mqueue.c

   1/*
   2 * POSIX message queues filesystem for Linux.
   3 *
   4 * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5 *                          Michal Wronski          (michal.wronski@gmail.com)
   6 *
   7 * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8 * Lockless receive & send, fd based notify:
   9 *			    Manfred Spraul	    (manfred@colorfullife.com)
  10 *
  11 * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12 *
  13 * This file is released under the GPL.
  14 */
  15
  16#include <linux/capability.h>
  17#include <linux/init.h>
  18#include <linux/pagemap.h>
  19#include <linux/file.h>
  20#include <linux/mount.h>
  21#include <linux/fs_context.h>
  22#include <linux/namei.h>
  23#include <linux/sysctl.h>
  24#include <linux/poll.h>
  25#include <linux/mqueue.h>
  26#include <linux/msg.h>
  27#include <linux/skbuff.h>
  28#include <linux/vmalloc.h>
  29#include <linux/netlink.h>
  30#include <linux/syscalls.h>
  31#include <linux/audit.h>
  32#include <linux/signal.h>
  33#include <linux/mutex.h>
  34#include <linux/nsproxy.h>
  35#include <linux/pid.h>
  36#include <linux/ipc_namespace.h>
  37#include <linux/user_namespace.h>
  38#include <linux/slab.h>
  39#include <linux/sched/wake_q.h>
  40#include <linux/sched/signal.h>
  41#include <linux/sched/user.h>
  42
  43#include <net/sock.h>
  44#include "util.h"
  45
  46struct mqueue_fs_context {
  47	struct ipc_namespace	*ipc_ns;
  48	bool			 newns;	/* Set if newly created ipc namespace */
  49};
  50
  51#define MQUEUE_MAGIC	0x19800202
  52#define DIRENT_SIZE	20
  53#define FILENT_SIZE	80
  54
  55#define SEND		0
  56#define RECV		1
  57
  58#define STATE_NONE	0
  59#define STATE_READY	1
  60
  61struct posix_msg_tree_node {
  62	struct rb_node		rb_node;
  63	struct list_head	msg_list;
  64	int			priority;
  65};
  66
  67/*
  68 * Locking:
  69 *
  70 * Accesses to a message queue are synchronized by acquiring info->lock.
  71 *
  72 * There are two notable exceptions:
  73 * - The actual wakeup of a sleeping task is performed using the wake_q
  74 *   framework. info->lock is already released when wake_up_q is called.
  75 * - The exit codepaths after sleeping check ext_wait_queue->state without
  76 *   any locks. If it is STATE_READY, then the syscall is completed without
  77 *   acquiring info->lock.
  78 *
  79 * MQ_BARRIER:
  80 * To achieve proper release/acquire memory barrier pairing, the state is set to
  81 * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
  82 * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
  83 *
  84 * This prevents the following races:
  85 *
  86 * 1) With the simple wake_q_add(), the task could be gone already before
  87 *    the increase of the reference happens
  88 * Thread A
  89 *				Thread B
  90 * WRITE_ONCE(wait.state, STATE_NONE);
  91 * schedule_hrtimeout()
  92 *				wake_q_add(A)
  93 *				if (cmpxchg()) // success
  94 *				   ->state = STATE_READY (reordered)
  95 * <timeout returns>
  96 * if (wait.state == STATE_READY) return;
  97 * sysret to user space
  98 * sys_exit()
  99 *				get_task_struct() // UaF
 100 *
 101 * Solution: Use wake_q_add_safe() and perform the get_task_struct() before
 102 * the smp_store_release() that does ->state = STATE_READY.
 103 *
 104 * 2) Without proper _release/_acquire barriers, the woken up task
 105 *    could read stale data
 106 *
 107 * Thread A
 108 *				Thread B
 109 * do_mq_timedreceive
 110 * WRITE_ONCE(wait.state, STATE_NONE);
 111 * schedule_hrtimeout()
 112 *				state = STATE_READY;
 113 * <timeout returns>
 114 * if (wait.state == STATE_READY) return;
 115 * msg_ptr = wait.msg;		// Access to stale data!
 116 *				receiver->msg = message; (reordered)
 117 *
 118 * Solution: use _release and _acquire barriers.
 119 *
 120 * 3) There is intentionally no barrier when setting current->state
 121 *    to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
 122 *    release memory barrier, and the wakeup is triggered when holding
 123 *    info->lock, i.e. spin_lock(&info->lock) provided a pairing
 124 *    acquire memory barrier.
 125 */
 126
 127struct ext_wait_queue {		/* queue of sleeping tasks */
 128	struct task_struct *task;
 129	struct list_head list;
 130	struct msg_msg *msg;	/* ptr of loaded message */
 131	int state;		/* one of STATE_* values */
 132};
 133
 134struct mqueue_inode_info {
 135	spinlock_t lock;
 136	struct inode vfs_inode;
 137	wait_queue_head_t wait_q;
 138
 139	struct rb_root msg_tree;
 140	struct rb_node *msg_tree_rightmost;
 141	struct posix_msg_tree_node *node_cache;
 142	struct mq_attr attr;
 143
 144	struct sigevent notify;
 145	struct pid *notify_owner;
 146	u32 notify_self_exec_id;
 147	struct user_namespace *notify_user_ns;
 148	struct ucounts *ucounts;	/* user who created, for accounting */
 149	struct sock *notify_sock;
 150	struct sk_buff *notify_cookie;
 151
 152	/* for tasks waiting for free space and messages, respectively */
 153	struct ext_wait_queue e_wait_q[2];
 154
 155	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
 156};
 157
 158static struct file_system_type mqueue_fs_type;
 159static const struct inode_operations mqueue_dir_inode_operations;
 160static const struct file_operations mqueue_file_operations;
 161static const struct super_operations mqueue_super_ops;
 162static const struct fs_context_operations mqueue_fs_context_ops;
 163static void remove_notification(struct mqueue_inode_info *info);
 164
 165static struct kmem_cache *mqueue_inode_cachep;
 166
 167static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 168{
 169	return container_of(inode, struct mqueue_inode_info, vfs_inode);
 170}
 171
 172/*
 173 * This routine should be called with the mq_lock held.
