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 250					   ipc_ns->mq_msg_default);
 251		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 252					    ipc_ns->mq_msgsize_default);
 253		if (attr) {
 254			info->attr.mq_maxmsg = attr->mq_maxmsg;
 255			info->attr.mq_msgsize = attr->mq_msgsize;
 256		}
 257		/*
 258		 * We used to allocate a static array of pointers and account
 259		 * the size of that array as well as one msg_msg struct per
 260		 * possible message into the queue size. That's no longer
 261		 * accurate as the queue is now an rbtree and will grow and
 262		 * shrink depending on usage patterns.  We can, however, still
 263		 * account one msg_msg struct per message, but the nodes are
 264		 * allocated depending on priority usage, and most programs
 265		 * only use one, or a handful, of priorities.  However, since
 266		 * this is pinned memory, we need to assume worst case, so
 267		 * that means the min(mq_maxmsg, max_priorities) * struct
 268		 * posix_msg_tree_node.
 269		 */
 270		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 271			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 272			sizeof(struct posix_msg_tree_node);
 273
 274		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 275					  info->attr.mq_msgsize);
 276
 277		spin_lock(&mq_lock);
 278		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 279		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 280			spin_unlock(&mq_lock);
 281			/* mqueue_evict_inode() releases info->messages */
 282			ret = -EMFILE;
 283			goto out_inode;
 284		}
 285		u->mq_bytes += mq_bytes;
 286		spin_unlock(&mq_lock);
 287
 288		/* all is ok */
 289		info->user = get_uid(u);
 290	} else if (S_ISDIR(mode)) {
 291		inc_nlink(inode);
 292		/* Some things misbehave if size == 0 on a directory */
 293		inode->i_size = 2 * DIRENT_SIZE;
 294		inode->i_op = &mqueue_dir_inode_operations;
 295		inode->i_fop = &simple_dir_operations;
 296	}
 297
 298	return inode;
 299out_inode:
 300	iput(inode);
 301err:
 302	return ERR_PTR(ret);
 303}
 304
 305static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 306{
 307	struct inode *inode;
 308	struct ipc_namespace *ns = data;
 309
 310	sb->s_blocksize = PAGE_SIZE;
 311	sb->s_blocksize_bits = PAGE_SHIFT;
 312	sb->s_magic = MQUEUE_MAGIC;
 313	sb->s_op = &mqueue_super_ops;
 314
 315	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 316	if (IS_ERR(inode))
 317		return PTR_ERR(inode);
 318
 319	sb->s_root = d_make_root(inode);
 320	if (!sb->s_root)
 321		return -ENOMEM;
 322	return 0;
 323}
 324
 325static struct dentry *mqueue_mount(struct file_system_type *fs_type,
 326			 int flags, const char *dev_name,
 327			 void *data)
 328{
 329	if (!(flags & MS_KERNMOUNT)) {
 330		struct ipc_namespace *ns = current->nsproxy->ipc_ns;
 331		/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
 332		 * over the ipc namespace.
 333		 */
 334		if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
 335			return ERR_PTR(-EPERM);
 336
 337		data = ns;
 338	}
 339	return mount_ns(fs_type, flags, data, mqueue_fill_super);
 340}
 341
 342static void init_once(void *foo)
 343{
 344	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 345
 346	inode_init_once(&p->vfs_inode);
 347}
 348
 349static struct inode *mqueue_alloc_inode(struct super_block *sb)
 350{
 351	struct mqueue_inode_info *ei;
 352
 353	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 354	if (!ei)
 355		return NULL;
 356	return &ei->vfs_inode;
 357}
