Linux Audio

Check our new training course

Loading...
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);
v3.15
   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		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);
 151insert_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
 158static 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
 164try_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
 216static 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;
 303out_inode:
 304	iput(inode);
 305err:
 306	return ERR_PTR(ret);
 307}
 308
 309static 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
 329static 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		struct ipc_namespace *ns = current->nsproxy->ipc_ns;
 335		/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
 336		 * over the ipc namespace.
 337		 */
 338		if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
 339			return ERR_PTR(-EPERM);
 340
 341		data = ns;
 342	}
 343	return mount_ns(fs_type, flags, data, mqueue_fill_super);
 344}
 345
 346static void init_once(void *foo)
 347{
 348	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 349
 350	inode_init_once(&p->vfs_inode);
 351}
 352
 353static struct inode *mqueue_alloc_inode(struct super_block *sb)
 354{
 355	struct mqueue_inode_info *ei;
 356
 357	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 358	if (!ei)
 359		return NULL;
 360	return &ei->vfs_inode;
 361}
 362
 363static void mqueue_i_callback(struct rcu_head *head)
 364{
 365	struct inode *inode = container_of(head, struct inode, i_rcu);
 366	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 367}
 368
 369static void mqueue_destroy_inode(struct inode *inode)
 370{
 371	call_rcu(&inode->i_rcu, mqueue_i_callback);
 372}
 373
 374static void mqueue_evict_inode(struct inode *inode)
 375{
 376	struct mqueue_inode_info *info;
 377	struct user_struct *user;
 378	unsigned long mq_bytes, mq_treesize;
 379	struct ipc_namespace *ipc_ns;
 380	struct msg_msg *msg;
 381
 382	clear_inode(inode);
 383
 384	if (S_ISDIR(inode->i_mode))
 385		return;
 386
 387	ipc_ns = get_ns_from_inode(inode);
 388	info = MQUEUE_I(inode);
 389	spin_lock(&info->lock);
 390	while ((msg = msg_get(info)) != NULL)
 391		free_msg(msg);
 392	kfree(info->node_cache);
 393	spin_unlock(&info->lock);
 394
 395	/* Total amount of bytes accounted for the mqueue */
 396	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 397		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 398		sizeof(struct posix_msg_tree_node);
 399
 400	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 401				  info->attr.mq_msgsize);
 402
 403	user = info->user;
 404	if (user) {
 405		spin_lock(&mq_lock);
 406		user->mq_bytes -= mq_bytes;
 407		/*
 408		 * get_ns_from_inode() ensures that the
 409		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 410		 * to which we now hold a reference, or it is NULL.
 411		 * We can't put it here under mq_lock, though.
 412		 */
 413		if (ipc_ns)
 414			ipc_ns->mq_queues_count--;
 415		spin_unlock(&mq_lock);
 416		free_uid(user);
 417	}
 418	if (ipc_ns)
 419		put_ipc_ns(ipc_ns);
 420}
 421
 422static int mqueue_create(struct inode *dir, struct dentry *dentry,
 423				umode_t mode, bool excl)
 424{
 425	struct inode *inode;
 426	struct mq_attr *attr = dentry->d_fsdata;
 427	int error;
 428	struct ipc_namespace *ipc_ns;
 429
 430	spin_lock(&mq_lock);
 431	ipc_ns = __get_ns_from_inode(dir);
 432	if (!ipc_ns) {
 433		error = -EACCES;
 434		goto out_unlock;
 435	}
 436
 437	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 438	    !capable(CAP_SYS_RESOURCE)) {
 439		error = -ENOSPC;
 440		goto out_unlock;
 441	}
 442	ipc_ns->mq_queues_count++;
 443	spin_unlock(&mq_lock);
 444
 445	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 446	if (IS_ERR(inode)) {
 447		error = PTR_ERR(inode);
 448		spin_lock(&mq_lock);
 449		ipc_ns->mq_queues_count--;
 450		goto out_unlock;
 451	}
 452
 453	put_ipc_ns(ipc_ns);
 454	dir->i_size += DIRENT_SIZE;
 455	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 456
 457	d_instantiate(dentry, inode);
 458	dget(dentry);
 459	return 0;
 460out_unlock:
 461	spin_unlock(&mq_lock);
 462	if (ipc_ns)
 463		put_ipc_ns(ipc_ns);
 464	return error;
 465}
 466
 467static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 468{
 469	struct inode *inode = dentry->d_inode;
 470
 471	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 472	dir->i_size -= DIRENT_SIZE;
 473	drop_nlink(inode);
 474	dput(dentry);
 475	return 0;
 476}
 477
 478/*
 479*	This is routine for system read from queue file.