 174 */
 175static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
 176{
 177	return get_ipc_ns(inode->i_sb->s_fs_info);
 178}
 179
 180static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 181{
 182	struct ipc_namespace *ns;
 183
 184	spin_lock(&mq_lock);
 185	ns = __get_ns_from_inode(inode);
 186	spin_unlock(&mq_lock);
 187	return ns;
 188}
 189
 190/* Auxiliary functions to manipulate messages' list */
 191static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
 192{
 193	struct rb_node **p, *parent = NULL;
 194	struct posix_msg_tree_node *leaf;
 195	bool rightmost = true;
 196
 197	p = &info->msg_tree.rb_node;
 198	while (*p) {
 199		parent = *p;
 200		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 201
 202		if (likely(leaf->priority == msg->m_type))
 203			goto insert_msg;
 204		else if (msg->m_type < leaf->priority) {
 205			p = &(*p)->rb_left;
 206			rightmost = false;
 207		} else
 208			p = &(*p)->rb_right;
 209	}
 210	if (info->node_cache) {
 211		leaf = info->node_cache;
 212		info->node_cache = NULL;
 213	} else {
 214		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
 215		if (!leaf)
 216			return -ENOMEM;
 217		INIT_LIST_HEAD(&leaf->msg_list);
 218	}
 219	leaf->priority = msg->m_type;
 220
 221	if (rightmost)
 222		info->msg_tree_rightmost = &leaf->rb_node;
 223
 224	rb_link_node(&leaf->rb_node, parent, p);
 225	rb_insert_color(&leaf->rb_node, &info->msg_tree);
 226insert_msg:
 227	info->attr.mq_curmsgs++;
 228	info->qsize += msg->m_ts;
 229	list_add_tail(&msg->m_list, &leaf->msg_list);
 230	return 0;
 231}
 232
 233static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
 234				  struct mqueue_inode_info *info)
 235{
 236	struct rb_node *node = &leaf->rb_node;
 237
 238	if (info->msg_tree_rightmost == node)
 239		info->msg_tree_rightmost = rb_prev(node);
 240
 241	rb_erase(node, &info->msg_tree);
 242	if (info->node_cache)
 243		kfree(leaf);
 244	else
 245		info->node_cache = leaf;
 246}
 247
 248static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 249{
 250	struct rb_node *parent = NULL;
 251	struct posix_msg_tree_node *leaf;
 252	struct msg_msg *msg;
 253
 254try_again:
 255	/*
 256	 * During insert, low priorities go to the left and high to the
 257	 * right.  On receive, we want the highest priorities first, so
 258	 * walk all the way to the right.
 259	 */
 260	parent = info->msg_tree_rightmost;
 261	if (!parent) {
 262		if (info->attr.mq_curmsgs) {
 263			pr_warn_once("Inconsistency in POSIX message queue, "
 264				     "no tree element, but supposedly messages "
 265				     "should exist!\n");
 266			info->attr.mq_curmsgs = 0;
 267		}
 268		return NULL;
 269	}
 270	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 271	if (unlikely(list_empty(&leaf->msg_list))) {
 272		pr_warn_once("Inconsistency in POSIX message queue, "
 273			     "empty leaf node but we haven't implemented "
 274			     "lazy leaf delete!\n");
 275		msg_tree_erase(leaf, info);
 276		goto try_again;
 277	} else {
 278		msg = list_first_entry(&leaf->msg_list,
 279				       struct msg_msg, m_list);
 280		list_del(&msg->m_list);
 281		if (list_empty(&leaf->msg_list)) {
 282			msg_tree_erase(leaf, info);
 283		}
 284	}
 285	info->attr.mq_curmsgs--;
 286	info->qsize -= msg->m_ts;
 287	return msg;
 288}
 289
 290static struct inode *mqueue_get_inode(struct super_block *sb,
 291		struct ipc_namespace *ipc_ns, umode_t mode,
 292		struct mq_attr *attr)
 293{
 294	struct inode *inode;
 295	int ret = -ENOMEM;
 296
 297	inode = new_inode(sb);
 298	if (!inode)
 299		goto err;
 300
 301	inode->i_ino = get_next_ino();
 302	inode->i_mode = mode;
 303	inode->i_uid = current_fsuid();
 304	inode->i_gid = current_fsgid();
 305	simple_inode_init_ts(inode);
 306
 307	if (S_ISREG(mode)) {
 308		struct mqueue_inode_info *info;
 309		unsigned long mq_bytes, mq_treesize;
 310
 311		inode->i_fop = &mqueue_file_operations;
 312		inode->i_size = FILENT_SIZE;
 313		/* mqueue specific info */
 314		info = MQUEUE_I(inode);
 315		spin_lock_init(&info->lock);
 316		init_waitqueue_head(&info->wait_q);
 317		INIT_LIST_HEAD(&info->e_wait_q[0].list);
 318		INIT_LIST_HEAD(&info->e_wait_q[1].list);
 319		info->notify_owner = NULL;
 320		info->notify_user_ns = NULL;
 321		info->qsize = 0;
 322		info->ucounts = NULL;	/* set when all is ok */
 323		info->msg_tree = RB_ROOT;
 324		info->msg_tree_rightmost = NULL;
 325		info->node_cache = NULL;
 326		memset(&info->attr, 0, sizeof(info->attr));
 327		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 328					   ipc_ns->mq_msg_default);
 329		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 330					    ipc_ns->mq_msgsize_default);
 331		if (attr) {
 332			info->attr.mq_maxmsg = attr->mq_maxmsg;
 333			info->attr.mq_msgsize = attr->mq_msgsize;
 334		}
 335		/*
 336		 * We used to allocate a static array of pointers and account
 337		 * the size of that array as well as one msg_msg struct per
 338		 * possible message into the queue size. That's no longer
 339		 * accurate as the queue is now an rbtree and will grow and
 340		 * shrink depending on usage patterns.  We can, however, still
 341		 * account one msg_msg struct per message, but the nodes are
 342		 * allocated depending on priority usage, and most programs
 343		 * only use one, or a handful, of priorities.  However, since
 344		 * this is pinned memory, we need to assume worst case, so
 345		 * that means the min(mq_maxmsg, max_priorities) * struct
 346		 * posix_msg_tree_node.
 347		 */
 348
 349		ret = -EINVAL;
 350		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
 351			goto out_inode;
 352		if (capable(CAP_SYS_RESOURCE)) {
 353			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
 354			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
 355				goto out_inode;
 356		} else {
 357			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
 358					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
 359				goto out_inode;
 360		}
 361		ret = -EOVERFLOW;
 362		/* check for overflow */
 363		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
 364			goto out_inode;
 365		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 366			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 367			sizeof(struct posix_msg_tree_node);
 368		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
 369		if (mq_bytes + mq_treesize < mq_bytes)
 370			goto out_inode;
 371		mq_bytes += mq_treesize;
 372		info->ucounts = get_ucounts(current_ucounts());
 373		if (info->ucounts) {
 374			long msgqueue;
 375
 376			spin_lock(&mq_lock);
 377			msgqueue = inc_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
 378			if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) {
 379				dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
 380				spin_unlock(&mq_lock);
 381				put_ucounts(info->ucounts);
 382				info->ucounts = NULL;
 383				/* mqueue_evict_inode() releases info->messages */
 384				ret = -EMFILE;
 385				goto out_inode;
 386			}
 387			spin_unlock(&mq_lock);
 388		}
 389	} else if (S_ISDIR(mode)) {
 390		inc_nlink(inode);
 391		/* Some things misbehave if size == 0 on a directory */
 392		inode->i_size = 2 * DIRENT_SIZE;
 393		inode->i_op = &mqueue_dir_inode_operations;
 394		inode->i_fop = &simple_dir_operations;
 395	}
 396
 397	return inode;
 398out_inode:
 399	iput(inode);
 400err:
 401	return ERR_PTR(ret);
 402}
 403
 404static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
 405{
 406	struct inode *inode;
 407	struct ipc_namespace *ns = sb->s_fs_info;
 408
 409	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
 410	sb->s_blocksize = PAGE_SIZE;
 411	sb->s_blocksize_bits = PAGE_SHIFT;
 412	sb->s_magic = MQUEUE_MAGIC;
 413	sb->s_op = &mqueue_super_ops;
 414	sb->s_d_flags = DCACHE_DONTCACHE;
 415
 416	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 417	if (IS_ERR(inode))
 418		return PTR_ERR(inode);
 419
 420	sb->s_root = d_make_root(inode);
 421	if (!sb->s_root)
 422		return -ENOMEM;
 423	return 0;
 424}
 425
 426static int mqueue_get_tree(struct fs_context *fc)
 427{
 428	struct mqueue_fs_context *ctx = fc->fs_private;
 429
 430	/*
 431	 * With a newly created ipc namespace, we don't need to do a search
 432	 * for an ipc namespace match, but we still need to set s_fs_info.