 358
 359static void mqueue_i_callback(struct rcu_head *head)
 360{
 361	struct inode *inode = container_of(head, struct inode, i_rcu);
 362	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 363}
 364
 365static void mqueue_destroy_inode(struct inode *inode)
 366{
 367	call_rcu(&inode->i_rcu, mqueue_i_callback);
 368}
 369
 370static void mqueue_evict_inode(struct inode *inode)
 371{
 372	struct mqueue_inode_info *info;
 373	struct user_struct *user;
 374	unsigned long mq_bytes, mq_treesize;
 375	struct ipc_namespace *ipc_ns;
 376	struct msg_msg *msg;
 377
 378	clear_inode(inode);
 379
 380	if (S_ISDIR(inode->i_mode))
 381		return;
 382
 383	ipc_ns = get_ns_from_inode(inode);
 384	info = MQUEUE_I(inode);
 385	spin_lock(&info->lock);
 386	while ((msg = msg_get(info)) != NULL)
 387		free_msg(msg);
 388	kfree(info->node_cache);
 389	spin_unlock(&info->lock);
 390
 391	/* Total amount of bytes accounted for the mqueue */
 392	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 393		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 394		sizeof(struct posix_msg_tree_node);
 395
 396	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 397				  info->attr.mq_msgsize);
 398
 399	user = info->user;
 400	if (user) {
 401		spin_lock(&mq_lock);
 402		user->mq_bytes -= mq_bytes;
 403		/*
 404		 * get_ns_from_inode() ensures that the
 405		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 406		 * to which we now hold a reference, or it is NULL.
 407		 * We can't put it here under mq_lock, though.
 408		 */
 409		if (ipc_ns)
 410			ipc_ns->mq_queues_count--;
 411		spin_unlock(&mq_lock);
 412		free_uid(user);
 413	}
 414	if (ipc_ns)
 415		put_ipc_ns(ipc_ns);
 416}
 417
 418static int mqueue_create(struct inode *dir, struct dentry *dentry,
 419				umode_t mode, bool excl)
 420{
 421	struct inode *inode;
 422	struct mq_attr *attr = dentry->d_fsdata;
 423	int error;
 424	struct ipc_namespace *ipc_ns;
 425
 426	spin_lock(&mq_lock);
 427	ipc_ns = __get_ns_from_inode(dir);
 428	if (!ipc_ns) {
 429		error = -EACCES;
 430		goto out_unlock;
 431	}
 432
 433	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 434	    !capable(CAP_SYS_RESOURCE)) {
 435		error = -ENOSPC;
 436		goto out_unlock;
 437	}
 438	ipc_ns->mq_queues_count++;
 439	spin_unlock(&mq_lock);
 440
 441	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 442	if (IS_ERR(inode)) {
 443		error = PTR_ERR(inode);
 444		spin_lock(&mq_lock);
 445		ipc_ns->mq_queues_count--;
 446		goto out_unlock;
 447	}
 448
 449	put_ipc_ns(ipc_ns);
 450	dir->i_size += DIRENT_SIZE;
 451	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 452
 453	d_instantiate(dentry, inode);
 454	dget(dentry);
 455	return 0;
 456out_unlock:
 457	spin_unlock(&mq_lock);
 458	if (ipc_ns)
 459		put_ipc_ns(ipc_ns);
 460	return error;
 461}
 462
 463static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 464{
 465	struct inode *inode = d_inode(dentry);
 466
 467	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 468	dir->i_size -= DIRENT_SIZE;
 469	drop_nlink(inode);
 470	dput(dentry);
 471	return 0;
 472}
 473
 474/*
 475*	This is routine for system read from queue file.
 476*	To avoid mess with doing here some sort of mq_receive we allow
 477*	to read only queue size & notification info (the only values
 478*	that are interesting from user point of view and aren't accessible
 479*	through std routines)
 480*/
 481static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 482				size_t count, loff_t *off)
 483{