 480*	To avoid mess with doing here some sort of mq_receive we allow
 481*	to read only queue size & notification info (the only values
 482*	that are interesting from user point of view and aren't accessible
 483*	through std routines)
 484*/
 485static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 486				size_t count, loff_t *off)
 487{
 488	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 489	char buffer[FILENT_SIZE];
 490	ssize_t ret;
 491
 492	spin_lock(&info->lock);
 493	snprintf(buffer, sizeof(buffer),
 494			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 495			info->qsize,
 496			info->notify_owner ? info->notify.sigev_notify : 0,
 497			(info->notify_owner &&
 498			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
 499				info->notify.sigev_signo : 0,
 500			pid_vnr(info->notify_owner));
 501	spin_unlock(&info->lock);
 502	buffer[sizeof(buffer)-1] = '\0';
 503
 504	ret = simple_read_from_buffer(u_data, count, off, buffer,
 505				strlen(buffer));
 506	if (ret <= 0)
 507		return ret;
 508
 509	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
 510	return ret;
 511}
 512
 513static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 514{
 515	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 516
 517	spin_lock(&info->lock);
 518	if (task_tgid(current) == info->notify_owner)
 519		remove_notification(info);
 520
 521	spin_unlock(&info->lock);
 522	return 0;
 523}
 524
 525static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 526{
 527	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 528	int retval = 0;
 529
 530	poll_wait(filp, &info->wait_q, poll_tab);
 531
 532	spin_lock(&info->lock);
 533	if (info->attr.mq_curmsgs)
 534		retval = POLLIN | POLLRDNORM;
 535
 536	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 537		retval |= POLLOUT | POLLWRNORM;
 538	spin_unlock(&info->lock);
 539
 540	return retval;
 541}
 542
 543/* Adds current to info->e_wait_q[sr] before element with smaller prio */
 544static void wq_add(struct mqueue_inode_info *info, int sr,
 545			struct ext_wait_queue *ewp)
 546{
 547	struct ext_wait_queue *walk;
 548
 549	ewp->task = current;
 550
 551	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 552		if (walk->task->static_prio <= current->static_prio) {
 553			list_add_tail(&ewp->list, &walk->list);
 554			return;
 555		}
 556	}
 557	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 558}
 559
 560/*
 561 * Puts current task to sleep. Caller must hold queue lock. After return
 562 * lock isn't held.
 563 * sr: SEND or RECV
 564 */
 565static int wq_sleep(struct mqueue_inode_info *info, int sr,
 566		    ktime_t *timeout, struct ext_wait_queue *ewp)
 567{
 568	int retval;
 569	signed long time;
 570
 571	wq_add(info, sr, ewp);
 572
 573	for (;;) {
 574		set_current_state(TASK_INTERRUPTIBLE);
 575
 576		spin_unlock(&info->lock);
 577		time = schedule_hrtimeout_range_clock(timeout, 0,
 578			HRTIMER_MODE_ABS, CLOCK_REALTIME);
 579
 580		while (ewp->state == STATE_PENDING)
 581			cpu_relax();
 582
 583		if (ewp->state == STATE_READY) {
 584			retval = 0;
 585			goto out;
 586		}
 587		spin_lock(&info->lock);
 588		if (ewp->state == STATE_READY) {
 589			retval = 0;
 590			goto out_unlock;
 591		}
 592		if (signal_pending(current)) {
 593			retval = -ERESTARTSYS;
 594			break;
 595		}
 596		if (time == 0) {
 597			retval = -ETIMEDOUT;
 598			break;
 599		}
 600	}
 601	list_del(&ewp->list);
 602out_unlock:
 603	spin_unlock(&info->lock);
 604out:
 605	return retval;
 606}
 607
 608/*
 609 * Returns waiting task that should be serviced first or NULL if none exists
 610 */
 611static struct ext_wait_queue *wq_get_first_waiter(
 612		struct mqueue_inode_info *info, int sr)
 613{
 614	struct list_head *ptr;
 615
 616	ptr = info->e_wait_q[sr].list.prev;
 617	if (ptr == &info->e_wait_q[sr].list)
 618		return NULL;
 619	return list_entry(ptr, struct ext_wait_queue, list);
 620}
 621
 622
 623static inline void set_cookie(struct sk_buff *skb, char code)
 624{