 433	 */
 434	if (ctx->newns) {
 435		fc->s_fs_info = ctx->ipc_ns;
 436		return get_tree_nodev(fc, mqueue_fill_super);
 437	}
 438	return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
 439}
 440
 441static void mqueue_fs_context_free(struct fs_context *fc)
 442{
 443	struct mqueue_fs_context *ctx = fc->fs_private;
 444
 445	put_ipc_ns(ctx->ipc_ns);
 446	kfree(ctx);
 447}
 448
 449static int mqueue_init_fs_context(struct fs_context *fc)
 450{
 451	struct mqueue_fs_context *ctx;
 452
 453	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
 454	if (!ctx)
 455		return -ENOMEM;
 456
 457	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
 458	put_user_ns(fc->user_ns);
 459	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 460	fc->fs_private = ctx;
 461	fc->ops = &mqueue_fs_context_ops;
 462	return 0;
 463}
 464
 465/*
 466 * mq_init_ns() is currently the only caller of mq_create_mount().
 467 * So the ns parameter is always a newly created ipc namespace.
 468 */
 469static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
 470{
 471	struct mqueue_fs_context *ctx;
 472	struct fs_context *fc;
 473	struct vfsmount *mnt;
 474
 475	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
 476	if (IS_ERR(fc))
 477		return ERR_CAST(fc);
 478
 479	ctx = fc->fs_private;
 480	ctx->newns = true;
 481	put_ipc_ns(ctx->ipc_ns);
 482	ctx->ipc_ns = get_ipc_ns(ns);
 483	put_user_ns(fc->user_ns);
 484	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 485
 486	mnt = fc_mount_longterm(fc);
 487	put_fs_context(fc);
 488	return mnt;
 489}
 490
 491static void init_once(void *foo)
 492{
 493	struct mqueue_inode_info *p = foo;
 494
 495	inode_init_once(&p->vfs_inode);
 496}
 497
 498static struct inode *mqueue_alloc_inode(struct super_block *sb)
 499{
 500	struct mqueue_inode_info *ei;
 501
 502	ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL);
 503	if (!ei)
 504		return NULL;
 505	return &ei->vfs_inode;
 506}
 507
 508static void mqueue_free_inode(struct inode *inode)
 509{
 510	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 511}
 512
 513static void mqueue_evict_inode(struct inode *inode)
 514{
 515	struct mqueue_inode_info *info;
 516	struct ipc_namespace *ipc_ns;
 517	struct msg_msg *msg, *nmsg;
 518	LIST_HEAD(tmp_msg);
 519
 520	clear_inode(inode);
 521
 522	if (S_ISDIR(inode->i_mode))
 523		return;
 524
 525	ipc_ns = get_ns_from_inode(inode);
 526	info = MQUEUE_I(inode);
 527	spin_lock(&info->lock);
 528	while ((msg = msg_get(info)) != NULL)
 529		list_add_tail(&msg->m_list, &tmp_msg);
 530	kfree(info->node_cache);
 531	spin_unlock(&info->lock);
 532
 533	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
 534		list_del(&msg->m_list);
 535		free_msg(msg);
 536	}
 537
 538	if (info->ucounts) {
 539		unsigned long mq_bytes, mq_treesize;
 540
 541		/* Total amount of bytes accounted for the mqueue */
 542		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 543			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 544			sizeof(struct posix_msg_tree_node);
 545
 546		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 547					  info->attr.mq_msgsize);
 548
 549		spin_lock(&mq_lock);
 550		dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
 551		/*
 552		 * get_ns_from_inode() ensures that the
 553		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 554		 * to which we now hold a reference, or it is NULL.
 555		 * We can't put it here under mq_lock, though.
 556		 */
 557		if (ipc_ns)
 558			ipc_ns->mq_queues_count--;
 559		spin_unlock(&mq_lock);
 560		put_ucounts(info->ucounts);
 561		info->ucounts = NULL;
 562	}
 563	if (ipc_ns)
 564		put_ipc_ns(ipc_ns);
 565}
 566
 567static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
 568{
 569	struct inode *dir = dentry->d_parent->d_inode;
 570	struct inode *inode;
 571	struct mq_attr *attr = arg;
 572	int error;
 573	struct ipc_namespace *ipc_ns;
 574
 575	spin_lock(&mq_lock);
 576	ipc_ns = __get_ns_from_inode(dir);
 577	if (!ipc_ns) {
 578		error = -EACCES;
 579		goto out_unlock;
 580	}
 581
 582	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 583	    !capable(CAP_SYS_RESOURCE)) {
 584		error = -ENOSPC;
 585		goto out_unlock;
 586	}
 587	ipc_ns->mq_queues_count++;
 588	spin_unlock(&mq_lock);
 589
 590	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 591	if (IS_ERR(inode)) {
 592		error = PTR_ERR(inode);
 593		spin_lock(&mq_lock);
 594		ipc_ns->mq_queues_count--;
 595		goto out_unlock;
 596	}
 597
 598	put_ipc_ns(ipc_ns);
 599	dir->i_size += DIRENT_SIZE;
 600	simple_inode_init_ts(dir);
 601
 602	d_make_persistent(dentry, inode);
 603	return 0;
 604out_unlock:
 605	spin_unlock(&mq_lock);
 606	if (ipc_ns)
 607		put_ipc_ns(ipc_ns);
 608	return error;
 609}
 610
 611static int mqueue_create(struct mnt_idmap *idmap, struct inode *dir,
 612			 struct dentry *dentry, umode_t mode, bool excl)
 613{
 614	return mqueue_create_attr(dentry, mode, NULL);
 615}
 616
 617static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 618{
 619	dir->i_size -= DIRENT_SIZE;
 620	return simple_unlink(dir, dentry);
 621}
 622
 623/*
 624*	This is routine for system read from queue file.