 484	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 485	char buffer[FILENT_SIZE];
 486	ssize_t ret;
 487
 488	spin_lock(&info->lock);
 489	snprintf(buffer, sizeof(buffer),
 490			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 491			info->qsize,
 492			info->notify_owner ? info->notify.sigev_notify : 0,
 493			(info->notify_owner &&
 494			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
 495				info->notify.sigev_signo : 0,
 496			pid_vnr(info->notify_owner));
 497	spin_unlock(&info->lock);
 498	buffer[sizeof(buffer)-1] = '\0';
 499
 500	ret = simple_read_from_buffer(u_data, count, off, buffer,
 501				strlen(buffer));
 502	if (ret <= 0)
 503		return ret;
 504
 505	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
 506	return ret;
 507}
 508
 509static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 510{
 511	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 512
 513	spin_lock(&info->lock);
 514	if (task_tgid(current) == info->notify_owner)
 515		remove_notification(info);
 516
 517	spin_unlock(&info->lock);
 518	return 0;
 519}
 520
 521static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 522{
 523	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 524	int retval = 0;
 525
 526	poll_wait(filp, &info->wait_q, poll_tab);
 527
 528	spin_lock(&info->lock);
 529	if (info->attr.mq_curmsgs)
 530		retval = POLLIN | POLLRDNORM;
 531
 532	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 533		retval |= POLLOUT | POLLWRNORM;
 534	spin_unlock(&info->lock);
 535
 536	return retval;
 537}
 538
 539/* Adds current to info->e_wait_q[sr] before element with smaller prio */
 540static void wq_add(struct mqueue_inode_info *info, int sr,
 541			struct ext_wait_queue *ewp)
 542{
 543	struct ext_wait_queue *walk;
 544
 545	ewp->task = current;
 546
 547	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 548		if (walk->task->static_prio <= current->static_prio) {
 549			list_add_tail(&ewp->list, &walk->list);
 550			return;
 551		}
 552	}
 553	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 554}
 555
 556/*
 557 * Puts current task to sleep. Caller must hold queue lock. After return
 558 * lock isn't held.
 559 * sr: SEND or RECV
 560 */
 561static int wq_sleep(struct mqueue_inode_info *info, int sr,
 562		    ktime_t *timeout, struct ext_wait_queue *ewp)
 563{
 564	int retval;
 565	signed long time;
 566
 567	wq_add(info, sr, ewp);
 568
 569	for (;;) {
 570		__set_current_state(TASK_INTERRUPTIBLE);
 571
 572		spin_unlock(&info->lock);
 573		time = schedule_hrtimeout_range_clock(timeout, 0,
 574			HRTIMER_MODE_ABS, CLOCK_REALTIME);
 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);
 595out_unlock:
 596	spin_unlock(&info->lock);
 597out:
 598	return retval;
 599}
 600
 601/*
 602 * Returns waiting task that should be serviced first or NULL if none exists
 603 */
 604static 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
 616static 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 */
 624static 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
 668static 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
 680static 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
 693static 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 */
 724static 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 */
 758static 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(d_inode(path->dentry), acc))
 767		return ERR_PTR(-EACCES);
 768	return dentry_open(path, oflag, current_cred());
 