 625	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 626}
 627
 628/*
 629 * The next function is only to split too long sys_mq_timedsend
 630 */
 631static void __do_notify(struct mqueue_inode_info *info)
 632{
 633	/* notification
 634	 * invoked when there is registered process and there isn't process
 635	 * waiting synchronously for message AND state of queue changed from
 636	 * empty to not empty. Here we are sure that no one is waiting
 637	 * synchronously. */
 638	if (info->notify_owner &&
 639	    info->attr.mq_curmsgs == 1) {
 640		struct siginfo sig_i;
 641		switch (info->notify.sigev_notify) {
 642		case SIGEV_NONE:
 643			break;
 644		case SIGEV_SIGNAL:
 645			/* sends signal */
 646
 647			sig_i.si_signo = info->notify.sigev_signo;
 648			sig_i.si_errno = 0;
 649			sig_i.si_code = SI_MESGQ;
 650			sig_i.si_value = info->notify.sigev_value;
 651			/* map current pid/uid into info->owner's namespaces */
 652			rcu_read_lock();
 653			sig_i.si_pid = task_tgid_nr_ns(current,
 654						ns_of_pid(info->notify_owner));
 655			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
 656			rcu_read_unlock();
 657
 658			kill_pid_info(info->notify.sigev_signo,
 659				      &sig_i, info->notify_owner);
 660			break;
 661		case SIGEV_THREAD:
 662			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 663			netlink_sendskb(info->notify_sock, info->notify_cookie);
 664			break;
 665		}
 666		/* after notification unregisters process */
 667		put_pid(info->notify_owner);
 668		put_user_ns(info->notify_user_ns);
 669		info->notify_owner = NULL;
 670		info->notify_user_ns = NULL;
 671	}
 672	wake_up(&info->wait_q);
 673}
 674
 675static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 676			   ktime_t *expires, struct timespec *ts)
 677{
 678	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 679		return -EFAULT;
 680	if (!timespec_valid(ts))
 681		return -EINVAL;
 682
 683	*expires = timespec_to_ktime(*ts);
 684	return 0;
 685}
 686
 687static void remove_notification(struct mqueue_inode_info *info)
 688{
 689	if (info->notify_owner != NULL &&
 690	    info->notify.sigev_notify == SIGEV_THREAD) {
 691		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 692		netlink_sendskb(info->notify_sock, info->notify_cookie);
 693	}
 694	put_pid(info->notify_owner);
 695	put_user_ns(info->notify_user_ns);
 696	info->notify_owner = NULL;
 697	info->notify_user_ns = NULL;
 698}
 699
 700static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 701{
 702	int mq_treesize;
 703	unsigned long total_size;
 704
 705	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 706		return -EINVAL;
 707	if (capable(CAP_SYS_RESOURCE)) {
 708		if (attr->mq_maxmsg > HARD_MSGMAX ||
 709		    attr->mq_msgsize > HARD_MSGSIZEMAX)
 710			return -EINVAL;
 711	} else {
 712		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 713				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 714			return -EINVAL;
 715	}
 716	/* check for overflow */
 717	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 718		return -EOVERFLOW;
 719	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
 720		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
 721		sizeof(struct posix_msg_tree_node);
 722	total_size = attr->mq_maxmsg * attr->mq_msgsize;
 723	if (total_size + mq_treesize < total_size)
 724		return -EOVERFLOW;
 725	return 0;
 726}
 727
 728/*
 729 * Invoked when creating a new queue via sys_mq_open
 730 */
 731static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
 732			struct path *path, int oflag, umode_t mode,
 733			struct mq_attr *attr)
 734{
 735	const struct cred *cred = current_cred();
 736	int ret;
 737
 738	if (attr) {
 739		ret = mq_attr_ok(ipc_ns, attr);
 740		if (ret)
 741			return ERR_PTR(ret);
 742		/* store for use during create */
 743		path->dentry->d_fsdata = attr;
 744	} else {
 745		struct mq_attr def_attr;
 746
 747		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 748					 ipc_ns->mq_msg_default);
 749		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 750					  ipc_ns->mq_msgsize_default);