 625*	To avoid mess with doing here some sort of mq_receive we allow
 626*	to read only queue size & notification info (the only values
 627*	that are interesting from user point of view and aren't accessible
 628*	through std routines)
 629*/
 630static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 631				size_t count, loff_t *off)
 632{
 633	struct inode *inode = file_inode(filp);
 634	struct mqueue_inode_info *info = MQUEUE_I(inode);
 635	char buffer[FILENT_SIZE];
 636	ssize_t ret;
 637
 638	spin_lock(&info->lock);
 639	snprintf(buffer, sizeof(buffer),
 640			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 641			info->qsize,
 642			info->notify_owner ? info->notify.sigev_notify : 0,
 643			(info->notify_owner &&
 644			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
 645				info->notify.sigev_signo : 0,
 646			pid_vnr(info->notify_owner));
 647	spin_unlock(&info->lock);
 648	buffer[sizeof(buffer)-1] = '\0';
 649
 650	ret = simple_read_from_buffer(u_data, count, off, buffer,
 651				strlen(buffer));
 652	if (ret <= 0)
 653		return ret;
 654
 655	inode_set_atime_to_ts(inode, inode_set_ctime_current(inode));
 656	return ret;
 657}
 658
 659static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 660{
 661	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 662
 663	spin_lock(&info->lock);
 664	if (task_tgid(current) == info->notify_owner)
 665		remove_notification(info);
 666
 667	spin_unlock(&info->lock);
 668	return 0;
 669}
 670
 671static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 672{
 673	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 674	__poll_t retval = 0;
 675
 676	poll_wait(filp, &info->wait_q, poll_tab);
 677
 678	spin_lock(&info->lock);
 679	if (info->attr.mq_curmsgs)
 680		retval = EPOLLIN | EPOLLRDNORM;
 681
 682	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 683		retval |= EPOLLOUT | EPOLLWRNORM;
 684	spin_unlock(&info->lock);
 685
 686	return retval;
 687}
 688
 689/* Adds current to info->e_wait_q[sr] before element with smaller prio */
 690static void wq_add(struct mqueue_inode_info *info, int sr,
 691			struct ext_wait_queue *ewp)
 692{
 693	struct ext_wait_queue *walk;
 694
 695	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 696		if (walk->task->prio <= current->prio) {
 697			list_add_tail(&ewp->list, &walk->list);
 698			return;
 699		}
 700	}
 701	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 702}
 703
 704/*
 705 * Puts current task to sleep. Caller must hold queue lock. After return
 706 * lock isn't held.
 707 * sr: SEND or RECV
 708 */
 709static int wq_sleep(struct mqueue_inode_info *info, int sr,
 710		    ktime_t *timeout, struct ext_wait_queue *ewp)
 711	__releases(&info->lock)
 712{
 713	int retval;
 714	signed long time;
 715
 716	wq_add(info, sr, ewp);
 717
 718	for (;;) {
 719		/* memory barrier not required, we hold info->lock */
 720		__set_current_state(TASK_INTERRUPTIBLE);
 721
 722		spin_unlock(&info->lock);
 723		time = schedule_hrtimeout_range_clock(timeout, 0,
 724			HRTIMER_MODE_ABS, CLOCK_REALTIME);
 725
 726		if (READ_ONCE(ewp->state) == STATE_READY) {
 727			/* see MQ_BARRIER for purpose/pairing */
 728			smp_acquire__after_ctrl_dep();
 729			retval = 0;
 730			goto out;
 731		}
 732		spin_lock(&info->lock);
 733
 734		/* we hold info->lock, so no memory barrier required */
 735		if (READ_ONCE(ewp->state) == STATE_READY) {
 736			retval = 0;
 737			goto out_unlock;
 738		}
 739		if (signal_pending(current)) {
 740			retval = -ERESTARTSYS;
 741			break;
 742		}
 743		if (time == 0) {
 744			retval = -ETIMEDOUT;
 745			break;
 746		}
 747	}
 748	list_del(&ewp->list);
 749out_unlock:
 750	spin_unlock(&info->lock);
 751out:
 752	return retval;
 753}
 754
 755/*
 756 * Returns waiting task that should be serviced first or NULL if none exists
 757 */
 758static struct ext_wait_queue *wq_get_first_waiter(
 759		struct mqueue_inode_info *info, int sr)
 760{
 761	struct list_head *ptr;
 762
 763	ptr = info->e_wait_q[sr].list.prev;
 764	if (ptr == &info->e_wait_q[sr].list)
 765		return NULL;
 766	return list_entry(ptr, struct ext_wait_queue, list);
 767}
 768
 769
 770static inline void set_cookie(struct sk_buff *skb, char code)
 771{
 772	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 773}
 774
 775/*
 776 * The next function is only to split too long sys_mq_timedsend
 777 */
 778static void __do_notify(struct mqueue_inode_info *info)
 779{
 780	/* notification
 781	 * invoked when there is registered process and there isn't process
 782	 * waiting synchronously for message AND state of queue changed from
 783	 * empty to not empty. Here we are sure that no one is waiting
 784	 * synchronously. */
 785	if (info->notify_owner &&
 786	    info->attr.mq_curmsgs == 1) {
 787		switch (info->notify.sigev_notify) {
 788		case SIGEV_NONE:
 789			break;
 790		case SIGEV_SIGNAL: {
 791			struct kernel_siginfo sig_i;
 792			struct task_struct *task;
 793
 794			/* do_mq_notify() accepts sigev_signo == 0, why?? */
 795			if (!info->notify.sigev_signo)
 796				break;
 797
 798			clear_siginfo(&sig_i);
 799			sig_i.si_signo = info->notify.sigev_signo;
 800			sig_i.si_errno = 0;
 801			sig_i.si_code = SI_MESGQ;
 802			sig_i.si_value = info->notify.sigev_value;
 803			rcu_read_lock();
 804			/* map current pid/uid into info->owner's namespaces */
 805			sig_i.si_pid = task_tgid_nr_ns(current,
 806						ns_of_pid(info->notify_owner));
 807			sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
 808						current_uid());
 809			/*
 810			 * We can't use kill_pid_info(), this signal should
 811			 * bypass check_kill_permission(). It is from kernel
 812			 * but si_fromuser() can't know this.
 813			 * We do check the self_exec_id, to avoid sending
 814			 * signals to programs that don't expect them.