 
 
 
 
 
 
 
 
 
 769}
 770
 771SYSCALL_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	struct filename *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	inode_lock(d_inode(root));
 799	path.dentry = lookup_one_len(name->name, root, strlen(name->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 (d_really_is_positive(path.dentry)) {	/* entry already exists */
 808			audit_inode(name, path.dentry, 0);
 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			audit_inode_parent_hidden(name, root);
 820			filp = do_create(ipc_ns, d_inode(root),
 821						&path, oflag, mode,
 822						u_attr ? &attr : NULL);
 823		}
 824	} else {
 825		if (d_really_is_negative(path.dentry)) {
 826			error = -ENOENT;
 827			goto out;
 828		}
 829		audit_inode(name, path.dentry, 0);
 830		filp = do_open(&path, oflag);
 831	}
 832
 833	if (!IS_ERR(filp))
 834		fd_install(fd, filp);
 835	else
 836		error = PTR_ERR(filp);
 
 
 
 
 
 
 837out:
 838	path_put(&path);
 
 839out_putfd:
 840	if (error) {
 841		put_unused_fd(fd);
 842		fd = error;
 843	}
 844	inode_unlock(d_inode(root));
 845	if (!ro)
 846		mnt_drop_write(mnt);
 847out_putname:
 848	putname(name);
 849	return fd;
 850}
 851
 852SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 853{
 854	int err;
 855	struct filename *name;
 856	struct dentry *dentry;
 857	struct inode *inode = NULL;
 858	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 859	struct vfsmount *mnt = ipc_ns->mq_mnt;
 860
 861	name = getname(u_name);
 862	if (IS_ERR(name))
 863		return PTR_ERR(name);
 864
 865	audit_inode_parent_hidden(name, mnt->mnt_root);
 866	err = mnt_want_write(mnt);
 867	if (err)
 868		goto out_name;
 869	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
 870	dentry = lookup_one_len(name->name, mnt->mnt_root,
 871				strlen(name->name));
 872	if (IS_ERR(dentry)) {
 873		err = PTR_ERR(dentry);
 874		goto out_unlock;
 875	}
 876
 877	inode = d_inode(dentry);
 878	if (!inode) {
 879		err = -ENOENT;
 880	} else {
 
 
 
 
 881		ihold(inode);
 882		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
 883	}
 
 
 
 
 884	dput(dentry);
 885
 886out_unlock:
 887	inode_unlock(d_inode(mnt->mnt_root));
 
 888	if (inode)
 889		iput(inode);
 890	mnt_drop_write(mnt);
 891out_name:
 892	putname(name);
 893
 894	return err;
 895}
 896
 897/* Pipelined send and receive functions.
 898 *
 899 * If a receiver finds no waiting message, then it registers itself in the
 900 * list of waiting receivers. A sender checks that list before adding the new
 901 * message into the message array. If there is a waiting receiver, then it
 902 * bypasses the message array and directly hands the message over to the
 903 * receiver. The receiver accepts the message and returns without grabbing the
 904 * queue spinlock:
 905 *
 906 * - Set pointer to message.
 907 * - Queue the receiver task for later wakeup (without the info->lock).
 908 * - Update its state to STATE_READY. Now the receiver can continue.
 909 * - Wake up the process after the lock is dropped. Should the process wake up
 910 *   before this wakeup (due to a timeout or a signal) it will either see
 911 *   STATE_READY and continue or acquire the lock to check the state again.
 912 *
 913 * The same algorithm is used for senders.
 914 */
 915
 916/* pipelined_send() - send a message directly to the task waiting in
 917 * sys_mq_timedreceive() (without inserting message into a queue).
 918 */
 919static inline void pipelined_send(struct wake_q_head *wake_q,
 920				  struct mqueue_inode_info *info,
 921				  struct msg_msg *message,
 922				  struct ext_wait_queue *receiver)
 923{
 924	receiver->msg = message;
 925	list_del(&receiver->list);
 926	wake_q_add(wake_q, receiver->task);
 927	/*
 928	 * Rely on the implicit cmpxchg barrier from wake_q_add such
 929	 * that we can ensure that updating receiver->state is the last
 930	 * write operation: As once set, the receiver can continue,
 931	 * and if we don't have the reference count from the wake_q,
 932	 * yet, at that point we can later have a use-after-free
 933	 * condition and bogus wakeup.
 934	 */
 935	receiver->state = STATE_READY;
 936}
 937
 938/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 939 * gets its message and put to the queue (we have one free place for sure). */
 940static inline void pipelined_receive(struct wake_q_head *wake_q,
 941				     struct mqueue_inode_info *info)
 942{
 943	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 944
 945	if (!sender) {
 946		/* for poll */
 947		wake_up_interruptible(&info->wait_q);
 948		return;
 949	}
 950	if (msg_insert(sender->msg, info))
 951		return;
 952
 953	list_del(&sender->list);
 954	wake_q_add(wake_q, sender->task);
 