 751		ret = mq_attr_ok(ipc_ns, &def_attr);
 752		if (ret)
 753			return ERR_PTR(ret);
 754	}
 755
 756	mode &= ~current_umask();
 757	ret = vfs_create(dir, path->dentry, mode, true);
 758	path->dentry->d_fsdata = NULL;
 759	if (ret)
 760		return ERR_PTR(ret);
 761	return dentry_open(path, oflag, cred);
 762}
 763
 764/* Opens existing queue */
 765static struct file *do_open(struct path *path, int oflag)
 766{
 767	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 768						  MAY_READ | MAY_WRITE };
 769	int acc;
 770	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 771		return ERR_PTR(-EINVAL);
 772	acc = oflag2acc[oflag & O_ACCMODE];
 773	if (inode_permission(path->dentry->d_inode, acc))
 774		return ERR_PTR(-EACCES);
 775	return dentry_open(path, oflag, current_cred());
 776}
 777
 778SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 779		struct mq_attr __user *, u_attr)
 780{
 781	struct path path;
 782	struct file *filp;
 783	struct filename *name;
 784	struct mq_attr attr;
 785	int fd, error;
 786	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 787	struct vfsmount *mnt = ipc_ns->mq_mnt;
 788	struct dentry *root = mnt->mnt_root;
 789	int ro;
 790
 791	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 792		return -EFAULT;
 793
 794	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 795
 796	if (IS_ERR(name = getname(u_name)))
 797		return PTR_ERR(name);
 798
 799	fd = get_unused_fd_flags(O_CLOEXEC);
 800	if (fd < 0)
 801		goto out_putname;
 802
 803	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
 804	error = 0;
 805	mutex_lock(&root->d_inode->i_mutex);
 806	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
 807	if (IS_ERR(path.dentry)) {
 808		error = PTR_ERR(path.dentry);
 809		goto out_putfd;
 810	}
 811	path.mnt = mntget(mnt);
 812
 813	if (oflag & O_CREAT) {
 814		if (path.dentry->d_inode) {	/* entry already exists */
 815			audit_inode(name, path.dentry, 0);
 816			if (oflag & O_EXCL) {
 817				error = -EEXIST;
 818				goto out;
 819			}
 820			filp = do_open(&path, oflag);
 821		} else {
 822			if (ro) {
 823				error = ro;
 824				goto out;
 825			}
 826			audit_inode_parent_hidden(name, root);
 827			filp = do_create(ipc_ns, root->d_inode,
 828						&path, oflag, mode,
 829						u_attr ? &attr : NULL);
 830		}
 831	} else {
 832		if (!path.dentry->d_inode) {
 833			error = -ENOENT;
 834			goto out;
 835		}
 836		audit_inode(name, path.dentry, 0);
 837		filp = do_open(&path, oflag);
 838	}
 839
 840	if (!IS_ERR(filp))
 841		fd_install(fd, filp);
 842	else
 843		error = PTR_ERR(filp);
 844out:
 845	path_put(&path);
 846out_putfd:
 847	if (error) {
 848		put_unused_fd(fd);
 849		fd = error;
 850	}
 851	mutex_unlock(&root->d_inode->i_mutex);
 852	if (!ro)
 853		mnt_drop_write(mnt);
 854out_putname:
 855	putname(name);
 856	return fd;
 857}
 858
 859SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 860{
 861	int err;
 862	struct filename *name;
 863	struct dentry *dentry;
 864	struct inode *inode = NULL;
 865	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 866	struct vfsmount *mnt = ipc_ns->mq_mnt;
 867
 868	name = getname(u_name);
 869	if (IS_ERR(name))
 870		return PTR_ERR(name);
 871
 872	audit_inode_parent_hidden(name, mnt->mnt_root);
 873	err = mnt_want_write(mnt);
 874	if (err)
 875		goto out_name;
 876	mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
 877	dentry = lookup_one_len(name->name, mnt->mnt_root,
 878				strlen(name->name));
 879	if (IS_ERR(dentry)) {
 880		err = PTR_ERR(dentry);
 881		goto out_unlock;
 882	}
 883
 884	inode = dentry->d_inode;
 885	if (!inode) {
 886		err = -ENOENT;
 887	} else {
 888		ihold(inode);
 889		err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
 890	}
 891	dput(dentry);
 892
 893out_unlock:
 894	mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
 895	if (inode)
 896		iput(inode);
 897	mnt_drop_write(mnt);
 898out_name:
 899	putname(name);
 900
 901	return err;
 902}
 903
 904/* Pipelined send and receive functions.