 815			 */
 816			task = pid_task(info->notify_owner, PIDTYPE_TGID);
 817			if (task && task->self_exec_id ==
 818						info->notify_self_exec_id) {
 819				do_send_sig_info(info->notify.sigev_signo,
 820						&sig_i, task, PIDTYPE_TGID);
 821			}
 822			rcu_read_unlock();
 823			break;
 824		}
 825		case SIGEV_THREAD:
 826			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 827			netlink_sendskb(info->notify_sock, info->notify_cookie);
 828			break;
 829		}
 830		/* after notification unregisters process */
 831		put_pid(info->notify_owner);
 832		put_user_ns(info->notify_user_ns);
 833		info->notify_owner = NULL;
 834		info->notify_user_ns = NULL;
 835	}
 836	wake_up(&info->wait_q);
 837}
 838
 839static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
 840			   struct timespec64 *ts)
 841{
 842	if (get_timespec64(ts, u_abs_timeout))
 843		return -EFAULT;
 844	if (!timespec64_valid(ts))
 845		return -EINVAL;
 846	return 0;
 847}
 848
 849static void remove_notification(struct mqueue_inode_info *info)
 850{
 851	if (info->notify_owner != NULL &&
 852	    info->notify.sigev_notify == SIGEV_THREAD) {
 853		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 854		netlink_sendskb(info->notify_sock, info->notify_cookie);
 855	}
 856	put_pid(info->notify_owner);
 857	put_user_ns(info->notify_user_ns);
 858	info->notify_owner = NULL;
 859	info->notify_user_ns = NULL;
 860}
 861
 862static int prepare_open(struct dentry *dentry, int oflag, int ro,
 863			umode_t mode, struct filename *name,
 864			struct mq_attr *attr)
 865{
 866	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 867						  MAY_READ | MAY_WRITE };
 868	int acc;
 869
 870	if (d_really_is_negative(dentry)) {
 871		if (!(oflag & O_CREAT))
 872			return -ENOENT;
 873		if (ro)
 874			return ro;
 875		audit_inode_parent_hidden(name, dentry->d_parent);
 876		return vfs_mkobj(dentry, mode & ~current_umask(),
 877				  mqueue_create_attr, attr);
 878	}
 879	/* it already existed */
 880	audit_inode(name, dentry, 0);
 881	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
 882		return -EEXIST;
 883	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 884		return -EINVAL;
 885	acc = oflag2acc[oflag & O_ACCMODE];
 886	return inode_permission(&nop_mnt_idmap, d_inode(dentry), acc);
 887}
 888
 889static struct file *mqueue_file_open(struct filename *name,
 890				     struct vfsmount *mnt, int oflag, int ro,
 891				     umode_t mode, struct mq_attr *attr)
 892{
 893	struct dentry *dentry;
 894	struct file *file;
 895	int ret;
 896
 897	dentry = start_creating_noperm(mnt->mnt_root, &QSTR(name->name));
 898	if (IS_ERR(dentry))
 899		return ERR_CAST(dentry);
 900
 901	ret = prepare_open(dentry, oflag, ro, mode, name, attr);
 902	file = ERR_PTR(ret);
 903	if (!ret) {
 904		const struct path path = { .mnt = mnt, .dentry = dentry };
 905		file = dentry_open(&path, oflag, current_cred());
 906	}
 907
 908	end_creating(dentry);
 909	return file;
 910}
 911
 912static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
 913		      struct mq_attr *attr)
 914{
 915	struct filename *name __free(putname) = NULL;;
 916	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
 917	int fd, ro;
 918
 919	audit_mq_open(oflag, mode, attr);
 920
 921	name = getname(u_name);
 922	if (IS_ERR(name))
 923		return PTR_ERR(name);
 924
 925	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
 926	fd = FD_ADD(O_CLOEXEC, mqueue_file_open(name, mnt, oflag, ro, mode, attr));
 927	if (!ro)
 928		mnt_drop_write(mnt);
 929	return fd;
 930}
 931
 932SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 933		struct mq_attr __user *, u_attr)
 934{
 935	struct mq_attr attr;
 936	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 937		return -EFAULT;
 938
 939	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
 940}
 941
 942SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 943{
 944	int err;
 945	struct filename *name;
 946	struct dentry *dentry;
 947	struct inode *inode;
 948	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 949	struct vfsmount *mnt = ipc_ns->mq_mnt;
 950
 951	name = getname(u_name);
 952	if (IS_ERR(name))
 953		return PTR_ERR(name);
 954
 955	audit_inode_parent_hidden(name, mnt->mnt_root);
 956	err = mnt_want_write(mnt);
 957	if (err)
 958		goto out_name;
 959	dentry = start_removing_noperm(mnt->mnt_root, &QSTR(name->name));
 960	if (IS_ERR(dentry)) {
 961		err = PTR_ERR(dentry);
 962		goto out_drop_write;
 963	}
 964
 965	inode = d_inode(dentry);
 966	ihold(inode);
 967	err = vfs_unlink(&nop_mnt_idmap, d_inode(mnt->mnt_root),
 968			 dentry, NULL);
 969	end_removing(dentry);
 970	iput(inode);
 971
 972out_drop_write:
 973	mnt_drop_write(mnt);
 974out_name:
 975	putname(name);
 976
 977	return err;
 978}
 979
 980/* Pipelined send and receive functions.
 981 *
 982 * If a receiver finds no waiting message, then it registers itself in the
 983 * list of waiting receivers. A sender checks that list before adding the new
 984 * message into the message array. If there is a waiting receiver, then it
 985 * bypasses the message array and directly hands the message over to the
 986 * receiver. The receiver accepts the message and returns without grabbing the
 987 * queue spinlock:
 988 *
 989 * - Set pointer to message.
 990 * - Queue the receiver task for later wakeup (without the info->lock).
 991 * - Update its state to STATE_READY. Now the receiver can continue.
 992 * - Wake up the process after the lock is dropped. Should the process wake up
 993 *   before this wakeup (due to a timeout or a signal) it will either see
 994 *   STATE_READY and continue or acquire the lock to check the state again.
 995 *
 996 * The same algorithm is used for senders.
 997 */
 998
 999static inline void __pipelined_op(struct wake_q_head *wake_q,
1000				  struct mqueue_inode_info *info,
1001				  struct ext_wait_queue *this)
1002{
1003	struct task_struct *task;
1004
1005	list_del(&this->list);
1006	task = get_task_struct(this->task);
1007
1008	/* see MQ_BARRIER for purpose/pairing */
1009	smp_store_release(&this->state, STATE_READY);
1010	wake_q_add_safe(wake_q, task);
1011}
1012
1013/* pipelined_send() - send a message directly to the task waiting in
1014 * sys_mq_timedreceive() (without inserting message into a queue).