 
 955	sender->state = STATE_READY;
 956}
 957
 958SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 959		size_t, msg_len, unsigned int, msg_prio,
 960		const struct timespec __user *, u_abs_timeout)
 961{
 962	struct fd f;
 963	struct inode *inode;
 964	struct ext_wait_queue wait;
 965	struct ext_wait_queue *receiver;
 966	struct msg_msg *msg_ptr;
 967	struct mqueue_inode_info *info;
 968	ktime_t expires, *timeout = NULL;
 969	struct timespec ts;
 970	struct posix_msg_tree_node *new_leaf = NULL;
 971	int ret = 0;
 972	WAKE_Q(wake_q);
 973
 974	if (u_abs_timeout) {
 975		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 976		if (res)
 977			return res;
 978		timeout = &expires;
 979	}
 980
 981	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 982		return -EINVAL;
 983
 984	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 985
 986	f = fdget(mqdes);
 987	if (unlikely(!f.file)) {
 988		ret = -EBADF;
 989		goto out;
 990	}
 991
 992	inode = file_inode(f.file);
 993	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
 994		ret = -EBADF;
 995		goto out_fput;
 996	}
 997	info = MQUEUE_I(inode);
 998	audit_file(f.file);
 999
1000	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1001		ret = -EBADF;
1002		goto out_fput;
1003	}
1004
1005	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1006		ret = -EMSGSIZE;
1007		goto out_fput;
1008	}
1009
1010	/* First try to allocate memory, before doing anything with
1011	 * existing queues. */
1012	msg_ptr = load_msg(u_msg_ptr, msg_len);
1013	if (IS_ERR(msg_ptr)) {
1014		ret = PTR_ERR(msg_ptr);
1015		goto out_fput;
1016	}
1017	msg_ptr->m_ts = msg_len;
1018	msg_ptr->m_type = msg_prio;
1019
1020	/*
1021	 * msg_insert really wants us to have a valid, spare node struct so
1022	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1023	 * fall back to that if necessary.
1024	 */
1025	if (!info->node_cache)
1026		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1027
1028	spin_lock(&info->lock);
1029
1030	if (!info->node_cache && new_leaf) {
1031		/* Save our speculative allocation into the cache */
 
1032		INIT_LIST_HEAD(&new_leaf->msg_list);
1033		info->node_cache = new_leaf;
 
1034		new_leaf = NULL;
1035	} else {
1036		kfree(new_leaf);
1037	}
1038
1039	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1040		if (f.file->f_flags & O_NONBLOCK) {
1041			ret = -EAGAIN;
1042		} else {
1043			wait.task = current;
1044			wait.msg = (void *) msg_ptr;
1045			wait.state = STATE_NONE;
1046			ret = wq_sleep(info, SEND, timeout, &wait);
1047			/*
1048			 * wq_sleep must be called with info->lock held, and
1049			 * returns with the lock released
1050			 */
1051			goto out_free;
1052		}
1053	} else {
1054		receiver = wq_get_first_waiter(info, RECV);
1055		if (receiver) {
1056			pipelined_send(&wake_q, info, msg_ptr, receiver);
1057		} else {
1058			/* adds message to the queue */
1059			ret = msg_insert(msg_ptr, info);
1060			if (ret)
1061				goto out_unlock;
1062			__do_notify(info);
1063		}
1064		inode->i_atime = inode->i_mtime = inode->i_ctime =
1065				CURRENT_TIME;
1066	}
1067out_unlock:
1068	spin_unlock(&info->lock);
1069	wake_up_q(&wake_q);
1070out_free:
1071	if (ret)
1072		free_msg(msg_ptr);
1073out_fput:
1074	fdput(f);
1075out:
1076	return ret;
1077}
1078
1079SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1080		size_t, msg_len, unsigned int __user *, u_msg_prio,
1081		const struct timespec __user *, u_abs_timeout)
1082{
1083	ssize_t ret;
1084	struct msg_msg *msg_ptr;
1085	struct fd f;
1086	struct inode *inode;
1087	struct mqueue_inode_info *info;
1088	struct ext_wait_queue wait;
1089	ktime_t expires, *timeout = NULL;
1090	struct timespec ts;
1091	struct posix_msg_tree_node *new_leaf = NULL;
1092
1093	if (u_abs_timeout) {
1094		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1095		if (res)
1096			return res;
1097		timeout = &expires;
1098	}
1099
1100	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1101
1102	f = fdget(mqdes);
1103	if (unlikely(!f.file)) {
1104		ret = -EBADF;
1105		goto out;
1106	}
1107
1108	inode = file_inode(f.file);
1109	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1110		ret = -EBADF;
1111		goto out_fput;
1112	}
1113	info = MQUEUE_I(inode);
1114	audit_file(f.file);
1115
1116	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1117		ret = -EBADF;
1118		goto out_fput;
1119	}
1120
1121	/* checks if buffer is big enough */
1122	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1123		ret = -EMSGSIZE;
1124		goto out_fput;
1125	}
1126
1127	/*
1128	 * msg_insert really wants us to have a valid, spare node struct so
1129	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1130	 * fall back to that if necessary.
1131	 */
1132	if (!info->node_cache)
1133		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1134
1135	spin_lock(&info->lock);
1136
1137	if (!info->node_cache && new_leaf) {
1138		/* Save our speculative allocation into the cache */
 