 905 *
 906 * If a receiver finds no waiting message, then it registers itself in the
 907 * list of waiting receivers. A sender checks that list before adding the new
 908 * message into the message array. If there is a waiting receiver, then it
 909 * bypasses the message array and directly hands the message over to the
 910 * receiver.
 911 * The receiver accepts the message and returns without grabbing the queue
 912 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 913 * are necessary. The same algorithm is used for sysv semaphores, see
 914 * ipc/sem.c for more details.
 
 
 
 
 915 *
 916 * The same algorithm is used for senders.
 917 */
 918
 919/* pipelined_send() - send a message directly to the task waiting in
 920 * sys_mq_timedreceive() (without inserting message into a queue).
 921 */
 922static inline void pipelined_send(struct mqueue_inode_info *info,
 
 923				  struct msg_msg *message,
 924				  struct ext_wait_queue *receiver)
 925{
 926	receiver->msg = message;
 927	list_del(&receiver->list);
 928	receiver->state = STATE_PENDING;
 929	wake_up_process(receiver->task);
 930	smp_wmb();
 
 
 
 
 
 
 931	receiver->state = STATE_READY;
 932}
 933
 934/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 935 * gets its message and put to the queue (we have one free place for sure). */
 936static inline void pipelined_receive(struct mqueue_inode_info *info)
 
 937{
 938	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 939
 940	if (!sender) {
 941		/* for poll */
 942		wake_up_interruptible(&info->wait_q);
 943		return;
 944	}
 945	if (msg_insert(sender->msg, info))
 946		return;
 
 947	list_del(&sender->list);
 948	sender->state = STATE_PENDING;
 949	wake_up_process(sender->task);
 950	smp_wmb();
 951	sender->state = STATE_READY;
 952}
 953
 954SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 955		size_t, msg_len, unsigned int, msg_prio,
 956		const struct timespec __user *, u_abs_timeout)
 957{
 958	struct fd f;
 959	struct inode *inode;
 960	struct ext_wait_queue wait;
 961	struct ext_wait_queue *receiver;
 962	struct msg_msg *msg_ptr;
 963	struct mqueue_inode_info *info;
 964	ktime_t expires, *timeout = NULL;
 965	struct timespec ts;
 966	struct posix_msg_tree_node *new_leaf = NULL;
 967	int ret = 0;
 
 968
 969	if (u_abs_timeout) {
 970		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 971		if (res)
 972			return res;
 973		timeout = &expires;
 974	}
 975
 976	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 977		return -EINVAL;
 978
 979	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 980
 981	f = fdget(mqdes);
 982	if (unlikely(!f.file)) {
 983		ret = -EBADF;
 984		goto out;
 985	}
 986
 987	inode = file_inode(f.file);
 988	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
 989		ret = -EBADF;
 990		goto out_fput;
 991	}
 992	info = MQUEUE_I(inode);
 993	audit_inode(NULL, f.file->f_path.dentry, 0);
 994
 995	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
 996		ret = -EBADF;
 997		goto out_fput;
 998	}
 999
1000	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1001		ret = -EMSGSIZE;
1002		goto out_fput;
1003	}
1004
1005	/* First try to allocate memory, before doing anything with
1006	 * existing queues. */
1007	msg_ptr = load_msg(u_msg_ptr, msg_len);
1008	if (IS_ERR(msg_ptr)) {
1009		ret = PTR_ERR(msg_ptr);
1010		goto out_fput;
1011	}
1012	msg_ptr->m_ts = msg_len;
1013	msg_ptr->m_type = msg_prio;
1014
1015	/*
1016	 * msg_insert really wants us to have a valid, spare node struct so
1017	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1018	 * fall back to that if necessary.