1015 */
1016static inline void pipelined_send(struct wake_q_head *wake_q,
1017				  struct mqueue_inode_info *info,
1018				  struct msg_msg *message,
1019				  struct ext_wait_queue *receiver)
1020{
1021	receiver->msg = message;
1022	__pipelined_op(wake_q, info, receiver);
1023}
1024
1025/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1026 * gets its message and put to the queue (we have one free place for sure). */
1027static inline void pipelined_receive(struct wake_q_head *wake_q,
1028				     struct mqueue_inode_info *info)
1029{
1030	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1031
1032	if (!sender) {
1033		/* for poll */
1034		wake_up_interruptible(&info->wait_q);
1035		return;
1036	}
1037	if (msg_insert(sender->msg, info))
1038		return;
1039
1040	__pipelined_op(wake_q, info, sender);
1041}
1042
1043static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1044		size_t msg_len, unsigned int msg_prio,
1045		struct timespec64 *ts)
1046{
1047	struct inode *inode;
1048	struct ext_wait_queue wait;
1049	struct ext_wait_queue *receiver;
1050	struct msg_msg *msg_ptr;
1051	struct mqueue_inode_info *info;
1052	ktime_t expires, *timeout = NULL;
1053	struct posix_msg_tree_node *new_leaf = NULL;
1054	int ret = 0;
1055	DEFINE_WAKE_Q(wake_q);
1056
1057	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1058		return -EINVAL;
1059
1060	if (ts) {
1061		expires = timespec64_to_ktime(*ts);
1062		timeout = &expires;
1063	}
1064
1065	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1066
1067	CLASS(fd, f)(mqdes);
1068	if (fd_empty(f))
1069		return -EBADF;
1070
1071	inode = file_inode(fd_file(f));
1072	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1073		return -EBADF;
1074	info = MQUEUE_I(inode);
1075	audit_file(fd_file(f));
1076
1077	if (unlikely(!(fd_file(f)->f_mode & FMODE_WRITE)))
1078		return -EBADF;
1079
1080	if (unlikely(msg_len > info->attr.mq_msgsize))
1081		return -EMSGSIZE;
1082
1083	/* First try to allocate memory, before doing anything with
1084	 * existing queues. */
1085	msg_ptr = load_msg(u_msg_ptr, msg_len);
1086	if (IS_ERR(msg_ptr))
1087		return PTR_ERR(msg_ptr);
1088	msg_ptr->m_ts = msg_len;
1089	msg_ptr->m_type = msg_prio;
1090
1091	/*
1092	 * msg_insert really wants us to have a valid, spare node struct so
1093	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1094	 * fall back to that if necessary.
1095	 */
1096	if (!info->node_cache)
1097		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1098
1099	spin_lock(&info->lock);
1100
1101	if (!info->node_cache && new_leaf) {
1102		/* Save our speculative allocation into the cache */
1103		INIT_LIST_HEAD(&new_leaf->msg_list);
1104		info->node_cache = new_leaf;
1105		new_leaf = NULL;
1106	} else {
1107		kfree(new_leaf);
1108	}
1109
1110	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1111		if (fd_file(f)->f_flags & O_NONBLOCK) {
1112			ret = -EAGAIN;
1113		} else {
1114			wait.task = current;
1115			wait.msg = (void *) msg_ptr;
1116
1117			/* memory barrier not required, we hold info->lock */
1118			WRITE_ONCE(wait.state, STATE_NONE);
1119			ret = wq_sleep(info, SEND, timeout, &wait);
1120			/*
1121			 * wq_sleep must be called with info->lock held, and
1122			 * returns with the lock released
1123			 */
1124			goto out_free;
1125		}
1126	} else {
1127		receiver = wq_get_first_waiter(info, RECV);
1128		if (receiver) {
1129			pipelined_send(&wake_q, info, msg_ptr, receiver);
1130		} else {
1131			/* adds message to the queue */
1132			ret = msg_insert(msg_ptr, info);
1133			if (ret)
1134				goto out_unlock;
1135			__do_notify(info);
1136		}
1137		simple_inode_init_ts(inode);
1138	}
1139out_unlock:
1140	spin_unlock(&info->lock);
1141	wake_up_q(&wake_q);
1142out_free:
1143	if (ret)
1144		free_msg(msg_ptr);
1145	return ret;
1146}
1147
1148static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1149		size_t msg_len, unsigned int __user *u_msg_prio,
1150		struct timespec64 *ts)
1151{
1152	ssize_t ret;
1153	struct msg_msg *msg_ptr;
1154	struct inode *inode;
1155	struct mqueue_inode_info *info;
1156	struct ext_wait_queue wait;
1157	ktime_t expires, *timeout = NULL;
1158	struct posix_msg_tree_node *new_leaf = NULL;
1159
1160	if (ts) {
1161		expires = timespec64_to_ktime(*ts);
1162		timeout = &expires;
1163	}
1164
1165	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1166
1167	CLASS(fd, f)(mqdes);
1168	if (fd_empty(f))
1169		return -EBADF;
1170
1171	inode = file_inode(fd_file(f));
1172	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1173		return -EBADF;
1174	info = MQUEUE_I(inode);
1175	audit_file(fd_file(f));
1176
1177	if (unlikely(!(fd_file(f)->f_mode & FMODE_READ)))
1178		return -EBADF;
1179
1180	/* checks if buffer is big enough */
1181	if (unlikely(msg_len < info->attr.mq_msgsize))
1182		return -EMSGSIZE;
1183
1184	/*
1185	 * msg_insert really wants us to have a valid, spare node struct so
1186	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1187	 * fall back to that if necessary.
1188	 */
1189	if (!info->node_cache)
1190		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1191
1192	spin_lock(&info->lock);
1193
1194	if (!info->node_cache && new_leaf) {
1195		/* Save our speculative allocation into the cache */
1196		INIT_LIST_HEAD(&new_leaf->msg_list);
1197		info->node_cache = new_leaf;
1198	} else {
1199		kfree(new_leaf);
1200	}
1201
1202	if (info->attr.mq_curmsgs == 0) {
1203		if (fd_file(f)->f_flags & O_NONBLOCK) {
1204			spin_unlock(&info->lock);
1205			ret = -EAGAIN;
1206		} else {
1207			wait.task = current;
1208
1209			/* memory barrier not required, we hold info->lock */
1210			WRITE_ONCE(wait.state, STATE_NONE);
1211			ret = wq_sleep(info, RECV, timeout, &wait);
1212			msg_ptr = wait.msg;
1213		}
1214	} else {
1215		DEFINE_WAKE_Q(wake_q);
1216
1217		msg_ptr = msg_get(info);
1218
1219		simple_inode_init_ts(inode);
1220
1221		/* There is now free space in queue. */
1222		pipelined_receive(&wake_q, info);
1223		spin_unlock(&info->lock);
1224		wake_up_q(&wake_q);
1225		ret = 0;
1226	}
1227	if (ret == 0) {
1228		ret = msg_ptr->m_ts;
1229
1230		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1231			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1232			ret = -EFAULT;
1233		}
1234		free_msg(msg_ptr);
1235	}
1236	return ret;
1237}
1238
1239SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1240		size_t, msg_len, unsigned int, msg_prio,
1241		const struct __kernel_timespec __user *, u_abs_timeout)
1242{
1243	struct timespec64 ts, *p = NULL;
1244	if (u_abs_timeout) {
1245		int res = prepare_timeout(u_abs_timeout, &ts);
1246		if (res)
1247			return res;
1248		p = &ts;
1249	}
1250	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1251}
1252
1253SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1254		size_t, msg_len, unsigned int __user *, u_msg_prio,
1255		const struct __kernel_timespec __user *, u_abs_timeout)
1256{
1257	struct timespec64 ts, *p = NULL;
1258	if (u_abs_timeout) {
1259		int res = prepare_timeout(u_abs_timeout, &ts);
1260		if (res)
1261			return res;
1262		p = &ts;
1263	}
1264	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1265}
1266
1267/*
1268 * Notes: the case when user wants us to deregister (with NULL as pointer)
1269 * and he isn't currently owner of notification, will be silently discarded.