1139		INIT_LIST_HEAD(&new_leaf->msg_list);
1140		info->node_cache = new_leaf;
 
1141	} else {
1142		kfree(new_leaf);
1143	}
1144
1145	if (info->attr.mq_curmsgs == 0) {
1146		if (f.file->f_flags & O_NONBLOCK) {
1147			spin_unlock(&info->lock);
1148			ret = -EAGAIN;
1149		} else {
1150			wait.task = current;
1151			wait.state = STATE_NONE;
1152			ret = wq_sleep(info, RECV, timeout, &wait);
1153			msg_ptr = wait.msg;
1154		}
1155	} else {
1156		WAKE_Q(wake_q);
1157
1158		msg_ptr = msg_get(info);
1159
1160		inode->i_atime = inode->i_mtime = inode->i_ctime =
1161				CURRENT_TIME;
1162
1163		/* There is now free space in queue. */
1164		pipelined_receive(&wake_q, info);
1165		spin_unlock(&info->lock);
1166		wake_up_q(&wake_q);
1167		ret = 0;
1168	}
1169	if (ret == 0) {
1170		ret = msg_ptr->m_ts;
1171
1172		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1173			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1174			ret = -EFAULT;
1175		}
1176		free_msg(msg_ptr);
1177	}
1178out_fput:
1179	fdput(f);
1180out:
1181	return ret;
1182}
1183
1184/*
1185 * Notes: the case when user wants us to deregister (with NULL as pointer)
1186 * and he isn't currently owner of notification, will be silently discarded.
1187 * It isn't explicitly defined in the POSIX.
1188 */
1189SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1190		const struct sigevent __user *, u_notification)
1191{
1192	int ret;
1193	struct fd f;
1194	struct sock *sock;
1195	struct inode *inode;
1196	struct sigevent notification;
1197	struct mqueue_inode_info *info;
1198	struct sk_buff *nc;
1199
1200	if (u_notification) {
1201		if (copy_from_user(&notification, u_notification,
1202					sizeof(struct sigevent)))
1203			return -EFAULT;
1204	}
1205
1206	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1207
1208	nc = NULL;
1209	sock = NULL;
1210	if (u_notification != NULL) {
1211		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1212			     notification.sigev_notify != SIGEV_SIGNAL &&
1213			     notification.sigev_notify != SIGEV_THREAD))
1214			return -EINVAL;
1215		if (notification.sigev_notify == SIGEV_SIGNAL &&
1216			!valid_signal(notification.sigev_signo)) {
1217			return -EINVAL;
1218		}
1219		if (notification.sigev_notify == SIGEV_THREAD) {
1220			long timeo;
1221
1222			/* create the notify skb */
1223			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1224			if (!nc) {
1225				ret = -ENOMEM;
1226				goto out;
1227			}
1228			if (copy_from_user(nc->data,
1229					notification.sigev_value.sival_ptr,
1230					NOTIFY_COOKIE_LEN)) {
1231				ret = -EFAULT;
1232				goto out;
1233			}
1234
1235			/* TODO: add a header? */
1236			skb_put(nc, NOTIFY_COOKIE_LEN);
1237			/* and attach it to the socket */
1238retry:
1239			f = fdget(notification.sigev_signo);
1240			if (!f.file) {
1241				ret = -EBADF;
1242				goto out;
1243			}
1244			sock = netlink_getsockbyfilp(f.file);