1019	 */
1020	if (!info->node_cache)
1021		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1022
1023	spin_lock(&info->lock);
1024
1025	if (!info->node_cache && new_leaf) {
1026		/* Save our speculative allocation into the cache */
1027		INIT_LIST_HEAD(&new_leaf->msg_list);
1028		info->node_cache = new_leaf;
1029		info->qsize += sizeof(*new_leaf);
1030		new_leaf = NULL;
1031	} else {
1032		kfree(new_leaf);
1033	}
1034
1035	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1036		if (f.file->f_flags & O_NONBLOCK) {
1037			ret = -EAGAIN;
1038		} else {
1039			wait.task = current;
1040			wait.msg = (void *) msg_ptr;
1041			wait.state = STATE_NONE;
1042			ret = wq_sleep(info, SEND, timeout, &wait);
1043			/*
1044			 * wq_sleep must be called with info->lock held, and
1045			 * returns with the lock released
1046			 */
1047			goto out_free;
1048		}
1049	} else {
1050		receiver = wq_get_first_waiter(info, RECV);
1051		if (receiver) {
1052			pipelined_send(info, msg_ptr, receiver);
1053		} else {
1054			/* adds message to the queue */
1055			ret = msg_insert(msg_ptr, info);
1056			if (ret)
1057				goto out_unlock;
1058			__do_notify(info);
1059		}
1060		inode->i_atime = inode->i_mtime = inode->i_ctime =
1061				CURRENT_TIME;
1062	}
1063out_unlock:
1064	spin_unlock(&info->lock);
 
1065out_free:
1066	if (ret)
1067		free_msg(msg_ptr);
1068out_fput:
1069	fdput(f);
1070out:
1071	return ret;
1072}
1073
1074SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1075		size_t, msg_len, unsigned int __user *, u_msg_prio,
1076		const struct timespec __user *, u_abs_timeout)
1077{
1078	ssize_t ret;
1079	struct msg_msg *msg_ptr;
1080	struct fd f;
1081	struct inode *inode;
1082	struct mqueue_inode_info *info;
1083	struct ext_wait_queue wait;
1084	ktime_t expires, *timeout = NULL;
1085	struct timespec ts;
1086	struct posix_msg_tree_node *new_leaf = NULL;
1087
1088	if (u_abs_timeout) {
1089		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1090		if (res)
1091			return res;
1092		timeout = &expires;
1093	}
1094
1095	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1096
1097	f = fdget(mqdes);
1098	if (unlikely(!f.file)) {
1099		ret = -EBADF;
1100		goto out;
1101	}
1102
1103	inode = file_inode(f.file);
1104	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1105		ret = -EBADF;
1106		goto out_fput;
1107	}
1108	info = MQUEUE_I(inode);
1109	audit_inode(NULL, f.file->f_path.dentry, 0);
1110
1111	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1112		ret = -EBADF;
1113		goto out_fput;
1114	}
1115
1116	/* checks if buffer is big enough */
1117	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1118		ret = -EMSGSIZE;
1119		goto out_fput;
1120	}
1121
1122	/*
1123	 * msg_insert really wants us to have a valid, spare node struct so
1124	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1125	 * fall back to that if necessary.
1126	 */
1127	if (!info->node_cache)
1128		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1129
1130	spin_lock(&info->lock);
1131
1132	if (!info->node_cache && new_leaf) {
1133		/* Save our speculative allocation into the cache */
1134		INIT_LIST_HEAD(&new_leaf->msg_list);
1135		info->node_cache = new_leaf;
1136		info->qsize += sizeof(*new_leaf);
1137	} else {
1138		kfree(new_leaf);
1139	}
1140
1141	if (info->attr.mq_curmsgs == 0) {
1142		if (f.file->f_flags & O_NONBLOCK) {
1143			spin_unlock(&info->lock);
1144			ret = -EAGAIN;
1145		} else {
1146			wait.task = current;
1147			wait.state = STATE_NONE;
1148			ret = wq_sleep(info, RECV, timeout, &wait);
1149			msg_ptr = wait.msg;
1150		}
1151	} else {
 
 
1152		msg_ptr = msg_get(info);
1153
1154		inode->i_atime = inode->i_mtime = inode->i_ctime =
1155				CURRENT_TIME;
1156
1157		/* There is now free space in queue. */
1158		pipelined_receive(info);
1159		spin_unlock(&info->lock);
 
1160		ret = 0;
1161	}
1162	if (ret == 0) {
1163		ret = msg_ptr->m_ts;
1164
1165		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1166			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1167			ret = -EFAULT;
1168		}
1169		free_msg(msg_ptr);
1170	}
1171out_fput:
1172	fdput(f);
1173out:
1174	return ret;
1175}
1176
1177/*
1178 * Notes: the case when user wants us to deregister (with NULL as pointer)
1179 * and he isn't currently owner of notification, will be silently discarded.