1270 * It isn't explicitly defined in the POSIX.
1271 */
1272static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1273{
1274	int ret;
1275	struct sock *sock;
1276	struct inode *inode;
1277	struct mqueue_inode_info *info;
1278	struct sk_buff *nc;
1279
1280	audit_mq_notify(mqdes, notification);
1281
1282	nc = NULL;
1283	sock = NULL;
1284	if (notification != NULL) {
1285		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1286			     notification->sigev_notify != SIGEV_SIGNAL &&
1287			     notification->sigev_notify != SIGEV_THREAD))
1288			return -EINVAL;
1289		if (notification->sigev_notify == SIGEV_SIGNAL &&
1290			!valid_signal(notification->sigev_signo)) {
1291			return -EINVAL;
1292		}
1293		if (notification->sigev_notify == SIGEV_THREAD) {
1294			long timeo;
1295
1296			/* create the notify skb */
1297			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1298			if (!nc)
1299				return -ENOMEM;
1300
1301			if (copy_from_user(nc->data,
1302					notification->sigev_value.sival_ptr,
1303					NOTIFY_COOKIE_LEN)) {
1304				kfree_skb(nc);
1305				return -EFAULT;
1306			}
1307
1308			/* TODO: add a header? */
1309			skb_put(nc, NOTIFY_COOKIE_LEN);
1310			/* and attach it to the socket */
1311retry:
1312			sock = netlink_getsockbyfd(notification->sigev_signo);
1313			if (IS_ERR(sock)) {
1314				kfree_skb(nc);
1315				return PTR_ERR(sock);
1316			}
1317
1318			timeo = MAX_SCHEDULE_TIMEOUT;
1319			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1320			if (ret == 1)
1321				goto retry;
1322			if (ret)
1323				return ret;
1324		}
1325	}
1326
1327	CLASS(fd, f)(mqdes);
1328	if (fd_empty(f)) {
1329		ret = -EBADF;
1330		goto out;
1331	}
1332
1333	inode = file_inode(fd_file(f));
1334	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations)) {
1335		ret = -EBADF;
1336		goto out;
1337	}
1338	info = MQUEUE_I(inode);
1339
1340	ret = 0;
1341	spin_lock(&info->lock);
1342	if (notification == NULL) {
1343		if (info->notify_owner == task_tgid(current)) {
1344			remove_notification(info);
1345			inode_set_atime_to_ts(inode,
1346					      inode_set_ctime_current(inode));
1347		}
1348	} else if (info->notify_owner != NULL) {
1349		ret = -EBUSY;
1350	} else {
1351		switch (notification->sigev_notify) {
1352		case SIGEV_NONE:
1353			info->notify.sigev_notify = SIGEV_NONE;
1354			break;
1355		case SIGEV_THREAD:
1356			info->notify_sock = sock;
1357			info->notify_cookie = nc;
1358			sock = NULL;
1359			nc = NULL;
1360			info->notify.sigev_notify = SIGEV_THREAD;
1361			break;
1362		case SIGEV_SIGNAL:
1363			info->notify.sigev_signo = notification->sigev_signo;
1364			info->notify.sigev_value = notification->sigev_value;
1365			info->notify.sigev_notify = SIGEV_SIGNAL;
1366			info->notify_self_exec_id = current->self_exec_id;
1367			break;
1368		}
1369
1370		info->notify_owner = get_pid(task_tgid(current));
1371		info->notify_user_ns = get_user_ns(current_user_ns());
1372		inode_set_atime_to_ts(inode, inode_set_ctime_current(inode));
1373	}
1374	spin_unlock(&info->lock);
1375out:
1376	if (sock)
1377		netlink_detachskb(sock, nc);
1378	return ret;
1379}
1380
1381SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1382		const struct sigevent __user *, u_notification)
1383{
1384	struct sigevent n, *p = NULL;
1385	if (u_notification) {
1386		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1387			return -EFAULT;
1388		p = &n;
1389	}
1390	return do_mq_notify(mqdes, p);
1391}
1392
1393static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1394{
1395	struct inode *inode;
1396	struct mqueue_inode_info *info;
1397
1398	if (new && (new->mq_flags & (~O_NONBLOCK)))
1399		return -EINVAL;
1400
1401	CLASS(fd, f)(mqdes);
1402	if (fd_empty(f))
1403		return -EBADF;
1404
1405	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1406		return -EBADF;
1407
1408	inode = file_inode(fd_file(f));
1409	info = MQUEUE_I(inode);
1410
1411	spin_lock(&info->lock);
1412
1413	if (old) {
1414		*old = info->attr;
1415		old->mq_flags = fd_file(f)->f_flags & O_NONBLOCK;
1416	}
1417	if (new) {
1418		audit_mq_getsetattr(mqdes, new);
1419		spin_lock(&fd_file(f)->f_lock);
1420		if (new->mq_flags & O_NONBLOCK)
1421			fd_file(f)->f_flags |= O_NONBLOCK;
1422		else
1423			fd_file(f)->f_flags &= ~O_NONBLOCK;
1424		spin_unlock(&fd_file(f)->f_lock);
1425
1426		inode_set_atime_to_ts(inode, inode_set_ctime_current(inode));
1427	}
1428
1429	spin_unlock(&info->lock);
1430	return 0;
1431}
1432
1433SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1434		const struct mq_attr __user *, u_mqstat,
1435		struct mq_attr __user *, u_omqstat)
1436{
1437	int ret;
1438	struct mq_attr mqstat, omqstat;
1439	struct mq_attr *new = NULL, *old = NULL;
1440
1441	if (u_mqstat) {
1442		new = &mqstat;
1443		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1444			return -EFAULT;
1445	}
1446	if (u_omqstat)
1447		old = &omqstat;
1448
1449	ret = do_mq_getsetattr(mqdes, new, old);
1450	if (ret || !old)
1451		return ret;
1452
1453	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1454		return -EFAULT;
1455	return 0;
1456}
1457
1458#ifdef CONFIG_COMPAT
1459
1460struct compat_mq_attr {
1461	compat_long_t mq_flags;      /* message queue flags		     */
1462	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1463	compat_long_t mq_msgsize;    /* maximum message size		     */
1464	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1465	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1466};
1467
1468static inline int get_compat_mq_attr(struct mq_attr *attr,
1469			const struct compat_mq_attr __user *uattr)
1470{
1471	struct compat_mq_attr v;
1472
1473	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1474		return -EFAULT;
1475
1476	memset(attr, 0, sizeof(*attr));
1477	attr->mq_flags = v.mq_flags;