1245			fdput(f);
1246			if (IS_ERR(sock)) {
1247				ret = PTR_ERR(sock);
1248				sock = NULL;
1249				goto out;
1250			}
1251
1252			timeo = MAX_SCHEDULE_TIMEOUT;
1253			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1254			if (ret == 1)
1255				goto retry;
1256			if (ret) {
1257				sock = NULL;
1258				nc = NULL;
1259				goto out;
1260			}
1261		}
1262	}
1263
1264	f = fdget(mqdes);
1265	if (!f.file) {
1266		ret = -EBADF;
1267		goto out;
1268	}
1269
1270	inode = file_inode(f.file);
1271	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1272		ret = -EBADF;
1273		goto out_fput;
1274	}
1275	info = MQUEUE_I(inode);
1276
1277	ret = 0;
1278	spin_lock(&info->lock);
1279	if (u_notification == NULL) {
1280		if (info->notify_owner == task_tgid(current)) {
1281			remove_notification(info);
1282			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1283		}
1284	} else if (info->notify_owner != NULL) {
1285		ret = -EBUSY;
1286	} else {
1287		switch (notification.sigev_notify) {
1288		case SIGEV_NONE:
1289			info->notify.sigev_notify = SIGEV_NONE;
1290			break;
1291		case SIGEV_THREAD:
1292			info->notify_sock = sock;
1293			info->notify_cookie = nc;
1294			sock = NULL;
1295			nc = NULL;
1296			info->notify.sigev_notify = SIGEV_THREAD;
1297			break;
1298		case SIGEV_SIGNAL:
1299			info->notify.sigev_signo = notification.sigev_signo;
1300			info->notify.sigev_value = notification.sigev_value;
1301			info->notify.sigev_notify = SIGEV_SIGNAL;
1302			break;
1303		}
1304
1305		info->notify_owner = get_pid(task_tgid(current));
1306		info->notify_user_ns = get_user_ns(current_user_ns());
1307		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1308	}
1309	spin_unlock(&info->lock);
1310out_fput:
1311	fdput(f);
1312out:
1313	if (sock)
1314		netlink_detachskb(sock, nc);
1315	else if (nc)
1316		dev_kfree_skb(nc);
1317
1318	return ret;
1319}
1320
1321SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1322		const struct mq_attr __user *, u_mqstat,
1323		struct mq_attr __user *, u_omqstat)
1324{
1325	int ret;
1326	struct mq_attr mqstat, omqstat;
1327	struct fd f;
1328	struct inode *inode;
1329	struct mqueue_inode_info *info;
1330
1331	if (u_mqstat != NULL) {
1332		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1333			return -EFAULT;
1334		if (mqstat.mq_flags & (~O_NONBLOCK))
1335			return -EINVAL;
1336	}
1337
1338	f = fdget(mqdes);
1339	if (!f.file) {
1340		ret = -EBADF;
1341		goto out;
1342	}
1343
1344	inode = file_inode(f.file);
1345	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1346		ret = -EBADF;
1347		goto out_fput;
1348	}
1349	info = MQUEUE_I(inode);
1350
1351	spin_lock(&info->lock);
1352
1353	omqstat = info->attr;
1354	omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1355	if (u_mqstat) {
1356		audit_mq_getsetattr(mqdes, &mqstat);
1357		spin_lock(&f.file->f_lock);
1358		if (mqstat.mq_flags & O_NONBLOCK)