1180 * It isn't explicitly defined in the POSIX.
1181 */
1182SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1183		const struct sigevent __user *, u_notification)
1184{
1185	int ret;
1186	struct fd f;
1187	struct sock *sock;
1188	struct inode *inode;
1189	struct sigevent notification;
1190	struct mqueue_inode_info *info;
1191	struct sk_buff *nc;
1192
1193	if (u_notification) {
1194		if (copy_from_user(&notification, u_notification,
1195					sizeof(struct sigevent)))
1196			return -EFAULT;
1197	}
1198
1199	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1200
1201	nc = NULL;
1202	sock = NULL;
1203	if (u_notification != NULL) {
1204		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1205			     notification.sigev_notify != SIGEV_SIGNAL &&
1206			     notification.sigev_notify != SIGEV_THREAD))
1207			return -EINVAL;
1208		if (notification.sigev_notify == SIGEV_SIGNAL &&
1209			!valid_signal(notification.sigev_signo)) {
1210			return -EINVAL;
1211		}
1212		if (notification.sigev_notify == SIGEV_THREAD) {
1213			long timeo;
1214
1215			/* create the notify skb */
1216			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1217			if (!nc) {
1218				ret = -ENOMEM;
1219				goto out;
1220			}
1221			if (copy_from_user(nc->data,
1222					notification.sigev_value.sival_ptr,
1223					NOTIFY_COOKIE_LEN)) {
1224				ret = -EFAULT;
1225				goto out;
1226			}
1227
1228			/* TODO: add a header? */
1229			skb_put(nc, NOTIFY_COOKIE_LEN);
1230			/* and attach it to the socket */
1231retry:
1232			f = fdget(notification.sigev_signo);
1233			if (!f.file) {
1234				ret = -EBADF;
1235				goto out;
1236			}
1237			sock = netlink_getsockbyfilp(f.file);
1238			fdput(f);
1239			if (IS_ERR(sock)) {
1240				ret = PTR_ERR(sock);
1241				sock = NULL;
1242				goto out;
1243			}
1244
1245			timeo = MAX_SCHEDULE_TIMEOUT;
1246			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1247			if (ret == 1)
1248				goto retry;
1249			if (ret) {
1250				sock = NULL;
1251				nc = NULL;
1252				goto out;
1253			}
1254		}
1255	}
1256
1257	f = fdget(mqdes);
1258	if (!f.file) {
1259		ret = -EBADF;
1260		goto out;
1261	}
1262
1263	inode = file_inode(f.file);
1264	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1265		ret = -EBADF;
1266		goto out_fput;
1267	}
1268	info = MQUEUE_I(inode);
1269
1270	ret = 0;
1271	spin_lock(&info->lock);
1272	if (u_notification == NULL) {
1273		if (info->notify_owner == task_tgid(current)) {
1274			remove_notification(info);
1275			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1276		}
1277	} else if (info->notify_owner != NULL) {
1278		ret = -EBUSY;
1279	} else {
1280		switch (notification.sigev_notify) {
1281		case SIGEV_NONE:
1282			info->notify.sigev_notify = SIGEV_NONE;
1283			break;
1284		case SIGEV_THREAD:
1285			info->notify_sock = sock;
1286			info->notify_cookie = nc;
1287			sock = NULL;
1288			nc = NULL;
1289			info->notify.sigev_notify = SIGEV_THREAD;
1290			break;
1291		case SIGEV_SIGNAL:
1292			info->notify.sigev_signo = notification.sigev_signo;
1293			info->notify.sigev_value = notification.sigev_value;
1294			info->notify.sigev_notify = SIGEV_SIGNAL;
1295			break;
1296		}
1297
1298		info->notify_owner = get_pid(task_tgid(current));
1299		info->notify_user_ns = get_user_ns(current_user_ns());
1300		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1301	}
1302	spin_unlock(&info->lock);
1303out_fput:
1304	fdput(f);
1305out:
1306	if (sock)
1307		netlink_detachskb(sock, nc);
1308	else if (nc)
1309		dev_kfree_skb(nc);
1310
1311	return ret;
1312}
1313
1314SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1315		const struct mq_attr __user *, u_mqstat,
1316		struct mq_attr __user *, u_omqstat)
1317{
1318	int ret;