1478	attr->mq_maxmsg = v.mq_maxmsg;
1479	attr->mq_msgsize = v.mq_msgsize;
1480	attr->mq_curmsgs = v.mq_curmsgs;
1481	return 0;
1482}
1483
1484static inline int put_compat_mq_attr(const struct mq_attr *attr,
1485			struct compat_mq_attr __user *uattr)
1486{
1487	struct compat_mq_attr v;
1488
1489	memset(&v, 0, sizeof(v));
1490	v.mq_flags = attr->mq_flags;
1491	v.mq_maxmsg = attr->mq_maxmsg;
1492	v.mq_msgsize = attr->mq_msgsize;
1493	v.mq_curmsgs = attr->mq_curmsgs;
1494	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1495		return -EFAULT;
1496	return 0;
1497}
1498
1499COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1500		       int, oflag, compat_mode_t, mode,
1501		       struct compat_mq_attr __user *, u_attr)
1502{
1503	struct mq_attr attr, *p = NULL;
1504	if (u_attr && oflag & O_CREAT) {
1505		p = &attr;
1506		if (get_compat_mq_attr(&attr, u_attr))
1507			return -EFAULT;
1508	}
1509	return do_mq_open(u_name, oflag, mode, p);
1510}
1511
1512COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1513		       const struct compat_sigevent __user *, u_notification)
1514{
1515	struct sigevent n, *p = NULL;
1516	if (u_notification) {
1517		if (get_compat_sigevent(&n, u_notification))
1518			return -EFAULT;
1519		if (n.sigev_notify == SIGEV_THREAD)
1520			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1521		p = &n;
1522	}
1523	return do_mq_notify(mqdes, p);
1524}
1525
1526COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1527		       const struct compat_mq_attr __user *, u_mqstat,
1528		       struct compat_mq_attr __user *, u_omqstat)
1529{
1530	int ret;
1531	struct mq_attr mqstat, omqstat;
1532	struct mq_attr *new = NULL, *old = NULL;
1533
1534	if (u_mqstat) {
1535		new = &mqstat;
1536		if (get_compat_mq_attr(new, u_mqstat))
1537			return -EFAULT;
1538	}
1539	if (u_omqstat)
1540		old = &omqstat;
1541
1542	ret = do_mq_getsetattr(mqdes, new, old);
1543	if (ret || !old)
1544		return ret;
1545
1546	if (put_compat_mq_attr(old, u_omqstat))
1547		return -EFAULT;
1548	return 0;
1549}
1550#endif
1551
1552#ifdef CONFIG_COMPAT_32BIT_TIME
1553static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1554				   struct timespec64 *ts)
1555{
1556	if (get_old_timespec32(ts, p))
1557		return -EFAULT;
1558	if (!timespec64_valid(ts))
1559		return -EINVAL;
1560	return 0;
1561}
1562
1563SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1564		const char __user *, u_msg_ptr,
1565		unsigned int, msg_len, unsigned int, msg_prio,
1566		const struct old_timespec32 __user *, u_abs_timeout)
1567{
1568	struct timespec64 ts, *p = NULL;
1569	if (u_abs_timeout) {
1570		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1571		if (res)
1572			return res;
1573		p = &ts;
1574	}
1575	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1576}
1577
1578SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1579		char __user *, u_msg_ptr,
1580		unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1581		const struct old_timespec32 __user *, u_abs_timeout)
1582{
1583	struct timespec64 ts, *p = NULL;
1584	if (u_abs_timeout) {
1585		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1586		if (res)
1587			return res;
1588		p = &ts;
1589	}
1590	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1591}
1592#endif
1593
1594static const struct inode_operations mqueue_dir_inode_operations = {
1595	.lookup = simple_lookup,
1596	.create = mqueue_create,
1597	.unlink = mqueue_unlink,
1598};
1599
1600static const struct file_operations mqueue_file_operations = {
1601	.flush = mqueue_flush_file,
1602	.poll = mqueue_poll_file,
1603	.read = mqueue_read_file,
1604	.llseek = default_llseek,
1605};
1606
1607static const struct super_operations mqueue_super_ops = {
1608	.alloc_inode = mqueue_alloc_inode,
1609	.free_inode = mqueue_free_inode,
1610	.evict_inode = mqueue_evict_inode,
1611	.statfs = simple_statfs,
1612};
1613
1614static const struct fs_context_operations mqueue_fs_context_ops = {
1615	.free		= mqueue_fs_context_free,
1616	.get_tree	= mqueue_get_tree,
1617};
1618
1619static struct file_system_type mqueue_fs_type = {
1620	.name			= "mqueue",
1621	.init_fs_context	= mqueue_init_fs_context,
1622	.kill_sb		= kill_anon_super,
1623	.fs_flags		= FS_USERNS_MOUNT,
1624};
1625
1626int mq_init_ns(struct ipc_namespace *ns)
1627{
1628	struct vfsmount *m;
1629
1630	ns->mq_queues_count  = 0;
1631	ns->mq_queues_max    = DFLT_QUEUESMAX;
1632	ns->mq_msg_max       = DFLT_MSGMAX;
1633	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1634	ns->mq_msg_default   = DFLT_MSG;
1635	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1636
1637	m = mq_create_mount(ns);
1638	if (IS_ERR(m))
1639		return PTR_ERR(m);
1640	ns->mq_mnt = m;
1641	return 0;
1642}
1643
1644void mq_clear_sbinfo(struct ipc_namespace *ns)
1645{
1646	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1647}
1648
1649static int __init init_mqueue_fs(void)
1650{
1651	int error;
1652
1653	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1654				sizeof(struct mqueue_inode_info), 0,
1655				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1656	if (mqueue_inode_cachep == NULL)
1657		return -ENOMEM;
1658
1659	if (!setup_mq_sysctls(&init_ipc_ns)) {
1660		pr_warn("sysctl registration failed\n");
1661		error = -ENOMEM;
1662		goto out_kmem;
1663	}
1664
1665	error = register_filesystem(&mqueue_fs_type);
1666	if (error)
1667		goto out_sysctl;
1668
1669	spin_lock_init(&mq_lock);
1670
1671	error = mq_init_ns(&init_ipc_ns);
1672	if (error)
1673		goto out_filesystem;
1674
1675	return 0;
1676
1677out_filesystem:
1678	unregister_filesystem(&mqueue_fs_type);
1679out_sysctl:
1680	retire_mq_sysctls(&init_ipc_ns);
1681out_kmem:
1682	kmem_cache_destroy(mqueue_inode_cachep);
1683	return error;
1684}
1685
1686device_initcall(init_mqueue_fs);