1359			f.file->f_flags |= O_NONBLOCK;
1360		else
1361			f.file->f_flags &= ~O_NONBLOCK;
1362		spin_unlock(&f.file->f_lock);
1363
1364		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1365	}
1366
1367	spin_unlock(&info->lock);
1368
1369	ret = 0;
1370	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1371						sizeof(struct mq_attr)))
1372		ret = -EFAULT;
1373
1374out_fput:
1375	fdput(f);
1376out:
1377	return ret;
1378}
1379
1380static const struct inode_operations mqueue_dir_inode_operations = {
1381	.lookup = simple_lookup,
1382	.create = mqueue_create,
1383	.unlink = mqueue_unlink,
1384};
1385
1386static const struct file_operations mqueue_file_operations = {
1387	.flush = mqueue_flush_file,
1388	.poll = mqueue_poll_file,
1389	.read = mqueue_read_file,
1390	.llseek = default_llseek,
1391};
1392
1393static const struct super_operations mqueue_super_ops = {
1394	.alloc_inode = mqueue_alloc_inode,
1395	.destroy_inode = mqueue_destroy_inode,
1396	.evict_inode = mqueue_evict_inode,
1397	.statfs = simple_statfs,
1398};
1399
1400static struct file_system_type mqueue_fs_type = {
1401	.name = "mqueue",
1402	.mount = mqueue_mount,
1403	.kill_sb = kill_litter_super,
1404	.fs_flags = FS_USERNS_MOUNT,
1405};
1406
1407int mq_init_ns(struct ipc_namespace *ns)
1408{
1409	ns->mq_queues_count  = 0;
1410	ns->mq_queues_max    = DFLT_QUEUESMAX;
1411	ns->mq_msg_max       = DFLT_MSGMAX;
1412	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1413	ns->mq_msg_default   = DFLT_MSG;
1414	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1415
1416	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1417	if (IS_ERR(ns->mq_mnt)) {
1418		int err = PTR_ERR(ns->mq_mnt);
1419		ns->mq_mnt = NULL;
1420		return err;
1421	}
1422	return 0;
1423}
1424
1425void mq_clear_sbinfo(struct ipc_namespace *ns)
1426{
1427	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1428}
1429
1430void mq_put_mnt(struct ipc_namespace *ns)
1431{
1432	kern_unmount(ns->mq_mnt);
1433}
1434
1435static int __init init_mqueue_fs(void)
1436{
1437	int error;
1438
1439	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1440				sizeof(struct mqueue_inode_info), 0,
1441				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1442	if (mqueue_inode_cachep == NULL)
1443		return -ENOMEM;
1444
1445	/* ignore failures - they are not fatal */
1446	mq_sysctl_table = mq_register_sysctl_table();
1447
1448	error = register_filesystem(&mqueue_fs_type);
1449	if (error)
1450		goto out_sysctl;
1451
1452	spin_lock_init(&mq_lock);
1453
1454	error = mq_init_ns(&init_ipc_ns);
1455	if (error)
1456		goto out_filesystem;
1457
1458	return 0;
1459
1460out_filesystem:
1461	unregister_filesystem(&mqueue_fs_type);
1462out_sysctl:
1463	if (mq_sysctl_table)
1464		unregister_sysctl_table(mq_sysctl_table);
1465	kmem_cache_destroy(mqueue_inode_cachep);
1466	return error;
1467}
1468
1469device_initcall(init_mqueue_fs);