1319	struct mq_attr mqstat, omqstat;
1320	struct fd f;
1321	struct inode *inode;
1322	struct mqueue_inode_info *info;
1323
1324	if (u_mqstat != NULL) {
1325		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1326			return -EFAULT;
1327		if (mqstat.mq_flags & (~O_NONBLOCK))
1328			return -EINVAL;
1329	}
1330
1331	f = fdget(mqdes);
1332	if (!f.file) {
1333		ret = -EBADF;
1334		goto out;
1335	}
1336
1337	inode = file_inode(f.file);
1338	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1339		ret = -EBADF;
1340		goto out_fput;
1341	}
1342	info = MQUEUE_I(inode);
1343
1344	spin_lock(&info->lock);
1345
1346	omqstat = info->attr;
1347	omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1348	if (u_mqstat) {
1349		audit_mq_getsetattr(mqdes, &mqstat);
1350		spin_lock(&f.file->f_lock);
1351		if (mqstat.mq_flags & O_NONBLOCK)
1352			f.file->f_flags |= O_NONBLOCK;
1353		else
1354			f.file->f_flags &= ~O_NONBLOCK;
1355		spin_unlock(&f.file->f_lock);
1356
1357		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1358	}
1359
1360	spin_unlock(&info->lock);
1361
1362	ret = 0;
1363	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1364						sizeof(struct mq_attr)))
1365		ret = -EFAULT;
1366
1367out_fput:
1368	fdput(f);
1369out:
1370	return ret;
1371}
1372
1373static const struct inode_operations mqueue_dir_inode_operations = {
1374	.lookup = simple_lookup,
1375	.create = mqueue_create,
1376	.unlink = mqueue_unlink,
1377};
1378
1379static const struct file_operations mqueue_file_operations = {
1380	.flush = mqueue_flush_file,
1381	.poll = mqueue_poll_file,
1382	.read = mqueue_read_file,
1383	.llseek = default_llseek,
1384};
1385
1386static const struct super_operations mqueue_super_ops = {
1387	.alloc_inode = mqueue_alloc_inode,
1388	.destroy_inode = mqueue_destroy_inode,
1389	.evict_inode = mqueue_evict_inode,
1390	.statfs = simple_statfs,
1391};
1392
1393static struct file_system_type mqueue_fs_type = {
1394	.name = "mqueue",
1395	.mount = mqueue_mount,
1396	.kill_sb = kill_litter_super,
1397	.fs_flags = FS_USERNS_MOUNT,
1398};
1399
1400int mq_init_ns(struct ipc_namespace *ns)
1401{
1402	ns->mq_queues_count  = 0;
1403	ns->mq_queues_max    = DFLT_QUEUESMAX;
1404	ns->mq_msg_max       = DFLT_MSGMAX;
1405	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1406	ns->mq_msg_default   = DFLT_MSG;
1407	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1408
1409	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1410	if (IS_ERR(ns->mq_mnt)) {
1411		int err = PTR_ERR(ns->mq_mnt);
1412		ns->mq_mnt = NULL;
1413		return err;
1414	}
1415	return 0;
1416}
1417
1418void mq_clear_sbinfo(struct ipc_namespace *ns)
1419{
1420	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1421}
1422
1423void mq_put_mnt(struct ipc_namespace *ns)
1424{
1425	kern_unmount(ns->mq_mnt);
1426}
1427
1428static int __init init_mqueue_fs(void)
1429{
1430	int error;
1431
1432	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1433				sizeof(struct mqueue_inode_info), 0,
1434				SLAB_HWCACHE_ALIGN, init_once);
1435	if (mqueue_inode_cachep == NULL)
1436		return -ENOMEM;
1437
1438	/* ignore failures - they are not fatal */
1439	mq_sysctl_table = mq_register_sysctl_table();
1440
1441	error = register_filesystem(&mqueue_fs_type);
1442	if (error)
1443		goto out_sysctl;
1444
1445	spin_lock_init(&mq_lock);
1446
1447	error = mq_init_ns(&init_ipc_ns);
1448	if (error)
1449		goto out_filesystem;
1450
1451	return 0;
1452
1453out_filesystem:
1454	unregister_filesystem(&mqueue_fs_type);
1455out_sysctl:
1456	if (mq_sysctl_table)
1457		unregister_sysctl_table(mq_sysctl_table);
1458	kmem_cache_destroy(mqueue_inode_cachep);
1459	return error;
1460}
1461
1462device_initcall(init_mqueue_fs);